Tibetan lakes have been persistent CO2 sources since the Last Glacial Maximum

Abstract

Lakes have great contributions to global CO₂ emissions, yet reconstructions of historical CO₂ sources/sinks in lakes remain limited, largely due to the lack of reliable indicators. This study addresses this gap by examining CO₂ concentrations and aquatic plant δ¹³C from 105 lakes in China. Our results demonstrate aquatic plant δ¹³C faithfully record CO₂ concentration variations, establishing it as a reliable indicator for paleolake CO₂. Building on this finding, we analyzed δ¹³C of aquatic plant remains in sediment cores from four Tibetan Plateau lakes. Integrating these data with published δ¹³C records from ten additional lakes has enabled a 26,000-year reconstruction of lake CO₂ concentrations. Our integrated record reveals that Tibetan lakes have functioned as persistent CO₂ sources since the Last Glacial Maximum, with enhanced emissions during the last deglaciation and early Holocene. Thus, inland lakes played a substantial role in the global CO₂ cycle during the deglaciation as compared to today.

Introduction

Carbon dioxide (CO₂) is a major greenhouse gas in Earth’s atmosphere, playing a critical role in the planet’s climate system (particularly temperature) by absorbing and re-emitting infrared radiation. Its increasing concentration due to human activities has greatly amplified global warming and climate change¹. Quantifying CO₂ sources/sinks across ecosystems is essential for understanding the global carbon cycles. The oceans, for example, act as a substantial carbon sink, absorbing approximately 2.3 Pg C yr⁻¹^2,3. Inland waters, particularly lakes, also play a key role by receiving and processing large quantities of terrestrially derived carbon, thereby influencing global carbon budgets⁴. Global estimates suggest that lakes (including reservoirs) emit between 0.32 and 0.64 Pg C yr⁻¹^2,4,5,6. Thus, accurate assessment of CO₂ emissions from lakes is crucial for clarifying their contribution to climate change and informing evidence-based climate policies, ecosystem conservation, and carbon management strategies.

Reconstructing paleo-CO₂ concentrations from sedimentary archives yield critical boundary conditions for climate models and reveals key nonlinear feedback mechanisms in Earth’s climate system. As such, paleo-CO₂ reconstruction continues to be a fundamentally important issue in climate science^7,8. Despite its significance, the reconstruction of past lacustrine CO₂ source-sink dynamics remains remarkably scarce. Our current understanding is largely limited to the Holocene period^9,10, based on δ¹³C signatures in pelagic cladocerans (phytoplankton-feeding zooplankton)⁹ or on reconstructions of water pH and alkalinity¹⁰. Notably, a major research gap exists in quantifying lake CO₂ concentrations across a full glacial-interglacial cycle, leaving a critical void in our understanding of long-term carbon cycle dynamics.

Carbon isotopic signatures (δ¹³C) have proven to be an indispensable tool for tracing carbon cycling processes and atmospheric CO₂ fluctuations^11,12, with terrestrial δ¹³C records providing particularly robust constraints for paleo-CO₂ reconstructions⁷. A growing body of evidence from lacustrine systems has consistently demonstrated an inverse relationship between dissolved CO₂ concentrations and δ¹³C values^{13,14,15,16,17}, which may be driven by variations in CO₂ sources^14,16. This well-documented relationship positions δ¹³C as a highly promising paleo-proxy for quantifying past CO₂ source-sink dynamics in lake ecosystems. Notably, despite this potential, the application of δ¹³C values to lacustrine CO₂ concentration reconstructions remains strikingly limited⁹, warranting a systematic investigation across diverse lake systems and longer timescales.

Aquatic plants constitute a major biomass component in lake ecosystems, playing crucial ecological roles. Their leaf remains exhibit exceptional preservation potential in lacustrine sedimentary records, comparable to terrestrial plant remains^{18,19,20,21,22,23,24,25,26,27,28,29,30}. This preservation quality, coupled with their carbon isotopic sensitivity to dissolved CO₂, positions aquatic plant remains as potentially powerful paleolimnological archives for reconstructing historical dissolved CO₂ concentrations. As primary producers, aquatic plants directly assimilate dissolved CO₂ during photosynthesis^31,32,33. Theoretically, δ¹³C of aquatic plants should reflect the isotopic composition of their dissolved CO₂ source. However, this theoretical framework currently lacks robust empirical validation through in-situ investigations. The critical knowledge gap regarding whether aquatic plant δ¹³C can serve as a quantitative proxy for paleo-CO₂ reconstruction underscores the need for comprehensive investigations combining modern process studies and paleo-applications.

The Tibetan Plateau, the world’s highest plateau, hosts an extensive lake system comprising over 1,400 lakes larger than 1 km², accounting for more than 50% of China’s total lake area³⁴. It is characterized by extreme cold climate (mean annual temperatures <5 °C in most areas), representing the largest mid-latitude cryosphere system on Earth. This unique environment makes the plateau a critical region for investigating cryosphere-atmosphere interactions³⁵, providing fundamental insights into global cryospheric CO₂ responses to climate change. Furthermore, CO₂ emissions from dry inland waters to the atmosphere play an important role in the global carbon cycles³⁶. As the planet’s largest high-altitude arid to semi-arid ecosystem, the Tibetan Plateau offers unparalleled opportunities to examine CO₂ dynamics in dryland systems. However, research on the historical CO₂ source/sink dynamics of Tibetan Plateau lakes remains limited, with only preliminary analyses of recent decades based on field monitoring or remote sensing data^37,38,39. Addressing the CO₂ dynamics over longer timescales is imperative, as such research would benefit our understanding of carbon cycle processes in both cryospheric and dryland lake systems worldwide.

Therefore, we undertook an extensive study of δ¹³C in aquatic plants collected from 63 lakes spanning China’s major geographic regions (Fig. 1): the Tibetan Plateau, Chinese Loess Plateau, Yangtze Plain, Inner Mongolia Plateau, Northeast Plain and Tarim Basin. Complementing these data, we performed concurrent measurements of dissolved CO₂ concentrations and δ¹³C in 79 lakes (including previously published CO₂ concentration datasets from 26 lakes in the Tibetan Plateau, Chinese Loess Plateau, and Yangtze Plain¹⁴; Methods). The primary objective is to establish crucial empirical relationships between aquatic plant δ¹³C and lake water CO₂ concentration—a vital step in validating aquatic plant δ¹³C as a robust paleolimnological proxy for historical CO₂ levels.

Fig. 1: Locations of the studied lakes.

Modern aquatic plant and lake water samples (blue triangle symbols) were collected for δ¹³C analysis from multiple regions in China, encompassing the Tibetan Plateau, Chinese Loess Plateau, Inner Mongolia Plateau, Yangtze Plain, Northeast Plain and Tarim Basin. Data on concentration and δ¹³C of CO₂ in lake water from the Northeastern Tibetan Plateau, Chinese Loess Plateau and Yangtze Plain were sourced from Liu et al. (2024)¹⁴. Core sediments for δ¹³C analysis in aquatic plant remains (circle symbols) were obtained from lakes on the Tibetan Plateau. Red symbols denote newly sampled lakes in this study (Lake Sugan, Qinghai, Erhai, and Kuhai), while green symbols represent previously studied lakes, including Lake Luanhaizi¹⁸, Lake Koucha¹⁹, Lake Keluke²⁰, Lake Genggahai²¹, Lake Heihai²², Lake Ahung Co²³, Lake Pumoyum Co²⁴, Lake Zhari Namco²⁶, Lake Karakul²⁷, and Lake Ngoring²⁸. Lake names and sample types (including modern lake water, modern aquatic plant and core sediment) for the lake numbers are listed in Supplementary Table 1.

To extend the findings to paleoenvironmental timescales, we analyzed δ¹³C in aquatic plant remains from sediment cores of four Tibetan Plateau lakes (Sugan, Qinghai, Erhai, and Kuhai; Methods), and compiled published δ¹³C records from ten additional lakes^{18,19,20,21,22,23,24,25,26,27,28} (Fig. 1). This synthesis enables the regional reconstruction of lacustrine CO₂ source-sink variability over the past 26 kyr, offering insights into the role of lakes in regional carbon budgets across a glacial-interglacial cycle and the evolution of lake carbon cycling during the late Quaternary.

Results

We measured dissolved CO₂ concentrations and carbon isotope compositions (δ¹³C-CO₂) in 53 lakes spanning the southwestern Tibetan Plateau, Yangtze Plain, Inner Mongolia Plateau, Northeast Plain and Tarim Basin, building upon our previously published dataset from 26 lakes from the northeastern Tibetan Plateau, Chinese Loess Plateau, and Yangtze Plain¹⁴. The lakes display considerable variability in their carbon parameters, with δ¹³C-CO₂ values ranging from −18.4‰ to 4.8‰ and CO₂ concentrations between 5.8 and 376 μmol/L (Supplementary Fig. 1). Statistical analyses show significant negative correlations between CO₂ concentrations and both δ¹³C-CO₂ values (R² = 0.36, p < 0.05; Supplementary Fig. 1) and lake water pH (R² = 0.52, p < 0.01; Supplementary Fig. 1).

Analysis of aquatic plants reveals distinct species-specific δ¹³C signatures: Potamogeton (from −23.2‰ to −2.6‰, avg. −10.6‰, n = 76), Myriophyllum (from −26.5‰ to −6.1‰, avg. −15.9‰, n = 31), Ruppiaceae (from −18.0‰ to −16.1‰, avg. −17.2‰, n = 4), Hydrilla (from −24.0‰ to −19.1‰, avg. −21.6‰, n = 6), Vallisneria (from −22.8‰ to −16.9‰, avg. −19.8‰, n = 6), Trapa (from −29.3‰ to −26.6‰, avg. −27.4‰, n = 13), Spirogyra (from −22.1‰ to −8.8‰, avg. −10.1‰, n = 16), and Chara (from −25.1‰ to −17.4‰, avg. −21.5‰, n = 18) (Fig. 2a). Importantly, despite the species-specific signatures, aquatic plant δ¹³C values with all plants combined display a strong positive correlation with δ¹³C-CO₂ (R² = 0.64, p < 0.01; Fig. 2b), negative correlation with CO₂ concentrations (R² = 0.93, p < 0.01; Fig. 2c), and positive correlation with pH (R² = 0.54, p < 0.01; Fig. 2d), establishing their utility as environmental proxies.

Fig. 2: δ¹³C variations in aquatic plants and their relationships with environmental factors.

A δ¹³C values of aquatic plants across different genera in submerged plants (including Potamogeton, Myriophyllum, Ruppiacea, Hydrilla, and Vallisneria), floating plants (Trapa), and algae (including Spirogyr and Chara); B Relationship between δ¹³C values in aquatic plants and δ¹³C values of dissolved CO₂ in lake water; C Relationship between δ¹³C values in aquatic plants and dissolved CO₂ concentrations in lake water; D: Relationship between δ¹³C values in aquatic plants and lake water pH. The grey shadings represent the 95% confidence interval.

Sediment core analyses provide paleolimnological perspectives, with aquatic plant remains from Lake Qinghai (9.5−13.2 ka) showing δ¹³C values of −15.3‰ to −10.3‰, Lake Kuhai (last 0.8 kyr) ranging from −18.6‰ to −13.3‰, Lake Erhai (last 1.4 kyr) varying between −15.4‰ and −6.2‰, and Lake Sugan (last 5.5 kyr) exhibiting values from −18.2‰ to −5.7‰ (Fig. 3). Combining new δ¹³C data from these four lakes with ten published records^{18,19,20,21,22,23,24,25,26,27,28}, we reconstructed lacustrine CO₂ concentrations over the last 26 kyrs, based on our new calibration dataset in Fig. 2C. The results reveal dynamic changes in CO₂ concentrations (0.1−303.3 μmol/L; Fig. 4a) and pH (7.3−10.6; Fig. 4c), demonstrating that Tibetan Plateau lakes have acted as regional CO₂ sources throughout the past 26 kyr, with particularly enhanced emissions during the last deglaciation and early Holocene (8−18 ka, Fig. 4). During the mid- and late Holocene (0−8 ka), most lakes act as CO₂ sources except Lake Koucha and display large variations in CO₂ concentrations among different lakes (Fig. 4a).

Fig. 3: δ¹³C variations in aquatic plant remains from Tibetan Plateau lakes during the past 26 kyr.

a δ¹³C records of aquatic plant remains from both newly investigated lakes in this study (Lake Sugan, Qinghai, Erhai, and Kuhai) and previously reported lakes, including Lake Luanhaizi¹⁸, Lake Koucha¹⁹, Lake Keluke²⁰, Lake Genggahai²¹, Lake Heihai²², Lake Ahung Co²³, Lake Pumoyum Co²⁴, Lake Zhari Namco²⁶, Lake Karakul²⁷, and Lake Ngoring²⁸; b Combined new and existing aquatic plant δ¹³C data. The solid black lines represent average δ¹³C values with a 300-year interval (grey shadings indicate the standard deviation).

Fig. 4: Reconstructed lake water CO₂ concentrations and pH values over the past 26 kyr based on δ¹³C data from Tibetan Plateau lakes (4 newly sampled lakes and 9 previously reported lakes).

Solid black lines represent mean CO₂ concentrations and pH values calculated at 300-year intervals and grey shadings indicate standard deviation. a Reconstructed CO₂ concentrations in lake water. The red line indicates atmospheric equilibrium CO₂ concentrations (derived from ice core data⁸). Values above this line represent lakes acting as CO₂ sources to the atmosphere, while values below indicate CO₂ sinks. b Calculated CO₂ concentration changes relative to modern lake water. For lakes with data extending to ~4 ka, we used the youngest data point as the modern reference. For lakes without data after 4 ka (Lake Qinghai, Zhari Namco, Pumoyum Co, and Heihai), we used contemporary measured CO₂ concentrations as the baseline. c Reconstructed pH values of lake water.

Discussion

Carbon isotopes in aquatic plants: a reliable proxy to reconstruct lake water CO₂ concentration

Through systematic analysis of the relationship between aquatic plant δ¹³C values and dissolved CO₂ concentrations in lake water, we identified a robust negative correlation (R² = 0.93, p < 0.01) between them across diverse lake ecosystems in China (Fig. 2c). This strong relationship likely stems from two key mechanisms. On the one hand, aquatic plants exhibit enhanced CO₂ uptake efficiency under elevated dissolved CO₂ conditions in lake water, which will simultaneously increase carbon isotope fractionation during photosynthesis^40,41 and decreases the relative HCO₃⁻ assimilation⁴², collectively resulting in more negative δ¹³C signatures in aquatic plants. On the other hand, both this study (Supplementary Fig. 1) and previous ones^{13,14,15,16,17} have indicated a well-established negative correlation between δ¹³C values of dissolved inorganic carbon and its concentration in lake water, which may be driven by variations in CO₂ input sources¹⁴ or respiration of dissolved organic carbon and decomposition of organic sediments¹⁶. Consequently, aquatic plants are known to utilize CO₂ as their carbon source for photosynthesis^31,32,33, with their δ¹³C values faithfully record the isotopic composition of their carbon source (Fig. 2b).

Although the δ¹³C composition of aquatic plants can be affected by factors such as water depth, productivity, salinity and specific species^{18,19,20,21,31,40,41}, which in turn potentially introduce uncertainty into the reconstruction of past CO₂ concentrations, it nevertheless captures the long-term isotopic shifts in CO₂ during their growth period. As shown by the strong correlation between aquatic plant δ¹³C values and CO₂ concentrations (Fig. 2c), these findings collectively establish aquatic plant δ¹³C as a reliable proxy for reconstructing historical lake water CO₂ concentrations, effectively addressing a critical methodological gap in paleolimnological research and providing a much-needed tool for understanding lacustrine carbon cycling dynamics.

Tibetan Lakes as persistent CO₂ sources since the Last Glacial Maximum

Building upon the robust relationship between modern aquatic plant δ¹³C and dissolved CO₂ concentrations (Fig. 2c), we integrated δ¹³C isotopic data of aquatic plant remains from four newly investigated Tibetan lakes (Sugan, Qinghai, Erhai, and Kuhai) with ten published lacustrine records^{18,19,20,21,22,23,24,25,26,27,28} to generate a continuous record of lake CO₂ dynamics spanning a full glacial-interglacial cycle (Fig. 4a). This paleo-CO₂ record demonstrates that: (1) Tibetan Plateau lakes have maintained CO₂ concentrations above atmospheric equilibrium levels throughout the past 26 kyr, confirming their persistent role as regional CO₂ sources; and (2) these CO₂ emissions were particularly enhanced during the last deglaciation and early Holocene (8−18 ka), when reconstructed CO₂ concentrations reached their maximum values.

Our paleo-CO₂ reconstructions reveal temporal variations in carbon dynamics on the Tibetan Plateau, with particularly striking patterns emerging during the mid-to-late Holocene (0−8 ka). This period is characterized by maximal heterogeneity in both reconstructed CO₂ concentrations and pH values among lakes (Fig. 4), reflecting the establishment of distinct biogeochemical equilibria as individual lake systems stabilized under relatively stable climatic conditions. Quantitative analysis of CO₂ concentration offsets relative to modern baselines (Fig. 4b) provides compelling evidence for temporal shifts in the plateau’s lacustrine carbon reservoir: the most pronounced positive offsets occurred during 8−18 ka (Fig. 4b), indicating these high-altitude lakes functioned as exceptionally strong CO₂ sources during the last deglaciation and early Holocene. In contrast, both the Last Glacial Maximum (18−26 ka) and mid-to-late Holocene (0−8 ka) maintained offset values indistinguishable from modern reference levels (Fig. 4b), suggesting similar carbon storage during these periods despite their differing climatic contexts.

Driving factors for CO₂ concentration variations in lake water since the Last Glacial Maximum

Ice core records show minimal variation in atmospheric CO₂ concentrations during 18−25 ka followed by a steady increase since 18 ka⁸ (Fig. 5b), while our lake records demonstrate distinctly different patterns, with peak CO₂ concentrations occurring during 8−18 ka (Fig. 5a). This temporal mismatch is further supported by isotopic evidence: atmospheric δ¹³C-CO₂ exhibited only minor fluctuations ( ~ 0.5‰) over the past 25 kyr⁸ (Fig. 5c), in stark contrast to the >20‰ variation observed in δ¹³C values of aquatic plant remains from Tibetan lakes (Fig. 3). These pronounced differences strongly suggest that atmospheric CO₂ changes had negligible influence on lacustrine CO₂ concentrations on the plateau. To systematically evaluate potential controls on lacustrine CO₂ variations, we conducted a comprehensive analysis of published sedimentary records from Tibetan lakes spanning the past 26 kyr, incorporating: (1) total organic carbon (TOC) content (reflecting productivity and organic matter sources)^{18,21,42,43,44,45,46,47,48,49,50} (Supplementary Fig. 2), (2) carbonate content (reflecting evaporation intensity and chemical equilibrium)^{18,21,45,51,52} (Supplementary Fig. 3), and (3) total nitrogen (TN) content (reflecting nutritional status)^{21,45,47,48,53} (Supplementary Fig. 4). Productivity, organic matter sources, evaporation intensity and nutritional status may influence the CO₂ concentrations in lake water^36,54,55,56. This multi-proxy synthesis reveals a consistent temporal pattern, showing higher values during the Holocene (0 − 12 ka) relative to glacial periods (12−26 ka) (Supplementary Fig. 2 − 4) − a trend that differs from our reconstructed lacustrine CO₂ maxima during 8 − 18 ka (Fig. 4a). Notably, a few lakes display high values of TOC content (Lake Chen Co, Tangra Yumco, Pumoyum Co), carbonate content (Lake Luanhaizi) or TN content (Lake Tangra Yumco) during deglacial intervals ( ~ 15−19 ka), likely due to regional climatic or hydrological influences^18,46,48,50. In short, the inconsistent relationship between these sedimentary indicators and CO₂ concentrations suggests that conventional explanations based on organic matter production, carbonate precipitation or nutrient availability cannot account for the observed CO₂ patterns in Tibetan lakes. Furthermore, although lake salinity on the Tibetan Plateau may have varied since the Last Glacial Maximum, modern studies indicate that salinity exerts only a limited influence on CO₂ levels^14,57,58. Consequently, changes in lake salinity are unlikely to determine the variations in CO₂ concentration recorded in these lakes.

Fig. 5: Comparison of reconstructed CO₂ concentrations with other paleoenvironmental records.

a Reconstructed lake water CO₂ concentrations from this study; b Atmospheric CO₂ concentration from ice cores⁸; c δ¹³C values of atmospheric CO₂ from ice cores⁸; d pH variations in Lake Nam Co⁶⁵; e HCO₃^- concentrations in Lake Koucha¹⁹, with higher values indicating lower pH; f Effective moisture reconstruction for Central Asia⁷⁰; g Annual mean precipitation reconstruction based on pollen data from the northeastern Tibetan Plateau⁷¹. The grey bar highlights the enhanced CO₂ emissions during the last deglaciation and early Holocene.

Both our modern survey (Supplementary Fig. 1) and previous studies^{59,60,61,62,63,64} consistently show an inverse relationship between lake water CO₂ concentration and pH values. Based on the established relationship between modern aquatic plant δ¹³C and lake water pH (Fig. 2d), we reconstructed pH variations over the past 26 kyr for 14 Tibetan lakes using aquatic plant remains. Our reconstructions reveal lower pH values during 8−18 ka (Fig. 4c). This pattern is corroborated by independent records from Lake Nam Co (lower pH during 7−17 ka and higher pH during 17−24 ka; Fig. 5d)⁶⁵ and Lake Koucha (lower pH during 8−17 ka and higher pH during 0−8 ka; Fig. 5e)¹⁹, showing remarkable consistency with our reconstructed pH trends.

On the Tibetan Plateau, atmospheric precipitation exhibits a near-neutral pH of approximately 7⁶⁶, while river water and meltwater typically maintain a slightly alkaline pH range of 7.5−8.5^67,68. In contrast, lake waters are generally more alkaline, with pH values generally exceeding 8.5^14,69. This pH gradient indicates that external water inputs (including precipitation, rivers and meltwater) would tend to decrease lake water pH. During the Last Glacial Maximum (18−26 ka), Tibetan lakes were predominantly very shallow and small, with limited moisture availability⁷⁰ (Fig. 5f), resulting in minimal precipitation and inflow inputs. The subsequent deglaciation and early Holocene (8−18 ka) witnessed increasing moisture availability⁷⁰ (Fig. 5f), precipitation amount⁷¹ (Fig. 5g), and meltwater inputs^72,73, while lakes remained relatively small in volume. This combination led to greater proportional influence of external water inputs, causing lower lake pH values during this period. During the mid-to-late Holocene (0−8 ka), the observed divergence in CO₂ levels likely resulted from differential responses of each lake’s carbonate system to Holocene environmental status. Lake conditions in this period were likely more comparable to those of modern lakes. Studies of CO₂ concentrations in contemporary lakes across the plateau reveal considerable inter-lake variability, which is caused by the variations of factors such as pH, water temperature, and phytoplankton biomass^{13,14,17,38,39,69}.

In addition to external water inputs, an alternative explanation for the observed pH decrease in lake water relates to enhanced bio-productivity, where elevated respiration rates relative to photosynthesis promote organic carbon decomposition and subsequent CO₂ release into lake water. During the last deglaciation and early Holocene (8 − 18 ka), numerous Tibetan lakes supported substantial aquatic plant biomass^18,19,74,75. Collectively, these findings indicate that pH fluctuations in Tibetan lakes over the past 26 kyr have been predominantly controlled by two key factors: (1) varying influences of external water inputs (precipitation and inflow discharge) and (2) changes in lake bio-productivity dynamics.

Variations in lake water pH influence the carbonate system equilibrium (CO₂, HCO₃⁻, CO₃²⁻), where a decrease in pH promotes CO₂ dissolution by shifting the chemical equilibrium toward free CO₂¹¹. Modern observations consistently demonstrate that riverine waters typically contain elevated CO₂ concentrations (due to soil respiration, organic matter decomposition, etc.)⁷⁶, and an increase in external riverine inputs can substantially increase lacustrine CO₂ levels⁷⁷. Therefore, during the last deglaciation and early Holocene (8−18 ka), the observed increase of CO₂ concentrations in Tibetan Plateau lakes (Fig. 4a) may result from the coupling factors: (1) pH-dependent chemical equilibrium, and (2) hydrological inputs (external carbon). These interconnected processes reflect the complex feedback of climate-hydrology-ecosystem interactions during the deglaciation period.

Potential influence of lake CO₂ emissions on global CO₂ cycles

Reconstructions of atmospheric CO₂ concentrations reveal a rapid increase in global atmospheric CO₂ levels accompanied by an approximately 0.5‰ negative shift in δ¹³C-CO₂ during the last deglaciation (12−18 ka)^8,78. Multiple hypotheses have been proposed to explain this deglacial CO₂ rise, including enhanced Southern Ocean nutrient utilization efficiency and dust-mediated iron fertilization⁷⁹, intensification of Southern Hemisphere Westerlies driving deep ocean ventilation⁸⁰, permafrost thawing⁸¹, increased CO₂ outgassing from the equatorial Pacific⁸², and circulation changes leading to deep Pacific water mass flushing⁸³. Modern estimates indicate that global lake CO₂ emissions constitute a critical component of the carbon cycle, accounting for approximately 14−28% of the oceanic carbon sink^2,3,4,5,6. Our study reveals higher CO₂ concentrations in Tibetan Plateau lakes during the last deglaciation (12−18 ka) compared to the Last Glacial Maximum (18 − 26 ka) (Fig. 4), indicating lakes served as substantial CO₂ sources and potentially released large quantities of CO₂ to the atmosphere. This finding suggests that enhanced lacustrine CO₂ emissions may have been an important contributing factor to the observed global atmospheric CO₂ rise during the deglaciation.

In addition, as a global representative region integrating both cryospheric and arid zone ecosystems, the Tibetan Plateau provides critical insights into past carbon dynamics³⁵. Our reconstructions show that the last deglaciation and early Holocene CO₂ concentrations in Tibetan lakes were 2−3 times higher than modern levels (Fig. 4a), suggesting that global lakes—particularly those in cryospheric and arid zones—played a more substantial role in the CO₂ cycle during deglaciation than they do today.

While our findings are currently constrained to the Tibetan Plateau, expanding CO₂ reconstructions to lakes in other regions is essential to better quantify the global impact of lacustrine CO₂ emissions on atmospheric CO₂ levels and carbon cycling. Nevertheless, the potential magnitude of these emissions underscores their importance in paleoclimate studies and Earth system models.

Conclusions

Through comprehensive investigations and correlation analyses of aquatic plant carbon isotopes and lake water CO₂ concentrations across diverse lake ecosystems in China, we show that aquatic plant δ¹³C can serve as a robust proxy for reconstructing historical lake CO₂ levels, effectively addressing the critical gap in lacustrine CO₂ reconstruction methodologies. By integrating δ¹³C data from aquatic plant remains in four newly studied Tibetan lakes (Sugan, Qinghai, Erhai, and Kuhai) with ten published lacustrine records, we present a continuous reconstruction of lake CO₂ dynamics spanning a full glacial-interglacial cycle over the past 26 kyr. Our results reveal that Tibetan Plateau lakes have consistently functioned as CO₂ sources throughout this period, with particularly enhanced emissions during the last deglaciation and early Holocene (8 − 18 ka). The pH variations in these lakes, which were primarily influenced by the external water inputs, appear to have been the dominant control on their CO₂ concentration changes over the past 26 kyr.

Methods

Materials and chronology

We collected 170 samples of various aquatic plant species (including Potamogeton, Myriophyllum, Ruppiaceae, Hydrilla, Vallisneria, Trapa, Spirogyra, and Chara) from 65 lakes across different regions of China (the Tibetan Plateau, Chinese Loess Plateau, Inner Mongolia Plateau, Yangtze Plain, Northeast Plain and Tarim Basin) (Fig. 1 and Supplementary Table 1, 2). Aquatic plants were sampled at varying depths using a grab sampler, with thorough rinsing using distilled water between collections to eliminate any sediment or dust particles. Additionally, we measured CO₂ concentrations and carbon isotopes in water samples from 53 lakes on the southwestern Tibetan Plateau, Inner Mongolia Plateau, Yangtze Plain, Northeast Plain and Tarim Basin (Fig. 1 and Supplementary Table 3), supplementing these with previously published CO₂ concentrations and carbon isotope data from 26 lakes on the northeastern Tibetan Plateau, Chinese Loess Plateau, and Yangtze Plain¹⁴. Water samples were collected in 2 L plastic bottles during June to September in 2022 to 2025. To prevent microbial activity, approximately 2 mL of mercuric chloride was added to each sample, followed by sealing with parafilm. The samples were then transported to the laboratory for analysis. Each measurement was conducted continuously for at least 10 minutes, with the mean value used for further analysis.

We retrieved sediment cores from four lakes on the Tibetan Plateau (Lake Sugan, Qinghai, Erhai, and Kuhai) (Fig. 1). Lake Sugan is a semi-closed saline lake situated on the northern margin of the Tibetan Plateau, with a current elevation of approximately 2,797 m and an area of 103.7 km². The lake has a maximum water depth of ~7.5 m (average depth: 2.5 m), with a pH of 8.5 and salinity of 32 g/L⁸⁴. In August 2023, a 733-cm-long sediment core (SGH2023-1) was obtained from Lake Sugan (38°51′15.12′′N, 93°53′26.16′′E) at a water depth of 7.5 m. Aquatic plant remains (Chara) were identified in the upper 587 cm of the core, and we analyzed the carbon isotopes of 72 such remains (Supplementary Table 4). Chronological control was established using radiocarbon (¹⁴C) dating by the acceleration mass spectrometry (AMS) from 7 bulk sediment samples⁸⁵. The age model of this core aligns well with previously reported cores (SG001227⁸⁶) from Lake Sugan, applying a reservoir effect of 2000 years for depths below 72 cm and 1200 years for depths above 72 cm.

Lake Qinghai, the largest inland brackish lake in China, is situated at an average elevation of 3,194 m. With a surface area of approximately 4,400 km², it features a maximum water depth of 27 m (mean depth: 21 m) and maintains an average salinity of 14.1 g/L. The lake is primarily fed by three major rivers: the Buha, Heima, and Shaliu Rivers. The sediment core LQDP05-1F (36°48′40.7″N, 100°08′13.5″E) was collected in 2005 using the ICDP GLAD800 drilling system under the auspices of the International Continental Drilling Programme (ICDP). The chronology was established through AMS ¹⁴C dating of 52 bulk samples of total organic carbon, 6 Ruppiaceae seeds samples, and 7 plant residue samples^46,87. The complete core spans approximately 1,800 cm, with a particularly well-preserved section between 416-499 cm containing Ruppiaceae seeds. From this section, we analyzed the carbon isotopic composition of 6 Ruppiaceae seeds (Supplementary Table 5).

Lake Erhai (36°32′−36°35′N, 100°42′−100°44′E; 3,200 m asl) is a small ( ~ 4 km²) satellite lake of Lake Qinghai, located in its southeastern margin. With an average water depth of 1.5 m and a catchment area of 6.8 km^2,88, the lake is primarily fed by the perennial Daotang River entering along its eastern shore. In January 2016, a 111.5-cm sediment core (EHLC) was collected from the southwestern basin at a water depth of 5 m. The chronological framework was established through a combination of AMS ¹⁴C and ¹³⁷Cs/²¹⁰Pb dating⁸⁸. 14 samples were selected from the top 27 cm for ¹³⁷Cs/²¹⁰Pb dating using a germanium well controller at the Institute of Polar Environment, USTC, Hefei, China. Five plant residues at 14, 27, 50, 80, and 108 cm depths were selected for AMS ¹⁴C dating at the Xi’an AMS Center, China⁸⁸. From this core, we analyzed the carbon isotopic composition of 168 Potamogeton remains (Supplementary Table 6), providing a high-resolution record of lacustrine carbon dynamics.

Lake Kuhai is a brackish/saline lake (20.4 g/L) situated in the eastern Tibetan Plateau, covering an area of ~45 km² with a maximum depth of 21 m. The lake exhibits alkaline conditions, with measured pH ranging from 8.9 to 9.1⁸⁹. In late September 2014, a 56-cm sediment core (KHC14-1) was collected from the northern basin at a water depth of 7.8 m (34.0108°N, 97.2422°E). ¹³⁷Cs/²¹⁰Pb dating was measured using an AMETEK germanium detector at the Institute of Polar Environment, USTC, Hefei, China⁸⁹. Two Potamogeton remains selected at approximately 11 cm and 54 cm depths were measured for AMS ¹⁴C dating at the Xi’an AMS Center, China⁸⁹. From this core, we conducted carbon isotope analysis on 86 well-preserved Potamogeton remains (Supplementary Table 7).

Additionally, we conducted a literature review on the preservation and carbon isotope values of aquatic plant remains in lakes across different regions in China. In the lakes located in Yangtze Plain, Inner Mongolia Plateau and Chinese Loess Plateau, pollen or aquatic plant remains in lake core sediments were discovered^30,90,91,92, but their carbon isotopic values were not reported. In contrast, aquatic plant remains are well preserved in the sediments of many lakes across the Tibetan Plateau. For instance, substantial amounts of aquatic plant remains have been found in the 4 lakes we mentioned (Lake Sugan, Qinghai, Erhai, and Kuhai), and carbon isotope values of aquatic plant remains have been reported for 10 lakes (Fig. 1), including Lake Luanhaizi¹⁸, Lake Koucha¹⁹, Lake Keluke²⁰, Lake Genggahai²¹, Lake Heihai²², Lake Ahung Co²³, Lake Pumoyum Co²⁴, Lake Zhari Namco²⁶, Lake Karakul²⁷, and Lake Ngoring²⁸. Therefore, in this study, we focus on the carbon isotopes of aquatic plant remains from the Tibetan Plateau and apply them to reconstruct historical CO₂ concentrations in lakes. It should be noted that the δ¹³C values in some lakes during certain periods are highly positive, leading to calculated CO₂ concentrations that are even below zero (for example, in Lake Koucha since 6 ka), which is impossible. These values are therefore defined as 0.1 μmol/L to represent its low CO₂ levels.

Carbon isotopic analysis in aquatic plants

Approximately 0.3 g aquatic plant remain samples were treated with 2 mol/L HCl for 24 h at room temperature to remove carbonates. The samples were subsequently rinsed with distilled water until the pH exceeded 6 and then dried at 40 °C. Combustion was conducted in vacuum-sealed quartz tubes at 860°C for 4 h, with Ag foil and CuO added to ensure complete oxidation. The resulting CO₂ was purified and analyzed for carbon isotope composition using a Finnigan MAT251 gas isotope ratio mass spectrometer. To ensure measurement accuracy, the national standard GBW04407 (δ¹³C_V-PDB = − 22.43 ± 0.07‰) was analyzed after every twelve samples. Isotopic ratios are reported in per mil (‰) relative to the V-PDB standard, with an analytical precision of 0.2‰.

Concentrations and δ¹³C values of dissolved CO₂ in lake water

We utilized a rapid air-water equilibration system for dissolved CO₂ extraction from lake water^14,93. The fast-response automated gas equilibrator operates on a flow-through principle, where a gas flow is introduced into a constant water flow to create a minimal headspace for real-time measurements. The system configuration consists of a mass flow controller (SIERRA C50L, Netherlands) maintaining a constant carrier gas flow rate of 150 mL/min, while a peristaltic pump simultaneously extracts water samples at 500 mL/min. The two streams are combined in a gas-water mixing unit (10 mL plastic syringe) before passing through a 2-meter Tygon equilibration tube, after which the mixture enters a gas-water separation unit (30 mL plastic syringe) where the equilibrated headspace gas is separated from the water phase.

The CO₂ concentration and δ¹³C compositions in lake water were measured using a Picarro G2131-i cavity ring-down spectrometer (Picarro Inc., USA). This high-precision analyzer operates by detecting the characteristic absorption lines of ¹²C¹⁶O₂ (6251.760 cm⁻¹) and ¹³C¹⁶O₂ (6251.315 cm⁻¹) within a temperature- and pressure-stabilized optical cavity (35 cm³ volume, maintained at 318.150 ± 0.002 K and 18.67 ± 0.02 kPa). The in-situ CO₂ concentrations were measured in ppm with a precision of ±0.2 ppm and subsequently corrected and converted to μmol/L⁹³. The δ¹³C values were determined from the ratio of the ¹³C¹⁶O₂ to ¹²C¹⁶O₂ peak heights. For instrument calibration, we used two certified reference gases (cylinder 1 with a CO₂ concentration of 395.49 ± 0.02 ppm and δ¹³C value of −8.980 ± 0.008‰, cylinder 2 with a CO₂ concentration of 491.43 ± 0.02 ppm and δ¹³C value of −10.395 ± 0.024‰) provided by the Chinese Academy of Meteorological Sciences⁹⁴. All δ¹³C values are reported relative to the V-PDB standard, with a measurement precision of 0.40‰ (30-s moving average) for CO₂ concentrations > 200 ppm.