Spatiotemporal patterns of methane fluxes across alpine permafrost region on the Tibetan Plateau

Abstract

Methane (CH₄) emissions from thawing permafrost could amplify climate warming. However, long-term trajectory of net CH₄ balance in permafrost regions, particularly high-altitude permafrost regions, remains unknown. Based on literature synthesis and CLM5.0 model, we evaluate the contemporary and future CH₄ fluxes across the Tibetan alpine permafrost region from 1989−2100. Here, we find that this permafrost region functions as a marginal CH₄ sink during 1989-2018 (−0.01 ± 0.01 Tg CH₄ yr⁻¹), and future trajectories diverge, with warming and wetting under low- and medium-emission scenarios (SSP1-2.6/SSP2-4.5) driving persistent CH₄ emissions (0.07 Tg CH₄ yr⁻¹). By contrast, under higher emission scenarios (SSP3-7.0/SSP5-8.5), the region shifts to net emissions by mid-century but enhanced atmospheric CH₄ concentrations strengthen sink, returning it to a net sink by century’s end (−0.06 ~ −0.02 Tg CH₄ yr⁻¹). These results demonstrate that climate change and atmospheric CH₄ dynamics jointly mediate the trajectory of alpine permafrost CH₄ balance.

Introduction

Permafrost (i.e., soil or rock that remains below 0 °C for at least two consecutive years) is primarily distributed in the high-latitude and high-altitude regions, covering ~24% of the land surface area in the Northern Hemisphere^1,2. Permafrost regions store large amounts of soil organic carbon (~1014 Pg C in the top three meters, including ~14 Pg C on the Tibetan Plateau^3,4; 1 Pg = 10¹⁵g). Recent warming (~0.6 °C per decade⁵) has accelerated permafrost thaw and microbial decomposition, which could enhance carbon dioxide (CO₂) and methane (CH₄) emissions and intensify global warming⁶. Of them, CH₄ is a potent greenhouse gas with a global warming potential (GWP) >27 times greater than CO₂ over a 100 year period⁷. Due to this point, it is projected to contribute one-third to the permafrost-climate feedback (on a 100 year GWP basis), despite CH₄ emissions accounting for only 2.3% of the total carbon release from permafrost regions in the Northern Hemisphere under the highest warming scenario (RCP8.5)⁸. Quantifying spatial and temporal patterns of CH₄ flux in these vulnerable ecosystems is thus critical for accurately predicting the direction and magnitude of permafrost carbon-climate feedback and informing climate change policy decisions⁹.

Given the critical role of CH₄ flux in mediating permafrost carbon-climate feedback, a number of experimental studies have examined short-term warming impacts (over several years) on CH₄ fluxes across various permafrost regions, including the high-latitude (e.g., Arctic/sub-Arctic)^10,11 and the high-altitude (e.g., the Tibetan Plateau) permafrost regions^12,13. Compared with manipulative experiments, process-based models are valuable tools for predicting long-term CH₄ dynamics under a changing environment^14,15. However, most projections of long-term warming effects (over the next century) have predominantly focused on high-latitude permafrost region^16,17. Until now, research on the long-term response of CH₄ fluxes to climate warming in the high-altitude permafrost region has remained poorly constrained, with only one study projecting a century-scale CH₄ budget under climate warming scenarios¹⁸. This single modeling study suffered from substantial uncertainties in predicting the spatial and temporal patterns of CH₄ fluxes, since model validation was limited to a non-permafrost marshland site¹⁸, leaving the high-altitude permafrost region being underrepresented and hindering efforts to quantify model uncertainty.

As the largest high-altitude permafrost region on Earth, the Tibetan Plateau constitutes 75% of the total alpine permafrost area in the Northern Hemisphere¹⁹. Unlike Arctic permafrost regions, which contain extensive marshlands that function as CH₄ sources¹⁶, the Tibetan alpine permafrost region is primarily located in arid and semi-arid areas²⁰. Only 3.2% of the region consists of marshlands in anaerobic conditions, while over 90% is comprised of arid uplands^20,21. Based on this situation, the Tibetan alpine permafrost region may act as a net CH₄ sink during the contemporary period^22,23. However, future climate warming and changes in rainfall could impact the CH₄-related biogeochemical processes²⁴. Climate warming can enhance CH₄ production and oxidation processes simultaneously^22,24,25, but may be more pronounced in anaerobic marshlands^22,25. Warming-induced permafrost thaw and increased rainfall can also lead to increased soil moisture, further benefiting CH₄ production²⁶. Consequently, the projected warmer and wetter trends on the plateau may alter the balance between CH₄ sinks and sources, favoring CH₄ emissions^27,28. Nevertheless, the spatial and temporal variations of CH₄ fluxes remain unclear across this alpine permafrost region, due to the lack of a comprehensive modeling analysis over the regional scale.

In this work, we investigate the spatiotemporal evolution of the CH₄ budget across the Tibetan alpine permafrost region during the contemporary period, and project the CH₄ cycling under future climate change scenarios, using the Community Land Model version 5.0 (CLM5.0) model^29,30 constrained by observational data. Specifically, we first construct a comprehensive dataset of regional CH₄ fluxes through a literature review to calibrate and validate the CLM5.0 model (Fig. 1; Supplementary Table 1; Supplementary Data 1; See Methods section). Based on the calibrated CLM5.0 model, we examine the spatial patterns, magnitudes, and temporal dynamics of CH₄ fluxes in marshland and non-marshland ecosystems across this alpine permafrost region during the contemporary period and under future climate change scenarios. The model is driven by the China Meteorological Forcing Dataset (CMFD; 0.1°$\times$0.1°) for the contemporary period of 1979−2018³¹, as well as the bias-corrected and downscaled climate forcings (0.25° × 0.25°)³² from four General Circulation Models (GCMs) in the Coupled Model Intercomparison Project Phase 6 (CMIP6) for both the contemporary period (1950-2014) and under four future Shared Socioeconomic Pathway (SSP) scenarios (i.e., SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5 for 2015-2100). Our results demonstrate that the Tibetan alpine permafrost region, which functions as a weak net CH₄ sink during the contemporary period, is projected to shift to a net source by around the mid-21st century under all four future climate scenarios. However, under both the intermediate-high (SSP3-7.0) and high (SSP5-8.5) emission scenarios, the region is projected to revert to a net CH₄ sink by the end of the century. Overall, this study aims to enhance our understanding of the spatiotemporal dynamics of soil CH₄ fluxes in high-altitude permafrost region.

Fig. 1: Spatial distribution of alpine permafrost on the Tibetan Plateau and the locations of the methane (CH₄) observation sites across the region.

a The red line outlines the Tibetan Plateau within the global permafrost distribution. b Yellow circles indicate the locations of CH₄ flux observation sites based on literature synthesis. The global permafrost distribution map is adapted from Obu et al.⁸⁹ and re-used with permission from Elsevier (License Number 6074130255122). The map of the Tibetan Plateau is adapted from Zou et al.³³, © Author(s) 2017, and is distributed under the Creative Commons Attribution 3.0 License (https://creativecommons.org/licenses/by/3.0/).

Results

Model performance over the Tibetan alpine permafrost region

We first compared the permafrost extent simulated by the CLM5.0 with the existing permafrost distribution map of the Tibetan Plateau from Zou et al.³³, and found a spatial agreement of ~60% within the top 3 meters (Fig. 1; Supplementary Fig. 1c). We then validated soil temperature, soil moisture and soil organic carbon density (carbon amount per area). The results revealed that the model reasonably represented soil temperature and moisture in this region (Supplementary Fig. 2a–f). The spatial pattern of soil organic carbon density matched the estimates by Ding et al.³⁴ (Supplementary Fig. 2i, j). We further validated the CLM5.0 model using monthly observations of CH₄ fluxes for 16 sites (7 marshlands and 9 non-marshlands) across the Tibetan alpine permafrost region. The model effectively captured the observed magnitudes of monthly CH₄ fluxes and their seasonal variations across different sites (Fig. 2a, b and Supplementary Figs. 3–5). Specifically, the average values and distributions of simulated monthly CH₄ fluxes aligned closely with observations at most sites for both marshlands and non-marshlands (Fig. 2a, b). Moreover, the seasonal variations of the simulated CH₄ fluxes matched the observed trends well (Supplementary Figs. 3, 5). In addition, the simulated and observed monthly mean CH₄ fluxes for marshlands were close to the 1:1 line (R² = 0.81, RMSE = 630 μg CH₄ m⁻² h⁻¹; Supplementary Fig. 4a). Similarly, most points were close to the 1:1 line for non-marshland sites, although some deviations were observed (R² = 0.45, RMSE = 10 μg CH₄ m⁻² h⁻¹; Supplementary Fig. 4b). These discrepancies between simulations and observations might be attributed to the relatively coarse spatial resolution of the gridded climate forcing (0.1° × 0.1°) and land surface data (e.g., soil texture and soil organic matter) used in the model. Overall, these evaluations indicated that the CLM5.0 model could reasonably capture CH₄ fluxes across the Tibetan alpine permafrost region.

Fig. 2: Comparison of simulated and observed CH₄ fluxes and spatial patterns of net methane (CH₄) fluxes in the Tibetan alpine permafrost region.

a,b Comparison of simulated and observed CH₄ fluxes at marshland and non-marshland sites. Green boxplots and scatter plots show monthly observed data (1-3 measurements/month), while yellow ones show daily CLM5.0 simulations for the same months. c Annual mean net CH₄ flux (positive for emission, negative for sink) in the Tibetan alpine permafrost region during 1989−2018. d−g Annual mean net CH₄ fluxes for 2071−2100 under four SSP scenarios: (d) SSP1-2.6, (e) SSP2-4.5, (f) SSP3-7.0, (g) SSP5-8.5. h Comparison of annual mean CH₄ emissions and sinks across all grid cells in marshland (left) and non-marshland (right) areas over different periods. In the boxplots, the solid line indicates the median; box edges represent the 25th (Q1) and 75th (Q3) percentiles; whiskers extend to the furthest data points within 1.5 × IQR (Q3–Q1).

Spatiotemporal patterns of the CH₄ fluxes during contemporary period

The results from the model, driven by the China Meteorological Forcing Dataset (CMFD)³¹, demonstrated considerable spatial variability in CH₄ fluxes across this alpine permafrost region during the contemporary period (1989-2018). Net CH₄ emissions were observed only in the central plateau (Fig. 2c; Supplementary Fig. 6a), primarily due to topographical factors, with their magnitude being mainly driven by soil moisture (Supplementary Note 1; Supplementary Table 2). In contrast, the remainder of the region acted as a net CH₄ sink (Fig. 2c; Supplementary Fig. 6a), displaying a spatial pattern that increased gradually from northwest to southeast, primarily driven by soil temperature (Supplementary Note 1; Supplementary Table 2). In addition, CH₄ fluxes varied among various ecosystem types (Supplementary Fig. 6b). All non-marshlands ecosystems consistently acted as a net CH₄ sink, with an overall average sink rate of -0.16 ± 0.12 mg CH₄ m⁻² d⁻¹ (mean ± SD, Fig. 2h and Supplementary Fig. 6a). The strength of CH₄ sink decreased in the following order: alpine shrub, alpine steppe, desert, forest, and alpine meadow (Supplementary Fig. 6b). By contrast, marshlands functioned as a net CH₄ source, with an average emission rate of 3.82 ± 1.04 mg CH₄ m⁻² d⁻¹, which was 24 times greater than the sink rate of non-marshlands (Fig. 2h and Supplementary Fig. 6). More importantly, the CH₄ emissions from marshlands (0.05 ± 0.005 Tg CH₄ yr⁻¹; consistent with previous studies²⁵; Supplementary Note 2; Supplementary Fig. 7a) were fully offset by the CH₄ sinks from non-marshlands (−0.06 ± 0.005 Tg CH₄ yr⁻¹; lower than previously reported^18,22,35; Supplementary Note 2; Supplementary Fig. 7b), resulting in the Tibetan alpine permafrost region as a marginal net CH₄ sink over the past three decades (−0.01 ± 0.01 Tg CH₄ yr⁻¹; Fig. 3).

Fig. 3: Temporal variation of net methane (CH₄) fluxes in the Tibetan alpine permafrost region based on contemporary data (black and gray lines) and future projections under SSP1-2.6 (green dashed line), SSP2-4.5 (dark green solid line), SSP3-7.0 (orange solid line), and SSP5-8.5 (blue dashed line).

The black solid line represents the net CH₄ fluxes during contemporary period (1989-2018), simulated using China Meteorological Forcing Dataset (CMFD). The gray solid line represents the net CH₄ fluxes during the contemporary period from 1989−2014, simulated using GCMs outputs. The colored solid and dashed lines represent multi-model mean CH₄ emissions under four SSP scenarios, and the shaded areas indicate the standard deviation of CH₄ emissions across multiple models. The mean and standard deviation of the multi-model data for the period 2071−2100 are shown in the right panel.

Temporal evolution of CH₄ fluxes under future climate scenarios

Our projections indicated that CH₄ emissions from marshlands would gradually increase under all SSP scenarios, whereas CH₄ sinks in non-marshlands were projected to steadily increase under most scenarios, except for SSP1-2.6, where they decline by the end of the century (Fig. 2h; Supplementary Fig. 8), primarily driven by a decrease in atmospheric CH₄ concentration (Supplementary Note 3). Specifically, during 2071-2100, CH₄ emissions from marshlands were expected to rise by 53.5-196.6%, while CH₄ sinks in non-marshlands were projected to decrease by 13.0% under SSP1-2.6 but increase by 60.9-265.2% under SSP2-4.5, SSP3-7.0, and SSP5-8.5, relative to the historical baseline (1985-2014) (Supplementary Fig. 8e–n). By integrating CH₄ emissions from marshlands and CH₄ sink from non-marshlands, our simulations revealed that the temporal evolution of the regional CH₄ budget exhibited divergent trajectories across SSP scenarios, with increasingly pronounced scenario-dependent differences over time (Fig. 3; Supplementary Fig. 9). Under the low- and medium-emission scenarios (SSP1-2.6 and SSP2-4.5), marshland CH₄ emissions were projected to exceed non-marshland CH₄ sinks throughout the 21^st century, thereby shifting the alpine permafrost region from a marginal net CH₄ sink to a net source (Fig. 3; Supplementary Figs. 7, 9). However, under the intermediate-high- and high-emission scenarios (SSP3-7.0 and SSP5-8.5), the region was projected to become a net CH₄ source by the mid-21st century (around 2040, with a regional warming of ~2.2 ± 0.25°C relative to the 1985−2014 average) (Fig. 3; Supplementary Figs. 7, 9). Notably, this net CH₄ source was projected to decline in the latter half of the century, ultimately transitioning back into a CH₄ sink by 2100 (Fig. 3; Supplementary Figs. 7, 9), with the reversal being particularly pronounced under SSP3-7.0 (Fig. 3; Supplementary Fig. 9), largely due to the increase in atmospheric CH₄ concentration (Supplementary Note 3). By the end of the 21st century (2071-2100), the net CH₄ fluxes over the Tibetan alpine permafrost region were projected to reach 0.07 ± 0.02, 0.07 ± 0.03, −0.06 ± 0.03, and -0.02 ± 0.02 Tg CH₄ yr⁻¹ under SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5, respectively (Fig. 3).

Our model prediction also showed that future changes in CH₄ fluxes exhibited both spatial and seasonal variability under future scenarios (Supplementary Figs. 10–12). Across all SSP scenarios, the central Tibetan Plateau was projected to experience the largest increase in CH₄ emissions, whereas the most pronounced enhancement of CH₄ sink occurred in the southwestern plateau (Supplementary Fig. 10b–e). This spatial pattern was significantly correlated with changes in soil temperature and moisture (Supplementary Table 3; Supplementary Figs. 13–14). The central region, which exhibited the greatest increase in soil moisture, provided favorable conditions for enhanced CH₄ emissions. In contrast, the southwestern plateau experienced a relatively larger increase in soil temperature and a smaller increase in soil moisture, conditions that were more conducive to CH₄ sink. As a result, CH₄ sink in the southwestern region was projected to be further enhanced under warming scenarios, thereby contributing to the spatial heterogeneity in the temporal dynamics of CH₄ fluxes (Supplementary Table 3; Supplementary Fig. 14). Furthermore, future climate changes were expected to alter the seasonal characteristics of CH₄ emissions in marshlands. Under all SSPs, the first emission peak—part of a bimodal seasonal cycle—was projected to occur earlier (from mid-May to late April; Supplementary Fig. 15), with the most pronounced shift under SSP5-8.5 and the least under SSP1-2.6, accompanied by a longer emission duration (Supplementary Figs. 11–12, 15). Peak CH₄ emission magnitudes for both seasonal peaks were highest under SSP5-8.5 and lowest under SSP1-2.6 (Supplementary Fig. 15).

Global relevance of CH₄ emissions over the Tibetan alpine permafrost region

Compared with the CH₄ fluxes in the Arctic permafrost region, the magnitude of the CH₄ fluxes across the Tibetan alpine permafrost region were significantly lower. Specifically, both the annual average CH₄ emissions from marshlands and the CH₄ sink from non-marshlands were lower than those in the Arctic (Supplementary Note 4). When only considering soil fluxes, the alpine permafrost region functioned as a weak net CH₄ sink (Fig. 3). However, including CH₄ emissions from natural lakes and thermokarst lakes (formed by the melting of excess ground ice during permafrost degradation, a typical process of abrupt thaw) shifted this alpine permafrost region to a marginal net CH₄ source (0.11 Tg CH₄ yr⁻¹; Supplementary Fig. 16), less than 2% of contemporary Arctic permafrost CH₄ emissions (6.7 Tg CH₄ yr⁻¹)³⁶. Furthermore, under the low- and medium-emission scenarios (SSP1-2.6 and SSP2-4.5), the future warming transitioned the Tibetan alpine permafrost region into a net CH₄ source, however, its contribution would remain negligible in the global permafrost CH₄ budget. The highest estimated soil-derived CH₄ emissions for 2100 (SSP2-4.5 scenario) reached 0.10 Tg CH₄ yr⁻¹ (Fig. 3). For context, adding historical emissions from natural and thermokarst lakes (~0.12 Tg CH₄ yr⁻¹) yielded a tentative total of ~0.22 Tg CH₄ yr⁻¹. Even this combined value—which likely underestimated future lake emissions due to warming-driven expansion of thermokarst systems—would still represent < 1% of the lowest-projection end-of-century Arctic emissions (28 Tg CH₄ yr⁻¹) under the low-emission scenario³⁶.

Discussion

Compared to the Arctic permafrost region which acts as a large CH₄ source¹⁶, our results demonstrated that the Tibetan alpine permafrost region currently functions as a marginal net CH₄ sink of −0.01 ± 0.01 Tg CH₄ yr⁻¹. The main reasons could be the much smaller proportion of marshlands and the lower CH₄ emission intensity in this alpine permafrost region. First, marshlands account for over 25% of the total area in the Arctic permafrost region³⁷, whereas they constitute only about 3.2% in the Tibetan alpine permafrost region (Supplementary Fig. 17)²¹. The extensive inundated areas in the Arctic create anaerobic conditions that promote CH₄ production and inhibit CH₄ oxidation, resulting in this high-latitude permafrost region as a net CH₄ source¹⁶. In contrast, the vast dry grasslands covering over 90% of the Tibetan alpine permafrost region favor CH₄ sink^20,21,23. Second, compared to Arctic wetlands, marshlands (swamp meadows) in the Tibetan alpine permafrost region exhibit substantially lower methanogen abundance and activity, primarily due to their hummock-hollow microtopography (Supplementary Fig. 18) and seasonal inundation regimes²⁵. On the one hand, the hummocks within swamp meadows create relatively well-drained and aerobic conditions, which are favorable for methanotroph communities and thereby enhance CH₄ sink²⁵. On the other hand, unlike the permanently inundated Arctic wetlands, swamp meadows are typically seasonally flooded²⁵. This markedly reduces the duration of soil saturation, further constraining the formation of anaerobic conditions necessary for methanogenesis and consequently suppressing CH₄ emissions²⁵.Together, these two factors result in a much weaker marshland CH₄ emission intensity than that in the Arctic (1.4 vs. 23.4 g CH₄ m⁻² yr⁻¹; Supplementary Fig. 6a)²⁵, favoring CH₄ sink across the entire study area. Overall, the Tibetan alpine permafrost region has remained a slight net CH₄ sink over the past 30 years, primarily due to its extensive arid uplands and lower CH₄ emissions from marshlands, which are shaped by hummock-hollow microtopography and characterized by seasonal inundation.

Future simulations revealed that the CH₄ budget of the Tibetan alpine permafrost region exhibited divergent trajectories under different SSP scenarios, shifting from a net sink to a source under low- and intermediate-emission pathways (SSP1-2.6 and SSP2-4.5), but returning to a sink under high-emission scenarios (SSP3-7.0 and SSP5-8.5) by the end of the 21st century (Fig. 3). These scenario-dependent changes were jointly regulated by climate forcing (e.g., warming and precipitation shifts), atmospheric CH₄ concentrations, and wetland dynamics, with climate and CH₄ concentrations dominating at different scenarios and periods. Atmospheric CH₄ concentrations were projected to play a dominant role in determining the sink trajectory in the late century, thereby influencing the net CH₄ flux, whereas wetland dynamics exerted a comparatively minor influence (see Supplementary Note 3 for details).

The attribution analysis (see Methods section for details) indicated that climate variables largely determined the persistence and magnitude of the net CH₄ source under low-to-moderate emission scenarios (Supplementary Fig. 19). First, warming-induced increases in soil temperature (Supplementary Fig. 13) could significantly stimulate methanogenic and methanotrophic activities^13,22,24, thereby enhancing CH₄ emissions from marshlands and sink in non-marshlands, as supported by strong correlations with soil temperature (r_emission = 0.74 and r_sink = −0.68, respectively; Supplementary Table 4). However, the higher temperature sensitivity (Q₁₀) of methanogens relative to methanotrophs led to CH₄ production increasing more than oxidation under warming (model-simulated Q₁₀: 3.0 for emissions, 2.0 for sink; similar to values reported in field studies; Supplementary Fig. 20; Supplementary Note 5)^25,38,39, further reinforcing the projected increase in net CH₄ emissions, although CLM5.0 represents these processes through bulk temperature sensitivities rather than explicitly modeling methanogens and methanotrophs. Additionally, warming could promote vegetation growth, increasing soil carbon inputs through litter and root exudates, which provided substrates for microbial decomposition and methanogenesis, while enhanced vascular plants facilitated CH₄ transport via aerenchyma, together amplifying CH₄ emissions (Supplementary Table 5)^40,41,42. Second, permafrost thaw and increased precipitation were expected to raise soil moisture levels (Supplementary Figs. 14, 21), which could in turn stimulate methanogenic activity in marshlands and suppress methanotrophic activity in non-marshlands^43,44 (Supplementary Table 4), thereby intensifying net CH₄ emissions. Higher soil moisture could also increase root exposure to anaerobic conditions and elevated CH₄ concentrations, further facilitating plant-mediated CH₄ transport (Supplementary Table 5)^41,42, which would then facilitate net CH₄ emissions. Third, future warming extended the CH₄ emission period by shortening snow cover duration, reducing microbial dormancy, and lengthening the growing season (Supplementary Fig. 11). Earlier snowmelt and prolonged snow-free periods enhanced soil moisture during the shoulder seasons, advancing anaerobic conditions and reducing diffusion barriers for CH₄ release (correlation coefficient = 0.84-0.97; Supplementary Figs. 11, 12b; Supplementary Table 5)^45,46, while warmer shoulder seasons reduced soil freezing and microbial dormancy, thereby extending heterotrophic decomposition and sustaining CH₄ release from soil organic matter (Supplementary Fig. 11; Fig. 12c)⁴⁷. In addition, longer growing seasons and enhanced vegetation productivity under warmer conditions enhanced substrate availability for anaerobic decomposition and plant-mediated CH₄ transport via aerenchyma (Supplementary Figs. 11m–p, 12d; Supplementary Table 5)^40,41, boosting marshland emissions¹⁰. The duration of CH₄ emissions was projected to extend under all SSP scenarios, contributing 6.6–8.6% of total emissions and accounting for 7.3−9.8% of the projected increase from 1989−2014 to 2071−2100 (Supplementary Table 6).

In contrast to SSP1-2.6 and SSP2-4.5, under SSP3-7.0 and SSP5-8.5 scenarios, although net CH₄ emissions increased during the mid-21st century, a reversal to a net CH₄ sink was projected by the end of the century. This reversal is predominantly driven by the high atmospheric CH₄ concentration under these two scenarios. The attribution analysis (see Methods section for details) showed that during the mid-century, the positive effect of climate variables (e.g., temperature and precipitation) on net CH₄ emissions—mediated by concurrent increases in soil temperature and moisture—outweighed the influence of atmospheric CH₄ concentration and inundation dynamics (Supplementary Figs. 22–24). These climate-induced effects, consistent with those observed under low-to-moderate emission scenarios, dominate the CH₄ balance and promote a net CH₄ source. However, by the end of the century, the sustained rise in atmospheric CH₄ concentration (Supplementary Fig. 22) would further enhance net CH₄ sink (Supplementary Fig. 23). Specifically, elevated atmospheric CH₄ concentration strengthens the CH₄ concentration gradient between the atmosphere and surface soils, facilitating diffusive CH₄ transport into well-aerated soils and stimulating methanotrophic activity⁴⁸. This enhanced sink partially offsets the warming-driven increase in CH₄ emission, ultimately shifting the regional CH₄ budget toward a net sink. Notably, SSP5-8.5 exhibited lower atmospheric CH₄ concentrations than SSP3-7.0 and featured a turning point after the 2080 s, when CH₄ levels began to decline. This resulted in a weaker net CH₄ sink under SSP5-8.5 compared to SSP3-7.0, with the sink strength gradually decreasing after the 2080 s and approaching zero (Supplementary Figs. 22–23).

Our study provides estimates of CH₄ fluxes in the Tibetan alpine permafrost region and the transition between CH₄ source and sink under both contemporary and future climate conditions, but several limitations remain. First, the uncertainties in future CH₄ flux projections arise from two aspects: the input climate data and model framework. For the input data, although we selected climate projections from four GCMs to roughly quantify the uncertainty range (Supplementary Fig. 25), substantial discrepancies remain among GCMs in simulating regional climate over the Tibetan Plateau⁴⁹. These differences introduce unavoidable variability into the atmospheric forcing of CH₄ flux simulations. Moreover, the spatial resolution of these GCM-derived datasets remains relatively coarse (~25 km), which limits their ability to capture the fine-scale climatic heterogeneity of the alpine permafrost region and further amplifies uncertainties in modeled CH₄ fluxes⁵⁰. In addition to input data limitations, model structural uncertainty also plays a significant role^51,52. Here we employed a single land surface model, CLM5.0, to simulate CH₄ dynamics. Relying solely on one model framework neglects the potentially divergent responses from other land models with differing process representations—such as LPJ-WSL or TEM—thereby limiting the robustness of the projections. Furthermore, CLM5.0 is built upon the CLM4Me framework, which does not explicitly represent the mechanistic processes of CH₄ production and oxidation⁵³, potentially limiting its ability to fully capture microbial-mediated responses under future climate change. Future efforts should thus prioritize the development of high-resolution regional climate forcing datasets and implement multi-model simulations, as well as incorporating microbial processes into model structures, to better constrain these sources of uncertainty.

Second, the uncertainty in CH₄ flux modeling arises from the use of grid-cell average decomposition rates and the simplified classification of plant functional types in CLM5.0 model. Particularly, marshland CH₄ flux modeling remains simplified, as using grid-cell average decomposition rates does not account for the fine-scale variability of marshlands³⁰. Given that marshlands exhibit substantial spatial variability, this averaging approach may obscure critical processes²⁷. Additionally, CLM5.0 represents Tibetan alpine grasslands (i.e., major vegetation type across this alpine permafrost region) with a single C3 grass plant functional type using static trait parameters (e.g., specific leaf area, maximum carboxylation rate of Rubisco, maximum electron transport rate)²⁹, failing to capture species composition and functional trait variations across grassland types (e.g., alpine meadow, alpine steppe) and their responses to environmental gradients⁵⁴. Similar plant functional type simplifications are also applied to shrubland and forest areas in the model, which may further contribute to uncertainty in the soil CH₄ flux. These oversimplified representations can bias the simulation of ecosystem structure and function, as key plant traits directly mediate vegetation productivity⁵⁵. Future improvements should thus incorporate a dedicated wetland sub-model that captures finer-scale wetland characteristics and a more detailed representation of plant functional types.

Third, our modeling study does not explicitly account for CH₄ emissions from natural lakes and thermokarst landforms (e.g., thermokarst lakes and thermo-erosion gullies), which likely results in an underestimation of total CH₄ fluxes. Previous studies have estimated that CH₄ emissions reach ~0.04 Tg CH₄ yr⁻¹ from natural lakes on the Tibetan Plateau (Supplementary Fig. 16), equivalent to 80% of the CH₄ emissions from marshlands in our study. Moreover, abrupt thaw processes, such as the formation of thermokarst lakes and thermos-erosion gullies, can trigger substantial increases in CH₄ emissions from permafrost regions. For instance, CH₄ fluxes from thermokarst lakes on the plateau are 50−150% higher than those from swamp meadows, while CH₄ emissions from thermos-erosion gullies increase by 3.5-fold during late thaw stage compared to undisturbed permafrost on the Tibetan Plateau^56,57. These rapid thaw features have already been estimated to release about 0.08 Tg CH₄ yr⁻¹, a magnitude comparable to the emissions from marshland areas⁵⁷. Additionally, given that thermokarst landscapes are projected to expand threefold under continued warming⁵⁸, their omission from our simulations likely results in an incomplete assessment of CH₄ fluxes. Future research should thus incorporate a thermokarst module and explicitly quantify CH₄ emissions from natural lakes, thermokarst lakes, and other thermokarst landforms (e.g., thermo-erosion gullies) to improve regional CH₄ budgets.

Fourth, while our modeling study focuses on the impacts of future climate change on CH₄ fluxes, it does not consider other key factors such as human activities (e.g., marshland conservation and livestock grazing) and vegetation succession, both of which can significantly alter the dynamics of CH₄ flux^59,60. Particularly, the implementation of Qinghai-Tibet Plateau Ecological Protection Law of the People’s Republic of China⁶¹ prohibits peat extraction and drainage of natural wetlands, which may lead to the expansion of wetland areas and increased anaerobic conditions, potentially enhancing CH₄ emissions⁶². Meanwhile, grazing intensity in the region has declined over the last several decades due to ecological conservation policies and projects (Supplementary Fig. 26)⁶², suggesting a weakening effect on CH₄ fluxes over time. Future study should thus focus on developing modules to characterize human impacts and vegetation shifts, which will be essential for improving the accuracy of projections and providing a more comprehensive understanding of CH₄ dynamics in permafrost regions.

Finally, observational constraints—including the scarcity of CH₄ flux measurements during non-growing season and the lack of site-specific climate input data—limit model validation and calibration, as the coarse resolution of the China Meteorological Forcing Dataset (CMFD) dataset cannot fully capture local climate variability. Expanding flux monitoring networks, particularly through the establishment of more eddy covariance flux towers, is thus essential to improve both CH₄ flux and climate observations, thereby reducing uncertainties. Despite these limitations, our study provides a crucial foundation for understanding CH₄ flux dynamics in high-altitude permafrost region and emphasizes the need for continued model and data improvements.

In summary, this study quantified and forecasted the spatiotemporal dynamics of soil CH₄ fluxes in the Tibetan alpine permafrost region based on the calibrated CLM5.0 model. Our results indicated that this alpine permafrost region currently functioned as a marginal net CH₄ sink (-0.01 ± 0.01 Tg CH₄ yr⁻¹), primarily due to its extensive arid uplands. However, future environmental changes were expected to alter the CH₄ budget, with divergent trends under different climate scenarios. Under low-emission scenarios (SSP1-2.6 and SSP2-4.5), the substantial increase in soil temperature and moisture levels, coupled with much longer snow-free periods and extended durations of microbial activity and growing seasons, could lead to shift from a net CH₄ sink to a net CH₄ source. In contrast, under medium-to-high emission scenarios (SSP3-7.0 and SSP5-8.5), the enhancement of the CH₄ sink by the continued rise in atmospheric CH₄ concentrations toward the end of this century was expected to outweigh the climate-driven increase in CH₄ emissions, resulting in a gradual decline of net CH₄ emissions and a transition back to a net CH₄ sink (Fig. 4). These findings address a critical knowledge gap in permafrost carbon cycle, and advance our understanding of the CH₄ budget in high-altitude permafrost region under climate change.

Fig. 4: Characteristics of methane (CH₄) fluxes in marshland and non-marshland areas across the Tibetan alpine permafrost region during the historical period and under different future climate scenarios.

During the historical period, non-marshlands acted as CH₄ sinks, while marshlands served as CH₄ sources. Overall, the region functioned as a weak net CH₄ sink. Under future climate change, CH₄ emissions from marshlands are projected to exceed CH₄ sink by non-marshlands throughout the twenty-first century under the low- and medium-emission scenarios (SSP1-2.6 and SSP2-4.5), shifting the region from a marginal CH₄ sink to a net source. Under the intermediate-high and high-emission scenarios (SSP3-7.0 and SSP5-8.5), the region is projected to become a net CH₄ source by the mid-twenty-first century. Notably, this net CH₄ source is expected to decline in the latter half of the century, eventually transitioning back to a CH₄ sink by 2100.

Methods

Study area

The alpine permafrost region on the Tibetan Plateau, covering ~1.06 × 10⁶km², is the largest high-altitude permafrost region in the Northern Hemisphere^19,33. This region has an average elevation exceeding 4,000 m, with a mean annual temperature of -5.9 ± 3.0°C and an average annual precipitation of 370.4 ± 176.0 mm (Supplementary Fig. 27)³¹. The permafrost in this region is predominantly classified as climate-driven type, which is vulnerable to rapid climate warming⁶³. The average active layer thickness is 2.3 ± 0.6 m⁶⁴. During the past decade, the plateau has undergone significant climate warming, leading to widespread permafrost degradation and an increase in active layer thickness at a rate of 0.9 cm yr^-15. The main vegetation types in this region are alpine steppe, alpine meadow, and swamp meadow. Of them, alpine steppe is primarily characterized by species such as Stipa purpurea and Carex moorcroftii, alpine meadow by Kobresia pygmaea and K. humilis, and swamp meadow by K. tibetica²¹. Additionally, the northwestern part of the permafrost region contains alpine deserts, while the southeastern part has a small amount of forest and shrubland (Supplementary Fig. 17b)²¹.

CH₄ flux dataset derived from in-situ observations

To explore the characteristics of CH₄ flux across the Tibetan alpine permafrost region, we constructed a field-observed CH₄ flux dataset covering various ecosystem types from peer-reviewed papers published before 2025, which were identified by searching Web of Science (http://apps.webofknowledge.com/) and the China National Knowledge Infrastructure (http://cnki.net/). The following keywords were utilized during the literature search: CH₄ flux/sink/source/exchange; greenhouse gases; Tibetan Plateau; Tibet; alpine; Qinghai-Xizang Plateau. To enhance data integrity, we systematically assessed the quality of each study included in the dataset, focusing on methodological consistency and the temporal and spatial resolution of reported measurements. We applied the following four criteria to filter the retrieved literature (Supplementary Table 1; Supplementary Data 1), ensuring data accuracy while minimizing omissions and biases during the selection process. First, the selected study sites had to be located within the permafrost region on the plateau, as described in the literature or indicated on the permafrost distribution map across this region³³. Second, only in-situ CH₄ flux measurements based on methods such as static chamber and eddy covariance techniques, were included. Third, for both single-factor and multi-factor experiments, only CH₄ flux measurements from control groups were considered. Fourth, CH₄ flux data could be extracted from text, tables, and figures, or obtained through calculations, and the observational data had to be clearly and accurately presented.

Following the above criteria, a total of 13 studies were obtained, covering 24 sites across the region (11 CH₄ source sites and 13 sink sites; Fig. 1). From these studies, we compiled a total of 1,112 CH₄ flux observations, with data obtained directly from tables and digitized from figures using GetData Graph Digitizer 2.2.4 (available online: http://getdata-graph-digitizer.com). All flux data were subsequently converted to a standardized unit (μg CH₄ m⁻² h⁻¹). In addition to the CH₄ fluxes, pertinent site information such as coordinates, elevation, climate, vegetation, and soil properties was also compiled from the original papers (Supplementary Table 1; Supplementary Data 1). The dataset comprised of five vegetation types: alpine steppe (−8.3 ± 7.2 μg CH₄ m⁻² h⁻¹; 5 sites), alpine meadow (−13.0 ± 10.1 μg CH₄ m⁻² h⁻¹; 7 sites), alpine desert (−6.0 μg CH₄ m⁻² h⁻¹; one site), alpine shrubland (−37.5 μg CH₄ m⁻² h⁻¹; one site) and swamp meadow (590.2 ± 689.1 μg CH₄ m⁻² h⁻¹; 10 sites) (Supplementary Table 1; Supplementary Table 7; Supplementary Data 1). This dataset provides essential data support for understanding the characteristics of CH₄ flux across the study area and offers validation data for model predictions over the regional scale.

Model description

The land surface model CLM5.0²⁹ was calibrated and validated (See Methods “Model calibration and validation” for details) for the Tibetan alpine permafrost region, and used to investigate the spatiotemporal characteristics of CH₄ fluxes across this region over the past three decades, as well as for future projections under different climate scenarios. CLM5.0 is the latest version of the land models developed through the Community Earth System Model project and can simulate terrestrial carbon and nitrogen cycles²⁹. The model integrates a variety of permafrost hydrological and thermal processes, including snow schemes, vertical dynamics of soil organic matter, and thermal properties of frozen and unfrozen soils²⁹. It accurately represents the physical and hydrological states of permafrost, effectively simulates the seasonal dynamics of active layer thickness, captures freeze-thaw cycles in cold regions, and improves predictions of permafrost carbon-climate feedback^29,65. The CLM model has been widely used to assess carbon dynamics in high-latitude permafrost region across the Northern Hemisphere⁶⁶. Overall, the choice of the CLM5.0 model stems from its advanced capability to simulate terrestrial carbon and nitrogen cycles, particularly in permafrost regions, which enables nuanced predictions of CH₄ dynamics.

The CH₄ module utilized in this study (CLM4Me, originally integrated into CLM4 by Riley et al.³⁰) has been further incorporated into the updated land model, CLM5.0. This module consists of five components: CH₄ production, oxidation, ebullition, aerenchyma transport and aqueous and gaseous diffusion. CH₄ production occurs in anaerobic soil layers, linked to the grid-cell estimate of heterotrophic respiration, with adjustments for temperature, pH, redox potential, and seasonal inundation fractions. In the absence of explicit wetland plant functional types and specific soil biogeochemical representations, the model uses averaged grid-cell decomposition rates as proxies. Upland heterotrophic respiration is used to estimate the wetland decomposition rate after first dividing off the O₂ limitation. CH₄ oxidation is primarily driven by heterotrophic methanotrophs, microbes that rely on CH₄ and oxygen concentrations for their activity. The process can be modeled using double Michaelis-Menten kinetics⁶⁷, which captures the dependency on both CH₄ and oxygen concentrations:

$$R_{{\mathrm{oxic}}}=R_{o,\max} \left( \frac{C_{{{\mathrm{CH}}}_4}}{K_{{{\mathrm{CH}}}_4}+C_{{{\mathrm{CH}}}_4}} \right) \left( \frac{C_{{{\mathrm{O}}}_2}}{K_{{{\mathrm{O}}}_2}+C_{{{\mathrm{O}}}_2}} \right) Q_{10} F_{\vartheta}$$

(1)

where $K_{{{\mathrm{CH}}}_4}$ and $K_{{{\mathrm{O}}}_2}$ represent the half saturation constants for CH₄ and O₂, respectively, while ${R}_{o,\max }$ is the maximum oxidation rate. The factor ${Q}_{10}$ accounts for the temperature sensitivity, and ${F}_{\vartheta }$ introduces a soil moisture limitation above the water table, representing water stress effects on methanotrophic activity. A key distinction between upland and wetland methanotrophs lies in their different affinities for CH₄. Following Whalen and Reeburgh⁶⁸, $K_{{{\mathrm{CH}}}_4}$ and ${R}_{o,\max }$ values for uplands are assumed to be 10 times greater than those for wetlands. Aerenchyma-mediated gas transport in CLM is modeled as diffusion across concentration gradients between soil layers and the atmosphere. Ebullition is based on Wania et al.⁶⁹, where the aqueous concentration of CH₄ informs the equilibrium gaseous partial pressure. If this pressure exceeds 15% of ambient levels, CH₄ is released through bubbling. Additionally, the diffusion rates of gases in both water and air are dependent on factors such as the gas type, soil structure, and temperature.

The inundation fraction is a key parameter influencing CH₄ production and emission. The default method for calculating the inundation fraction involves applying a simple inversion model, which relies on three optimized parameters (e.g., ${p}_{1},{p}_{2}$ and ${p}_{3}$) and simulated water table depth (${z}_{w}$) and surface runoff (${Q}_{r}$) to determine spatial inundation (${f}_{s}$)³⁰.

$${f}_{s}={p}_{1}{e}^{-{z}_{w}/{p}_{2}}+{p}_{3}{Q}_{r}$$

(2)

The three parameters (e.g., ${p}_{1},{p}_{2}$ and ${p}_{3}$) have been evaluated at relatively low spatial resolution (0.25° × 0.25°)³⁰. Given the strong spatial heterogeneity across our study area, the flood inundation map generated using this method does not align well with the wetland distribution derived from Niu et al.⁷⁰ and the 1:1,000,000 vegetation map (Supplementary Fig. 17)²¹. To address this issue, we modified the method for calculating the inundation fraction by setting “finundation_method = h2osfc”, which calculates the spatial inundation based on the amount of surface water storage and the microtopography variability of each grid cell⁷¹. Additionally, we specifically set dynamic inundation fractions for grid cells identified as containing marshland (Supplementary Note 6; Supplementary Fig. 17a), based on the wetland distribution map from Niu et al.⁷⁰ and estimates of inundation fraction derived from the calibrated TOPMODEL (Supplementary Fig. 28) to simulate CH₄ emissions under dynamic wetland extent (See Methods section: Dynamic wetland model for details).

Model inputs

CLM5.0 was driven by land surface data, climate forcing, atmospheric CO₂ concentrations, and atmospheric CH₄ concentrations (Supplementary Table 8). In this study, the land surface input data includes soil properties and the distribution of plant functional types. The soil data (i.e., silt, sand content and soil organic carbon content) were obtained from the National Earth System Science Data Center (http://www.geodata.cn). The vegetation distribution was derived from China’s Vegetation Atlas at a scale of 1:1,000,000²¹. The Tibetan alpine permafrost region encompasses forests, shrublands, alpine steppes, alpine meadows, marshlands, and deserts. Given that the CLM model does not classify plant functional types in a detailed manner, we designated alpine steppes, alpine meadows, and marshlands (wet meadows) as C3 grass vegetation types, while alpine shrublands and forests were represented using the default shrub and forest plant functional types in CLM5.0. All simulations and analyses for marshland and non-marshland areas were conducted using fixed grid cells (Supplementary Note 6; Supplementary Fig. 17a) throughout both historical and future period. Within the marshland grid cells, the inundation fraction varied dynamically according to the TOPMODEL-predicted wetland inundation fraction during both historical and future periods (See Methods section: Dynamic wetland model for details). The climate forcing data comprise seven meteorological elements: air temperature, precipitation rate, downward longwave radiation, downward shortwave radiation, wind speed, specific humidity, and surface pressure. For the retrospective simulation period (1979 to 2018), these seven meteorological inputs were derived from the China meteorological forcing dataset (CMFD; 0.1° spatial resolution and 3-h temporal resolution)³¹. For projecting the future changes of CH₄ fluxes, the bias-corrected and downscaled climate forcings (0.25°$\times$0.25°)³² from four GCM runs in the Coupled Model Intercomparison Project Phase 6 (CMIP6) were used for both the historical period (1950−2014) and future projections (2015-2100) under four SSP scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5). The four GCMs used are NorESM2-LM, EC-Earth3-Veg-LR, CanESM5, and ACCESS-CM2³². The selection of these GCMs was based on four criteria: First, we selected GCMs which provide projections under all four warming scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5) and deliver all seven climate variables (air temperature, precipitation rate, downward longwave radiation, downward shortwave radiation, wind speed, specific humidity, and surface pressure) required by the CLM5.0 model. Second, we chose GCMs that are widely used in global studies^72,73,74. Third, we prioritized GCMs from different institutions to avoid the potential for highly correlated results within models developed by the same Earth System Model (ESM) institute⁷⁵. Finally, we reviewed the literature to identify GCMs that show relatively better performance and lower biases in simulating temperature and precipitation over the Tibetan Plateau^76,77. The forcings were provided by the NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP-CMIP6; 0.25° spatial resolution and daily temporal resolution)³². It should be noted that as the NEX-GDDP-CMIP6 provided most meteorological variables needed for running CLM5.0 except surface pressure, surface pressure was obtained from the original GCM runs (the same variants as in the NEX-GDDP-CMIP6) from CMIP6 and resampled to a spatial resolution of 0.25°. Historical atmospheric CO₂ and CH₄ concentrations were derived from a combination of ice core records and atmospheric observations for the period 1860−2018⁷⁸, while the atmospheric CO₂ and CH₄ concentrations under different scenarios were obtained from Meinshausen et al.⁷⁸. Atmospheric N deposition is fixed at 1850 levels from Lamarque et al.⁷⁹.

Model calibration and validation at site level

We conducted model calibration and validation for marshland and non-marshland ecosystems using CH₄ observations from the dataset compiled in this study (see Methods section: CH₄ flux dataset derived from in-situ observations). For each site, the climate of the grid cell covering the site location was derived from the China Meteorological Forcing Dataset (CMFD). Within the range of parameter thresholds, the 21 CH₄ sensitivity parameters provided by Muller et al.⁸⁰ were manually adjusted using a trial-and-error approach to minimize the differences between simulated and observed CH₄ emission and sink rates (Supplementary Table 9). Subsequently, the optimized parameters were used by the CLM5.0 model for simulating CH₄ fluxes at all sites with observations.

The CH₄ simulations from the optimized model were compared with observations from the aforementioned CH₄ dataset to validate the model (see Methods section: CH₄ flux dataset derived from in-situ observations). First, we selected sites with observation durations of at least five consecutive months and observation periods spanning from 1989−2018 (16 sites) to compare the seasonal dynamics of CH₄ fluxes between actual observations and model simulations (Supplementary Fig. 3). Model performance was assessed based on the 1:1 line plot and four indicators: R² (coefficient of determination; Eq. 3), RMSE (root mean square error; Eq. 4), BIAS (Eq. 5), and MAE (mean absolute error; Eq.6).

$${R}^{2}=\frac{{\left[{\sum }_{i=1}^{n}\left({O}_{i}-\bar{O}\right)\left({S}_{i}-\bar{S}\right)\right]}^{2}}{{\sum }_{i=1}^{n}{\left({O}_{i}-\bar{O}\right)}^{2}{\sum }_{i=1}^{n}{\left({S}_{i}-\bar{S}\right)}^{2}}$$

(3)

$${{\rm{RMSE}}}=\sqrt{\frac{{\sum }_{i=1}^{n}{({S}_{i}-{O}_{i})}^{2}}{n}}$$

(4)

$${{\rm{BIAS}}}=\frac{1}{n}\sum ({S}_{i}-{O}_{i})$$

(5)

$${{\rm{MAE}}}=\frac{1}{n}{\sum}_{i=1}^{n}|{S}_{i}-{O}_{i}|$$

(6)

where ${S}_{i}$ and ${O}_{i}$ indicate simulated and observed values, respectively; $\bar{S}$ and $\bar{O}$ are their averages; $n$ denotes the number of simulation-observation pairs. The R² values indicate a strong agreement between observed and simulated fluxes, while the RMSE values reflect minimal average differences, suggesting reliable model performance. The model validation indicated that the data points for simulated and observed CH₄ fluxes were closely aligned along the 1:1 line (Supplementary Fig. 4), demonstrating effective model performance across the study area. Additionally, we validated the surface soil temperature (0−5 cm), soil volumetric water content (0−5 cm), and soil organic carbon density (0–1 m). The simulations and observations for soil temperature and soil organic carbon density were well-aligned around the 1:1 line, and the seasonal dynamics of soil moisture simulations were also consistent with observations (Supplementary Fig. 2). This indicated that the model reliably simulated soil temperature, moisture, and organic carbon stock at these sites.

Simulation protocol

Using the validated model, we conducted two sets of simulations across the Tibetan alpine permafrost region: (i) retrospective simulations driven by observation-based meteorological dataset (1979−2018) to investigate contemporary CH₄ fluxes; and (ii) a set of simulations driven by bias-corrected and downscaled climate forcings from four CMIP6 GCMs derived from the NEX-GDDP-CMIP6 (1950–2100) to investigate the future changes in CH₄ fluxes under different climate scenarios. For the retrospective simulation, a spin-up simulation of 1000 years (consisted of 400 years in accelerated mode and 600 years in standard mode) was first conducted using the initial 10 years of climate data (i.e., 1979–1988), which were cycled, and with atmospheric CO₂ and CH₄ concentration held constant at 1850 levels to allow the modeled ecosystem to achieve a quasi-equilibrium status. Subsequently, another pre-transient run was performed with the initial 10 years of climate data (i.e., 1979–1988) cycled, varying atmospheric CO₂ and CH₄ concentrations from 1851–1988. Finally, a formal transient run was carried out for the contemporary period of 1989 to 2018, with varied climate conditions, atmospheric CO₂ and CH₄ concentrations. The retrospective simulations were conducted at a resolution of 0.1°. For the future simulations driven by GCM-based climate forcings, a 1000-year spin-up (consisted of 400 years in accelerated mode and 600 years in normal mode) was conducted using the first 10 years of climate data cycled (1950−1959), with atmospheric CO₂ and CH₄ concentration fixed at 1850 levels to allow the modeled ecosystem to achieve a quasi-equilibrium status. Subsequently, another pre-transient run was performed with the initial 10 years of climate data (i.e., 1950–1959) cycled, varying atmospheric CO₂ and CH₄ concentrations from 1851 to 1959. A transient run was then performed for the historical period as defined by CMIP6 (1960-2014) with varied climate and atmospheric CO₂ and CH₄ concentrations, and this simulation was extended into the future (2015−2100) with varying climate and atmospheric CO₂ and CH₄ concentrations according to different SSP scenarios. These simulations were performed at a resolution of 0.25°. In both sets of simulations, CH₄ emissions from marshlands and CH₄ sinks from non-marshlands were separately modeled, and then aggregated to yield total regional fluxes through area weighting for each grid cell, based on the estimated marshland distribution map (Supplementary Note 6; Supplementary Fig. 17a). For marshlands, CH₄ fluxes were calculated as the area-weighted sum of emissions from inundated and non-inundated portions, with the inundation extent varying over time from a TOPMODEL-based diagnostic model calibrated over this region (see Methods section: Dynamic wetland model).

Dynamic wetland model

To simulate wetland extent over the study area, we employed a diagnostic model based on TOPMODEL (TOPography-based hydrological MODEL), which has been successfully used to capture the spatiotemporal dynamics of global natural wetlands (Xi et al.⁸¹_’⁸²). As topography is the key determinant of lateral water flow, the classical TOPMODEL utilizes a compound topographic index (CTI) to redistribute soil water within sub-grid elements of a larger land surface grid cell or a catchment. Sub-grid areas that are always or regularly saturated–based on CTI–are identified as inundated areas (or wetlands). However, the application of TOPMODEL at large scales and over long time series is computationally intensive due to the high spatial resolution of CTI datasets (e.g., 1 km for the HYDRO1k dataset from the U.S. Geological Survey and 500 m for the CTI dataset from Marthews et al.⁸³). To address this limitation, Xi et al.^81,82 developed a computationally efficient diagnostic model using an asymmetric sigmoid function, originally proposed by Stocker et al.⁸⁴. This function (Eq. 7) directly related the wetland fraction to the grid-cell mean water table depth (WTD). The simulated wetland fraction was further constrained by a parameter, f_x^max, representing the maximum wetland fraction (Eq. 8).

$$\varPsi ({\varGamma }_{x})={(1+{v}_{x}\, \,{{\rm{\cdot }}}\,{{{\rm{e}}}}^{-{k}_{x}({\varGamma }_{x}-{q}_{x})})}^{-1/{v}_{x}}$$

(7)

$${f}_{x}=\min ({\varPsi }_{x}({\varGamma }_{x}),{f}_{x}^{\max })$$

(8)

where the ${\varGamma }_{x}$ is the water table in the grid x, and the f_x is the fitted inundated fraction. The v_x, k_x, q_x, f_x^max are four parameters. The optimized parameter combination (v, k, q, f ^max)_x is determined by selecting minimum root-mean-square-error (RMSE, defined in Eq. (4) of long-term maximum wetland extent between simulations from TOPMODEL and observations). Once the value of $\left({v}_{x},{k}_{x},{q}_{x}\right)$ are determined, the wetland fraction ${f}_{x}$ can be directly derived from the monthly water table ${\varGamma }_{x}$ according to Eqs. (7) and (8).

In this study, we followed the same methodology as Xi et al.^81,82 to calculate monthly WTD based on monthly total soil moisture (SM) and soil temperature (ST). Model parameters in Eqs. (7) and (8) were calibrated using the wetland map developed by Niu et al.⁷⁰, which represents one of the most recent and comprehensive reconstructions of wetland spatial distribution and temporal dynamics in China³⁵. This dataset, with a spatial resolution of 240 × 240 m, was derived through systematic visual interpretation of a series of Landsat MSS images (1978) and CBERS-CCD images (2008), covering four time periods: 1978, 1990, 2000, and 2008. The map included both natural and artificial wetlands; however, for model calibration, we used only inland natural wetlands, excluding lakes and artificial wetland types. The spatial distribution of wetland extent over the study area derived from this map was shown in Supplementary Fig. 28.

The parameter calibration of the TOPMODEL-based diagnostic model and the prediction of wetland extent were conducted at two spatial resolutions (0.1° and 0.25°), to match with the spatial resolutions of the simulations forced by CMFD³¹ and CMIP6 climate data³², respectively. For the 0.1° case, monthly SM and ST from the simulation forced by CMFD were used to calculate WTD and predict the wetland extent over the period 1989−2018. For the 0.25° case, parameter calibration was conducted using monthly SM and ST data from ERA5⁸⁵ (1989-2018), due to large uncertainties and inconsistencies in soil moisture across CMIP6 climate models. The selected optimized parameter combination (v, k, q, f ^max)_x was then applied to all CMIP6 models to simulate wetland extent for both the historical period (1980-2018) and future projections under four shared socioeconomic pathway (SSP) scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5), preserving inter-model variability. For CMIP6 outputs, we first regridded them to a spatial resolution of 0.25° × 0.25°. Then grid-scale correction factors were calculated for SM and ST separately, following Hersbach et al.⁸⁵, using ERA5 as a reference over the historical period (1989−2018) to correct CMIP6 biases. The bias-corrected SM and ST were subsequently used to compute WTD and predict wetland extent.

Model validation showed that our historical simulation effectively reproduced the spatial pattern of wetland extent in the Tibetan alpine permafrost region, with a RMSE of less than 3% across most regions (Supplementary Fig. 29). Using the calibrated model, we simulated current and future wetland dynamics across the region. The results suggest a slightly gradual decline in wetland fraction throughout the 21st century under all four SSP scenarios (Supplementary Fig. 30). Due to the unique topographic and climatic conditions of the Tibetan Plateau, high-altitude marshlands are not fully inundated but are instead dominated by swamp meadows. These ecosystems typically consist of ~70% hummocks and 30% inundated hollows⁸⁶. Our in-situ CH₄ flux measurements were collected from such swamp meadows, where CH₄ dynamics are strongly influenced by this distinct microtopography. To ensure consistency with the actual conditions at these sites, we derived a marshland distribution map based on the satellite-derived wetland datasets from Niu et al.⁷⁰. Specifically, we extracted, for each grid cell, the maximum inundation fraction across the available time periods in Niu et al.⁷⁰, and then divided this maximum value by 0.3 to estimate the spatial extent of marshland. A detailed description of this approach was provided in Supplementary Note 6. All subsequent analyses were conducted for marshland and non-marshland regions (See Supplementary Note 6; Supplementary Fig. 17a for details).

Attribution analysis

To attribute the relative contributions of climate forcing, dynamic wetland extent, and atmospheric CH₄ concentration to the projected net CH₄ budget over the Tibetan alpine permafrost region, we designed a set of sensitivity experiments in addition to the default simulation (${{{{\rm{FCH}}}}_{4}}_{{{\rm{default}}}}$), which incorporated time-varying climate, atmospheric CH₄ concentration, and wetland extent during 1989-2100. First, to isolate the effect of atmospheric CH₄ concentration, we conducted simulations with a fixed atmospheric CH₄ level (1.7 ppm)⁸⁷ throughout 2015−2100, denoted as ${{{{\rm{FCH}}}}_{4}}_{{{\rm{fixatmch}}}4}$. Second, to isolate the effect of wetland dynamics, we fixed wetland extent at its 2015 distribution and conducted simulations for 2015−2100, denoted as ${{{{\rm{FCH}}}}_{4}}_{{{\rm{fixinundation}}}}$. The individual contributions of each driver to the net CH₄ flux were estimated as follows:

$${\Delta }_{{{\rm{atmch}}}4}={{{{\rm{FCH}}}}_{4}}_{{{\rm{default}}}}-{{{{\rm{FCH}}}}_{4}}_{{{\rm{fixatmch}}}4}$$

(9)

$${\Delta }_{{{\rm{inundation}}}}={{{{\rm{FCH}}}}_{4}}_{{{\rm{default}}}}-{{{{\rm{FCH}}}}_{4}}_{{{\rm{fixinundation}}}}$$

(10)

$${\Delta }_{{{\rm{climate}}}\_{{\rm{forcing}}}}={{{{\rm{FCH}}}}_{4}}_{{{\rm{default}}}}-{\Delta }_{{{\rm{atmch}}}4}-{\Delta }_{{{\rm{inundation}}}}$$

(11)

where ${\Delta }_{{{\rm{atmch}}}4}$, ${\Delta }_{{{\rm{inundation}}}}$ and ${\Delta }_{{{{\rm{climate}}}}\_{{{\rm{forcing}}}}}$ represent the contributions of atmospheric CH₄ concentration, wetland extent, and climate forcing, respectively.

To quantify the contributions of each factor to changes in CH₄ flux across three future periods (2015−2040, 2041−2070, 2071−2100), we calculated the change in default CH₄ flux relative to the historical period (1989−2014):

$${\Delta {\overline{{{\rm{F}}}{{{\rm{CH}}}}_{4}}}}_{{{\rm{default}}}}^{{{\rm{period}}}}={\overline{{{\rm{F}}}{{{\rm{CH}}}}_{4}}}_{{{\rm{default}}}}^{{{\rm{period}}}}-{\overline{{{\rm{F}}}{{{\rm{CH}}}}_{4}}}_{{{\rm{default}}}}^{{{\rm{his}}}}$$

(12)

$${\Delta {\overline{{{\rm{F}}}{{{\rm{CH}}}}_{4}}}_{{{\rm{atmch}}}4}^{{{\rm{period}}}}}={\overline{{{\rm{F}}}{{{\rm{CH}}}}_{4}}}_{{{\rm{atmch}}}4}^{{{\rm{period}}}}-{\overline{{{\rm{F}}}{{{\rm{CH}}}}_{4}}}_{{{\rm{atmch}}}4}^{{{\rm{his}}}}$$

(13)

$${\Delta {\overline{{{\rm{F}}}{{{\rm{CH}}}}_{4}}}}_{{{\rm{inundation}}}}^{{{\rm{period}}}}={\overline{{{\rm{F}}}{{{\rm{CH}}}}_{4}}}_{{{\rm{inundation}}}}^{{{\rm{period}}}}-{\overline{{{\rm{F}}}{{{\rm{CH}}}}_{4}}}_{{{\rm{inundation}}}}^{{{\rm{his}}}}$$

(14)

where “period” refers to one of the three future time intervals, and the overbars (e.g., ${\overline{{{\rm{F}}}{{{\rm{CH}}}}_{4}}}$) indicate the mean CH₄ flux over the corresponding time period. The differences represent the changes in net CH₄ flux in future periods relative to the historical mean under the default, fixed-atmospheric-CH₄, and fixed-wetland-extent scenarios. Based on flux changes, the individual contributions of each driver in a given future period were quantified as:

$${\Delta }_{{{\rm{atmch}}}4}^{{{\rm{period}}}}=\Delta {\overline{{{\rm{F}}}{{{\rm{CH}}}}_{4}}}_{{{\rm{default}}}}^{{{\rm{period}}}}-{\Delta \overline{{{\rm{F}}}{{{\rm{CH}}}}_{4}}}_{{{\rm{atmch}}}4}^{{{\rm{period}}}}$$

(15)

$${\Delta }_{{{\rm{inundation}}}}^{{{\rm{period}}}}=\Delta {\overline{{{\rm{F}}}{{{\rm{CH}}}}_{4}}}_{{{\rm{default}}}}^{{{\rm{period}}}}-{\Delta \overline{{{\rm{F}}}{{{\rm{CH}}}}_{4}}}_{{{\rm{inundation}}}}^{{{\rm{period}}}}$$

(16)

$${\Delta }_{{{\rm{climate}}}}^{{{\rm{period}}}}=\Delta {\overline{{{\rm{F}}}{{{\rm{CH}}}}_{4}}}_{{{\rm{default}}}}^{{{\rm{period}}}}-{\Delta }_{{{\rm{atmch}}}4}^{{{\rm{period}}}}-{\Delta }_{{{\rm{inundation}}}}^{{{\rm{period}}}}$$

(17)

where ${\Delta }_{{{\rm{atmch}}}4}^{{{\rm{period}}}}$, ${\Delta }_{{{\rm{inundation}}}}^{{{\rm{period}}}}$ and ${\Delta }_{{{\rm{climate}}}}^{{{\rm{period}}}}$ represent the individual contributions of atmospheric methane concentration, inundation extent, and climate factors, respectively, to changes in methane flux during a given future period.

Statistical analyses

To analyze the relationships between simulated CH₄ fluxes and soil properties under various environmental changes, we calculated Pearson correlation coefficients between monthly CH₄ fluxes and soil properties (soil temperature and moisture). Additionally, to further explore the response of CH₄ emissions and sinks to changes in top-soil temperature (0−10 cm), we employed an exponential equation to fit variations in CH₄ emission and sink rates with soil temperature.

$$R=A{e}^{({BT})}$$

(18)

where $R$ denotes CH₄ emission or sink rates (mg CH₄ m^-2 d^-1), $T$ is soil temperature (0−10 cm), and $A$ and $B$ are parameters. The temperature sensitivity (${Q}_{10}$) value was further calculated based on the obtained parameter B.

$${Q}_{10}={e}^{(10B)}$$

(19)

Finally, we calculated the coefficient of variation based on CH₄ fluxes from the CLM5.0 model driven by climate forcings from different GCMs to quantify the spatiotemporal uncertainty.

$${C}_{v}=\frac{\sigma }{\mu }$$

(20)

where ${C}_{v}$ denotes the coefficient of variation, $\sigma$ is the standard deviation, and $\mu$ represents the average of the simulated CH₄ fluxes using the CLM5.0 model driven by the four GCM outputs.