Introduction

Lakes are active hotspots critical for watershed-scale carbon transport, transformation, and storage, as well as aquatic-atmospheric carbon (CO2 and CH4) exchange1,2,3,4. Dilute unproductive lakes, particularly small lentic freshwaters5, typically function as net carbon sources, due to allochthonous carbon inputs driving emissions to exceed fixation by photosynthesis2,6,7. In contrast, solute-rich alkaline lakes with elevated pH and salinity can sequester atmospheric CO28,9 and suppress CH4 emissions10,11. This occurs because high pH converts dissolved CO2 to bicarbonate (HCO3−) and carbonate (CO32−) ions12, as well as high salinity inhibits methanogen abundance and enhances sulfate-mediated anaerobic CH4 oxidation13,14. Despite progress in understanding carbon exchange in dilute unproductive lakes, data scarcity from remote alkaline lakes impedes robust quantification.

The Qinghai Tibet Plateau (QTP), known as Earth’s Third Pole15, encompasses plenty of alpine alkaline lakes16, accounting for roughly half of China’s total lake area17. Infrequent manual sampling on large alpine lakes of the QTP has produced contrasting estimates of CO2 exchange, with some studies reporting both source and sink behavior. For example, relying on low-frequency manual sampling, Jia et al. 18 and Xiao et al. 19 reported pronounced temporal variability in CO2 exchange over the past two decades, ranging from 1.16 to 6.87 Tg C yr−1 and from 0.34 to −0.34 Tg C yr−1, respectively. In contrast, eddy-covariance measurements consistently indicate that these large alpine alkaline lakes act as annual atmospheric CO2 sinks and display strong seasonal variability20,21, a pattern likely influenced by the prolonged observation period that constrains invasive CO2 fluxes. Additionally, smaller lakes generally emit CO2 and exhibit stronger CH4 emissions4,11,22,23, due to greater terrestrial carbon loads per unit volume5. These recent studies have improved our general understanding of carbon flux in alpine lakes. However, comprehensive assessments of annual carbon exchange across diverse alpine lakes remain lacking, impeding accurate regional carbon budgeting24.

As one of the most climate-sensitive regions, the alpine lakes on the QTP are highly exposed to climate change15,25. In contrast to widespread declines in global arid/semi-arid lakes due to climate warming and human activities26, QTP alpine lakes are expanding rapidly under a wetter and warmer climate25,27,28 and attracted worldwide attention29,30,31. Furthermore, projections indicate over 50% increase in alpine lake area by 2100 in response to climate change, potentially causing over 20,000 km2 of alpine grasslands to be submerged32. Thus, projecting future changes in carbon exchange for these expanding alpine lakes and evaluating their impacts on regional carbon budgets is essential29. This study aimed to quantify the carbon exchange between alpine lakes and the atmosphere over the last decade, project their future trends by 2100 under Shared Socioeconomic Pathways (SSP) scenarios, and further assess the impact of lake expansion on the regional carbon budget (Fig. 1a).

Fig. 1: Overview of the study design, study area, and alpine lake water chemistry.
Fig. 1: Overview of the study design, study area, and alpine lake water chemistry.
Full size image

a Flow diagram of the study design. b Study area and CO2 and CH4 fluxes alpine small (<0.1 km2), midsize (0.1–10 km2) and large lake (>10 km2) sites. c and d Measurements of water chemistry parameters of different lake size classes during the open-water period.

Results

Areal CO2 and CH4 flux rates in the QTP alpine lakes

From 2011 to 2022, the average annual CO2 fluxes were 122.2 g CO2 m−2 yr−1 for small lakes (<0.1 km2) and 25.3 g CO2 m−2 yr−1 for midsize lakes (0.1–10 km2) (Table 1). Approximately 1% of annual CO2 emissions occurred during the ice-melt period. Water pH showed a strong negative correlation with CO2 flux in our sampling lakes, both annually and during the ice-melt period (Supplementary Fig. 1c). Based on hydrological classification (i.e., closed, upstream, and outflow lakes) and eddy covariance measurements in alpine large lakes, we assessed their average annual CO2 flux at −611.3 g CO2 m−2 yr−1. Meanwhile, applying the pH-CO2 flux equation for global large lakes with pH > 8 (CO2 flux = −1026 pH + 8935, r2 = 0.80, p < 0.01; Supplementary Fig. 2) and the median pH of different hydrological types of alpine large lakes (Supplementary Fig. 3e), we estimated the average annual CO2 flux of large lakes to be −557.2 g m−2 CO2 yr−1. The estimates from the two methods are close, and their average (−584.2 g m−2 CO2  yr−1) was considered the annual CO2 flux in alpine large lakes, indicating that lakes act as a CO2 sink. Among large lakes of different hydrological types, closed lakes exhibited higher CO2 uptake rates than outflow lakes (Supplementary Fig. 4a).

Table 1 Average alpine lake surface area and its proportion over the last decade, annual CO2 and CH4 fluxes, and total carbon exchange for alpine lakes on the QTP
Full size table

The average CH4 emission rates from small and midsize lakes were 10.4 and 1.1 g CH4 m−2 yr−1, respectively (Table 1). Marked spatial variability was observed in annual CH4 flux and the contribution of CH4 flux during ice melt to annual emissions across alpine grasslands (Supplementary Fig. 1d, e). By combining our measurements with compiled CH4 flux data from previous studies and hydrological classification, we found that alpine large lakes had a lower average annual CH4 emission rate of 0.3 g CH4 m−2 yr−1, and outflow lakes exhibited relative higher CH4 emission rate (1.0 g CH4 m−2 yr−1; Supplementary Fig. 4b).

Annual net CO2 and CH4 exchanges for the QTP alpine lakes over the last decade

Over the last decade, alpine lakes were estimated to have a net CO2 exchange of −6.7 Tg C yr−1 (−10.8 to −2.5 Tg C yr−1, 95% Confidence Interval, hereinafter as 95% CI). The estimates fall within the range of previous studies20,21, with values from −10.3 Tg C yr−1 to between −3.0 and −1.9 Tg C yr−1. Small and midsize lakes acted as net carbon sources (0.05 and 0.03 Tg C yr−1, respectively), while large lakes served as a strong net CO2 sink of 6.8 Tg C yr−1 (Fig. 2a).

Fig. 2: Lake carbon and CO2 equivalent (CO2-eq) fluxes on GWP100 over the last decade.
Fig. 2: Lake carbon and CO2 equivalent (CO2-eq) fluxes on GWP100 over the last decade.
Full size image

The error bars in panels a and b indicate the 95% confidence interval.

Alpine lakes constituted a CH4 source of 0.03 Tg C yr−1 (95% CI: 0.02–0.07 Tg C yr−1) over the last decade. Covering 12% of the total alpine lake area, small and mid-sized lakes account for 60% of CH4 emissions. Integrated CO2 and CH4 exchanges, the net carbon exchange in alpine lakes was −6.7 Tg C yr−1 (95% CI: −10.8 to −2.4 Tg C yr−1), indicating alpine lakes on the QTP acted as a net carbon sink over the last decade (Table 1). Although small and midsize lakes functioned as carbon sources, their emissions counteracted only ~2% of the substantial carbon sequestration achieved by large lakes (Fig. 2a). Alpine lakes CH4 emissions reduced the net climate benefit of CO2 sequestration by merely 0.4% in carbon mass balance terms and by 4% using the CO2 equivalent (CO2-eq) flux on100-year global warming potential (GWP100) (Fig. 2). Overall, CO2 sequestration in large lakes determined the magnitude and direction of net carbon exchange (Fig. 2).

Impacts of future lake expansion on carbon exchanges

In 2100, under SSP scenarios, higher water temperatures are projected to reduce average annual CO2 uptake rates per unit area in alpine lakes by 5–33%, despite an extended open-water duration. However, lake expansion is projected to increase the CO2 sink of alpine lakes by from 46% (SSP1-2.6) to 9% (SSP5-8.5), with values ranging from 7.3 to 9.8  Tg C yr−1 in 2100 (Table 2). Higher water temperatures are also projected to increase annual CH4 emission rates from alpine lakes by 46–135%, with total emissions rising by 122–279% in 2100 (Table 2). Overall, the carbon sink of alpine lakes is expected to grow by 8–46%, reaching 7.2–9.7 Tg C yr−1 in 2100 (Fig. 3b and Table 2). However, when using CO2-eq fluxes on GWP100, the carbon sink of alpine lakes will increase by 43% and 25% in 2100 under SSP1-2.6 and SSP2-4.5, respectively, but decrease by 2% under SSP5-8.5 (Fig. 3c).

Fig. 3: Projected carbon exchanges of expanding Qinghai-Tibet Plateau lakes by 2100.
Fig. 3: Projected carbon exchanges of expanding Qinghai-Tibet Plateau lakes by 2100.
Full size image

Projected changes in lake area (a), carbon exchange (b), CO2-eq fluxes on GWP100 (c), and the impact of lake expansion and water temperature changes on the regional carbon sink through 2100 (d–f). The initial divergence in panels b, c, and f is primarily attributed to differing lake surface warming rates across SSP scenarios during the 2012–2022 to 2024 initialization period.

Table 2 Projected lake area and carbon exchanges in 2100 under SSP scenarios
Full size table

The projected expansion of alpine lakes will submerge over 25,000 km2 of grasslands (Fig. 3a), potentially altering the regional carbon budget. Future variations in net ecosystem productivity (NEP) and CH4 fluxes in alpine grasslands must be evaluated to assess changes in the regional carbon sink by lake expansion. NEP in alpine grasslands is projected to increase by 1.35–3.42 times under SSP scenarios by 2100, except in swamp meadow zones, where NEP is expected to decline under SSP5-8.5 (Supplementary Table 1). Meanwhile, process-based ecosystem models predicted a 20–154% increase in annual CH4 emission rates from alpine swamp meadows, and a 21–63% increase in CH4 uptake rates for alpine upland grasslands (Supplementary Table 2)33,34,35. In the absence of land cover changes, the projected carbon sink in the inundated zone would range from 2.3 to 2.9 Tg C yr−1 in 2100 across SSP scenarios (Fig. 3d). Furthermore, by comparing future alpine lake carbon exchange to baseline carbon exchange in projected inundation zones under scenarios without land cover changes, we quantified the impact of lake expansion on the regional carbon sink. Under the SSP1-2.6 scenario, lake expansion is projected to enhance the regional carbon sink through 2100, with an annual increase of 0.6 Tg C in 2100 (Fig. 3e). Conversely, under SSP2-4.5 and SSP5-8.5 scenarios, the regional carbon sink is expected to increase initially but declines after 2080 and 2035, respectively, yielding annual reductions of 0.3 and 2.3 Tg C in 2100 (Fig. 3e). By 2100, lake expansion is projected to drive a cumulative increase in the regional carbon sink of 54.2 and 12.2 Tg C under the SSP1-2.6 and SSP2-4.5 scenarios, respectively, but a total decrease of 61.4 Tg C under SSP5-8.5 (Fig. 3f).

Discussion

Carbon exchange for the QTP alpine lakes over the last decade

Our study demonstrates that alpine lakes on the QTP acted as a net carbon sink, sequestering 6.7 Tg C yr−1, in contrast to the net carbon emissions observed in most global freshwater lakes2,3,6. This difference is primarily due to the strong CO2 sequestration of alpine large lakes, which constitute the majority of alpine lake areas20,21. Specifically, alpine large lakes receive lower loads of terrestrial carbon relative to water volume, owing to their low perimeter to surface area ratios and greater depth5. In Nam Co, the third largest alpine lake on the QTP, annual external riverine carbon inputs account for <1% of its total lake carbon pool36. Consequently, carbon fluxes are predominantly driven by internal biogeochemical processes in these alpine large lakes rather than allochthonous input. Biological factors likely have a limited influence on lake CO2 dynamics, as most alpine large lakes have low nutrient status16. Instead, chemical characteristics could strongly influence carbon dynamics in large lakes. Higher water pH in large lakes leads to a predominance of dissolved inorganic carbon (DIC) in the form of bicarbonate (HCO3−) and carbonate ions (CO32−), rather than dissolved CO28,9. This results in water surface dissolved CO2 concentration in large lakes being lower than atmospheric CO2 levels37,38, contributing to net CO2 sequestration. Among different hydrological large lakes, large closed lakes as hydrological terminals accumulate large mineral ions and salts, coupled with strong evaporation, leading to higher water pH compared to outflow lakes. Meanwhile, outflow lakes receive more terrestrial labile carbon inputs, which can enhance CO2 decomposition. Thus, both eddy covariance and manual measurements have found that closed lakes have higher annual CO2 uptake rates than outflow lakes (Supplementary Fig. 5). Given that large closed lakes represent over 80% of the area of alpine large lakes, large lakes serve as a strong CO2 sink.

In contrast to large lakes, small and midsize lakes acted as CO2 sources (Table 1), primarily due to their higher terrestrial carbon inputs relative to water volume5. The decomposition and conversion of allochthonous carbon could outweigh the negative effects of high water pH on dissolved CO2 concentrations. Globally, alpine small and midsize lakes exhibit lower annual CO2 emission rates than lakes in other regions with lower water pH (Supplementary Fig. 6). Additionally, most annual CO2 emissions from small and midsize lakes occurred during the open-water period (Supplementary Fig. 1b). This pattern likely results from increased allochthonous carbon input from soil to the aquatic systems39,40 (Supplementary Fig. 7), CaCO3 precipitation during groundwater recharge41,42, and rising water temperature. However, during winter, external organic and inorganic carbon inputs are reduced due to soil freezing, and thus higher water pH promotes the conversion of dissolved CO2 to HCO3− and CO32− under ice8,43. Consequently, only about 1% of annual CO2 emissions occurred during the ice-melt period (Supplementary Fig. 1b). Given their smaller area and lower CO2 emission rates, small and midsize lakes offset only about 1% of the net CO2 sequestration in alpine large lakes (Fig. 2), indicating alpine lakes have collectively acted as a CO2 sink over the last decade.

Over the last decade, the average annual CH4 emission rate from alpine lakes was 0.7 g CH4 m−2 yr−1, amounting to only 8% CH4 flux from high-latitude lakes3. This is primarily due to the lower coverage of small lakes (3%) and midsize lakes (9%) with higher CH4 emission rates in the QTP alpine region compared to high latitude regions (17% and 38% for small and midsize lakes, respectively)44. Additionally, the QTP alpine regions exhibit lower soil organic carbon density45, resulting in less terrestrial carbon input into lakes46. Among different-sized alpine lakes, small lakes show higher annual CH4 emission rates than midsize and large lakes (Table 1), due to their shallow depths and rich carbon substrates11,47. For small and midsize lakes, spatial variations in carbon substrates influence the quantity of mcrA and the ratio of mcrA to pmoA11, leading to a decline in CH4 emission rates from swamp meadow lakes, meadow lakes, steppe zone lakes, to desert lakes (Supplementary Fig. 1d). Large outflow lakes, with major rivers running, receive higher terrestrial labile carbon inputs and exhibit higher CH4 emission rates than those from closed lakes11 (Supplementary Fig. 4b). Collectively, alpine lakes act as a weak CH4 source and their CH4 emissions offsetting the minor net CO2 sequestrations in alpine lakes (Fig. 2). Thus, alpine lakes in the QTP serve as a carbon sink (6.7 Tg C yr−1) over the last decade, predominantly driven CO2 uptake in large lakes (Fig. 2).

Future changes in carbon exchange for the QTP alpine lakes

Based on projected future lake areas, water temperatures, and ice cover duration through 2100 under SSP scenarios, we predict future changes in carbon exchange for alpine lakes and their influence on the regional carbon budget (Fig. 3). Rising water temperatures are expected to reduce CO2 solubility and enhance carbon decomposition48,49, thereby diminishing net CO2 uptake rates during the open-water period in alpine large lakes20,21. In these large lakes, the highest monthly CO2 uptake typically occurs during the freezing months due to decreased water temperature, increased salinity, and higher pH20,21, while the seasonal ice barrier can reduce CO2 exchange to nearly zero during the frozen period21. Although warmer climates are projected to shorten the completely frozen duration, the freezing period would persist for alpine lakes by 2100 under SSP scenarios. Thus, reduced annual areal CO2 sequestration in alpine large lakes will primarily result from higher temperatures during the open water months. The decline in the average annual areal CO2 sequestration rate in alpine large lakes is projected to be modest under SSP1-2.6 (1%), more considerable under SSP2-4.5 (13%), and highest under SSP5-8.5 (31%) in 2100. However, the projected expansion of large lake areas (more than 50%) could offset the negative effects of rising water temperatures on net CO2 sequestration in 2100. Compared to the last decade, we project that net CO2 sequestration in large lakes will increase of 47%, 31%, and 10% by 2100 under SSP1-2.6, SSP2-4.5, and SSP5-8.5 scenarios, respectively (Table 2). Additionally, the expansion of small and midsize lakes is expected to increase their CO2 emissions by 1.6 and 0.6 times, respectively (Table 2). The increased CO2 emissions from small and midsize lakes will be outweighed by the predicted rise in net CO2 sequestration in large lakes, as large lakes dominate the total alpine lake area (Table 2). Therefore, the net lake CO2 sink is expected to exceed its contemporary level by 2100 under all SSP scenarios.

Water temperature positively influences CH4 emissions from lakes by promoting organic carbon decomposition50,51,52. Projected increases in water temperature, longer open water duration, and expanding lake areas are expected to elevate CH4 emissions from alpine lakes in 2100, especially under the high-warming SSP5-8.5 scenario53. CH4 emissions from alpine lakes are expected to peak at 0.11 Tg C yr−1 in 2100 under SSP5-8.5, followed by SSP2-4.5 and SSP1-2.6 scenarios (Table 2). Integrated predicted CO2 and CH4 exchanges, the carbon sink of alpine lakes is projected to increase in 2100 (Fig. 3b). This is primarily attributed to greater net CO2 uptake by expanded large lakes, which outweighs the rise in CH4 emissions. Under SSP5-8.5, the carbon sink of alpine lakes is projected to grow minimally (8%) by 2100, with a smaller increase in CO2 sequestration and a larger increase in CH4 emissions. Additionally, the CO2-eq fluxes on GWP100 are expected to decrease by 2% under SSP5-8.5 due to predicted higher CH4 emissions from alpine lakes (GWP100 of CH4 is 27). In contrast, under SSP1-2.6, both the net carbon sink and CO2-eq fluxes of alpine lakes are expected to increase more appreciably in 2100 (Fig. 3b, c).

Impacts of projected lake expansion on the regional carbon budget

The projected expansion of alpine lakes by 2100, which is expected to submerge surrounding grasslands32, necessitates understanding changes in alpine grassland carbon fluxes to assess impacts on the regional carbon budget. Alpine grasslands functioned as a net carbon sink24,54, though with substantial variability across grassland zones (Supplementary Fig. 8a). Alpine steppe and desert zones, comprising ~80% of the submerged area (Supplementary Fig. 8b), have lower net carbon sequestration rates than the lake average over the last decade (Supplementary Fig. 8a). Meanwhile, alpine grasslands’ future carbon sequestration capacity is expected to increase under a warmer and wetter climate35,55, thus narrowing the gap with lakes (Supplementary Fig. 8a). The future enhanced carbon sink capacity of alpine grasslands is primarily driven by increased temperature and precipitation, the primary climatic constraints on CO2 sequestration in these ecosystems54,55,56; however, under the SSP5-8.5 scenario, extreme warming may reduce NEP in high soil organic carbon (SOC) density swamp meadow ecosystems (Supplementary Table 1), potentially due to accelerated SOC decomposition. Future CH4 uptake in alpine upland grasslands is projected to increase35, driven primarily by warming, with precipitation increases unlikely to alter the warming effects57. Thus, under the SSP2-4.5 and SSP5-8.5 scenarios, annual regional carbon sink dynamics associated with lake expansion are projected to first increase and then decline (Fig. 3e), a pattern which is driven by a rise in the areal grassland carbon sequestration rate and a substantial decrease in the lake carbon sequestration rate. In contrast, under SSP1-2.6, the slower warming rate leads to a smaller reduction in the areal lake carbon sequestration rate, enabling lake expansion can continue to augment the regional carbon sink through 2100 (Fig. 3b, d).

This study enhances the accuracy of alpine lake carbon exchange assessments over the past decade and provides initial projections for their future changes, and further evaluates lake expansion impacts on regional carbon budgets. However, several factors may constrain the evaluation, particularly for projections of future alpine lake carbon exchange. Specifically, limited continuous carbon flux measurements from midsize and large alpine lakes could increase uncertainty in carbon exchange estimates. Climate change cascading through aquatic and terrestrial ecosystems could directly and indirectly alter lake carbon exchange dynamics. Increased temperatures and precipitation can alter lake water pH, salinity, and carbon inputs, potentially influencing lake carbon dynamics. The vulnerability of permafrost and the mobilization of labile carbon following thaw could further affect lake carbon exchanges. Additionally, in-phase PDO and El Niño events can alter carbon flux patterns by influencing temperature and precipitation58. However, quantifying these future changes and further clarifying their effects on lake carbon exchanges remains challenging. Given the difficulty in projecting future carbon flux changes, we excluded 95% CIs for future alpine lake carbon exchanges. By leveraging the growing availability of measurement data, improving our understanding of carbon exchange mechanisms in alpine lakes, and advancing ESMs, future studies can employ process-based lake biogeochemical models to project variations in carbon exchanges in alpine lakes.

Our study demonstrates that alpine lakes acted as a carbon sink over the last decade (Fig. 4), sequestering 6.7 Tg C yr−1 (95% CI: 2.4–10.8 Tg C yr−1). CO2 sequestration in alpine large lakes determined the magnitude and direction of carbon exchange between alpine lakes and the atmosphere. While small and midsize lakes acted as carbon sources, their contribution to total carbon exchanges in alpine lakes was limited. Even though climate warming is projected to reduce the mean carbon sequestration rate of alpine large alkaline lakes and increase carbon emissions from small lakes, the anticipated expansion of large lake areas would augment lakes’ total carbon sink (Fig. 4). Due to lake expansion, the regional carbon sink is projected to increase continuously through 2100 under the SSP1-2.6, whereas it is expected to decline under the SSP5-8.5, as higher warming rates substantially reduce the carbon sequestration rate of large alkaline lakes. Therefore, lake carbon exchange must be integrated into QTP carbon budget assessments to better constrain and predict future conditions in this climate-sensitive area.

Fig. 4: Diagram illustrating the carbon exchange of alpine lakes over the last decade and 2100 under the SSP2-4.5 scenario, as well as the impacts of lake expansion on the regional carbon sink.
Fig. 4: Diagram illustrating the carbon exchange of alpine lakes over the last decade and 2100 under the SSP2-4.5 scenario, as well as the impacts of lake expansion on the regional carbon sink.
Full size image

The negative and positive fluxes denote a carbon sink and a carbon source, respectively.

Methods

Study region

The QTP is the highest plateau in the world, with an average elevation exceeding 4000 m above sea level. Precipitation on the QTP is primarily concentrated from June to September in the summer. Across the QTP alpine regions, swamp meadows, meadows, steppes, and deserts are distributed from east to west (Fig. 1b). Lakes on the QTP cover about 50,000 km2, primarily comprising alpine lakes typically surrounded by alpine grasslands at elevations above 3190 m. Most alpine lakes are located in the inner basin of the QTP, surrounded by alpine desert and steppe (Fig. 1b). Meanwhile, lakes larger than 10 km2 account for ~88% of the total alpine lake area on the QTP (Table 1). The measured alpine lakes are generally alkaline, with 71% large lakes, 67% midsize lakes, and 27% small lakes exhibiting pH values above 9 (Supplementary Fig. 3). The elevated water pH in these lakes primarily results from the extensive distribution of carbonate bedrock and strong evaporation59,60.

Data sources

The data for our analyses were derived from field observations, obtained either by our team or from previously published literature. Given the impact of lake size on carbon fluxes5,61,62, we first categorized alpine lakes into three size groups in this study: small lakes (<0.1 km2), midsize lakes (0.1–10 km2), and large lakes (>10 km2). We selected 25 small lakes and 4 midsize lakes across alpine grasslands during the ice-free and ice-melt periods to measure their annual CO2 and CH4 fluxes (Supplementary Fig. 9), with these lakes representing perennial small and midsize lakes on the QTP, ranging in area from 0.0006 to 2.26 km2 (Supplementary Fig. 10 and Supplementary Table 3). The ice-covered period for these sampling alpine lakes is typically from late October to April or May of the following year. The maximum freezing depth of the sampling lakes ranges from approximately 70 to 100 cm, typically shallower than their maximum depths. Additionally, we collected diffusive CH4 flux in 14 alpine large lakes in June 2022 and October 2023 (Supplementary Fig. 9).

Dissolved CO2 and CH4 concentrations were sampled from shore to center in each lake during three periods: (1) the open-water period (May–October 2021 and 2022), (2) the pre-ice melt period (April to early May 2022, before 1 week of ice melt), and (3) the ice-melt period (late April to May 2022, during ice melt to after 1 week of ice melt), using the headspace equilibration method11. CO2 and CH4 diffusive fluxes between the lakes and atmosphere during the open-water period were estimated using Fick’s first law of diffusion combined with a wind-based model to calculate piston velocity63,64,65,66, as detailed in the Supplementary Text. Approximately one week before and after lake ice melt in April or May, we collected water samples using a tube equipped with a rubber sleeve, a three-way stopcock, and a 100 mL syringe to measure dissolved CO2 and CH4 concentrations. These samples were taken at depths of 0.4, 2 m, and/or at the sediment surface to estimate the CO2 and CH4 storage variations before and after ice melt, representing the CO2 and CH4 fluxes during the ice-melt period. We determined the monthly CH4 ebullitive fluxes during the open-water period using bubble traps11,67. Three to six traps were deployed in each sampling lake for 24–72 h, and the accumulated gas was manually collected using a syringe and a gas-sampling foil bag, with the sampling time and gas volume recorded. The dissolved and ebullitive carbon concentrations were analyzed using a gas chromatograph (GC-7890B, Agilent, USA) and determined with Henry’s law (Supplementary Methods). We also measured water temperature, dissolved oxygen, pH, salinity, and electric conductivity in the water using a portable multi-parameter meter (HQ40D, Hach, USA) during each measurement.

To evaluate carbon exchange in QTP alpine lakes, we further compiled CO2 and CH4 flux data from 80 and 111 lakes, respectively (Fig. 1b and Supplementary Table 4). The compiled alpine lakes with measured carbon fluxes range from 0.00004 to 4432 km2 across different vegetation zones (Fig. 1b)11. For compiled small lakes, our dataset includes only those with carbon flux measurements taken at least three times during the ice-free period to ensure accuracy. Due to limited observations for midsize lakes, we also included flux data from Xun et al. 68 despite single measurements during the ice-free period. Given the pronounced seasonal variability in CO2 flux from large lakes20,21, manual sampling with low frequencies cannot adequately characterize their temporal patterns. For instance, previous studies using manual sampling to measure CO2 fluxes in Qinghai Lake across different months reported highly divergent results (Supplementary Fig. 11). Therefore, we focused on measuring CO2 flux data from large lakes using eddy covariance, which provides continuous flux data throughout the year. Meanwhile, we measured and compiled water pH data from 219 alpine lakes and water salinity data from 206 alpine lakes. We also collected carbon exchange data for alpine grasslands from previous studies to understand their carbon exchange patterns and assess the impact of future alpine grassland inundation by lake expansion on the regional carbon budget (Supplementary Tables 5 and 6).

Estimates of lake carbon exchange for the last decade

Carbon flux data in alpine lakes were measured between 2011 and 2022, reflecting conditions over the last decade, during which water temperature and ice phenology exhibited minimal variability69,70. Total carbon exchanges (TCE) were calculated by multiplying the areas of different-sized lakes by their respective annual carbon fluxes.

$${{{\rm{T}}}}{{{\rm{C}}}}{{{\rm{E}}}}={\Sigma }_{i}\,{A}_{i}\times {F}_{i}={A}_{{{{\rm{S}}}}}\times {F}_{{{{\rm{S}}}}}+{A}_{{{{\rm{M}}}}}\times {F}_{{{{\rm{M}}}}}+{A}_{{{{\rm{L}}}}}\times {F}_{{{{\rm{L}}}}}$$
(1)

where Ai is the surface area of alpine lakes in size class i (m2), and Fi is the carbon flux (g CO2 m−2 yr−1, g CH4 m−2 yr−1, or g C m−2 yr−1) for the corresponding class. Subscripts denote lake size classes: small (S), midsize (M), and large (L).

We determined alpine lake areas in the QTP to calculate carbon exchange over the last decade. Pi et al. 71 reported global lake dynamics for lakes larger than 0.03 km2 (GLAKES) in three periods (1985–1990s, 2000s, and 2010s). Given measurements of carbon flux in alpine large lakes were observed during the 2010s, we selected the probability-weighted alpine lake areas larger than 10 km2 in the QTP during the 2010s, covering 42,457 km2. Xu and Zhang72 provided a dataset of lakes larger than 0.01 km2 on the Tibetan Plateau in 2023. Using the area proportion of QTP alpine lakes ranging from 0.01 to 0.03 km2 in 2023, we estimated their area for the 2010s to be 292.5 km2. In addition, Wei et al. 73 provided lake data for the QTP permafrost region, covering areas from 0.0005 to 3 km2 in 2020. We further used the proportion of lakes between 0.0005 and 0.01 km2 to estimate their corresponding areas for alpine lakes in the 2010s and 2023. Carbon fluxes for small and midsize lakes were measured from 2012 to 2022 (Supplementary Table 4), and their average areas in the 2010s and 2023 were used for analysis. Finally, the average area of alpine lakes larger than 500 m2 on the QTP over the last decade is 48,435 km2, comprising 1589 km2 of small lakes, 4390 km2 of midsize lakes, and 42,457 km2 of large lakes (Table 1). Lakes smaller than 0.0005 km2 are excluded from this analysis due to their negligible contribution to the total lake area and challenges in identification73,74.

The surrounding vegetation cover of smaller alpine lakes can indirectly affect their carbon fluxes by altering water pH and carbon input11,22. We further classified the vegetation zones of small and midsize lakes by merging two vegetation distribution maps75,76. We employed a bootstrapping method to upscale regional carbon fluxes for small lakes during the ice-free period. This non-parametric method is a type of Monte Carlo Simulation approach widely applied for upscaling1,4. We used the medians of each resampling and iterated the resampling 50 times to generate a set of statistics. Subsequently, we multiplied the bootstrapped lake carbon fluxes for each vegetation zone by the corresponding waterbody areas and the average ice-free duration of 168 days as observed. We derived the annual carbon exchanges in small lakes by incorporating contributions from the ice-melt period. We estimated annual CO2 exchanges for midsize lakes using the median fluxes in different vegetation zones and their respective lake areas.

The hydrological network determines water pH and salinity in alpine large lakes (Supplementary Fig. 3e, f). Given the impact of water pH on CO2 exchange in alpine large lakes20,21, we classified the hydrological networks of large lakes using data from Gao et al. 77 and visual interpretation of Google images (Supplementary Fig. 12). We then used the average annual CO2 flux data measured by eddy covariance across alpine large lakes with different hydrological networks to estimate their CO2 exchanges. The eddy covariance method has been used to measure CO2 fluxes in four alpine large lakes (Qinghai Lake, Siling Co, Nam Co, and Ngoring Lake), providing insights into the CO2 flux mechanisms of these representative alpine large alkaline lakes. CO2 flux data from the four alpine lakes also align closely with those from global large lakes (Supplementary Fig. 6). Thus, we conclude that combining eddy covariance with hydrological classification offers an accurate method for evaluating CO2 exchange in alpine large lakes. Although CO2 fluxes were not measured in the alpine large upstream lakes, we inferred that their annual CO2 fluxes are similar to those of large closed lakes, as both have a pH above 9 (Supplementary Fig. 3e). Additionally, we established a flux-pH equation derived from global large lakes with pH greater than 8 (Supplementary Fig. 2). Using this equation and the pH values of different hydrological alpine lakes, we calculated their annual CO2 exchanges. The hydrological network also influences CH4 emission rates from alpine large lakes by altering the characteristics of water dissolved organic matter11, prompting us to evaluate CH4 emissions using network classification. Using available CH4 flux data from different hydrological lakes, we applied the bootstrapping method to upscale and calculate CH4 emissions from alpine large lakes.

Projections of future carbon exchange changes

Projections of future alpine lake area changes

We categorized lakes into two classes based on area (> or <1 km2) to predict future changes in lake area by 2100. First, Xu et al. 32 projected area changes for closed alpine lakes larger than 1 km2 on the Tibetan Plateau by 2100 under SSP scenarios. These alpine closed lakes, acting as hydrological terminals, have driven ~90% of the total area increase in lakes larger than 1 km2 from 2000 to 202372. Thus, we assumed that the expansion of alpine lakes larger than 1 km2 and grassland submersion can be represented by the findings of Xu et al. 32.

Using the GLAKES dataset71, we analyzed area changes in alpine lakes smaller than 1 km2 from the 2000s to the 2010s across alpine regions with varying permafrost stability (Supplementary Fig. 13). Our findings revealed higher lake expansion rates in semi-stable, transitional, and unstable permafrost regions (Supplementary Fig. 13), despite these regions having a lower precipitation-to-evapotranspiration ratio compared to stable and very stable permafrost regions, as well as seasonal frost areas (Supplementary Fig. 14). This highlights the dominant role of permafrost stability in the dynamics of smaller alpine lakes (<1 km2), consistent with patterns observed in high-latitude regions78,79,80. Precipitation projections for the Tibetan Plateau (QTP) indicate a continuous increase, while thermodynamical and dynamical components still cause uncertainty in the projection81. Given the uncertainty in precipitation and the dominant influence of permafrost stability on smaller lake expansion, we focused on the impact of permafrost stability shifts on smaller lake area changes by 2100 under SSP scenarios.

Permafrost stability changes by 2100 were projected using mean annual land surface temperature (MAST), a reliable indicator for evaluating permafrost distribution patterns82. Using current MAST spatial patterns for the QTP83 and its future changes84, we projected future permafrost area changes (Supplementary Fig. 15). Notably, we grouped semi-stable, transitional, and unstable permafrost regions into a single “active permafrost” category, representing the thermal stability of permafrost in these zones conducive to active lake expansion (Supplementary Fig. 13). We also combined very stable and stable permafrost regions, where higher permafrost stability restricts lateral lake expansion. Consequently, we classified the QTP alpine region into three categories: very stable and stable permafrost, active permafrost, and seasonal frost regions, and projected their future spatial distribution (Supplementary Fig. 15). The projected permafrost loss by 2100 is comparable with the findings of Zhang et al. 85 and Li et al. 86, confirming the reliability of our future permafrost change projections. Furthermore, we projected future changes in smaller lakes based on permafrost change projections and decadal lake area change rates within each alpine region from the 2000s to the 2010s. Detailed methods for projecting alpine lake areas are provided in the Supplementary Text.

Projections of future changes in the carbon fluxes of alpine lakes

Several previous studies have found a positive relationship between temperature and organic carbon mineralization48,87, while others reported a weak relationship between water temperature and lake dissolved CO2 concentration88,89. Kuhn et al. 90 observed a shift from CO2 source to sink as lakes transition from north permafrost regions to warmer areas. Additionally, CO2 efflux from Canadian alkaline lakes could decrease with increasing atmospheric warming and shortened ice cover duration8. In alpine small and midsize lakes, we only found a weak positive correlation between water temperature and CO2 fluxes in desert lakes, but no significant relationship in lakes from swamp meadow, meadow, and steppe zones (Supplementary Fig. 16). This suggests water temperature may not be the primary driver of CO2 flux in alpine small and midsize lakes. Although water pH and allochthonous carbon inputs can influence lake CO2 flux23,39, projecting their future changes in alpine permafrost regions remains challenging. Moreover, given the small surface area of alpine small and midsize lakes (12% of total lake area) and their relatively low CO2 emission rates (Supplementary Fig. 6), changes in their CO2 fluxes are expected to have minimal impact on the future total lake and regional carbon budgets. We further assumed that annual CO2 fluxes between alpine small and midsize lakes and the atmosphere would remain comparable to current levels. CH4 efflux from alpine lakes is strongly influenced by water temperature4,11. However, the typically once-a-month sampling frequency of CH4 flux data in alpine small and midsize lakes is insufficient to resolve their seasonal and interannual variability using temperature as a proxy50. Consequently, we used sediment temperature sensitivities (1.36 eV for ebullition and 1.00 eV for diffusion activation energy) based on multiyear, high-frequency CH4 measurements in three high-latitude lakes to project future CH4 flux changes in alpine small and midsize lakes during the open-water period50. Considering the generally shallow depths of alpine small and midsize lakes (Supplementary Table 3), we assumed the magnitude of future changes in sediment temperature would match that of projected surface temperatures on long-term scales91. Furthermore, we multiplied the estimated carbon fluxes in small and mid-sized lakes by their predicted areas to foresee their future carbon exchanges during the open water period.

Unlike CO2 flux in small and midsize lakes, water temperature strongly affects the monthly variation of CO2 flux in large lakes during open-water and freezing periods (Supplementary Fig. 17). To predict future changes in CO2 flux for large lakes under SSP scenarios, we focused on continuous, multiyear CO2 flux measurements via eddy covariance and corresponding temperature data from Qinghai Lake and Nam Co20,21,58. Meanwhile, Wu et al.53 reported the projected change rates in large lake surface water temperatures and ice duration by 2100. Based on these projected temperature changes and the two fitted equations between monthly CO2 flux and water surface temperature (CO2 flux = 0.25 × T−3.70, r2 = 0.88 for Qinghai Lake; CO2 flux = 0.21 × T−2.02, r2 = 0.70 for Nam Co) (Supplementary Fig. 17), we calculated the response of large lake CO2 flux during open-water and freezing periods to surface water temperature changes by 2100 under SSP scenarios. Due to limited temporal CH4 emission data from alpine large lakes11, we applied a Q10 value of 3.4 derived from deep boreal lakes in Quebec, Canada51, given their typically great depths. Although this value is slightly lower than the typical Q10 of around 4 for methanogenesis reported in large-scale meta-analyses52, it is considered robust, as most alpine large lakes are oligotrophic and saline16. Overall, future changes in carbon fluxes for alpine large lakes during the open water period were predicted based on the temperature sensitivity of carbon fluxes.

Lake ice phenology is expected to feature delayed freezing and advanced thawing by 210053. We first assumed that future carbon fluxes during the transitional periods—from ice-covered to freezing or early open-water phases—would resemble those observed in the current freezing and early open-water phases. Meanwhile, ice-cover duration, rather than temperature, is a crucial driver of carbon flux during the ice-melt period92. We assumed that daily carbon accumulation under ice would remain consistent with recent measured levels. By multiplying the predicted ice duration53, we estimated future carbon fluxes during the ice-melt period for different alpine lakes. Overall, future carbon fluxes in alpine lakes of different sizes were projected by incorporating changes in water temperature and ice duration.

Projections of the impacts of lake expansion on regional carbon sinks

Given the differences in carbon fluxes between alpine lakes and grasslands, we believe future lake expansion could influence the regional carbon budget. Thus, we also estimated carbon exchanges in alpine grasslands by 2100. As mentioned above, we have collected carbon fluxes of alpine grasslands (Supplementary Tables 5 and 6). We compared observed NEP values with NEP outputs from nine Coupled Model Intercomparison Project Phase 6 (CMIP6) Earth system models (ESMs) and selected three ESMs (ACCESS-ESM1-5, CMCC-CM2-SR5, and MPI-ESM1-2-LR) due to their better representation of NEP variations across alpine grasslands. Using the average NEP outputs from the three ESMs, we calculated the NEP change ratios in 2100 under SSP scenarios relative to the 2000–2014 period (Supplementary Table 1). We assumed that the projected NEP change ratios from the three ESMs could reflect future NEP variations. Furthermore, we calculated the NEP of alpine grasslands in 2100 under SSP scenarios based on the average projected change ratios and the collected NEP data (Supplementary Table 7).

A warmer, wetter, and CO2-enriched climate by 2100 is expected to enhance organic matter availability and create a more humid soil environment in alpine swamp meadow ecosystems, leading to higher CH4 production and emissions33,93. CH4 emissions from swamp meadows were projected under representative concentration pathways (RCP) 2.6, 4.5, and 8.5 scenarios33,34,35. We assumed that the CH4 emission changes in swamp meadows by 2100 under SSP1-2.5, SSP2-4.5, and SSP5-8.5 would be comparable to those under RCP2.6, RCP4.5, and RCP8.5 scenarios. Using the predicted average CH4 emission increase rates from previous studies and the collected average CH4 emission rates over recent years (Supplementary Table 2), we estimated the CH4 emission rate for alpine swamp meadows in 2100 (Supplementary Table 8). Meanwhile, Jin et al.35 projected increased CH4 uptake rates in alpine upland grasslands (alpine meadow, steppe, and desert zones) under warmer and wetter climates using a terrestrial ecosystem model, which is supported by the controlling experiment57. We also classified inundated alpine vegetation types resulting from lake expansion using the recent vegetation maps and projected future lake area32,64,76, assuming no change in grassland types by 2100. Furthermore, we quantified future grassland carbon sequestration losses and finally assessed the net regional carbon budget changes caused by lake expansion through a comparative analysis of enhanced lake carbon sink versus diminished grassland carbon sink.

Uncertainty analysis

For regional upscaling, we applied bootstrapping, a nonparametric Monte Carlo approach1,4,11, to upscale CO2 and CH4 fluxes from small lakes (stratified by vegetation type) and to CH4 fluxes from large lakes (stratified by hydrological network type) over the last decade. The bootstrapping generated median fluxes and their 95% CIs. For CO2 fluxes in alpine large lakes, we derived medians and 95% CIs by combining an established CO2 flux–pH relationship with measured pH data (median and 95% CIs), as well as eddy covariance flux data from four large lakes. Finally, we estimated the uncertainty of alpine lake carbon exchanges in the QTP over the last decade by integrating the upscaled fluxes with respective lake areas.

Reporting summary

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