Ice cores were collected from BY in a transect roughly perpendicular to the shore (Fig. ). Ice growth slows when heat flux to the atmosphere decreases in spring and is negligible by the end of April at our study site. Ice was cored at close to its maximum thickness to include almost the entire winter record of freezing. Ice cores were drilled on 8 April 2017 (LK), 10 April 2017 (LG) and 11 April 2017 (BY). For LG, two cores were recovered near the centre of the lagoon. The sites of the five cores from LK were located along an approximately west–east transect across the lake. Mean water body ice thicknesses measured during ice core sampling lie within 0.22 m of each other (Table ).

The ice cores were taken with a Kovacs Mark II ice coring system (0.09 m diameter), between 5 and 12 April 2017. Cores were collected in triplicate from each sampling site. One core was only used for temperature measurements. Immediately after sampling, vertical temperature profiles were obtained by drilling small holes every 0.1 m, at the boundaries between sample sections, and inserting a digital thermometer. Average temperature values were calculated for each section from the top and bottom depths. The other cores were wrapped in sealed plastic tubes and packed in thermally insulated boxes for transport in a frozen state to the climate-controlled room where sub-sampling took place. Water column temperature and electrical conductivity (EC) were measured with a SonTek CastAway instrument at each coring site. The accuracy and resolution of the device were ±0.05 and 0.01 ∘C, respectively, for temperature and ±5 and 1 µScm-1 for EC. Further details on fieldwork and sampling can be found in the expedition report .

Cores were stored in freezers after drilling and in a permafrost tunnel (lednik) while waiting for transport, and then they were transported by refrigerated truck at freezer temperatures (-18 ∘C). We did not record temperature during transport, but based on a comparison of photos of the cores after drilling and in the laboratory the structure and ice morphology of the cores were preserved during transport. We are therefore confident that not even surficial melting took place. The ice cores were cut with a band saw in a climate-controlled room at -15 ∘C. Cores were cut at 0.1 m intervals, and samples were stored for melting in gas-tight TEDLAR bags at 4 to 7 ∘C for 1–2 d. The closed bags were evacuated with a vacuum pump prior to melting, which effectively sealed the inner bag surface to the ice. After melting, the water was gently mixed and poured through tubing, without producing bubbles, into 100 mL glass bottles for the analysis of CH4 concentrations and δ13C in CH4. The remaining water was distributed into other sample bottles for hydrochemical measurements of pH, EC, dissolved organic carbon (DOC), δ18O and δD isotopes of water, as well as major anion and cation concentrations. Data points in the graphs of variation against core depth thus represent a mean value for a section of usually 0.1 m within an ice core.

EC and temperature of LK and LG were measured in the field. EC and pH of ice core samples were measured with a WTW Multilab 540 device directly after bottling. The salinity was calculated from the values of the EC referenced to 25 ∘C after . The samples of DOC were filtered with 0.7 µm pore size glass fibre filters (the filters were first rinsed with 20 mL of the sample), filled in 20 mL glass-headspace vials and closed with aluminum crimp caps. For preservation, 50 µL of 30 % supra-pure HCl was added to the sample before closing the vials, which were stored at 4 ∘C until measurement. DOC was measured with a Shimadzu Total Organic Carbon Analyzer (TOC-VCPH). An average of three to five injections per sample was used as the measured value. The detection limit for the DOC measurement is 0.25 mg L-1, and the uncertainty of the measurement was ±10 % for values higher than 1.5 mg L-1 and ±15–20 % for values lower than 1.5 mg L-1.

To measure stable water isotopes (δ18O, δD), 10 mL of the untreated water sample was filled in 10 mL PE narrow-neck bottles. Seven samples with EC higher than 3 mS cm-1 were measured with an Isotope Ratio Mass Spectrometer (IRMS: Finnigan Delta-S), using equilibration techniques . All other samples had lower EC and were measured with an Ultra High-Precision Isotopic Water Analyzer (PICARRO L2130-i, coupled with an autosampler and vaporizer using cavity ring-down spectroscopy). The internal precision of the H and O isotope measurements is better than ±0.8 and ±0.10 ‰, respectively. The oxygen and hydrogen isotopic compositions are given relative to Vienna Standard Mean Ocean Water (VSMOW) using the conventional δ notation.

Stable water isotopes have been widely used in palaeoclimate and palaeohydrological research as isotope fractionation is temperature-dependent. The mean annual δ18O of precipitation is positively correlated with the mean annual air temperature, and hence snow typically has a strong isotope variability as well as relatively low (or light) values, particularly at high latitudes . Stable water isotopes can also be used to trace water phase changes, i.e., prior to freezing and during freezing as these are accompanied by kinetic isotope fractionation processes . The comparison of δ18O to δD provides valuable information about the freezing characteristics, as the regression line (freezing line) is usually lower than that of the global meteoric water line (GMWL, slope =8). Slope values in the range of 6 to 7.3 can be interpreted as freezing under equilibrium freezing conditions . The extent of fractionation in the water–ice system critically depends on the velocity and rate of freezing , which in turn is directly connected to the thermal conditions and the water availability of a given system. Isotopic fractionation during freezing is accompanied by heavier isotope composition for the first ice and lighter isotope composition for the last ice formed . In this study, we differentiate between open- and closed-system freezing. In an open system, the water source under the ice and hence the isotope composition of the ice formed both remain largely constant, differing from closed-system freezing where the isotopic composition of the water pool changes as freezing progresses. Furthermore, the water isotopic signature may be indicative of the mixing of different water masses (endmembers), i.e., precipitation with surface water. The isotopic signature is then indicative of the relative contribution of each endmember and preceding natural isotopic fractionation processes, which changes if the endmembers' specific isotope composition differs. The δ18O-δD values in ice may also be directly influenced by precipitation if liquid or solid precipitation falls on the ice layer and freezes to a part of the ice. In a co-isotope plot, δ18O and δD of precipitation generally have values that lie on or near to the GMWL .

Meltwater from the TEDLAR bags was filled until overflowing into 100 mL glass bottles, sealed with butyl stoppers and crimped with aluminum plugs. The samples were kept cool (4 ∘C) and dark until the measurements in the lab (max 2 months between sampling and measurement). For CH4 concentration, 5 mL of N2 was added into the vials, whereby 5 mL of water was removed from the vials. Afterward, the water sample equilibrated with the N2 by shaking for 1 h at room temperature. Then, 1.5 mL of the equilibrated air was removed from the vial and injected into a gas chromatograph (GC; Agilent 8900) with a flame ionization detector (FID). For gas chromatographic separation, a packed column (Porapak Q 80/100 mesh) was used. The GC was operated isothermally (60 ∘C), and the FID was held at 250 ∘C. Standards of different gas mixtures between 1.665 and 100 ppm that were used for calibration yielded standard deviations of <0.3 ppm. The GC precision had an error of 1 %. Forty-six percent (n=74) of the samples had CH4 concentrations below the lowest standard of 1.665 ppm (ranging from 0.7 to 1.6 ppm), while 11 % (n=18) had concentrations exceeding the highest standard (ranging from 105 to 2723 ppm).

The carbon isotopic composition of CH4 (δ13CCH4) was measured on the same day and from the same bottle as the CH4 concentrations, to assure comparability of the data. After measuring the CH4 concentration, 20 mL of N2 was added to the sample bottle to increase the headspace of the bottle for stable carbon isotope measurements. The bottle was shaken for at least 30 min. Twenty millilitres of gas was removed with a glass syringe by adding 20 mL of Milli-Q water at the same time to equilibrate pressure. δ13CCH4 was determined using a Finnigan Delta plus XP mass spectrometer. The extracted gas was purged and trapped with PreCon equipment (Finnigan) to pre-concentrate the sample. The carbon isotopic ratios are given relative to the Vienna Pee Dee Belemnite (VPDB) standard using the conventional δ notation. The analytical error of the analyses is ±1.5 ‰ for δ13CCH4 values. CH4 concentration in nanomolar (nM) was calculated with the Bunsen solubility coefficient of .

For all water bodies, a Rayleigh distillation model of the type discussed by and used by was calculated: 1δ13CCH4=10001α-1ln⁡f+δ13CCH40, where α is the isotope fractionation factor, f is the fraction of the CH4 remaining and (δ13CCH4)0 is the initial isotopic composition. For the Rayleigh model, bacterial oxidation was assumed to be the only CH4 sink, with no further inputs or mixing that would affect the isotopic composition of CH4.

To gain insight into the type and spatial distribution of CH4 bubbles trapped in the ice of LK, a CH4 bubble transect was mapped. Snow was cleared from a lake ice transect of 70 m length and 1.4 m width, starting 30 m from the northwest shoreline aiming towards the lake centre (Figs. , ). A GoPro Hero 7 camera was used to take overlapping photos of the transect from approx. 1.7 m vertical height along the entire transect to photographically record the CH4 bubble patches captured in the lake ice that are associated with seep types. A measurement tape at the side of the cleared transect area served as a scale in the images and to measure the sizes of seeps and bubble types. The images were orthorectified and mosaicked in the image rectification software AGISoft Professional, and the georeferenced mosaic was imported to a desktop geographical information system (ArcGIS, version 10.4). CH4 seep types were classified and mapped following . The distance to the shoreline of seep classes was calculated in ArcGIS using the “near” function. We chose 14 distance classes in a kernel density estimation guided by two criteria: (1) only allow bin size in which each bin is represented by seep data, and (2) maximize visualization of the density trend over the profile. The kernel density estimation was calculated with R (a free software environment with interpreted computer language for statistical computing and graphics, https://www.r-project.org/, last access: 15 February 2021).

We compared the variability of measured parameters within each ice core as a function of depth below the ice surface and between sets of cores from each water body. For a simple comparison between ice cores and the three water bodies, mean values and the range (min and max values) were calculated for every location (Table ). The length of the ice cores ranged from 1.1 to 1.97 m, corresponding to total ice thicknesses. Mean lengths were 1.44 m in BY, 1.66 m in LG and 1.6 m in LK. The ice of all cores from BY was nontransparent, with a high concentration of enclosed gas bubbles. The uppermost 0.03 m (BY-4) to 0.1 m (BY-2) was identified as regelation ice from snowmelt based on the appearance of the ice during sampling.

On the ice of BY, the snow thickness ranged from 0.11 m (BY-4) to 0.32 m (BY-2). For LG, the ice of the two cores was clear and contained inclusions of elongated gas bubbles. The uppermost 0.03 m of core LG-1 appeared milky white. The difference in snow thickness above the two cores was quite large: 0.08 m (LG-2) and 0.2 m (LG-1). For LK, the ice morphology was heterogeneous. LK-2 appeared transparent white, while LK-1 included small, elongated air bubbles, and the uppermost 0.05 m was regelation ice from snowmelt. The ice of core LK-4 appeared whiter below a depth of 1.11 m, while LK-5 was milky white from the surface down to a depth of ca. 1.12 to 1.14 m. The snow cover was hard-packed and characterized by different melt forms. Snow was either absent (LK-2 and LK-3) or 0.15 m (LK-4), 0.23 m (LK-1) and 0.92 m (LK-5) thick. None of the cores contained algae inclusions .

In the total snow-cleared transect area of 89.2 m2 we found 29 seeps of class A and 5 seeps of class B (Fig. ), but none of the larger class C or hot spots . The average density of class A seeps in the observed area was 0.33 seeps m-2. Seep density of class B was more than 6 times lower (0.05 seeps m-2), and the total seep density over the transect (classes A+B) was 0.39 seeps m-2. The distribution of seep density along the transect showed no linear or homogeneous pattern.

Tikski Bay is part of Buor Khaya Bay and via the central Laptev Sea perennially connected to the Arctic Ocean. The marine setting delays the onset of ice formation compared to the terrestrial aquatic system shown by satellite images from ESA Sentinel-1 and Sentinel-2. In addition to seawater, snowmelt in spring, small coastal catchments and the Lena River enter into BY south of Cape Muostakh. Lena River discharge follows a nival discharge regime, with very high discharge in the spring and early summer months , while in winter, when the connection between Buor Khaya Bay and BY is restricted by sea ice, the contribution of Lena discharge must be much smaller. When BY freezes, it is supposed to be a continuously open system; i.e., water exchange is ongoing during winter below the ice. Both aspects (open system and the mixing of fresh and brackish water) are corroborated in the stable water isotope composition of the ice cores. A mean δ18O value of -15 ‰ for BY is well below full marine conditions and displays the continuous and strong influence of freshwater supply through the Lena River to the Laptev Sea.

Firstly, in an open system such as BY, the water circulates freely beneath the ice cover. The isotope composition of the ice should remain more or less constant over the winter and reflect the fractionation resulting from the freezing of Tiksi Bay water. Accordingly, the water isotopic composition and salinity values are constant until the ice is approximately 0.8–0.9 m thick (Figs. , ). The small variation at the top might be due to variability in isotopic fractionation, i.e., related to a change of the freezing rate after ice formation started. Oxygen isotope fractionation during the freezing of seawater has been addressed by through laboratory experiments and field observations. These authors demonstrate a general dependency of increasing isotope fractionation with decreasing ice growth rate. Therefore, faster freezing induces less isotope fractionation when compared to slowly formed ice at a later stage of sea ice formation.

Secondly, BY is influenced by marine and river water, the ratio of which affects δ18O and δD captured in the ice. The advection of mixed water below the ice imprints on the freezing process. Hence, water mixing is likely reflected in the small shift towards -17 ‰ in δ18O and -130 ‰ in δD in the lower part of the ice (>0.9 m) (Figs. , and ), even though the δ18O-δD data behave like isotopic fractionation in a closed system, where the last ice that formed displays the lightest isotope composition. The change of the water isotope composition without a change in the d-excess, as well as changing EC and DOC, suggests a substantial contribution of another water mass and indicates the reduction of marine influence or an increased river water impact in the lower part of the ice compared to the upper part. Reduced EC concentrations corroborate the enhanced freshwater content in the bottom segments of the ice cores, further substantiated by lower δ18O and δD values. Overall, the upper part of the BY ice cores indicates freezing from brackish water, whereas the lower part may capture a slightly higher proportion of river water, probably reflecting changes to the isotopic and hydrochemical composition of BY. This is reflected in the regression lines of the δ18O-δD plot for BY, which differ in slope for the two sections (Fig. ) and indicate a shift in fractionation . Episodic advection of meteoric water during the winter season was also detected in landfast sea ice cores from Barrow, Alaska . The samples of the upper portion of the ice plot along a regression slope of 6.9, demonstrating open-system freezing under equilibrium conditions . In contrast, the lower portion of the ice plots along a regression line of 7.9 (Table ), close to the slope of the GMWL of 8. Lena River water from much further south is carried into Tiksi Bay throughout the year and plots close to the local meteoric water line slope 7.3, . In April this flux is still at base flow levels but contributes to the least dense surface water layer beneath the ice. Hence, we assumed this portion of the ice to be mixed with water of meteoric origin. The change in the water isotopes between the upper and lower ice was not related to closed-system freezing, as this would also be indicated by a deviation from the GMWL and a change in the d-excess.

CH4 concentrations in BY ice are small but still supersaturated relative to the atmospheric background concentration and probably the result of freezing of CH4-supersaturated brackish water in BY. CH4 supersaturation has been reported for the Buor Khaya Bay , and large amounts of CH4 are stored in sediments of this region . Supersaturation in the shallow water of BY is probably caused by CH4 release from sediments , which likely occurs in the fall when cooling induces convection and resuspension events . In comparison, during calm sub-ice winter conditions resuspension events cease and CH4 concentration decreases in the brackish water from which the ice forms, shown by the slightly reduced concentration in the lower part of the ice (>0.9 m) (Fig. 5). The slight decrease in concentration with increasing distance from shore and a homogeneous carbon isotopic signature also point to calm water conditions during freezing and a local source of CH4. Very similar processes during freezing for all BY cores are corroborated by the homogeneous (δ13CCH4) carbon isotopic signature along the onshore–offshore gradient.

LG is connected to BY through a wide, shallow and winding channel. Storm surges and high-water events in the summer force seawater, driftwood and sediment into LG. ESA Sentinel-1 and Sentinel-2 imagery indicate that the ice on LG started to form in mid-October, and hence earlier than at BY. During the early freezing period, the connecting channel allowed water exchange between LG and BY until the growing ice reached the channel bed (less than 0.5 m deep), where it exits LG in 2017. At that point, LG changed from a connected to an isolated system, affecting water chemistry below the ice. The timing of closure, interpreted from the change of δ18O and δD values with depth, corresponds to an ice thickness of 0.6 m. At the time of coring the ice thickness on LG was about 1.6 m. The ice on more than 75 % of the lagoon's area was probably be frozen to the bed . In the lagoon centre, ice was floating and the lagoon had a talik, although it contained a thin ice-bearing layer 4.8 to 8.3 m below the sediment surface, .

For ice that formed before the lagoon was separated from the sea (above 0.6 m), the isotopic signature indicates freezing under non-equilibrium conditions , with a slope of 8.2 between δ18O and δD (Fig. , Table ). As the freezing was influenced by mixing of the initial water with a different water mass (water from the bay), probably no equilibrium freezing occurred. Below 0.6 m, however, isotopic values decrease, indicating freezing under closed conditions. A slope of 6.5 (Fig. , Table ) suggests freezing under equilibrium conditions . This freezing process is influenced not by a changing initial isotope composition of water but by isotopic fractionation during freezing. Freezing leads to a lighter isotope signature of the remaining water, as the heavy isotopes are preferentially incorporated into the ice . This results in lighter isotope composition and an increase of d-excess in deeper ice (below 0.6 m) compared to the upper ice. The concentrations of dissolved constituents (indicated by EC and DOC) also increase with increasing ice thickness (Fig. ) and corroborate this interpretation.

Thermokarst lake basins that are transformed into thermokarst lagoons are increasingly affected by seawater, at least intermittently, during high-water events such as storm surges, resulting in changes to their temperature and salinity regimes . Increasing salinity in turn also affects subsurface permafrost thaw dynamics and may therefore result in different CH4 production rates . In the uppermost ice – i.e., when freezing started – CH4 concentration in LG water was more than 10 times higher than in BY. The isotopic signature was in the range of microbially produced CH4 (-70 to -80 ‰) . This large excess of δ13C-depleted CH4 clearly points to CH4 from the unfrozen talik, released into the water body and stored therein during the ice-free season. When LG switched from a connected to a closed system at around 0.6 m ice thickness, ongoing CH4 oxidation beneath the ice lowered the CH4 concentration captured in the ice to ≤10 nM, comparable to those in BY ice. At the same time, the CH4 isotopic signature became comparably enriched in δ13C (Figs. , ). Ongoing ice formation under closed-system conditions (below 0.6 m), as indicated by stable water isotopes, induced a continuous increase in ice salinity (Fig. ), which in turn favoured the shift of the horizon where CH4 oxidation could occur from the water to the bottom of the ice. This is corroborated by the Rayleigh fractionation curves calculated for ice that grew under closed conditions, using as initial CH4 isotopic signature the uppermost value, i.e., the signature when freezing began (Figs. , ).

In addition, the temperature increased towards the bottom of the ice (Fig. ). The bottom ice offers a protected environment with favourable conditions for microbial metabolism: relatively warm temperatures, contact with liquid water and permeable ice. The latter permits migration of gases and nutrients, similar to marine ice, where most bacteria are located in the lowest centimetres of the ice . At LG, the bottom ice temperature decreases during the winter. This occurs because the temperature of the underlying water remains in equilibrium with a dynamic freezing point that decreases with increasing salinity when LG is cut off from Tiksi Bay. The ice surface temperature at the time of coring was primarily a function of snow cover and air temperature. The ice coring locations for LG exhibited colder ice surface temperatures and steeper gradients compared to ice coring locations at LK. Ice has a high thermal conductivity and is susceptible to quick temperature changes. Since ice temperatures were also observed for windswept areas at LK, decreasing air temperatures from 8 April 2017 (final LK coring day) to 11 April 2017 (LG coring day) explain the generally colder ice temperature profiles at LG. How salty the sub-ice water becomes is probably highly variable from year to year and will depend on how much seawater is forced into the lagoon during the fall storm season.

During freezing of the ice cover, its growth rate decreases cf., providing more time and space for bacterial metabolism. CH4 uptake from the water into the bottom of the ice and its oxidation there may have continued over the winter until the ice break-up. CH4 oxidation ceases when concentrations are too low for oxidation to be efficient , at values ranging from 0.6 to 10 nM. CH4 concentrations in the ice above 1.3 m (Fig. ) were less than 10 nM, suggesting that ice is an effective sink for CH4 in this type of water body during winter.

The ice core hydrochemistry from LK reflected a water body that froze in equilibrium with atmospheric CH4 concentrations, with two cores that showed the influence of snow loading and of ebullition.

LK is an isolated thermokarst basin surrounded by ice-rich Yedoma uplands to the west and east and partially degraded Yedoma uplands to the north and south. The lake is underlain by continuous permafrost some hundreds of metres deep and has a thaw bulb (talik) beneath its bed due to the positive temperatures at the lake bottom. Water in LK derives from precipitation, most of which falls as snow, and reaches the lake either directly, via overland flow, or via groundwater flow through the shallow active layer. Thus, the concentration of dissolved constituents remains small in lake water and ice, as reflected in very low electrical conductivities of less than 50 µScm-1 (Fig. ).

Water depths range from <1 m to about 8.5 m from the western to the eastern shore, respectively (Fig. ). This asymmetric shape influences the progress of ice growth in winter. Ice formation typically starts from the lake shore, most likely at the very shallow west shore, and leads to bedfast ice formation at the position of LK-1, i.e., lake ice frozen to the lake bed. Hence, it is likely that LK-1 began to form earlier in the season and that the upper ice in this core, and probably also in LK-4, reflected the chemistry of the summer/autumn lake water. All cores except LK-5 had similar δ18O (around -16.5 ‰) and δD (-140 ‰) values, with a regression slope of 6.6 (Fig. , Table ), pointing to freezing under equilibrium conditions . At this location both the CH4 concentrations and the uniform δ13C signature indicate equilibrated values with respect to the atmospheric background values (Fig. ). Our data demonstrate that lake water was not supersaturated during the ice-free season. The other two cores show clear differences and are described in the following.

For LK-5 stable water isotopes in the upper 1.2 m follow the GMWL (Fig. ), indicating equilibrium fractionation (in the vapour–water system, as snow is involved) due to different precipitation sources. δ18O values as low as -28 ‰ indicate snow as a source. The proximity of snow samples to the GMWL is typical for northern Siberia and has been also found at snow patches on the Bykovsky Peninsula . LK-5 lay close to LK's steepest shoreline, where active thermo-erosion results in shoreline retreat, and that lies in the lee of prevailing winds , where deep snow is expected to accumulate and load the lake ice. Snow of 0.92 m thickness lay on the ice at this location at the time of drilling, and water streamed out of the hole after coring , indicating an ice cover under positive hydrostatic pressure. The ice of LK-5 was milky white from the surface to about 1.12 to 1.14 m depth, consistent with a mixture of snow and water. Thus, we conclude a snow signal evident from the stable isotope composition in the upper 1.2 m of LK-5. The higher EC and DOC in the same interval (relative to lower ice and to the other LK cores) are unlikely to have derived from snow, however, and imply heterogeneous ice development above 1.2 m. These may be the result of mixing of uprising lake water with snow within ice cracks. describe the formation of white ice (or snow ice) by water percolation through thermally induced cracks to the surface of the ice, where the water mixes with snow and forms another ice layer above the former ice. This was observed by when a snow load depressed the surface of the ice cover below the hydrostatic water level. The high and highly variable CH4 concentrations over this interval, together with the high EC and DOC, suggest that resuspension events, for example slope failure, occurred during ice formation. The carbon stable isotope signature in LK-5 may reflect microbial CH4 oxidation in the sediment rather than oxidation in or beneath the ice.

The highest concentrations of CH4 were found in LK-3 (up to 15 000 nM), with the lowest δ13C signatures (down to -150 ‰, Fig. ). LK-3 was drilled above the steepest portion of the lake bed, where ebullition may be a by-product of thermokarst processes beneath the bed. The 12C-enriched signal is consistent with CH4 that has not been oxidized in the sediment and is released in gas bubbles. The earlier release of the lightest CH4 may be the result of fractionation during migration in the sediment and may be an indication the ebullition and ice formation are related, i.e., that ebullition was initiated as a result of pressure changes owing to the onset of freezing.