Introduction

A large portion of the Arctic region (Siberia, Alaska, Canada) is covered by late Pleistocene ice-rich syngenetic permafrost (50–90% in volume), referred to as “Yedoma”1. It contains between 71 and 150 Gt of organic carbon (OC)2, most of which is well-preserved OC coming from vegetation accumulation and sequestration during permafrost formation3. In Central Yakutia (Eastern Siberia) where this study was conducted, an increase of 2.9 °C in air temperature between 1980 and 20194 has led to intense Yedoma degradation, resulting in ground subsidence (thermokarst) and the formation and rapid development of thermokarst lakes5.

Permafrost degradation increases hydrological connectivity and the potential transfer of elements previously stored in permafrost for millennia to aquatic ecosystems6,7,8,9,10. Specifically, the transfer of old organic matter (OM) from permafrost soils to aquatic ecosystems and its mineralization into greenhouse gases (GHGs; carbon dioxide ‒ CO2 ‒ and methane ‒ CH4) reintroduces millennial-old carbon into the active carbon cycle and leads to the amplification of air temperature increasing, a process known as the permafrost carbon feedback (PCF)11. The magnitude of this process is highly uncertain12. It depends to some extent on the relative biolability and photolability of the OM stored in permafrost. In the context of Yedoma, permafrost dissolved OM (DOM) has been shown to contain a high proportion of low molecular weight and aromaticity molecules and to be preferentially degraded by microbial communities compared to modern DOM13,14,15,16,17,18,19. Sunlight effect on Yedoma DOM is poorly understood, some study on Alaskan Yedoma leachates reporting high photolability20, while no DOM loss has been highlighted in Siberian Yedoma thaw after light irradiation21.

In this context of permafrost degradation, radiocarbon (14C) measurements of the different carbon forms in aquatic ecosystems provide a powerful and indispensable tool to distinguish old permafrost-derived carbon from recent carbon (either produced in the hydrosystem or coming from plant material and litterfall from the banks) and therefore better understand processes involved in the PCF. Most studies on thermokarst lakes have measured the 14C content of GHGs (CO2, CH4)22. Contrasting results have emerged across the Arctic, including strong contribution of permafrost thaw to diffusive or ebullitive GHG emissions, either from Pleistocene-aged permafrost in Yedoma areas23,24,25,26,27,28,29, or from Holocene-aged permafrost30,31. Other studies have found that GHG emissions in thermokarst lakes of Pleistocene-aged permafrost areas were dominated by recently fixed or late Holocene carbon, indicating minimal transfer of Pleistocene-aged carbon in the GHG form32,33. In thermokarst lakes, the 14C content of dissolved OC (DOC) and particulate OC (POC) remains understudied. In a synthesis done in 2020, only about thirty measurements have been referenced on the 14C of DOC (10 times less than GHGs) and almost none for POC in Arctic lakes22. Since 2020, in the Yedoma area, 14C-DOC and 14C-POC were studied in one thermokarst lake near the Kolyma River mouth23, and 14C-DOC was reported in thermokarst lakes of the Lena Delta34. No data is available for Central Yakutia.

While lakes are common across the Arctic landscapes, they can have very distinct origins, with possible strong implications for their carbon sources and cycling. In Central Yakutia (Fig. 1a, b), we showed that both recent thermokarst lakes (developing in relation to recent permafrost degradation; Fig. 1c) and alas-lakes (residual lakes in alas basins formed by permafrost thaw during the early Holocene, Fig. 1d) shape the landscape35,36. Different hydrological connectivity relationships within the watershed are observed, with most alas basins being endorheic, while others are connected to the hydrographic network (Fig. 1e). Alas-lakes can also be modified by recent thermokarst activity along their banks in the form of retrogressive thaw slumps (RTS), thermo-erosion gullies in between polygons37 or recent lake spills (Fig. 1f).

Fig. 1: Study area context.
figure 1

a Study site localization in Siberia (SRTM Plus elevation data at 1 km resolution). b Studied area in Central Yakutia near Yakutsk (Spot 7 2019‒08). Different sampled lake types: c recent lake d alas-lake, e connected alas-lake, f thermokarst-modified alas-lake, g permafrost meltwater inside a retrogressive thaw slump.

Full size image

Here, in order to investigate how permafrost dynamics rules carbon cycling in lakes with different origin and hydrological connectivity, we report on the OC concentrations, stable isotopic signature and radiocarbon content of DOM and particulate OM (POM) in recent thermokarst lakes, alas-lakes, hydrologically connected alas-lakes and thermokarst-modified alas-lakes. The study area is located in Central Yakutia in a continuous Yedoma permafrost area, a key permafrost region in the circumpolar Arctic witnessing rapid environmental changes35,38 (Fig. 1a, Table 1). We also sampled permafrost meltwater inside a RTS (Fig. 1g). Samples were collected in fall 2018 and winter, spring and summer 2019 (Table S1) in order to investigate seasonal variability in relation to lake geomorphological and hydrological characteristics (Fig. 1b).

Table 1 Mean dissolved and particulate organic matter characteristics by lake types and in the permafrost meltwater sampled during summer 2019
Full size table

Results

Seasonality of lake organic matter

In all lake types and seasons, OC was mostly present in the dissolved form (Table 1, Fig. 2a-b). DOC represented 83% to 99% of the total OC (TOC; Table 1, Fig.2), except for two lakes. In permafrost meltwater, DOC corresponded to 96% of the TOC (Table 1). We measured extremely high DOC concentrations, up to 838 mg.L-1, and except for two samples, low POC concentrations (Fig. 2a-b, Table 1, ref. 39). POC concentrations were the highest for alas-lakes and thermokarst-modified alas-lakes at each season (no data for alas-lakes in winter due to complete freezing), and permafrost meltwater sampled in summer 2019 (Table 1, ref. 39). The DOC concentration of permafrost meltwater was very high (341 mg.L-1), a level reached by a few recent lakes and one thermokarst-modified alas-lake during this period. In contrast, connected alas-lakes showed the lowest DOC concentrations during the whole year (29 to 51 mg.L-1).

Fig. 2: Seasonal evolution of particulate and dissolved organic matter-related parameters in studied lakes between fall 2018 and summer 2019.
figure 2

a Particulate organic carbon (POC) concentrations b Dissolved organic carbon (DOC) concentrations c Specific ultra-violet absorbance (SUVA) of dissolved organic matter and d Dissolved organic carbon and nitrogen ratio (DOC/DON). Data is missing in alas-lakes in winter due to complete freezing of these shallow ( < 1 m deep) lakes. Particulate organic matter-related information is represented by circles; Dissolved organic matter-related information is represented by triangles.

Full size image

A strong seasonal signal was observed for DOC concentrations, with a general concentration increase in winter and a decrease in spring (Fig. 2, Kruskal-Wallis, KW, p = 4.85 × 10-9, Table S2). Despite these seasonal patterns in DOC concentration, no such seasonal trend was observed for POC concentrations (Table S2-S3). In addition, almost no changes in DOM optical properties (specific ultraviolet absorbance (SUVA), proxy of aromaticity and E2/E3, ratio of molecular weight) and DOC/DON (ratio of dissolved organic carbon and dissolved organic nitrogen concentrations) were detected during the different sampling seasons, suggesting a steady dominant DOM origin in lakes (Table S2-S3). Only recent lakes showed significantly lower values for DOC/DON in winter compared to fall and summer, and higher E2/E3 in winter compared to fall and spring (Table S3). We attribute the seasonal variations of DOC concentrations to hydro-climatic parameters. In winter, the ice cover is about 1 m thick, leading to over-concentration of DOC by reduced water volume in lakes40. On the contrary in spring, the ice and snow melt brings carbon-depleted water to the lakes, resulting in under-concentration of DOC relative to other seasons36,41. From summer to fall, high evaporation rates42,43,44 lead to increasing concentrations of DOC. Connected alas-lakes had very low amplitude variations, with the median DOC concentrations being statistically similar (Table S2). Their much bigger size and depth ( > 10 m) compared to other lake types (only a few meters deep) might buffer the effects of seasonality.

Organic matter composition and origin

For all lake types, POM composition reflected strong autochthonous origin (Fig. 3a), i.e. internal primary production, with little influence of allochthonous OM. POC/PON (particulate organic nitrogen) ratio values ranged between 4.6 and 7.9, with no pattern related to lake types, while δ13C-POC values covered a large range, between −34.5 and −18.8‰. The δ13C values associated with low POC/PON values were consistent with autochthonous production45,46,47,48,49 (Fig. 3a). The highest δ13C-POC values were found for an alas-lake and thermokarst-modified alas-lakes and could correspond to algal blooms observed during the summer (Fig. 3a; refs. 45,46). F14C (fraction modern) of POC values ranged from modern to 0.694, indicating a varying proportion of OC originating from permafrost (median value between 1 and 39% in each lake, Fig. 4, Table S4) as a source of OC for microorganisms contributing to POC. Recent lakes and thermokarst-modified alas-lakes had older F14C-POC values (0.918 and 0.694; i.e. 690 to 2 930 BP (before present); Table 1, Fig. 3b, ref. 39), suggesting permafrost-derived POC inputs into these lakes (median value between 18 and 39%, Fig. 4, Table S4). The alas-lake and one thermokarst-modified alas-lake showed modern F14C values, indicating no significant influence of permafrost (median value between 1 and 3% of POM derived from permafrost, Fig. 4, Table S4).

Fig. 3: Dissolved and particulate organic matter (DOM and POM) composition and origin in the studied lakes and permafrost meltwater in summer 2019.
figure 3

a Summer 2019 DOM and POM δ13C and C/N signatures in lakes and permafrost meltwater. Common ranges in terrestrial plants and autochthonous sources from refs. 45,46,47,48 common range in δ13C and C/N in Central Yakutia Yedoma permafrost from ref. 51. C/N ratio corresponds to DOC/DON and POC/PON respectively for DOM and POM samples of this study, as well as for common ranges in terrestrial plants and autochthonous sources. C/N ratio corresponds to TOC/TN for the Yedoma common range. b F14C of dissolved and particulate organic carbon (DOC and POC) (modern pool corresponds to the post-1950 F14C range from ref. 85 and Yedoma pool age range from this study and refs. 50,68,69). The dashed line represents an equivalent DOC and POC age.

Full size image
Fig. 4: Contribution of permafrost-derived organic matter to dissolved and particulate organic carbon (DOC and POC) in the four lake types.
figure 4

To each lake type are associated a pie chart and two histograms. The pie charts represent the comparative proportion of DOC vs. POC in %. The whisker box represents the relative contribution of atmosphere, topsoil and permafrost to DOC (top) and POC (bottom), modeled with the three-pole model. The source distribution of each lake obtained with the modeling (Table S4) within a category was combined by Monte Carlo modeling to account for the uncertainty range of each lake. For each lake type, the thick horizontal lines represent the median, limits of the boxes represent upper and lower quartiles, and whiskers extend to 1.5 times the interquartile range.

Full size image

DOM composition differed between lake types, reflecting various contributions of autochthonous and allochthonous sources in lakes. The δ13C-DOC values showed lower variability compared to POM, between −28.1 and −23.6‰ (Fig. 3a, ref. 39). High DOC/DON ratio (19.5 to 39.2 in summer) for recent lakes, connected alas-lakes and most thermokarst-modified alas-lakes in association with δ13C values between −28.1 and −24.0‰ suggested that DOM mostly originated from terrestrial vegetation47 (Fig. 3a). In contrast, a significantly lower DOC/DON ratio was measured for alas-lakes for all seasons (Figs. 2d-3a, 13.0 to 15.5, one alas-lake reaching 27.9 in summer) indicating a higher proportion of DOM produced by autochthonous primary production47. The 14C content of DOC differed strongly between lake types. DOC in alas-lakes and connected alas-lakes remained close to modern, indicating low influence of permafrost-derived DOC (Figs. 3b-4, Table S4, median value between 4 and 14% of permafrost derived OM), connected alas-lakes receiving modern terrestrial DOM from watersheds. Contrastingly, recent lake F14C-DOC ranged from 0.69 to 0.37 (3 045 to 7 940 BP; Table 1, Fig. 3b, ref. 39), suggesting substantial permafrost OM contribution to lake DOM (median value between 40 and 73%, Fig. 4, Table S4). Thermokarst-modified alas-lakes also showed low F14C values (0.67 to 0.32; 3 230 to 9 065 BP), indicating that inputs of permafrost meltwater strongly affected their DOM composition (median value between 41 and 75% of permafrost derived OM, Fig. 4, Table S4).

Permafrost meltwater sampled in the RTS represented the signature of recently thawed Yedoma permafrost. F14C values were 0.20 for DOC and 0.18 for POC, corresponding to average 14C ages of 12 950 and 13 790 BP, respectively (Table 1, Fig. 3b, ref. 39). These values were fairly similar to the F14C values of bulk OM from permafrost sediment measured at 1.5 m (0.28) and 1.8 m depth (0.10), corresponding to 10 110, 10 150 and 18 315 BP, respectively (this study), and consistent with other published results in the area (23 600 cal BP, i.e. ~19 500 BP or F14C ~ 0.09 at 3.5 m depth; ref. 50). The δ13C and DOC/DON values of DOM (−24.6‰ and 2.7) were close to the δ13C and POC/PON values POM (−26.5‰ and 6.1). DOC/DON and POC/PON ratio in our permafrost meltwater sample were slightly lower than C/N (carbon/nitrogen) values reported in Yedoma cores in Central Yakutia (ref. 51; Fig. 3a). Ref. 51 used the TOC/TN (total nitrogen) ratio. The inorganic N fraction, if not negligible, would artificially decrease the TOC/TN ratio from ref. 51 relative to a ratio calculated only with the organic fraction of N. Consequently, the use of TN in ref. 51 cannot explain the difference on C/N values between our observations and Yedoma soils. This difference can be attributed to other factors. Permafrost meltwater can originate from ice wedge thawing, with a different OM signature from Yedoma soil. In addition, microbial degradation processes can have lowered the DOC/DON ratio measured on our samples.

Permafrost degradation impacts on lake DOM composition

DOC and DON concentrations as well as OM optical parameters were strongly related to the DOC radiocarbon content. Alas-lakes and connected alas-lakes had the highest SUVA (high aromaticity) and lowest E2/E3 (high molecular weight) on average (Table 1, ref. 39). Recent thermokarst lakes had lower SUVA and higher E2/E3 values, meaning lower aromaticity and molecular weight on average than other lake types (Table 1). There was a statistically significant linear relationship between lower F14C-DOC values and higher DOC concentrations (R² = 0.80; Fig. 5a), higher DON concentrations (R² = 0.71; Fig. S1) and higher E2/E3 values (R² = 0.51; Fig. S1), as well as with lower SUVA values (R² = 0.59; Fig. 5b). Higher permafrost contribution to lake DOC (reflected by lower F14C values) was associated with the highest DOC concentrations, particularly in recent lakes and thermokarst-modified alas-lakes (Fig. 5a). In these lakes, old OM showed a distinct composition, with low aromaticity and low molecular weight, as previously reported in Yedoma areas14. However, the permafrost meltwater signature did not fit with the F14C linear relationship observed for lakes between F14C and optical properties (1.4 L.mg-1.m-1 for SUVA; 9.0 for E2/E3; Table 1), suggesting that very early transformations of DOM and equilibrium between DOC and POC occurred quickly after permafrost thaw and were not conserved in lakes. These values were consistent with other studies on Yedoma thaw water (SUVA between 1.0 and 2.1 L.mg-1.m-1 for F14C values between 0.02 to 0.24; ref. 15; Fig. S2).

Fig. 5: Relationship between 14C content (F14C on left axis and conventional 14C age in BP on right axis) of dissolved organic carbon (DOC) and DOC concentrations and specific ultraviolet absorbance (SUVA) of dissolved organic matter in the sampled lakes and permafrost meltwater in summer 2019.
figure 5

a Relationship between 14C content of DOC and DOC concentrations b Relationship between 14C content of DOC and SUVA. Trend lines were calculated without the permafrost meltwater sample.

Full size image

Within the categories of recent lakes and thermokarst-modified alas-lakes, the radiocarbon content of DOC was old but with some variation from 0.68 to 0.32 (3 045 to 9 065 BP; Fig. 5b), indicating different magnitudes of permafrost contribution. We could not associate this variability in DOC concentration and F14C content with the timing of lake formation (formation in 1945, 1955, 1977 or 1999 ‒ 2000) or lake size (1000 ‒ 11 000 m²; Fig. S3). However, all recent lakes showed higher concentrations and older DOC when lakes were actively developing due to permafrost degradation, as observed by collapsing banks and the presence of many drunken trees.

GHG bubble age

We analyzed F14C of CO2 and CH4 ebullitive samples from near-shore bottom sediments of seven lakes in summer 2019 (Fig. S4, ref. 39). Our results show that CO2 ebullition originated partly from permafrost-derived DOM (Fig. S4a). F14C of CO2 bubbles was modern for the connected alas-lake while it was older for recent lakes and thermokarst-modified alas-lakes (0.98 ‒ 0.81; 140 ‒ 1 685 BP). F14C-CO2 showed lower values and a relationship with F14C-DOC (R² = 0,64; Fig. S4a). When removing the connected alas-lake value from the F14C-CO2 and F14C-DOC relationship, the correlation coefficient is even higher (R² = 0.91; Fig. S4a), which indicates that CO2 bubbles are partly derived from permafrost DOM mineralization in recent lakes and thermokarst-modified alas-lakes. This mineralization could occur in the water column and diffusive CO2 transferred to the lake sediments, but old DOC and CO2 could also be released into the sediment when the recent lakes are deepening due to thermokarst (Fig. 6). In contrast, F14C-CH4 was modern except for one recent lake (F14C 0.90; 765 BP) and showed no relationship with F14C-DOC and F14C-POC data (Fig. S4a-b). This indicated a modern source of CH4, probably from POC and macrophyte biomass deposited in the lake bottom (Fig. 6; refs. 33,36,52). Nevertheless, in the bottom center of one recent lake with high thermokarst activity (Fig. 1c), both CH4 and CO2 bubbles were millennial-old. F14C was of late Pleistocene age for CO2 (0.16; 14825 BP) and Holocene age for CH4 (0.66; 3 345 BP), indicating more permafrost contribution to GHGs in lake bottom center compared to the shores.

Fig. 6: Conceptual representation of the origin of carbon fractions (dissolved and particulate organic carbon; DOC and POC; CO2 and CH4) in the studied recent thermokarst lakes.
figure 6

The fluxes are not proportional due to lack of quantitative data. The carbon fractions age was assigned a color from red (late-Pleistocene age) to green (modern) depending on the mean age of each fraction in recent lakes (Table 1, ref. 39). Dissolved CO2 age was not analyzed but inferred for the representation. CO2 and CH4 bubble ages correspond to the measurements conducted near the shore (n = 4; Fig. S4).

Full size image

Discussion

We report direct evidence of permafrost degradation influence on DOC origin and DOM composition in both old and recent thermokarst lakes from Central Yakutia.

In alas-lakes, most of the OC (median value of 99% for POC and 86% for DOC) originated from recently fixed carbon (Fig. 4, Table S4). Most of the recent DOC and POC came from primary production. We observed algae mats at the lake surface, high chlorophyll-a quantity and high dissolved oxygen ( > 150%) in alas-lakes, corroborating our former conclusions36. A high primary production source of OC in these lakes is also consistent with the modern ages of dissolved inorganic carbon (DIC)53 and undersaturation in CO2 (CO2 sink) due to strong photosynthesis36.

In recent thermokarst lakes, permafrost-derived carbon contributed to 40 – 73% of DOC and 18 ‒ 39% of POC (median values, Figs. 4, 6, Table S4). Our results on permafrost contribution to recent lake OC are consistent with the single study on an endorheic recent thermokarst lake in the Yedoma area near the Kolyma delta, where 50 ± 35% of carbon (OC and GHG) was ancient (Pleistocene-aged, Fig. S5)23. As in our study, POC was younger than DOC for this one recent lake (DOC 29.4 kBP and POC 4.2 kBP; ref. 23). As POC in our recent lakes also showed autochthonous δ13C and C/N characteristics, older POC values might partly originate from the recycling of respired old OM (Fig. 6; ref. 47). These strong inputs of old OC from permafrost in recent lakes occur year-round, except in winter, when DOC/DON and E2/E3 showed significantly different values (lower for DOC/DON and higher for E2/E3). This could be due to the freezing of the lakes, limiting inputs of permafrost OM from the banks at this period. Recent lakes showed a diversity of OM optical parameters and ages, reflecting differences in thermokarst lake development stages, and potential evolution of DOM composition in relation to biodegradation and photodegradation13,14,15,16,17,20,21. DOC concentrations and F14C variability of recent lakes was probably driven by a difference in intensity of permafrost OC inputs to the lake, either by subsidence of the lake floor or lateral erosion of their banks (or a combination of both). Recent lakes where thermokarst activity seemed to have stopped showed the lowest permafrost-derived DOC inputs. The impact of the intensity and magnitude of bank erosion has already been reported on GHGs in Arctic ponds54, where diffusive and ebullitive CH4 and CO2 showed higher permafrost and active layer contribution through a gradient of erosion intensity (25% for CO2 in high-erosion ice-wedge trough ponds).

Permafrost degradation (thaw slumping and thermo-erosion) along the banks of inactive alas-lakes had an important impact on the OC origin of these lakes. In thermokarst-modified alas-lakes, permafrost-derived carbon contributed to 41 – 75% of DOC and 3 – 31% of POC (median values, Fig. 4, Table S4), which is much higher than in unaffected alas-lakes. The radiocarbon content reported here for thermokarst-modified alas-lakes OM highlights that even alas-lakes, generally considered as “inactive” (no more thermokarst degradation), exhibit strong changes in OM origin and properties when experiencing active thermokarst along their banks. Studies on the impact of thermokarst on streams showed similar results13,55. OM characteristics of our studied lakes are a blend of those found in alas-lakes (high primary production) and permafrost meltwater (very high concentrations of old DOC), with their variability appearing to be related to the duration of degradation. The highest permafrost contribution was observed in a lake where thermo-erosion gullying had been active for a longer time (a few years).

Hydrologically connected alas-lakes had low Yedoma-derived carbon inputs (median values respectively more than 91 and 86% of modern DOC and POC; Fig. 4; Table S4) and showed river-like functioning. In our study reporting baseflow (end of the summer) conditions, DOC originated mainly from modern terrestrial vegetation, indicating low Yedoma-derived DOC inputs even during baseflow conditions (F14C between 0.951 and 0.991). In all large Arctic rivers, including the Lena River, sampled DOC originates mainly from young terrestrial vegetation and soils; permafrost-origin DOC exports are very limited, due to rapid mineralization by microorganisms15,19 or photo-oxidation20 and occur during the baseflow when more groundwater feeds the river56,57,58,59. The POC composition we report is indicative of important autochthonous production. POC was older (F14C between 0.913 and 0.916) than DOC but its modern fraction was not as low as what has been reported in the Lena River (Δ14C between −200 to −600‰ with a mean at −327 ± 71‰ in summer corresponding to F14C about 0.67 ± 0.0760). This could be explained by the lower discharge of the local stream, generating less erosion. Moreover, there has been no active RTS that could provide old POC along the banks of these connected alas-lakes since the 2010s37. To explain the concomitant autochthonous characteristics and not completely modern F14C values, we suspect the recycling of old OM as previously reported in the Lena River basin47. In Arctic rivers, it has recently been shown that 39 – 60% of the POC originates from autochthonous primary production, the remaining derived mainly from allochthonous inputs60. In the Lena River, 25% of POC originates from Yedoma60.

In our study in Central Yakutia, we report much higher DOC concentrations than previously measured in Arctic lakes. In a synthesis of summer DOC concentrations in Arctic lakes61, the highest concentrations for Yedoma lakes were 50.6 mg.L-1 with a median value of 11.8 mg.L-1, which is much lower than our observations (median value of 158 mg.L-1). Isolated high DOC concentration (1 130 mg.L-1) have already been reported from lakes of interior Alaska, in a non-Yedoma area covered by eolian deposits but with low ground-ice content61.

For our Yakutian lakes, high concentrations of DOM suggest important transfers from the banks, in line with the high DOC concentrations in permafrost pore water in the area62, and concentration processes linked to evaporation and low connectivity with the hydrographic network. Indeed, in Central Yakutia, evapotranspiration in summer is higher than precipitation42, causing overconcentration of dissolved elements. Another hypothesis is that the lack of connectivity could prevent flushing of OC to rivers during the spring and lead to accumulation of DOC in recent lakes61, which could also be applied to alas-lakes and thermokarst-modified alas-lakes in our case because they are hydrologically isolated. These two processes could explain the very high concentrations of DOC in fall and summer in unconnected lakes, and the lower concentrations and lower seasonal amplitude in connected alas-lakes, which are hydrologically connected to rivers and are less influenced by evaporation due to their larger volume.

Thawing of Yedoma permafrost releases considerable amounts of DOC with a specifically low aromaticity and molecular weight, indicative of high biolability as shown in other studies14,15,56. In rivers, previous studies could not trace permafrost-derived DOC downstream of Yedoma thaw, because of a preferential mineralization of permafrost-origin DOC by microorganisms15. However, for our lakes in Central Yakutia, despite characteristics of highly biolabile DOM, substantial concentrations of old DOC were detected in recent lakes and in thermokarst-modified alas-lakes. This suggests that the rate of old DOC input to lakes is higher than the mineralization rate (by both biodegradation and photodegradation). We hypothesize that permafrost-derived DOM is not, in contrast to rivers, the most labile source of DOM for micro-organisms, and therefore can accumulate. We suggest primary production provides more biolabile OC to microbial communities in lakes.

In the studied Yakutian lakes, DOM originating from permafrost partly contributes to ebullitive CO2 emissions (no data on diffusive GHGs origin), whereas CH4 emissions are more likely fueled by modern macrophyte remains deposited in the lake bottom as a labile and rapidly degradable source of OC for methanogens (Fig. 6; refs. 33,52). In contrast, most of the previous studies on thermokarst lakes in areas of Yedoma permafrost or ice-rich Holocene permafrost showed older diffusive and ebullitive CH4 compared to CO2, corresponding to the influence of taliks that fuel CH4 production with old permafrost OM at the bottom of the lakes23,24,25,26,30,31. However, older CO2 compared to CH4 was previously found in thermokarst ponds with no taliks where CH4 also originated from modern aquatic vegetation and algae31. We also found a spatial pattern of GHG origin in lakes, with higher permafrost degradation influence in the lake center bottom, where it undergoes maximal subsidence, compared to lake shores. Similarly, the same study hypothesized for CH4 that the age can be younger next to the banks because the submerged deposits are younger, unlike in the deeper lake center31.

Previous study of diffusive GHG emissions from the same lakes during 4 seasons between fall 2018 and summer 2019 revealed strong seasonal patterns and notable differences between lake types36. All lakes were super-saturated in CH4 relative to the atmosphere during the ice-free seasons36. As we showed, CH4 in most of the bubbles seems to be modern in our lakes, so these emissions should have almost no impact on the PCF. For CO2, all lakes except alas-lakes were strong emitters during ice-free seasons36. With part of their CO2 bubbles (and even CH4 where the subsidence is maximal) originating from Yedoma-derived carbon, recent lake emissions should contribute to the PCF. Moreover, the transformation of alas-lakes from carbon sink to source of old CO2 when modified by thermokarst processes on their banks should be non-negligible amplification of the PCF. However, as we do not report on the relative contribution of diffusive vs. ebullitive GHG fluxes, we are not able to assess the magnitude of the impact on the PCF.

In this area covered by Yedoma permafrost, recent lakes represented 15% of lakes by number but only about 1% of the total lake surface area (for summer 2019; ref. 35). Their number and area increased between 1989 and 2019 due to thermokarst and was positively correlated to air temperature35,38. With continued warming, permafrost thawing will extend, leading to more inputs of old carbon in the atmosphere. Moreover, we should consider not only carbon transferred from permafrost but also changes in ecosystems at the landscape scale. Here, recent thermokarst lakes replace taiga forests, which are carbon sinks (−3.44 tC/ha; ref. 63). Then, even if not entirely fueled by old carbon, recent thermokarst lakes are participating in the PCF, since they turn patches of the landscapes from carbon sinks to sources. Finally, precipitation is expected to rise in Central Yakutia43, so hydrological connectivity might increase, allowing inputs of not yet degraded old carbon in rivers, where DOC is not over-concentrated and could rapidly be transformed into GHGs.

The observed DOC- and POC-related processes in thermokarst lakes in Syrdakh are likely also occurring in thermokarst lakes throughout the whole Yedoma region. Considering that the Yedoma domain represents 12% of the terrestrial surface underlined by permafrost in the Arctic64, the implications for global carbon cycle can be considered at the circumpolar Arctic scale. In this study, we show that old, Yedoma derived DOM accumulates in thermokarst lakes, with the hypothesis that primary production in Yakutian lakes, supported by joint inputs of DOC and DON with old DOM inputs (Fig. S1a), provides even more labile DOM as a substrate for microbial activity. Yedoma deposits have been shown to be relatively nitrogen-rich compared to other features across the Arctic64. As nitrogen and other nutrients can be limiting factors to primary production65, we can expect less primary production in thermokarst lakes of the non-Yedoma region where degrading permafrost contains less nitrogen, and therefore higher rates of permafrost-derived DOM biodegradation.

Methods

Study site

The study site is the watershed of the Syrdakh Creek (about 3000 km²), a tributary of the Lena River in Central Yakutia (Sakha Republic), Eastern Siberia, Russia. It is located roughly 100 km north-east of the city of Yakutsk (62.5°/62.6°N ‒ 130.8°/131.2°E; Fig. 1). The regional climate is hyper-continental with air temperatures ranging between +20 °C in July and −40 °C in January, and an annual mean of −10 °C. Temperatures remain under 0 °C between October and April, leading to ice-cover development on lakes during this period. Increasing air temperatures have been recorded at Yakutsk meteorological station since 196666. Precipitations are low, about 230 mm/year43. The evapotranspiration rate is higher than precipitations in summer42.

This area is located in a large lowland interfluve between the Lena River in the west and the Aldan River in the north. It is underlain by 300 to 500 m thick continuous permafrost67, whose temperature is between −2 °C and −4 °C at 10 ‒ 20 m deep5. The uppermost permafrost is composed of syngenetic ice-rich Pleistocene sediment of various origins (fluvial, eolian, lacustrine; ~70 – 90% of ice by volume; refs. 1,5). This permafrost is referred to as “Yedoma ice-complex”, and is composed of three fluvial terraces: Tyungyulyu, Abalakh and Magan. The study area is located in the Tyungyulyu terrace, where 10–20 m thick ice wedges can compose 50% of the ground volume5. At Syrdakh, syngenetic ice wedges have been dated at 21 710 ± 680 BP68 and 26 570 ± 160 BP69 with organic remains, and at 21 380 ± 80 BP with trapped CO2 bubbles69. Another study estimated the Yedoma age from plant remains at 3.5 m depth to be 23 600 cal. BP (ref. 50, based on ref. 70). Values of OC content have been estimated to 2.1 ‒ 2.2% of TOC by weight in a Syrdakh outcrop in the Tyungyulyu terrace50, and of less than 2% on average in a 22 m thickness core at Yukechi site in the Abalakh terrace51.

Sampling

The watershed is characterized by a large number of lakes35. Four lake types have been identified in the area in terms of geomorphology, age, thermokarst process, and hydrological connection36:

  • Recent thermokarst lakes, referred to as recent lakes, formed by permafrost thaw since the 1950s, due to air temperature increase or anthropogenic disturbances (e.g., forest cutting), and which are not hydrologically connected to the Syrdakh Creek38,71,72. Their depth ranges between 1 and 5 m.

  • Alas-lakes, residual endorheic lakes in basins (alas) formed by permafrost thaw during the early Holocene5,73,74,75, hydrologically unconnected, and not affected by recent permafrost thaw. They are shallow, between tens of centimeters to 1 m deep for the alas-lakes we sampled in this study.

  • Connected alas-lakes, i.e. alas-lakes that are connected to the Syrdakh Creek35, up to 10 m deep.

  • Thermokarst-modified alas-lakes, i.e. alas-lakes unconnected to the hydrographic system, that undergo permafrost degradation (RTS, thermo-erosion gullies, recent lake spill) along their banks due to recent thermokarst (maximum 2 m deep).

Several lakes of each type have been sampled between 2017 and 2021, however we focus on those collected in August 2019, which corresponds to the end of the hydrologic year, with the lowest base flow and a deep active layer (~ 1.5 – 2 m). For seasonal study, samples were taken in September 2018, March-April 2019, May 2019 and August 2019, corresponding to fall, winter, spring and summer seasons (Table S1). The sampling occurred at 30 cm under the surface of the lake and a few meters from the shore. In winter, alas-lakes were frozen to the bottom, precluding sampling. Moreover, permafrost meltwater was collected in August 2019 from a pool inside an active RTS formed in July 2019, to have the permafrost end-member signature. Plant macrofossil remains were sampled from an actively thawing outcrop of Yedoma permafrost from the Tyungyulyu terrace and stored in sterile ziploc bags. Samples were taken twice at 1.5 m and once at 1.8 m deep, where the active layer was 1.4 m thick.

Sample analyses

For each lake, concentrations, 13C and 14C isotopes, and C/N ratio of DOC and POC were analyzed, as well as absorbance-related parameters of DOM and F14C of ebullitive GHGs (CO2, CH4) (see below). Total dissolved nitrogen (TDN), NO3-, NH4+ and major anions and cations were also quantified. DOC and POC in water samples from lakes were separated through a pre-combusted Whatman glass fiber filter with 0.7 µm pore diameter, using a vacuum pump.

Water samples for DOC concentration analyses were stored in 30 mL burned amber glass vials after acidification with HCl (11 N). DOC and TDN concentrations were measured by catalytic oxidation with a TOC analyzer (TOC-L, Shimadzu, Japan, in PAPC Toulouse). 13C of DOC was analyzed on water samples stored in 10 mL amber glass bottles by DOC catalytic oxidation at 100 °C followed by liquid chromatography-IRMS coupling (Silvatech, Nancy). Values are expressed in relative ratio (δ13C) versus the Pee-Dee Belemnite international standard in ‰. F14C of DOC was measured on ECHoMICADAS (LSCE, Gif-sur-Yvette) using the gas source dedicated to very small samples of some dozens of µg of carbon76 and coupled to an elemental analyzer through a gas interface system. Prior measurement, frozen-dried DOM is leached in the Sn capsule used for EA and freeze-dried again77,78. 14C activities are expressed in F14C79 and also in BP80 facilitating comparison with previous work. The same methodology as for DOC was applied to analyze radiocarbon from plant macrofossils from the permafrost.

GF/F glass fiber filters with POC were dried at 40 °C, acidified with HCl to remove inorganic carbon and dried again. Concentration of POC and particulate nitrogen (PN) were measured on filter punch with an elemental analyzer (Flash NA1112 Thermofisher, LSCE, Gif-sur-Yvette). Particulate inorganic nitrogen is assumed negligible in our water samples, so PON concentrations corresponded to analyzed PN concentrations. 13C of POC was analyzed on filter punch with mass spectrometer ThermoFiningan Delta + XP (LSCE, Gif-sur-Yvette). Values are expressed in relative ratio (δ13C) versus the Pee-Dee Belemnite international standard in ‰. F14C of POC was measured on filter punch with ECHoMICADAS (LSCE, Gif-sur-Yvette), either through the gas source or the solid source depending on the sample size. 14C activities are expressed in F14C79 and in BP80.

Major ions were quantified in water samples collected in lakes using a HPLC (Dionex, USA), a Dionex DX-120 analyzer for cations except NH4+ (Thermo Fisher Scientific, Toulouse, France), a Dionex ICS 1000 analyzer for anions (Thermo Fisher Scientific, Paris-Saclay, France) and a Dionex ICS-5000 + analyzer for NH4+ (Thermo Fisher Scientific, Toulouse, France) following recommendations from ref. 81 The limit of quantification was 0.5 mg.L−1 for Ca2+; 0.1 mg.L−1 for Mg2+; 0.05 mg.L−1 for F-, Br- and NO2-; 0.03 mg.L−1 for K+; 0.02 mg.L−1 for Cl- and SO42-; 0.01 mg.L−1 for Na+, NH4+ and NO3-. DON concentrations were calculated as follows [DON] = [TDN]-([NO3-] + [NH4+]) with all concentrations in mg.L-1. C/N ratio on DOM and POM were calculated respectively as [DOC]/[DON] and [POC]/[PON] ratio.

The Absorption spectrum of DOM was measured with an absorption spectrophotometer (UV-vis Light XT5 SECOMAM, PAPC Toulouse) with a 1 cm optical path. In 2017 and 2018 campaigns, Fe concentrations were measured and shown to represent less than 0.05% of the absorbance at 254 nm, using ref. 82 correction. Consequently, iron concentrations were not measured during the 2019 campaign, and absorbance values were not corrected from iron interference. Specific ultraviolet absorbance (SUVA, L.mg-1.m-1) was calculated as the ratio between absorbance value at the wavelength 254 nm (A254, no unit) and DOC concentration (in mg.L-1) multiplied by the optical path (in m). The E2/E3 (ratio of the absorption coefficient at 250 and 365 nm) parameter was calculated as the ratio of A250 and A365. SUVA and E2/E3 are respectively proxies for aromaticity and molecular weight83,84.

Ebullitive GHG samples were collected after water sampling, several meters away to the side, by gently poking the untouched lake bottom sediments and using submerged funnels (as in ref. 30) equipped with a 140-ml plastic syringe to retrieve the ascending bubbles. Samples were then transferred into 50 mL glass bottles with butyl rubber stoppers (bottles acid-washed, precombusted, helium flushed and vacuumed) for subsequent 14C dating. δ13C values of ebullitive GHG were measured on Gas Bench coupled with IRMS (Finningan Delta Plus, UC IRVINE) and 14C with AMS spectrometer (UC IRVINE) and expressed in F14C and BP80. Diffusive GHG sampling was also conducted during fieldwork and data are reported in a previous publication36.

Statistical analysis

As the data was continuous but non normal, we used the Kruskal-Wallis test to identify statistical differences of variables between the seasons and lake types. These tests were conducted with python code (version 3.9.13) using the kruskal function of the scipy.stats package. The Dunn post-hoc test was performed with the posthoc_dunn function of the scikit_posthocs package to compare data groups in pairs by season or lake types. For both tests, the hypothesis that the medians of data groups were similar was rejected if the p-value was under the threshold of 0.05.

Source appointment model

To evaluate the proportional contribution of permafrost derived OC in lakes, we used a Bayesian three-component mixing model. Multi-pole models with only one variable, in this case F14C, were previously used in environmental studies23,32,77.

We considered that three sources could provide OC to lakes: living aquatic or terrestrial plants (atmospheric), topsoil first centimeters eroded from lake banks, and deeper permafrost. It should be noted that in the area, the litter thickness is about ten centimeters, lying directly on silts, with almost no organic horizon. With available data, it was not possible to determine other sources. Readers need to take into account that endmembers were chosen with available data, a different endmember choice would affect the relative contribution of each source.

The atmospheric source endmember, representing living vegetation, corresponded to the atmospheric signature in 2019, i.e. 1.011 ± 0.001 (NH1 zone, ref. 85). The topsoil first centimeters that can be eroded from the lake banks correspond to larch needles that are expected to last ten years, i.e. 1.043 ± 0.002 (NH1 zone, ref. 85). For the permafrost source, based on the assumption that permafrost meltwater inputs can come from various depths, the permafrost pool was therefore considered as an average value between the maximum (upper permafrost layer, 0.284 ± 0.005, this study) and minimum (0.0366 ± 0.0007; ref. 69) permafrost F14C reported in Syrdakh, i.e., 0.160 ± 0.128.

We used the Stable Isotope Mixing Models (simmr 0.5.1.216) package, coupled to Just Another Gibbs Sampler (JAGS) package23, available in R studio, to estimate the relative contribution of different sources to the 14C signature of lake OC. The code performed a mass balance (Eq. 1), and potential contribution (p1, p2, p3) of each source was assessed through Monte Carlo chains, exploring all possible combinations that may result in the observed sample radiocarbon signature. The contribution of each source was ≥0 and their sum was 1. The uncertainty of each potential source was propagated, and median source contribution was expressed within a 95% confidence interval (Table S4). We configured the model with four chains for greater reliability and 100,000 iterations, allowing for model convergence of all samples. The model was considered trustworthy when the Gelman-Rubin convergence index was 1.00 ± 0.10 23. The same endmembers and model parameters (chains, iterations) were used for all samples for consistency. We did not apply a preferential source contribution.

$${{{{\rm{F}}}}}^{14}{{{{\rm{C}}}}}_{{{{\rm{sample}}}}}={{{{\rm{p}}}}}_{\,1}{* \,{{{\rm{F}}}}}^{14}{{{{\rm{C}}}}}_{{{{\rm{atmosphere}}}}}+{{{{{\rm{p}}}}}_{2}* {{{\rm{F}}}}}^{14}{{{{\rm{C}}}}}_{{{{\rm{topsoil}}}}}+{{{{\rm{p}}}}}_{\,3}{* \,{{{\rm{F}}}}}^{14}{{{{\rm{C}}}}}_{{{{\rm{permafrost}}}}}$$
(1)

Reporting summary

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