Limnol Oceanogr 9999 2020 1ndash15copy 2020 The Authors Limnology and Oceanography published by Wiley Periodicals LLC on

behalf of Association for the Sciences of Limnology and Oceanographydoi 101002lno11560

Carbon emission from thermokarst lakes in NE European tundra

Svetlana A Zabelina1 Liudmila S Shirokova12 Sergey I Klimov1 Artem V Chupakov 1 Artem G Lim3

Yuri M Polishchuk45 Vladimir Y Polishchuk67 Alexander N Bogdanov4 Ildar N Muratov4

Frederic Guerin 2 Jan Karlsson 8 Oleg S Pokrovsky 21N Laverov Federal Center for Integrated Arctic Research IEPS Russian Academy of Science Arkhangelsk Russia2Geoscience and Environment Toulouse UMR 5563 CNRS University of Toulouse IRD Toulouse France3BIO-GEO-CLIM Laboratory Tomsk State University Tomsk Russia4Ugra Research Institute of Information Technology Khanty-Mansiysk Russia5Institute of Petroleum Chemistry SB RAS Tomsk Russia6Tomsk Polytechnic University Tomsk Russia7Institute of Monitoring of Climatic and Ecological Systems SB RAS Tomsk Russia8Climate Impacts Research Centre (CIRC) Department of Ecology and Environmental Science Umearing University Umearing Sweden

AbstractEmission of greenhouse gases (GHGs) from inland waters is recognized as highly important and an understudied

part of the terrestrial carbon (C) biogeochemical cycle These emissions are still poorly quantified in subarcticregions that contain vast amounts of surface C in permafrost peatlands This is especially true in NE Europeanpeatlands located within sporadic to discontinuous permafrost zones which are highly vulnerable to thaw Initialmeasurements of C emissions from lentic waters of the Bolshezemelskaya Tundra (BZT 200000 km2) demonstratedsizable CO2 and CH4 concentrations and fluxes to the atmosphere in 98 depressions thaw ponds and thermokarstlakes ranging from 05 times 106 to 5 times 106 m2 in size CO2 fluxes decreased by an order of magnitude as waterbody sizeincreased by gt 3 orders of magnitude while CH4 fluxes showed large variability unrelated to lake size By using acombination of Landsat-8 and GeoEye-1 images we determined lakes cover 4 of BZT and thus calculated overallC emissions from lentic waters to be 38 065 Tg C yrminus1 (99 C-CO2 1 C-CH4) which is two times higher thanthe lateral riverine export Large lakes dominated GHG emissions whereas small thaw ponds had a minor contribu-tion to overall water surface area and GHG emissions These data suggest that if permafrost thaw in NE Europeresults in disappearance of large thermokarst lakes and formation of new small thaw ponds and depressions GHGemissions from lentic waters in this region may decrease

Arctic and subarctic lakes emit sizeable amounts of green-house gases (GHGs) to the atmosphere (Tranvik et al 2009Kosten et al 2010) High CO2 and CH4 fluxes are partly drivenby microbiological and physicochemical processing of labileorganic carbon released from thawing permafrost (vanHuissteden et al 2011 Vonk et al 2015) Because climatewarming and permafrost thaw may strongly enhance thisimpact (Pokrovsky et al 2013 Tan and Zhuang 2015

Bartosiewicz et al 2016) numerous works have measured car-bon emissions from inland waters in circumpolar territories(Rautio et al 2011 Langer et al 2015 Wik et al 2016) Inthese regions abundant thermokarst lakes are formed viathawing of frozen organic or mineral soils (Shirokovaet al 2013 Sjoumlberg et al 2013 Bouchard et al 2017) Thereare also instances where certain lakes disappear entirely due todrainage (Smith et al 2005) However there is significant biasin the geographical coverage of available field measurements(see Fig 1) The vast majority of seasonally resolved data onthermokarst lakes and ponds were collected in NorthernEurope (Lundin et al 2013 de Wit et al 2018 Kuhnet al 2018) Alaska (Sepulveda-Jauregui et al 2015 Stackpooleet al 2017) and Canada (Tank et al 2009 Laurion et al 2010Arsenault et al 2018 Matveev et al 2018) only sparse mea-surements were conducted in Eastern Siberia (Walteret al 2006 2007 Abnizova et al 2012 Walter Anthonyet al 2014 Tan et al 2017) Specifically as Matveev (2019)also note there are a limited number of C measurements in

Correspondence olegpokrovskygetompeu

This is an open access article under the terms of the Creative CommonsAttribution License which permits use distribution and reproduction inany medium provided the original work is properly cited

Additional Supporting Information may be found in the online version ofthis article

Special Issue Biogeochemistry and Ecology across Arctic Aquatic Ecosys-tems in the Face of ChangeEdited by Peter J Hernes Suzanne Tank and Ronnie N Glud

1

peatland thermokarst lakes as compared to those for yedoma-like permafrost

Recently a multiseasonal study with large geographic cover-age in western Siberian peatlands demonstrated high C emis-sions from both rivers and thermokarst lakes (Serikovaet al 2018 2019) While western Siberia can serve as a proxy forpeatlands in northern Eurasia the lack of data on continentalplanes with permafrost peatlands prevents the direct extrapola-tion of western Siberia lake GHG concentrations and fluxes toother regions Among large territories of frozen peatbogs accessi-ble all year round for sampling the Bolshezemelskaya Tundra(BZT) of northeastern Europe (area ~ 200000 km2) is especiallyinteresting as it contains a vast amount of thermokarst lakes(Goldina 1972 Davydova et al 1994) combines isolated spo-radic and discontinuous permafrost zones and exhibits a gener-ally milder climate and with a higher terrestrial biomeproductivity compared to western Siberia (Zamolodchikov andKarelin 2001) There is very limited information on GHG emis-sions from inland waters of the European Russian tundra CO2

and CH4 emissions have been assessed for a single lake and river

in the Vorkuta region (Heikkinen et al 2004) however seasonalvariations of CO2 emissions have been measured in threethermokarst lakes of this same region (Marushchak et al 2013)There is recent interest in the BZT as over past decades lakes inthis territory exhibited sizeable increases in summer temperatureand pCO2 presumably due to enhanced bacterial respiration ofallochthonous dissolved organic matter (DOM) from thawingpermafrost (Drake et al 2019)

In order to quantify GHG emissions from inland waters ofthe BZT we measured CO2 and CH4 concentrations and atmo-spheric fluxes in thermokarst lakes thaw ponds and depres-sions of various sizes during the three main hydrologicalseasons (spring summer and autumn) We further measuredthe size distributions and water surface areas of lakes for thefull BZT territory using remote sensing We aimed to (1) assessthe effect of waterbody size (from 1 to 105 m2) and seasonalityon C emissions and (2) quantify the overall emissions flux ofCO2 and CH4 to the atmosphere from lakes of the BZT terri-tory We believe assessment of GHG pools and exchangefluxes in thermokarst waterbodies of the NE European tundra

Fig 1 Map of BZT In the isolated and sporadic permafrost zone the thickness of permafrost is regionally variable and generally does not exceed 3 mA detailed map of research areas (shown by red circles 1ndash3) is provided in Supporting Information Fig S1 Geographical coverage of available GHG mea-surements (citations in the left panel) in thermokarst lakes is shown on the global permafrost map

Zabelina et al C emission from thermokarst lakes

2

combined with evolving lake sizes distributions and surfaceareas over past decades can allow forsight into potential futureC emissions to the atmosphere from lentic aquatic ecosystemsin northern Eurasia

Study site and methodsThermokarst lakes and thaw ponds of BZT NE Europe

BZT (NE Europe) is the largest frozen peatland in Europe Itis covered by permafrost discontinuous in the eastern partand sporadic to isolated in the western part (Brittainet al 2009) Peat thickness in the BZT ranges from 05 to 5 mdepending on the landscape context (Maksimova andOspennikov 2012 Iglovsky 2016) The dominant elevationhere is between 100 and 150 m these altitudes are comprisedof hills and moraine ridges which are mainly composed ofsands and silts with intermixed boulders Thermokarst lakesare located between these moraines and ridges In the south-ern forest-tundra zone the dominant soils are histosols of peatbogs and podzol-gleys The mean annual temperature isminus31C with a mean annual precipitation of 503 mm Thedominant vegetation in the tundra zone is mosses lichensand dwarf shrubs

Permafrost thaw in high latitude peatlands leads to forma-tion of ldquothermokarstrdquo (thaw) lakes due to frozen peat andmineral soil subsidence The formation and development ofthermokarst lakes has been investigated over the years andappears to follow a cycle (Matthews et al 1997 Hinkelet al 2003 Audry et al 2011 Kirpotin et al 2011 Pokrovskyet al 2011) The cycle starts with frozen peat subsidence cau-sed by warming and formation of a depression in mosslichencover that is filled by thaw water (Kirpotin et al 2011Pokrovsky et al 2011) The depression grows due to ongoingsubsidence of peat in the soil and instability of the edges untilit reaches lake size A mature thermokarst lake can eventuallybe drained into another lake or into the hydrological system(local streams and rivers) Gramineae-like vegetation colonizesalong the drained lake bottom followed by mosses andlichens Gradually the renewed permafrost heaves the soilcreating small isolated mounds that merge into a flat uniformpalsa plateau Ultimately the cycle can starts over againthrough formation of another wet depression This sequenceis an important part of the hydrological continuum in theBZT region and it strongly influences organic carbon andGHG dynamics in surface waters (Shirokova et al 2019)

Sampled waterbodies and basic physical and chemical char-acteristics and parameters of the CO2 system are listed inSupporting Information Table S1 For convenience all sam-pled waterbodies (generally referred to as ldquolakesrdquo) were classi-fied according to their surface area depressions (lt 10 m2)thaw ponds (10ndash1000 m2) and thermokarst lakes (gt 1000 m2)Altogether 48 thermokarst lakes 10 thaw ponds and40 depressions were sampled for hydrochemical parametersdissolved C CO2 and CH4 concentrations GHG emissions for

these bodies of water were calculated from CO2 and CH4 con-centrations and wind speed (see ldquoCO2 and CH4 flux measure-ments and calculationsrdquo section) One key site (~ 43 km SEfrom the town of Naryan-Mar) was selected for seasonal andannual monitoring of lakes with two additional sites(Shapkino and Khorey-Ver) to account for spatial heterogene-ity over summer (July 2016) these sites are shown in Fig 1and Supporting Information Fig S1a In order to reveal sea-sonal and annual variability of GHG concentrations andfluxes into the atmosphere in BZT depressions thaw pondsand thermokarst lakes we visited the Naryan-Mar site inspring (16ndash17 June 2017 sampling 7 lakes and 3 depressions)summer (17ndash23 July 2017 sampling 11 lakes and 13 depres-sions) and autumn (02ndash04 October 2017 sampling 11 lakesand 12 depressions) To account for interannual variabilitythree lakes from the Naryan-Mar site were sampled in July of2015 2016 2017 and 2018 These four sampled years wererather contrasting in mean monthly and annual temperatureand precipitation (see Supporting Information Table S2) Thesummer of 2015 was cold (minus48C below normal for July)while summers of 2016 2017 and 2018 were warm (+50C+40C and +25C respectively above normal for July) Thesummers of 2016 and 2018 were dry (31 and 36 of normalprecipitation for July) whereas the summers of 2015 and 2017were more humid (83 and 56 of normal precipitationfor July)

The spatial variability of dissolved organic carbon (DOC)and GHG concentrations was assessed in July 2016 A total of19 lakes thaw ponds and depressions were sampled in theShapkino site Another 33 lakes and ponds were visited in theKhorey-ver site Both of these sites were compared with sixlakes in the Naryan-Mar site sampled during the same period

Thermokarst lakes and thaw ponds in the BZT similar tothose in western Siberia (Shirokova et al 2013 Manasypovet al 2014 Serikova et al 2019) are mainly shallow (largelakes are only 05ndash15 m deep and small thaw ponds arelt 05 m deep) and are not stratified with respect to tempera-ture There are only a few exceptions where lakes are15ndash20 m deep but these lakes constituted less than 1 of allstudied (see Supporting Information Table S1) For simplicitywe refer to all waterbodies studied in this work as ldquolakesrdquoalthough their size ranges from lt 1 m2 to gt 7 km2

Lake size and area inventory using satellite imageryRemote sensing data utilized for this study include

Landsat-8 medium resolution (MR) and GeoEye-1 ultra-highresolution (UHR) imagery imagery were used to map distribu-tion of lakes throughout the BZT territory (Supporting Infor-mation Fig S1b) Landsat-8 images were taken in the late Julythrough early August 2017 and 2018 GeoEye-1 images werecollected in the summer months of 2011 through 2017 Thesesummer periods when the lakes are ice-free correspond tominimum coverage of the territory by lakes and minimumseasonal change in water level A combination of Landsat-8

Zabelina et al C emission from thermokarst lakes

3

images over 2 yr was necessary as observation over the courseof 1 yr could not provide full coverage of the territory

For the entire BZT territory affected by permafrost(195000 km2) we used a mosaic of 50 Landsat-8 MR imagesImages were processed using standard ArcGIS 103 softwaretools using the Fmask algorithm (see Polishchuk et al 20142017 2018 for method description) Given the sizable arealcoverage and accessibility challenges no ground calibrationhas been performed in BZT and we rely on ground truthingfrom former studies of western Siberia where there is a similardistribution of thermokarst lakes Taking into account the30 m resolution of MR imagery the minimum size of lakesexamined is limited to 5000 m2 This corresponds to six pixels30 times 30 m (900 m2) in size which is sufficient for reliable sep-aration of lakes from background noise The uncertainty ofmeasuring the surface area of a lake using Landsat-8 is 35for lakes gt 105 m2 surface area and 7 for lakes gt 2 times 104 m2

(Bryksina and Polishchuk 2013 Kornienko 2014) UHRGeoEye-1 imagery evenly distributed throughout the terri-tory were used to quantify the number and total areas ofsmall lakes and ponds For test sites we chose representativepermafrost peatland landscape similar to those sampled inthis work With a spatial resolution of 06 m pixel size inthese images is 036 m2 providing a minimal pond size of2 m2 GeoEye images were processed using ArcGIS 103 at11 test sites distributed throughout the study area

To plot the distribution of lakes in a wide range of sizes weused 23 partial size ranges of lakes taken on arithmetic pro-gression scale from 2 to 5 m2 5 to 10 m2 10 to 20 m2 and soon up to 108 m2 Note that remote sensing did not allow forquantification of the proportion of smallest depressions(lt 2 m2) However based on our previous work on similar per-mafrost peatlands in western Siberia the overall contributionof such small waterbodies to water volume surface area andGHG pools is negligible (Polishchuk et al 2018)

To combine the MR and UHR imagery we used a techniquedeveloped for thermokarst lakes in western Siberia (Polishchuket al 2018) First we constructed diagrams for the numberand areal distribution of all thermokarst lakes detectable onMR images Then we built similar diagrams for smallthermokarst lakes and very small thaw ponds using HR imagesof the 11 test sites across the BZT This allowed for extrapola-tion of the number and area of lakes across the entire BZT ter-ritory as determined by UHR images from the 11 test sitesFinally we combined primary MR and UHR histograms intointegral histograms for lake numbers and overall areas Thesehistograms included lakes and small and very small thawponds spread over five orders of magnitude in size We esti-mated uncertainty in lake size and area distribution stemmedfrom the process of overlapping images up to 15 followingthe methods applied for much larger territory of western Sibe-ria (Polishchuk et al 2018) The average depth was measuredfor all sampled lakes (see Supporting Information Table S1)For the regional estimation of water volume we used an

empirical approach developed for similar peatlands of westernSiberia (Polishchuk et al 2018) where we approximated thedependence of lake sizendashlake depth and used it for each lakesize bin

Sampling and chemical analysesThe surface waters were sampled from the shore (depression

and small thaw ponds) or via small PVC boats which gentlymoved across the lake surface On small ponds we used a ropeheld by a person standing on shore This was done to avoid anydisturbance of lake sediments during sampling At each loca-tion we measured water temperature and dissolved oxygen con-centration (WTW Oxi 3205 DurOx 325-3) pH and specificconductivity (WTW Multi 3320 multiparameter) as well as thewater depth (Cole-Parmer) air temperature and atmosphericpressure (Silva) Dissolved oxygen saturation was measured at50 cm intervals until reaching the lake sediment interfaceWater samples for DOC dissolved inorganic carbon (DIC) anddissolved CO2 and CH4 were collected from 02 to 03 m depthsand analyzed following methods of Manasypov et al (2015)and Serikova et al (2019) In brief for DOC analyses collectedwaters were filtered on-site in prewashed 30-mL polypropyleneNalgenereg bottles through single-use Minisart filter units(045 μm pore size Sartorius acetate cellulose filter) DOC andDIC were analyzed using a Carbon Total Analyzer (ShimadzuTOC VSCN) with an uncertainty better than 3 (see Prokushkinet al 2011 for methodology) Note that in small acidic (pH lt 5)thaw ponds with high DOC (gt 30 ppm) and low DIC(lt 1ndash2 ppm) the DIC measurements could be subjected to sig-nificant uncertainty For GHG analyses unfiltered water wassampled in 60-mL Serum bottles and closed without air bubblesusing vinyl stoppers and aluminum caps and immediately poi-soned by adding 02 mL of saturated HgCl2 via a two-way nee-dle system The CH4 and CO2 concentrations were measuredwith 3ndash5 uncertainty using a Bruker GC-456 gas chromato-graph equipped with flame ionization and thermal conductivitydetectors Replicate samplings from the same lake yielded anaverage reproducibility of 6ndash8 Dissolved (lt 045 μm) nutrients

(NOminus3 NH+

4 PO3minus4 ) and trace elements were measured using

spectrophotometry and Inductively Coupled Plasma MassSpectrometry (ICP_MS) respectively as described elsewherefor humic lake waters (Manasypov et al 2015 Chupakovet al 2017)

CO2 and CH4 flux measurements and calculationsCO2 and CH4 fluxes were calculated from wind speed and

surface-water gas concentrations This technique is based onthe two-layer model of Liss and Slater (1974) and widely usedfor GHG flux assessment (Repo et al 2007 Juutinen et al 2009Laurion et al 2010 Elder et al 2018) The gas transfer coeffi-cient was taken from Cole and Caraco (1998)

k600 = 207 +0215 U1710 eth1THORN

Zabelina et al C emission from thermokarst lakes

4

where U10 is the wind speed taken at 10 m height withoutexplicitly accounting for lake size and shape (ie Vachon andPrairie 2013) Average daily wind speed was retrieved fromofficial data of the nearest weather station (Naryan-Mar) aspublished by Rosgidromet for the day of sampling Further8 and 15 waterbodies were used for measuring emissions ofCO2 and CH4 respectively using floating chambers in orderto validate modeled wind-based fluxes

The direct CO2 fluxes were measured using small (30 cmdiameter) floating chambers equipped with nondispersiveinfrared CO2 logger (ELG SenseAir Alin et al 2011 Bastvikenet al 2015) These CO2 loggers were calibrated in the labagainst pure N2ndashCO2 mixtures before each sampling cam-paign We placed 2ndash3 chambers per lake along the transectfrom the shore to the center of the lake if the lake sizeallowed For small thaw ponds (lt 100 m2) and permafrostdepressions (lt 10 m2) two chambers were typically deployedThe CO2 accumulation rate inside each chamber was recordedat 5 min interval Chambers were ventilated every 10ndash24 hand CO2 accumulation rate inside each chamber was calcu-lated via linear regression for about 30 min accumulation time(Serikova et al 2019) All measurements had a linear CO2

increase over time with R2 gt 098CH4 flux measurements from the surface of several

waterbodies (7 lakes and 1 permafrost depression in 20163 lakes and 5 depressions in 2017) were performed using circu-lar chambers (surface area = 0154 m2 volume = 200 L) fol-lowing the design described in Guerin et al (2007) Insummary within 45 min four air samples were collected witha valve and a syringe from the chambers over 15 min inter-vals Air samples for CH4 were collected in 15 mL glass vialswhich contained 6 mol Lminus1 NaCl solution and capped withhigh density butyl stoppers and aluminum seals Methanefluxes were calculated from the slope of the linear regressionof gas concentration in the chamber vs time The fluxes wereaccepted when the coefficient R2 of the linear regression wasgt 090 Emissions were considered as the mix of ebullition anddiffusive when one or more abrupt increases of the floatingchamber headspace methane concentration were observedand the R2 of linear regression between methane concentra-tion and the elapsed time was below 090 Only diffusiveCH4 fluxes were used to calculate the overall CH4 flux Forexample any strong deviations from linear increase of CH4

concentration linked to potential bubble evasion were dis-carded Therefore obtained numbers represent the lowestthreshold of total methane emission It should be noted thatbubble evasion was strongly pronounced in all studiedponds and lakes this could be the main reason for thestrong difference between calculated and measured methanefluxes (see below) Each direct flux measurement was donetogether with a determination of the CH4 concentration insurface water

Although chamber measurements are the most accurate toolfor obtaining direct flux in remote locations (Serikova et al 2019)

we used calculated fluxes for regional C emission upscalinggiven that direct chamber measurements represented only 7and 19 of total CO2 and CH4 fluxes assessed To estimateregional C emission we used median values of CO2 and CH4

fluxes for each category of lake size averaged over seasonsWe exclusively considered lake size as the main controllingparameter on GHG DOC and DIC concentration becausetemporal and spatial variations of concentrations and fluxeswere smaller than variations due to lake size (see ldquoDOC DICCO2 and CH4 concentration and C emissions of thermokarstlakes as a function of lake areardquo section and ldquoSeasonalannual and spatial variability of C concentrations and emis-sionrdquo section) For this we used six size categories ofwaterbodies lt 10 m2 10ndash102 m2 102ndash103 m2 103ndash104 m2104ndash105 m2 and gt 105 m2 Note that similar to previousworks in other Arctic lakes GHG flux numbers represent aminimum estimate for the total annual CO2 and CH4 fluxesin Eastern European tundra thaw lakes because we did notinclude CH4 ebullition (ie Sabrekov et al 2017 Elderet al 2018) however we accounted for potential springrelease of CO2 and CH4 that accumulated during the winterseason under ice (Karlsson et al 2013)

Statistical treatment of the dataA nonparameric MannndashWhitney U-test was taken for

treating the data due to the short series of observation non-normal distribution of number of parameters and someamount of outliers Regressions were used to examine the rela-tionships between DOC CO2 CH4 concentrations and GHGfluxes to the atmosphere of the lake area The normality ofDOC GHG concentration and fluxes was assessed via theShapirondashWilk test Because the data were not normally distrib-uted we used nonparametric statistics Thus the median 1stand 3rd quartiles were used to trace the dependence of DOCCO2 CH4 concentration and GHG fluxes to the atmosphereon the area of the waterbody season and the year of sam-pling This is consistent with other studies on GHG in lakeswhen dealing with non-normal data distribution (Erkkilaumlet al 2018 Serikova et al 2019) Differences in concentrationsand fluxes between lake size classes seasons and years at thesame and among different sites for the same seasons and lakesize classes were tested using MannndashWhitney U-test for paireddata sets with significance level at 005 The Spearman rankorder correlation coefficient (Rs) (p lt 005) was used to deter-mine the relationship between GHG concentrations andfluxes and climatic variables Further statistical treatment of acomplete set of C concentration drivers in thermokarst lakewaters included a stepwise regression which allowed to testthe effect of various hydrochemical and climatic parameterson behavior of DOC and GHG All graphs and drawings wereplotted using MS Excel 2016 and Statistica-8 software package(httpwwwstatsoftcom)

Zabelina et al C emission from thermokarst lakes

5

combined with evolving lake sizes distributions and surfaceareas over past decades can allow forsight into potential futureC emissions to the atmosphere from lentic aquatic ecosystemsin northern Eurasia

Study site and methodsThermokarst lakes and thaw ponds of BZT NE Europe

BZT (NE Europe) is the largest frozen peatland in Europe Itis covered by permafrost discontinuous in the eastern partand sporadic to isolated in the western part (Brittainet al 2009) Peat thickness in the BZT ranges from 05 to 5 mdepending on the landscape context (Maksimova andOspennikov 2012 Iglovsky 2016) The dominant elevationhere is between 100 and 150 m these altitudes are comprisedof hills and moraine ridges which are mainly composed ofsands and silts with intermixed boulders Thermokarst lakesare located between these moraines and ridges In the south-ern forest-tundra zone the dominant soils are histosols of peatbogs and podzol-gleys The mean annual temperature isminus31C with a mean annual precipitation of 503 mm Thedominant vegetation in the tundra zone is mosses lichensand dwarf shrubs

Permafrost thaw in high latitude peatlands leads to forma-tion of ldquothermokarstrdquo (thaw) lakes due to frozen peat andmineral soil subsidence The formation and development ofthermokarst lakes has been investigated over the years andappears to follow a cycle (Matthews et al 1997 Hinkelet al 2003 Audry et al 2011 Kirpotin et al 2011 Pokrovskyet al 2011) The cycle starts with frozen peat subsidence cau-sed by warming and formation of a depression in mosslichencover that is filled by thaw water (Kirpotin et al 2011Pokrovsky et al 2011) The depression grows due to ongoingsubsidence of peat in the soil and instability of the edges untilit reaches lake size A mature thermokarst lake can eventuallybe drained into another lake or into the hydrological system(local streams and rivers) Gramineae-like vegetation colonizesalong the drained lake bottom followed by mosses andlichens Gradually the renewed permafrost heaves the soilcreating small isolated mounds that merge into a flat uniformpalsa plateau Ultimately the cycle can starts over againthrough formation of another wet depression This sequenceis an important part of the hydrological continuum in theBZT region and it strongly influences organic carbon andGHG dynamics in surface waters (Shirokova et al 2019)

Sampled waterbodies and basic physical and chemical char-acteristics and parameters of the CO2 system are listed inSupporting Information Table S1 For convenience all sam-pled waterbodies (generally referred to as ldquolakesrdquo) were classi-fied according to their surface area depressions (lt 10 m2)thaw ponds (10ndash1000 m2) and thermokarst lakes (gt 1000 m2)Altogether 48 thermokarst lakes 10 thaw ponds and40 depressions were sampled for hydrochemical parametersdissolved C CO2 and CH4 concentrations GHG emissions for

these bodies of water were calculated from CO2 and CH4 con-centrations and wind speed (see ldquoCO2 and CH4 flux measure-ments and calculationsrdquo section) One key site (~ 43 km SEfrom the town of Naryan-Mar) was selected for seasonal andannual monitoring of lakes with two additional sites(Shapkino and Khorey-Ver) to account for spatial heterogene-ity over summer (July 2016) these sites are shown in Fig 1and Supporting Information Fig S1a In order to reveal sea-sonal and annual variability of GHG concentrations andfluxes into the atmosphere in BZT depressions thaw pondsand thermokarst lakes we visited the Naryan-Mar site inspring (16ndash17 June 2017 sampling 7 lakes and 3 depressions)summer (17ndash23 July 2017 sampling 11 lakes and 13 depres-sions) and autumn (02ndash04 October 2017 sampling 11 lakesand 12 depressions) To account for interannual variabilitythree lakes from the Naryan-Mar site were sampled in July of2015 2016 2017 and 2018 These four sampled years wererather contrasting in mean monthly and annual temperatureand precipitation (see Supporting Information Table S2) Thesummer of 2015 was cold (minus48C below normal for July)while summers of 2016 2017 and 2018 were warm (+50C+40C and +25C respectively above normal for July) Thesummers of 2016 and 2018 were dry (31 and 36 of normalprecipitation for July) whereas the summers of 2015 and 2017were more humid (83 and 56 of normal precipitationfor July)

The spatial variability of dissolved organic carbon (DOC)and GHG concentrations was assessed in July 2016 A total of19 lakes thaw ponds and depressions were sampled in theShapkino site Another 33 lakes and ponds were visited in theKhorey-ver site Both of these sites were compared with sixlakes in the Naryan-Mar site sampled during the same period

Thermokarst lakes and thaw ponds in the BZT similar tothose in western Siberia (Shirokova et al 2013 Manasypovet al 2014 Serikova et al 2019) are mainly shallow (largelakes are only 05ndash15 m deep and small thaw ponds arelt 05 m deep) and are not stratified with respect to tempera-ture There are only a few exceptions where lakes are15ndash20 m deep but these lakes constituted less than 1 of allstudied (see Supporting Information Table S1) For simplicitywe refer to all waterbodies studied in this work as ldquolakesrdquoalthough their size ranges from lt 1 m2 to gt 7 km2

Lake size and area inventory using satellite imageryRemote sensing data utilized for this study include

Landsat-8 medium resolution (MR) and GeoEye-1 ultra-highresolution (UHR) imagery imagery were used to map distribu-tion of lakes throughout the BZT territory (Supporting Infor-mation Fig S1b) Landsat-8 images were taken in the late Julythrough early August 2017 and 2018 GeoEye-1 images werecollected in the summer months of 2011 through 2017 Thesesummer periods when the lakes are ice-free correspond tominimum coverage of the territory by lakes and minimumseasonal change in water level A combination of Landsat-8

Zabelina et al C emission from thermokarst lakes

3

images over 2 yr was necessary as observation over the courseof 1 yr could not provide full coverage of the territory

For the entire BZT territory affected by permafrost(195000 km2) we used a mosaic of 50 Landsat-8 MR imagesImages were processed using standard ArcGIS 103 softwaretools using the Fmask algorithm (see Polishchuk et al 20142017 2018 for method description) Given the sizable arealcoverage and accessibility challenges no ground calibrationhas been performed in BZT and we rely on ground truthingfrom former studies of western Siberia where there is a similardistribution of thermokarst lakes Taking into account the30 m resolution of MR imagery the minimum size of lakesexamined is limited to 5000 m2 This corresponds to six pixels30 times 30 m (900 m2) in size which is sufficient for reliable sep-aration of lakes from background noise The uncertainty ofmeasuring the surface area of a lake using Landsat-8 is 35for lakes gt 105 m2 surface area and 7 for lakes gt 2 times 104 m2

(Bryksina and Polishchuk 2013 Kornienko 2014) UHRGeoEye-1 imagery evenly distributed throughout the terri-tory were used to quantify the number and total areas ofsmall lakes and ponds For test sites we chose representativepermafrost peatland landscape similar to those sampled inthis work With a spatial resolution of 06 m pixel size inthese images is 036 m2 providing a minimal pond size of2 m2 GeoEye images were processed using ArcGIS 103 at11 test sites distributed throughout the study area

To plot the distribution of lakes in a wide range of sizes weused 23 partial size ranges of lakes taken on arithmetic pro-gression scale from 2 to 5 m2 5 to 10 m2 10 to 20 m2 and soon up to 108 m2 Note that remote sensing did not allow forquantification of the proportion of smallest depressions(lt 2 m2) However based on our previous work on similar per-mafrost peatlands in western Siberia the overall contributionof such small waterbodies to water volume surface area andGHG pools is negligible (Polishchuk et al 2018)

To combine the MR and UHR imagery we used a techniquedeveloped for thermokarst lakes in western Siberia (Polishchuket al 2018) First we constructed diagrams for the numberand areal distribution of all thermokarst lakes detectable onMR images Then we built similar diagrams for smallthermokarst lakes and very small thaw ponds using HR imagesof the 11 test sites across the BZT This allowed for extrapola-tion of the number and area of lakes across the entire BZT ter-ritory as determined by UHR images from the 11 test sitesFinally we combined primary MR and UHR histograms intointegral histograms for lake numbers and overall areas Thesehistograms included lakes and small and very small thawponds spread over five orders of magnitude in size We esti-mated uncertainty in lake size and area distribution stemmedfrom the process of overlapping images up to 15 followingthe methods applied for much larger territory of western Sibe-ria (Polishchuk et al 2018) The average depth was measuredfor all sampled lakes (see Supporting Information Table S1)For the regional estimation of water volume we used an

empirical approach developed for similar peatlands of westernSiberia (Polishchuk et al 2018) where we approximated thedependence of lake sizendashlake depth and used it for each lakesize bin

Sampling and chemical analysesThe surface waters were sampled from the shore (depression

and small thaw ponds) or via small PVC boats which gentlymoved across the lake surface On small ponds we used a ropeheld by a person standing on shore This was done to avoid anydisturbance of lake sediments during sampling At each loca-tion we measured water temperature and dissolved oxygen con-centration (WTW Oxi 3205 DurOx 325-3) pH and specificconductivity (WTW Multi 3320 multiparameter) as well as thewater depth (Cole-Parmer) air temperature and atmosphericpressure (Silva) Dissolved oxygen saturation was measured at50 cm intervals until reaching the lake sediment interfaceWater samples for DOC dissolved inorganic carbon (DIC) anddissolved CO2 and CH4 were collected from 02 to 03 m depthsand analyzed following methods of Manasypov et al (2015)and Serikova et al (2019) In brief for DOC analyses collectedwaters were filtered on-site in prewashed 30-mL polypropyleneNalgenereg bottles through single-use Minisart filter units(045 μm pore size Sartorius acetate cellulose filter) DOC andDIC were analyzed using a Carbon Total Analyzer (ShimadzuTOC VSCN) with an uncertainty better than 3 (see Prokushkinet al 2011 for methodology) Note that in small acidic (pH lt 5)thaw ponds with high DOC (gt 30 ppm) and low DIC(lt 1ndash2 ppm) the DIC measurements could be subjected to sig-nificant uncertainty For GHG analyses unfiltered water wassampled in 60-mL Serum bottles and closed without air bubblesusing vinyl stoppers and aluminum caps and immediately poi-soned by adding 02 mL of saturated HgCl2 via a two-way nee-dle system The CH4 and CO2 concentrations were measuredwith 3ndash5 uncertainty using a Bruker GC-456 gas chromato-graph equipped with flame ionization and thermal conductivitydetectors Replicate samplings from the same lake yielded anaverage reproducibility of 6ndash8 Dissolved (lt 045 μm) nutrients

(NOminus3 NH+

4 PO3minus4 ) and trace elements were measured using

spectrophotometry and Inductively Coupled Plasma MassSpectrometry (ICP_MS) respectively as described elsewherefor humic lake waters (Manasypov et al 2015 Chupakovet al 2017)

CO2 and CH4 flux measurements and calculationsCO2 and CH4 fluxes were calculated from wind speed and

surface-water gas concentrations This technique is based onthe two-layer model of Liss and Slater (1974) and widely usedfor GHG flux assessment (Repo et al 2007 Juutinen et al 2009Laurion et al 2010 Elder et al 2018) The gas transfer coeffi-cient was taken from Cole and Caraco (1998)

k600 = 207 +0215 U1710 eth1THORN

Zabelina et al C emission from thermokarst lakes

4

where U10 is the wind speed taken at 10 m height withoutexplicitly accounting for lake size and shape (ie Vachon andPrairie 2013) Average daily wind speed was retrieved fromofficial data of the nearest weather station (Naryan-Mar) aspublished by Rosgidromet for the day of sampling Further8 and 15 waterbodies were used for measuring emissions ofCO2 and CH4 respectively using floating chambers in orderto validate modeled wind-based fluxes

The direct CO2 fluxes were measured using small (30 cmdiameter) floating chambers equipped with nondispersiveinfrared CO2 logger (ELG SenseAir Alin et al 2011 Bastvikenet al 2015) These CO2 loggers were calibrated in the labagainst pure N2ndashCO2 mixtures before each sampling cam-paign We placed 2ndash3 chambers per lake along the transectfrom the shore to the center of the lake if the lake sizeallowed For small thaw ponds (lt 100 m2) and permafrostdepressions (lt 10 m2) two chambers were typically deployedThe CO2 accumulation rate inside each chamber was recordedat 5 min interval Chambers were ventilated every 10ndash24 hand CO2 accumulation rate inside each chamber was calcu-lated via linear regression for about 30 min accumulation time(Serikova et al 2019) All measurements had a linear CO2

increase over time with R2 gt 098CH4 flux measurements from the surface of several

waterbodies (7 lakes and 1 permafrost depression in 20163 lakes and 5 depressions in 2017) were performed using circu-lar chambers (surface area = 0154 m2 volume = 200 L) fol-lowing the design described in Guerin et al (2007) Insummary within 45 min four air samples were collected witha valve and a syringe from the chambers over 15 min inter-vals Air samples for CH4 were collected in 15 mL glass vialswhich contained 6 mol Lminus1 NaCl solution and capped withhigh density butyl stoppers and aluminum seals Methanefluxes were calculated from the slope of the linear regressionof gas concentration in the chamber vs time The fluxes wereaccepted when the coefficient R2 of the linear regression wasgt 090 Emissions were considered as the mix of ebullition anddiffusive when one or more abrupt increases of the floatingchamber headspace methane concentration were observedand the R2 of linear regression between methane concentra-tion and the elapsed time was below 090 Only diffusiveCH4 fluxes were used to calculate the overall CH4 flux Forexample any strong deviations from linear increase of CH4

concentration linked to potential bubble evasion were dis-carded Therefore obtained numbers represent the lowestthreshold of total methane emission It should be noted thatbubble evasion was strongly pronounced in all studiedponds and lakes this could be the main reason for thestrong difference between calculated and measured methanefluxes (see below) Each direct flux measurement was donetogether with a determination of the CH4 concentration insurface water

Although chamber measurements are the most accurate toolfor obtaining direct flux in remote locations (Serikova et al 2019)

we used calculated fluxes for regional C emission upscalinggiven that direct chamber measurements represented only 7and 19 of total CO2 and CH4 fluxes assessed To estimateregional C emission we used median values of CO2 and CH4

fluxes for each category of lake size averaged over seasonsWe exclusively considered lake size as the main controllingparameter on GHG DOC and DIC concentration becausetemporal and spatial variations of concentrations and fluxeswere smaller than variations due to lake size (see ldquoDOC DICCO2 and CH4 concentration and C emissions of thermokarstlakes as a function of lake areardquo section and ldquoSeasonalannual and spatial variability of C concentrations and emis-sionrdquo section) For this we used six size categories ofwaterbodies lt 10 m2 10ndash102 m2 102ndash103 m2 103ndash104 m2104ndash105 m2 and gt 105 m2 Note that similar to previousworks in other Arctic lakes GHG flux numbers represent aminimum estimate for the total annual CO2 and CH4 fluxesin Eastern European tundra thaw lakes because we did notinclude CH4 ebullition (ie Sabrekov et al 2017 Elderet al 2018) however we accounted for potential springrelease of CO2 and CH4 that accumulated during the winterseason under ice (Karlsson et al 2013)

Statistical treatment of the dataA nonparameric MannndashWhitney U-test was taken for

treating the data due to the short series of observation non-normal distribution of number of parameters and someamount of outliers Regressions were used to examine the rela-tionships between DOC CO2 CH4 concentrations and GHGfluxes to the atmosphere of the lake area The normality ofDOC GHG concentration and fluxes was assessed via theShapirondashWilk test Because the data were not normally distrib-uted we used nonparametric statistics Thus the median 1stand 3rd quartiles were used to trace the dependence of DOCCO2 CH4 concentration and GHG fluxes to the atmosphereon the area of the waterbody season and the year of sam-pling This is consistent with other studies on GHG in lakeswhen dealing with non-normal data distribution (Erkkilaumlet al 2018 Serikova et al 2019) Differences in concentrationsand fluxes between lake size classes seasons and years at thesame and among different sites for the same seasons and lakesize classes were tested using MannndashWhitney U-test for paireddata sets with significance level at 005 The Spearman rankorder correlation coefficient (Rs) (p lt 005) was used to deter-mine the relationship between GHG concentrations andfluxes and climatic variables Further statistical treatment of acomplete set of C concentration drivers in thermokarst lakewaters included a stepwise regression which allowed to testthe effect of various hydrochemical and climatic parameterson behavior of DOC and GHG All graphs and drawings wereplotted using MS Excel 2016 and Statistica-8 software package(httpwwwstatsoftcom)

Zabelina et al C emission from thermokarst lakes

5

images over 2 yr was necessary as observation over the courseof 1 yr could not provide full coverage of the territory

For the entire BZT territory affected by permafrost(195000 km2) we used a mosaic of 50 Landsat-8 MR imagesImages were processed using standard ArcGIS 103 softwaretools using the Fmask algorithm (see Polishchuk et al 20142017 2018 for method description) Given the sizable arealcoverage and accessibility challenges no ground calibrationhas been performed in BZT and we rely on ground truthingfrom former studies of western Siberia where there is a similardistribution of thermokarst lakes Taking into account the30 m resolution of MR imagery the minimum size of lakesexamined is limited to 5000 m2 This corresponds to six pixels30 times 30 m (900 m2) in size which is sufficient for reliable sep-aration of lakes from background noise The uncertainty ofmeasuring the surface area of a lake using Landsat-8 is 35for lakes gt 105 m2 surface area and 7 for lakes gt 2 times 104 m2

(Bryksina and Polishchuk 2013 Kornienko 2014) UHRGeoEye-1 imagery evenly distributed throughout the terri-tory were used to quantify the number and total areas ofsmall lakes and ponds For test sites we chose representativepermafrost peatland landscape similar to those sampled inthis work With a spatial resolution of 06 m pixel size inthese images is 036 m2 providing a minimal pond size of2 m2 GeoEye images were processed using ArcGIS 103 at11 test sites distributed throughout the study area

To plot the distribution of lakes in a wide range of sizes weused 23 partial size ranges of lakes taken on arithmetic pro-gression scale from 2 to 5 m2 5 to 10 m2 10 to 20 m2 and soon up to 108 m2 Note that remote sensing did not allow forquantification of the proportion of smallest depressions(lt 2 m2) However based on our previous work on similar per-mafrost peatlands in western Siberia the overall contributionof such small waterbodies to water volume surface area andGHG pools is negligible (Polishchuk et al 2018)

To combine the MR and UHR imagery we used a techniquedeveloped for thermokarst lakes in western Siberia (Polishchuket al 2018) First we constructed diagrams for the numberand areal distribution of all thermokarst lakes detectable onMR images Then we built similar diagrams for smallthermokarst lakes and very small thaw ponds using HR imagesof the 11 test sites across the BZT This allowed for extrapola-tion of the number and area of lakes across the entire BZT ter-ritory as determined by UHR images from the 11 test sitesFinally we combined primary MR and UHR histograms intointegral histograms for lake numbers and overall areas Thesehistograms included lakes and small and very small thawponds spread over five orders of magnitude in size We esti-mated uncertainty in lake size and area distribution stemmedfrom the process of overlapping images up to 15 followingthe methods applied for much larger territory of western Sibe-ria (Polishchuk et al 2018) The average depth was measuredfor all sampled lakes (see Supporting Information Table S1)For the regional estimation of water volume we used an

empirical approach developed for similar peatlands of westernSiberia (Polishchuk et al 2018) where we approximated thedependence of lake sizendashlake depth and used it for each lakesize bin

Sampling and chemical analysesThe surface waters were sampled from the shore (depression

and small thaw ponds) or via small PVC boats which gentlymoved across the lake surface On small ponds we used a ropeheld by a person standing on shore This was done to avoid anydisturbance of lake sediments during sampling At each loca-tion we measured water temperature and dissolved oxygen con-centration (WTW Oxi 3205 DurOx 325-3) pH and specificconductivity (WTW Multi 3320 multiparameter) as well as thewater depth (Cole-Parmer) air temperature and atmosphericpressure (Silva) Dissolved oxygen saturation was measured at50 cm intervals until reaching the lake sediment interfaceWater samples for DOC dissolved inorganic carbon (DIC) anddissolved CO2 and CH4 were collected from 02 to 03 m depthsand analyzed following methods of Manasypov et al (2015)and Serikova et al (2019) In brief for DOC analyses collectedwaters were filtered on-site in prewashed 30-mL polypropyleneNalgenereg bottles through single-use Minisart filter units(045 μm pore size Sartorius acetate cellulose filter) DOC andDIC were analyzed using a Carbon Total Analyzer (ShimadzuTOC VSCN) with an uncertainty better than 3 (see Prokushkinet al 2011 for methodology) Note that in small acidic (pH lt 5)thaw ponds with high DOC (gt 30 ppm) and low DIC(lt 1ndash2 ppm) the DIC measurements could be subjected to sig-nificant uncertainty For GHG analyses unfiltered water wassampled in 60-mL Serum bottles and closed without air bubblesusing vinyl stoppers and aluminum caps and immediately poi-soned by adding 02 mL of saturated HgCl2 via a two-way nee-dle system The CH4 and CO2 concentrations were measuredwith 3ndash5 uncertainty using a Bruker GC-456 gas chromato-graph equipped with flame ionization and thermal conductivitydetectors Replicate samplings from the same lake yielded anaverage reproducibility of 6ndash8 Dissolved (lt 045 μm) nutrients

(NOminus3 NH+

4 PO3minus4 ) and trace elements were measured using

spectrophotometry and Inductively Coupled Plasma MassSpectrometry (ICP_MS) respectively as described elsewherefor humic lake waters (Manasypov et al 2015 Chupakovet al 2017)

CO2 and CH4 flux measurements and calculationsCO2 and CH4 fluxes were calculated from wind speed and

surface-water gas concentrations This technique is based onthe two-layer model of Liss and Slater (1974) and widely usedfor GHG flux assessment (Repo et al 2007 Juutinen et al 2009Laurion et al 2010 Elder et al 2018) The gas transfer coeffi-cient was taken from Cole and Caraco (1998)

k600 = 207 +0215 U1710 eth1THORN

Zabelina et al C emission from thermokarst lakes

4

where U10 is the wind speed taken at 10 m height withoutexplicitly accounting for lake size and shape (ie Vachon andPrairie 2013) Average daily wind speed was retrieved fromofficial data of the nearest weather station (Naryan-Mar) aspublished by Rosgidromet for the day of sampling Further8 and 15 waterbodies were used for measuring emissions ofCO2 and CH4 respectively using floating chambers in orderto validate modeled wind-based fluxes

The direct CO2 fluxes were measured using small (30 cmdiameter) floating chambers equipped with nondispersiveinfrared CO2 logger (ELG SenseAir Alin et al 2011 Bastvikenet al 2015) These CO2 loggers were calibrated in the labagainst pure N2ndashCO2 mixtures before each sampling cam-paign We placed 2ndash3 chambers per lake along the transectfrom the shore to the center of the lake if the lake sizeallowed For small thaw ponds (lt 100 m2) and permafrostdepressions (lt 10 m2) two chambers were typically deployedThe CO2 accumulation rate inside each chamber was recordedat 5 min interval Chambers were ventilated every 10ndash24 hand CO2 accumulation rate inside each chamber was calcu-lated via linear regression for about 30 min accumulation time(Serikova et al 2019) All measurements had a linear CO2

increase over time with R2 gt 098CH4 flux measurements from the surface of several

waterbodies (7 lakes and 1 permafrost depression in 20163 lakes and 5 depressions in 2017) were performed using circu-lar chambers (surface area = 0154 m2 volume = 200 L) fol-lowing the design described in Guerin et al (2007) Insummary within 45 min four air samples were collected witha valve and a syringe from the chambers over 15 min inter-vals Air samples for CH4 were collected in 15 mL glass vialswhich contained 6 mol Lminus1 NaCl solution and capped withhigh density butyl stoppers and aluminum seals Methanefluxes were calculated from the slope of the linear regressionof gas concentration in the chamber vs time The fluxes wereaccepted when the coefficient R2 of the linear regression wasgt 090 Emissions were considered as the mix of ebullition anddiffusive when one or more abrupt increases of the floatingchamber headspace methane concentration were observedand the R2 of linear regression between methane concentra-tion and the elapsed time was below 090 Only diffusiveCH4 fluxes were used to calculate the overall CH4 flux Forexample any strong deviations from linear increase of CH4

concentration linked to potential bubble evasion were dis-carded Therefore obtained numbers represent the lowestthreshold of total methane emission It should be noted thatbubble evasion was strongly pronounced in all studiedponds and lakes this could be the main reason for thestrong difference between calculated and measured methanefluxes (see below) Each direct flux measurement was donetogether with a determination of the CH4 concentration insurface water

Although chamber measurements are the most accurate toolfor obtaining direct flux in remote locations (Serikova et al 2019)

we used calculated fluxes for regional C emission upscalinggiven that direct chamber measurements represented only 7and 19 of total CO2 and CH4 fluxes assessed To estimateregional C emission we used median values of CO2 and CH4

fluxes for each category of lake size averaged over seasonsWe exclusively considered lake size as the main controllingparameter on GHG DOC and DIC concentration becausetemporal and spatial variations of concentrations and fluxeswere smaller than variations due to lake size (see ldquoDOC DICCO2 and CH4 concentration and C emissions of thermokarstlakes as a function of lake areardquo section and ldquoSeasonalannual and spatial variability of C concentrations and emis-sionrdquo section) For this we used six size categories ofwaterbodies lt 10 m2 10ndash102 m2 102ndash103 m2 103ndash104 m2104ndash105 m2 and gt 105 m2 Note that similar to previousworks in other Arctic lakes GHG flux numbers represent aminimum estimate for the total annual CO2 and CH4 fluxesin Eastern European tundra thaw lakes because we did notinclude CH4 ebullition (ie Sabrekov et al 2017 Elderet al 2018) however we accounted for potential springrelease of CO2 and CH4 that accumulated during the winterseason under ice (Karlsson et al 2013)

Statistical treatment of the dataA nonparameric MannndashWhitney U-test was taken for

treating the data due to the short series of observation non-normal distribution of number of parameters and someamount of outliers Regressions were used to examine the rela-tionships between DOC CO2 CH4 concentrations and GHGfluxes to the atmosphere of the lake area The normality ofDOC GHG concentration and fluxes was assessed via theShapirondashWilk test Because the data were not normally distrib-uted we used nonparametric statistics Thus the median 1stand 3rd quartiles were used to trace the dependence of DOCCO2 CH4 concentration and GHG fluxes to the atmosphereon the area of the waterbody season and the year of sam-pling This is consistent with other studies on GHG in lakeswhen dealing with non-normal data distribution (Erkkilaumlet al 2018 Serikova et al 2019) Differences in concentrationsand fluxes between lake size classes seasons and years at thesame and among different sites for the same seasons and lakesize classes were tested using MannndashWhitney U-test for paireddata sets with significance level at 005 The Spearman rankorder correlation coefficient (Rs) (p lt 005) was used to deter-mine the relationship between GHG concentrations andfluxes and climatic variables Further statistical treatment of acomplete set of C concentration drivers in thermokarst lakewaters included a stepwise regression which allowed to testthe effect of various hydrochemical and climatic parameterson behavior of DOC and GHG All graphs and drawings wereplotted using MS Excel 2016 and Statistica-8 software package(httpwwwstatsoftcom)

Zabelina et al C emission from thermokarst lakes

5

where U10 is the wind speed taken at 10 m height withoutexplicitly accounting for lake size and shape (ie Vachon andPrairie 2013) Average daily wind speed was retrieved fromofficial data of the nearest weather station (Naryan-Mar) aspublished by Rosgidromet for the day of sampling Further8 and 15 waterbodies were used for measuring emissions ofCO2 and CH4 respectively using floating chambers in orderto validate modeled wind-based fluxes

The direct CO2 fluxes were measured using small (30 cmdiameter) floating chambers equipped with nondispersiveinfrared CO2 logger (ELG SenseAir Alin et al 2011 Bastvikenet al 2015) These CO2 loggers were calibrated in the labagainst pure N2ndashCO2 mixtures before each sampling cam-paign We placed 2ndash3 chambers per lake along the transectfrom the shore to the center of the lake if the lake sizeallowed For small thaw ponds (lt 100 m2) and permafrostdepressions (lt 10 m2) two chambers were typically deployedThe CO2 accumulation rate inside each chamber was recordedat 5 min interval Chambers were ventilated every 10ndash24 hand CO2 accumulation rate inside each chamber was calcu-lated via linear regression for about 30 min accumulation time(Serikova et al 2019) All measurements had a linear CO2

increase over time with R2 gt 098CH4 flux measurements from the surface of several

waterbodies (7 lakes and 1 permafrost depression in 20163 lakes and 5 depressions in 2017) were performed using circu-lar chambers (surface area = 0154 m2 volume = 200 L) fol-lowing the design described in Guerin et al (2007) Insummary within 45 min four air samples were collected witha valve and a syringe from the chambers over 15 min inter-vals Air samples for CH4 were collected in 15 mL glass vialswhich contained 6 mol Lminus1 NaCl solution and capped withhigh density butyl stoppers and aluminum seals Methanefluxes were calculated from the slope of the linear regressionof gas concentration in the chamber vs time The fluxes wereaccepted when the coefficient R2 of the linear regression wasgt 090 Emissions were considered as the mix of ebullition anddiffusive when one or more abrupt increases of the floatingchamber headspace methane concentration were observedand the R2 of linear regression between methane concentra-tion and the elapsed time was below 090 Only diffusiveCH4 fluxes were used to calculate the overall CH4 flux Forexample any strong deviations from linear increase of CH4

concentration linked to potential bubble evasion were dis-carded Therefore obtained numbers represent the lowestthreshold of total methane emission It should be noted thatbubble evasion was strongly pronounced in all studiedponds and lakes this could be the main reason for thestrong difference between calculated and measured methanefluxes (see below) Each direct flux measurement was donetogether with a determination of the CH4 concentration insurface water

Although chamber measurements are the most accurate toolfor obtaining direct flux in remote locations (Serikova et al 2019)

we used calculated fluxes for regional C emission upscalinggiven that direct chamber measurements represented only 7and 19 of total CO2 and CH4 fluxes assessed To estimateregional C emission we used median values of CO2 and CH4

fluxes for each category of lake size averaged over seasonsWe exclusively considered lake size as the main controllingparameter on GHG DOC and DIC concentration becausetemporal and spatial variations of concentrations and fluxeswere smaller than variations due to lake size (see ldquoDOC DICCO2 and CH4 concentration and C emissions of thermokarstlakes as a function of lake areardquo section and ldquoSeasonalannual and spatial variability of C concentrations and emis-sionrdquo section) For this we used six size categories ofwaterbodies lt 10 m2 10ndash102 m2 102ndash103 m2 103ndash104 m2104ndash105 m2 and gt 105 m2 Note that similar to previousworks in other Arctic lakes GHG flux numbers represent aminimum estimate for the total annual CO2 and CH4 fluxesin Eastern European tundra thaw lakes because we did notinclude CH4 ebullition (ie Sabrekov et al 2017 Elderet al 2018) however we accounted for potential springrelease of CO2 and CH4 that accumulated during the winterseason under ice (Karlsson et al 2013)

Statistical treatment of the dataA nonparameric MannndashWhitney U-test was taken for

treating the data due to the short series of observation non-normal distribution of number of parameters and someamount of outliers Regressions were used to examine the rela-tionships between DOC CO2 CH4 concentrations and GHGfluxes to the atmosphere of the lake area The normality ofDOC GHG concentration and fluxes was assessed via theShapirondashWilk test Because the data were not normally distrib-uted we used nonparametric statistics Thus the median 1stand 3rd quartiles were used to trace the dependence of DOCCO2 CH4 concentration and GHG fluxes to the atmosphereon the area of the waterbody season and the year of sam-pling This is consistent with other studies on GHG in lakeswhen dealing with non-normal data distribution (Erkkilaumlet al 2018 Serikova et al 2019) Differences in concentrationsand fluxes between lake size classes seasons and years at thesame and among different sites for the same seasons and lakesize classes were tested using MannndashWhitney U-test for paireddata sets with significance level at 005 The Spearman rankorder correlation coefficient (Rs) (p lt 005) was used to deter-mine the relationship between GHG concentrations andfluxes and climatic variables Further statistical treatment of acomplete set of C concentration drivers in thermokarst lakewaters included a stepwise regression which allowed to testthe effect of various hydrochemical and climatic parameterson behavior of DOC and GHG All graphs and drawings wereplotted using MS Excel 2016 and Statistica-8 software package(httpwwwstatsoftcom)

Zabelina et al C emission from thermokarst lakes

5

ResultsDOC DIC CO2 and CH4 concentration and C emissions ofthermokarst lakes as a function of lake area

The full set of measured DOC CO2 CH4 concentrationsand CO2 and CH4 fluxes into atmosphere are provided inSupporting Information Table S1 DOC concentrations typi-cally ranged for different lake classes from 10 to 100 mg Lminus1

(median of 20ndash50 mg Lminus1) and decreased almost 10-fold as sur-face area increased from lt 1 m2 to 1 km2 (Fig 2a andSupporting Information Fig S2a) The DIC concentration wasless than 10 of DOC and was invariant to lake size (Fig 2bSupporting Information Fig S2b) CO2 concentrationdecreased up to 10-fold with increasing lake size (Fig 2c) Themedian values of CO2 concentration (100ndash500 μmol Lminus1)

Fig 2 DOC (a) and DIC (b) concentrations CO2 (c) concentrations and fluxes to atmosphere (d) CH4 concentrations (e) and fluxes to atmosphere (f)as a function of lake surface area in spring summer and autumn 2016 at the Naryan-Mar site Note that these shallow lakes are frozen solid in winter

Zabelina et al C emission from thermokarst lakes

6

demonstrated a factor of 5 decrease with increasing lake size(from lt 10 m2 to gt 1000 m2 Supporting Information Fig S2c)The CO2 fluxes progressively decreased from 500 to1000 mmol mminus2 dminus1 in depressions (lt 10 m2) to about100 mmol mminus2 dminus1 in lakes larger than 1000 m2 (Fig 2d) Athreefold decrease in median CO2 fluxes from ponds lt 100 m2

to lakes ge 1000 m2 was also observed (Supporting InformationFig S2d) A general decrease of DOC CO2 concentrationFCO2 and optical density at 254 nm (UV254) with lake surfacearea across all sites and all seasons is confirmed by strong(p = 00000) power dependence (Supporting InformationTable S3) In contrast to CO2 CH4 concentrations and fluxesto atmosphere exhibited strong variability ranging from 01to 100 μmol Lminus1 and mmol mminus2 dminus1 respectively There was alocal maximum of CH4 concentration and emission in thawponds of 10ndash1000 m2 surface area (Fig 2ef Supporting Infor-mation Fig S2ef respectively) Considering all calculatedfluxes to atmosphere (n = 135) the contribution of C-СН4 tothe total C emissions calculated by wind-based modelingranged from 002 to 19 (with a mean of 26 and amedian of 11) Chamber-measured CH4 fluxes in eight bod-ies of water constituted between 02 and 58 of the totalCO2 + CH4 fluxes (Supporting Information Table S4a)

The CO2 fluxes to atmosphere obtained by using wind-basedmodels were on average 25 times (range of 08ndash189 times n = 8)higher than those directly measured using chambers (SupportingInformation Table S4a) It should be noted that Laurionet al (2010) reported an average factor of 38 of the excess of cal-culated fluxes in comparison to measured CO2 fluxes for similarpeatland thaw ponds The difference between measured and cal-culated CH4 fluxes to atmosphere was significantly (p lt 005)higher than for CO2 (range 15ndash194 mean 68) In thermokarstlakes of western Siberian permafrost peatlands CO2 and CH4

fluxes to atmosphere calculated with wind-based models (eitherCole and Caraco 1998 or Vachon and Prairie 2013) were on aver-age three and four times lower than measured fluxes (Serikovaet al 2019) These authors explained this discrepancy as due tothe remoteness of the studied lakes in relation to their respectivemeteorological stations and the lack of local wind data This isalso true for most BZT sites located approximately 40 km fromthe Naryan-Mar station The most likely cause for CH4 differencesis bubble evasion (Casper et al 2003) as it was not possible tosample shallow waterbodies (lt 1 m depth) with zero disturbanceof the organic detritus at lake bottom and in surrounding coastalwaterlogged zones Note that the agreement between the twomethods was much better in lakes with mineral banks wherevisual bubble disturbance during sampling was minimal Anotherreason for observed discrepancies may be due to the wind effecton the gas transfer coefficient being overestimated in smallaquatic systems with small fetch and sheltered sampling pointsalong the shoreline (Alin et al 2011)

As a measure of variability associated with model parame-ters we calculated the effects of variable wind speed for

estimates of diffusive GHG fluxes to atmosphere using Eq 1for 25 lakes gt 1000 m2 sampled in July 2015ndash2018 at theNarian-Mar site (Supporting Information Table S4b) Typicallyvariation of the wind speed by a factor of 2 yielded a factor of13ndash15 variation in relevant median values of GHG fluxes toatmosphere Only drastic changes in wind speed (by a factorof 5 or more) had sizable effect (exceeding the interquartilerange [IQR]) on CH4 and CO2 fluxes Overall this suggeststhat the offset between measured and calculated fluxes is notcaused by uncertainties in wind speed alone

Seasonal annual and spatial variability of Cconcentrations and emission

GHG concentrations and fluxes to atmosphere showedlarger spatial variability across lake sizes than temporal vari-ability across seasons (June July and October) Differences inDOC DIC and GHG between seasons (2015ndash2018) wereassessed for lakes gt 1000 m2 from the Naryan-Mar site(Supporting Information Table S5) For these lakes the DICand CO2 concentrations were sizably lower in spring com-pared to summer (p lt 005) however there was no differencebetween summer and autumn (p gt 005 Supporting Informa-tion Table S5 and Fig S3bc) The DOC showed an inverse pat-tern with spring lt summer = autumn (p lt 005 SupportingInformation Fig S3a) CH4 concentrations and CO2 and CH4

fluxes to atmosphere followed the order ldquospring gt summerrdquo(p lt 005) but there was no difference (p gt 005) between sum-mer and autumn (Supporting Information Fig S3dndashf) Therewere no seasonal differences (p gt 005) in DIC DOC CO2 andCH4 concentrations and fluxes for depressions (lt 10 m2 area)Interannual variations in DOC CO2 CH4 concentrations andCO2 and CH4 fluxes (Supporting Information Fig S4) did notexhibit any link to potential drivers like annual andor sea-sonal temperature or precipitation for the specific year ofobservation (Supporting Information Table S2) In fact despitesizeable contrast in mean June and July temperature and pre-cipitation between 2015 and 2018 (p le 001 MannndashWhitneyU-test) no difference in DOC DIC CO2 and CH4 was evidentfor these years (p gt 005)

Using Spearman rank order correlation and multiple regres-sion approach we examined the impact of water temperatureand cumulative precipitation and temperature during thesummer prior to the sampling campaign on the GHG DOCand DIC concentrations and GHG fluxes to atmosphere(Supporting Information Table S6a) Water temperature neg-atively impacted CO2 and positively impacted DOC(RSpearman = minus020 and 022 respectively at p lt 005) TheSpearman rank order correlation yielded positive depen-dence of CH4 and CO2 on cumulative air temperature priorto sampling (RSpearman = 020ndash028) as well as on the meanair temperature during June and July (RSpearman = 026ndash030in lakes RSpearman = 043ndash046 in small thaw ponds)(Supporting Information Table S6a) Precipitation exerted a

Zabelina et al C emission from thermokarst lakes

7

negative impact on GHG concentration and fluxes to atmo-sphere (RSpearman = minus023 to minus055) but a positive impact onspecific UV absorbance (SUVA RSpearman = 024ndash038) Over-all the impact of both temperature and precipitation wasmuch more pronounced for small thaw ponds (lt 500 m2)compared to thermokarst lakes (gt 500 m2) The multipleregression approach demonstrated that the GHG concentra-tion variability is partially determined by waterbody areaand cumulative climatic parameters such as air temperatureand precipitation (Supporting Information Table S6b) How-ever the multiple regression was not capable of explainingmore than 20ndash40 of the overall variability in CO2 CH4DOC and DIC concentrations

The variability of C concentration and emissions acrosssites was assessed for Naryan-Mar Shapkino and Khorey-Versites in July 2016 This comparison was conducted separatelyfor thaw ponds with surface areas lt 50 m2 and 50ndash500 m2 andfor lakes with surface areas gt 1000 m2 (Supporting Informa-tion Fig S5) The difference between sites was assessed via thenonparametric MannndashWhitney U-test (Supporting Informa-tion Table S7) All sites were similar in DOC CO2 CH4 con-centrations and C emissions to atmosphere except lakes inShapkino which had three times higher DIC and CO2 concen-trations than those of Naryan-Mar and Khorey-Ver This isconsistent with the shallower position of mineral soils onclays and fluvioglacial deposits at the Shapkino site wheresome carbonate shells have been reported (Andreicheva 2007)

In order to assess possible control of hydrochemical param-eters of waterbodies on CO2 and CH4 concentrations we cal-culated Spearman rank correlation coefficients (SupportingInformation Table S8) The concentrations of CO2 and CH4

negatively correlated with pH and O2 and positively corre-lated with DIC DOC and Total Dissolved Solids (TDS) CO2

concentration positively correlated with PO3minus4 Ptot NOminus

3 NtotSi DIC DOC Fe Co Ni and negatively correlated with NH+

4 CH4 concentration positively correlated with Ntot Si DICDOC Fe Co and Ni and negatively correlated with NH+

4 and

SO2minus4 (Supporting Information Table S8)

Regional-scale estimation of C emissions from BZTterritory (based on satellite imagery)

For the full BZT territory (195000 km2) the total numberof inventoried lakes ranging from 2 m2 to 5 times 107 m2 in sizewas 896 times 106 with a total water area of 7800 km2 (Table 1and Supporting Information Fig S6ab) Lakes lt 5000 m2 pro-vided about 10 of overall lake water area in BZT While smalllakes (lt 2000 m2) provided 97 of the total lake numberstheir area contributed to only about 6 of total area coverageThermokarst lakes between 5000 and 108 m2 in areaaccounted for only 15 of the lake count but provided gt 90of total lake area Note that from an optical remote sensingview point there is no difference between thermokarst andnonthermokarst lakes because their reflections yield similar T

able

1Lake

characteristic

sC

pools

andGHG

flux

esinventory[m

edianan

d(IQR)]forlakesof

EasternEu

rope

an(Bolshezem

elskaya)

Tund

ra(195

000

km2)

Themax

imal

contrib

utionto

both

CO

2an

dCH4po

olisprov

ided

bylakesof

gt10

5m

2surfaceareaLakeslarger

than

105m

2prov

idegt43

of

totalC

H4am

ount

whe

reas

thesharefrom

smalllakes

(lt10

00m

2)islt20

N

otethat

smalllakes

contrib

uted

less

toov

erallC

O2an

dDOClevelsthan

toCH4level

Lake

size

range(m

2)

Dep

th(m

)La

kenum

ber

Area

(m2)

Volume

(m3)

C-СO

2(g

)C-СН4(g

)DIC

(g[С

])DOC(g

[С])

C-CO

2

flux

(gСdminus1)

C-CH4

flux

(gСdminus1)

lt10

010

482

times10

6235

times10

7235

times10

6818

times10

6

(201times

107)

460

times10

4

(425times

105)

267

times10

6

(207times

106)

133

times10

8

(726times

107)

720

times10

7

(996times

107)

334

times10

5

(264times

106)

10ndash10

2029

303

times10

6915

times10

7265

times10

7156

times10

8

(228times

108)

514

times10

6

(100times

107)

539

times10

7

(691times

107)

105

times10

9

(125times

109)

448

times10

8

(403times

108)

149

times10

7

(183times

107)

102ndash10

3050

783

times10

5231

times10

8115

times10

8292

times10

8

(365times

108)

127

times10

7

(453times

107)

117

times10

8

(128times

108)

414

times10

9

(304times

109)

450

times10

8

(435times

108)

200

times10

7

(529times

107)

103ndash10

4067

246

times10

5829

times10

8556

times10

8801

times10

8

(445times

108)

106

times10

7

(138times

107)

499

times10

8

(341times

108)

107

times10

10

(109times

1010)

963

times10

8

(979times

108)

137

times10

7

(118times

107)

104ndash10

5077

653

times10

4213

times10

9164

times10

9186

times10

9

(275times

109)

216

times10

7

(180times

107)

110

times10

9

(272times

109)

226

times10

10

(738times

109)

282

times10

9

(252times

109)

210

times10

7

(244times

107)

gt105

083

130

times10

4457

times10

9379

times10

9566

times10

9

(543times

109)

386

times10

7

(903times

107)

453

times10

9

(279times

109)

518

times10

10

(177times

1010)

545

times10

9

(620times

109)

334

times10

7

(651times

107)

Total

896

times10

6784

times10

9613

times10

9878

times10

9887

times10

7631

times10

9903

times10

10

102

times10

10

103

times10

8

Zabelina et al C emission from thermokarst lakes

8

signals for Landsat-8 OLI However according to publishedexpert estimations the majority of lakes in the BZT are ofthermokarst origin (Kravtzova and Bystrova 2009) this originis also confirmed by their generally isometric shape as com-pared to elongated lakes with glacial origins Both the lakeabundance and area distribution obtained for the permafrost-affected part of BZT are in fairly good agreement (SupportingInformation Fig S7) with those reported for permafrost-affected parts of western Siberia (Polishchuk et al 2017 2018)and Swedish territory (Cael and Seekell 2016) In the BZT thepower law extrapolation (Pareto law) strongly overestimatesthe number of small lakes (lt 1000 m2) compared to empiricaldata (Supporting Information Fig S7) which is in agreementwith other regions of the world

The maximal contribution to overall GHG emissions fromthe territory of 195000 km2 (38 Tg yrminus1) was provided bylakes gt 100000 m2 whereas contributions from small thawponds (lt 1000 m2) were below 20 (Table 1 Fig 3) The dis-persity of the areal GHG fluxes to atmosphere is quite highachieving one order of magnitude in small thaw ponds and afactor of 2ndash5 for thermokarst lakes (Fig 2) These uncertaintiesvisible as IQR in Supporting Information Figs S2ndashS5 S8 do notoriginate from analytical procedures and rather reflect variabil-ity between sites and seasons It is important to note that thecontribution of small thaw ponds to overall C balance in BZT isbelow the actual uncertainties for total pools and emissions

DiscussionC pools fluxes to atmosphere and possible mechanisms

The highest DOC and CO2 concentrations and CO2 fluxesto atmosphere were observed in small (lt 1000 m2) thawponds Overall CO2 and DOC concentrations and CO2 emis-sion decreased by an order of magnitude when lake sizeincreased by gt 3 orders of magnitude Assuming that the sizegradient of thermokarst lakes in permafrost peatlands reflectsthe age gradient (Shirokova et al 2013 Manasypov et al 2014)high CO2 flux to atmosphere from small waterbodies

corresponding to the beginning of permafrost thaw and lakeformation is likely largely sustained by relatively high input ofallochthonous DOM DOC concentrations in studiedwaterbodies were quite high (median 60ndash20 mg Lminus1 for differ-ent size bins and decreasing with lake size) and exceededvalues observed in soil pore waters of wetlands in the southernpart of the studied territory (20ndash30 mg Lminus1 Avagyanet al 2014) Also given the likely short residence time ofwater in these small thaw ponds and depressions (Manasypovet al 2015) this DOC input from the peat soil would bequickly replenished as well as rapidly bio- and photodegradedas was recently suggested for small waterbodies of easternEuropean tundra (Shirokova et al 2019) The very low qualityof DOC that remains after fast initial degradation of DOM inwaterbodies of the BZT is confirmed by the high values ofSUVA254 (from 3 to 6 L mgminus1 mminus1) which was independent ofthe season and size of the waterbody (Supporting InformationFig S9) Note that Fe concentrations in studied lakes (500ndash-1000 μg Lminus1 depending on the lake size) are too low to inter-fere with SUVA values (ie Poulin et al 2014)

In large thermokarst lakes both sediment source andsuprapermafrost input of contemporary DOM likely supportCO2 fluxes to atmosphere Allochthonous (terrestrial) organiccarbon from peat and vegetation leachates is delivered intolarge lakes via shallow subsurface flow in summer and autumnand via surface overflow during spring (Manasypov et al 2015Ala-Aho et al 2017) High concentrations of DOC and somemacro- and micronutrients (P N Si Fe Co Ni) positively cor-relating with pCO2 in small thaw ponds are due to intensiveabrasion of peat at the shore and concomitant C and elementdelivery to the waterbodies via suprapermafrost flow(Shirokova et al 2013 Raudina et al 2018) Upon maturationof lakes and decrease in ratio of the circumference length tolake surfacevolume delivery of terrestrial DOM per surfacearea of lake or lake volume is decreasing This increases the pHto above 4ndash45 thereby favoring phytoplankton and periphy-ton development as it is known in lakes of frozen peatlands ofwestern Siberia (Pavlova et al 2016) This may result indecreased CO2 concentration and fluxes to atmosphere forthese lakes Furthermore lake growth leads to certain steady-state conditionsmdashwhen delivery of DOM matches itsprocessing in the water column and burial in the sedimentsmdashas was demonstrated in western Siberia (Manasypovet al 2015) Thus after some threshold lake area (103ndash104 m2)DOC and CO2 concentrations remain constant with furtherincrease in the lake area

However the lack of information on C content and thebiodegradability of BZT lake sediments require in-depth stud-ies of C processing in benthic ecosystems of the BZT In partic-ular in order to distinguish the origin of CO2 and CH4 inthermokarst lake waters radiocarbon dating of dissolved CO2

and dissolved vs ebullited CH4 (ie Walter et al 2006Matveev et al 2016 Elder et al 2019) would be useful how-ever such research is beyond the scope of this study

Fig 3 Partial contributions of different size range lakes to overall CO2

(light gray columns) and CH4 (black columns) emissions for lakes in theBZT (200000 km2) To calculate these contributions we used medianconcentrations and fluxes to atmosphere (averaged over all sites and allseasons Supporting Information Fig S2) and lake size distribution fromGeoEye-1 and Landsat data (Supporting Information Fig S6)

Zabelina et al C emission from thermokarst lakes

9

Regional yields and upscale of fluxesThe total carbon (CO2 + CH4) flux from the water surface

of BZT lakes ranged from 100 to 400 mmol C mminus2 dminus1 or 12to 48 g C mminus2 dminus1 (with median values corresponding to lakesof different size bins) This value should be considered withcaution given a factor of 25 lower values obtained by cham-ber measurements compared to those calculated by wind-based model Given the scarcity of direct chamber measure-ments the regional estimation is possible only based on calcu-lated fluxes Nevertheless these numbers are comparable to Cfluxes measured for western Siberian lakes (17 g C mminus2 dminus1range 06ndash53 g C mminus2 dminus1 Serikova et al 2019) and withfluxes from small (4ndash150 m2) thaw ponds of NorthernEuropean peatlands (34 g C-CO2 mminus2 dminus1 01 g C-CH4 mminus2 dminus1 Kuhn et al 2018) as summarized in SupportingInformation Table S9 At the same time C fluxes in the BZTare higher than those of lakes in Arctic Alaska (Kling et al 1992Elder et al 2018) thermokarst lakes in the Vorkuta region(Heikkinen et al 2004 Marushchak et al 2013) palsapeatlands of Canada (Matveev et al 2016) and runnel pondsin continuous permafrost zone of the Canadian High Arctic(Negandhi et al 2013) It should be noted that recent model-ing of CO2 emissions from Arctic (yedoma thermokarst andglacial) lakes across the globe recommended values between0036 and 008 g C-CO2 mminus2 dminus1 (Tan et al 2017) which is10ndash100 times lower than CO2 fluxes measured in BZT There-fore we caution against use of simulated values for all perma-frost regions especially peatlands Western Siberia lowland(WSL) NE European tundra and presumably other peat-bearing lowlands of Northern Eurasia might turn out to beprominent exceptions to these simulations thus they mayrequire detailed spatial and seasonal analysis of C storage andemissions as their contribution to the global Arctic C cyclemay be strongly underestimated For example even thoughthe BZT represents 1 of overall permafrost-affected land areaits contribution alone can be comparable to the whole Arcticas estimated by Tan et al (2017)

Upscaling the results to the full territory of BZT(195443 km2) yields present day CO2 + CH4 emission of 38( 065) Tg C yrminus1 (median IQR) This value is higher thanthe total annual CO2 emissions from the North Slope ofAlaska (095 Tg C yrminus1) across a similar territory (231157 km2Elder et al 2018) Also C emission flux to atmosphere fromBZT lakes is 30 higher than the annual DOC + DIC exportfor the Pechora River (Swatershed = 324000 km2) to the ArcticOcean (11 times 106 t DIC yrminus1 17 times 106 t DOC yrminus1 Gordeevet al 1996) When compared to the same land (watershed)surface area of this region the C emission exceeds the lateralC export by a factor of 22 C emission from the BZT lakes rep-resents 30 of that reported for the 1300000 km2 of WSL(12 26 Tg C yrminus1 Serikova et al 2019) a territory of similarwetlands and peatbogs Normalized to total land area the spe-cific aquatic C yields to the atmosphere are however twotimes higher in the BZT than in the WSL (195 and

(92 20) times 10minus6 Tg kmminus2 yrminus1 respectively) It is importantto note that the mean annual air temperature is 2ndash6C higherin the BZT compared to the WSL Precipitation for bothregions is relatively similar This suggests more active terres-trial C processing in inland waters of permafrost peatlands inEuropean tundra compared to western Siberia Because thepermafrost in BZT is essentially discontinuous to sporadic andthe warming is more pronounced than in the WSL (Vasilievet al 2020) lakes of BZT could be expected to experience stron-ger and faster response to warming and permafrost thaw com-pared to those in western Siberia Upscaling the average Cemissions of European tundra and western Siberia peatlands([14 5] times 10minus6 Tg kmminus2 yrminus1) to the overall area of permafrost-affected peatlands (2864000 km2 Smith et al 2007) yields avalue of 40 14 Tg yrminus1 However this value should be consid-ered with caution given the significant spatial heterogeneity ofthermokarst lakes in permafrost landscapes (Vonk et al 2019)Another issue is distinguishing between the GHG pattern ofthermokarst and nonthermokarst ecosystems including thewinter period Here Synthetic Aperture Radar (ie Engramet al 2013) can be extremely useful

A critical issue is the emission from thaw ponds andthermokarst lakes compared to the overall C balance of perma-frost wetlands None of the BZT catchment has been subjectedto extensive C exchange studies (ie by chamber and eddycovariance techniques coupled with aquatic export and emis-sion estimation) in ways which it was performed for otherpermafrost-affected regions such as Northern Sweden (Lundinet al 2016 Kuhn et al 2018) or Alaska (Bogard et al 2019)The C balance in Eastern European tundra (BZT) is poorly con-strained due to strong seasonal and spatial variations that arenot captured in existing measurements (ie Heikkinenet al 2002 Marushchak et al 2013) The upscaled value forterrestrial C uptake for this territory is minus72 Tg yrminus1 (Heikkinenet al 2004) which strongly contrasts with the mean annualterrestrial C balance for the eastern part of the BZT(minus08 113 Tg yrminus1 Zamolodchikov and Karelin 2001)Because the magnitude of these values is comparable to net Cemissions from the waterbodies estimated in this study(+38 Tg C yrminus1) lakes may play a crucial role in overall C bal-ance of the European permafrost peatlands

The ongoing climate warming and permafrost thaw in fro-zen Siberian wetlands leads to the drainage of largethermokarst lakes (Smith et al 2005) and appearance ofnumerous new small thaw ponds (Polishchuk et al 2014Bryksina and Polishchuk 2015) Our results suggest that thismay lead to a decrease in overall regional C emissions frominland water surfaces At the same time this change of regimeis not universal across the Eurasian Arctic (Polishchuket al 2020) loss of lakes in discontinuoussporadic permafrostzones may be compensated by increases in the number oflakes in regions of continuous permafrost (Smith et al 2005)Moreover interannual fluctuations may well mask the netchanges in surface hydrology (Karlsson et al 2014) Although

Zabelina et al C emission from thermokarst lakes

10

remote sensing-based lake dynamics are not available for theBZT by analogy with western Siberia and Alaska which arealso located in the discontinuous permafrost zone one mayexpect that permafrost thaw in NE Europe will reduce the totalarea of lakes Such lake drainage occurs via soil disturbanceand increase of hydrological connectivity due to permafrostthaw and active layer thickness increase phenomena observedboth in Alaska (Hinkel et al 2003 Riordan et al 2006) andwestern Siberia (Kirpotin et al 2008) Massive drainage of largelakes may have an unexpected impact on C balance becauseair-exposed sediments have high CO2 effluxes that greatlyexceed those of surface waters (Martinsen et al 2019) Thishighlights the need for systematic assessment of C emissionsin large lakes including freshly drained lake basins

ConclusionsThis multiseasonal study of thermokarst lakes in the perma-

frost peatlands of NE European tundra (BZT) demonstrated thehighest DOC and CO2 concentrations and CO2 emissions indepressions and small thaw ponds (lt 1000 m2 water surface area)However CH4 concentrations and fluxes to atmosphere exhibiteda local maximum in ponds of 10ndash1000 m2 surface area The CO2

and CH4 concentrations and fluxes to atmosphere were highestin spring and the lowest in autumn External spatial and temporalfactors exhibited the following order of control on GHG parame-ters lake size gt season gt year Overall identification of these fac-tors allowed straightforward upscaling of field results obtainedfrom several test sites in the BZT to the full territory ofpermafrost-affected NE European tundra zone The estimated lakewater surface emission flux to atmosphere from the BZT territoryis two times higher than the lateral export of C (DOC + DIC)documented for the Pechora River to the Arctic Ocean Howeverdue to temporal and spatial variations of C fluxes uncertaintieson area GHG emissions are at least one order of magnitude forsmall thaw ponds and a factor of 3ndash5 for thermokarst lakes Possi-ble decrease in lake water area of Northern Eurasia due to lakedrainage induced by permafrost thaw may decrease overall Cemissions from the surface of lentic waters in this territory

ReferencesAbnizova A J Siemens M J Langer and J Boike 2012 Small

ponds with major impact The relevance of ponds and lakes inpermafrost landscapes to carbon dioxide emissions Global Bio-geochem Cycles 26 GB2041 doi1010292011GB004237

Ala-Aho P and others 2017 Using stable isotopes to assess sur-face water source dynamics and hydrological connectivity ina high-latitude wetland and permafrost influenced landscapeJ Hydrol 556 279ndash293 doi101016jjhydrol201711024

Alin S R M F F L Rasera C I Salimon J E Richey G WHoltgrieve A V Krusche and A Snidvongs 2011 Physicalcontrols on carbon dioxide transfer velocity and flux inlow-gradient river systems and implications for regional

carbon budgets J Geophys Res 116 G01009 doi1010292010JG001398

Andreicheva L N 2007 Pleistocene deposits in r Shapkinabasin (Bolshezemelskaya tundra) Lithol Miner Resour 193ndash110 doi101134S0024490207010099

Arsenault J J Talbot and T R Moore 2018 Environmentalcontrols of C N and P biogeochemistry in peatland pondsSci Total Environ 631ndash632 714ndash722 doi101016jscitotenv201803064

Audry S O S Pokrovsky L S Shirokova S N Kirpotin andB Dupreacute 2011 Organic matter mineralization and traceelement post-depositional redistribution in Western Siberiathermokarst lake sediments Biogeosciences 8 3341ndash3358doi105194bg-8-3341-2011

Avagyan A B R K Runkle J Hartmann and L Kutzbach2014 Spatial variations in pore-water biogeochemistrygreatly exceed temporal changes during baseflow condi-tions in a boreal river valley mire complex NorthwestRussia Wetlands 34 1171ndash1182 doi101007s13157-014-0576-4

Bartosiewicz M I Laurion F Clayer and R Maranger 2016Heat-wave effects on oxygen nutrients and phytoplanktoncan alter global warming potential of gases emitted from asmall shallow lake Environ Sci Technol 50 6267ndash6275doi101021acsest5b06312

Bastviken D I Sundgren S Natchimuthu H Reyier and MGaringlfalk 2015 Technical note Cost-efficient approaches tomeasure carbon dioxide (CO2) fluxes and concentrations interrestrial and aquatic environments using mini loggers Bio-geosciences 12 3849ndash3859 doi105194bg-12-3849-2015

Bogard M J and others 2019 Negligible cycling of terrestrialcarbon in many lakes of the arid circumpolar landscapeNat Geosci 12 180ndash185 doi101038s41561-019-0299-5

Bouchard F and others 2017 Paleolimnology ofthermokarst lakes A window into permafrost landscapeevolution Arct Sci 3 91ndash117 doi101139AS-2016-0022

Brittain J F and others 2009 Arctic rivers (chapter 9)p 337ndash379 In K Tockner U Uehlinger and C T Robinson[eds] Rivers of Europe Academic Press httpswwwelseviercombooksrivers-of-europetockner978-0-12-369449-2

Bryksina N A and Y M Polishchuk 2013 Research ofremote measurement accuracy of lake areas using spaceimages Geoinformatika 1 64ndash68 Available from httpselibraryrudownloadelibrary_18882698_14046761pdf

Bryksina N A and Y M Polishchuk 2015 Analysis of changes inthe number of thermokarst lakes in permafrost of Western Sibe-ria on the basis of satellite images Kriosfera Zemli 19 114ndash120httpwwwizdatgeorupdfkrio2015-2114pdf

Cael B B and D A Seekell 2016 The size-distribution ofEarthrsquos lakes Sci Rep 6 29633 doi101038srep29633

Casper P O Chan A Furtado and D Adams 2003 Meth-ane in an acidic bog lake The influence of peat in thecatchment on the biogeochemistry of methane Aquat Sci65 36ndash46 doi101007s000270300003

Zabelina et al C emission from thermokarst lakes

11

demonstrated a factor of 5 decrease with increasing lake size(from lt 10 m2 to gt 1000 m2 Supporting Information Fig S2c)The CO2 fluxes progressively decreased from 500 to1000 mmol mminus2 dminus1 in depressions (lt 10 m2) to about100 mmol mminus2 dminus1 in lakes larger than 1000 m2 (Fig 2d) Athreefold decrease in median CO2 fluxes from ponds lt 100 m2

to lakes ge 1000 m2 was also observed (Supporting InformationFig S2d) A general decrease of DOC CO2 concentrationFCO2 and optical density at 254 nm (UV254) with lake surfacearea across all sites and all seasons is confirmed by strong(p = 00000) power dependence (Supporting InformationTable S3) In contrast to CO2 CH4 concentrations and fluxesto atmosphere exhibited strong variability ranging from 01to 100 μmol Lminus1 and mmol mminus2 dminus1 respectively There was alocal maximum of CH4 concentration and emission in thawponds of 10ndash1000 m2 surface area (Fig 2ef Supporting Infor-mation Fig S2ef respectively) Considering all calculatedfluxes to atmosphere (n = 135) the contribution of C-СН4 tothe total C emissions calculated by wind-based modelingranged from 002 to 19 (with a mean of 26 and amedian of 11) Chamber-measured CH4 fluxes in eight bod-ies of water constituted between 02 and 58 of the totalCO2 + CH4 fluxes (Supporting Information Table S4a)

The CO2 fluxes to atmosphere obtained by using wind-basedmodels were on average 25 times (range of 08ndash189 times n = 8)higher than those directly measured using chambers (SupportingInformation Table S4a) It should be noted that Laurionet al (2010) reported an average factor of 38 of the excess of cal-culated fluxes in comparison to measured CO2 fluxes for similarpeatland thaw ponds The difference between measured and cal-culated CH4 fluxes to atmosphere was significantly (p lt 005)higher than for CO2 (range 15ndash194 mean 68) In thermokarstlakes of western Siberian permafrost peatlands CO2 and CH4

fluxes to atmosphere calculated with wind-based models (eitherCole and Caraco 1998 or Vachon and Prairie 2013) were on aver-age three and four times lower than measured fluxes (Serikovaet al 2019) These authors explained this discrepancy as due tothe remoteness of the studied lakes in relation to their respectivemeteorological stations and the lack of local wind data This isalso true for most BZT sites located approximately 40 km fromthe Naryan-Mar station The most likely cause for CH4 differencesis bubble evasion (Casper et al 2003) as it was not possible tosample shallow waterbodies (lt 1 m depth) with zero disturbanceof the organic detritus at lake bottom and in surrounding coastalwaterlogged zones Note that the agreement between the twomethods was much better in lakes with mineral banks wherevisual bubble disturbance during sampling was minimal Anotherreason for observed discrepancies may be due to the wind effecton the gas transfer coefficient being overestimated in smallaquatic systems with small fetch and sheltered sampling pointsalong the shoreline (Alin et al 2011)

As a measure of variability associated with model parame-ters we calculated the effects of variable wind speed for

estimates of diffusive GHG fluxes to atmosphere using Eq 1for 25 lakes gt 1000 m2 sampled in July 2015ndash2018 at theNarian-Mar site (Supporting Information Table S4b) Typicallyvariation of the wind speed by a factor of 2 yielded a factor of13ndash15 variation in relevant median values of GHG fluxes toatmosphere Only drastic changes in wind speed (by a factorof 5 or more) had sizable effect (exceeding the interquartilerange [IQR]) on CH4 and CO2 fluxes Overall this suggeststhat the offset between measured and calculated fluxes is notcaused by uncertainties in wind speed alone

Seasonal annual and spatial variability of Cconcentrations and emission

GHG concentrations and fluxes to atmosphere showedlarger spatial variability across lake sizes than temporal vari-ability across seasons (June July and October) Differences inDOC DIC and GHG between seasons (2015ndash2018) wereassessed for lakes gt 1000 m2 from the Naryan-Mar site(Supporting Information Table S5) For these lakes the DICand CO2 concentrations were sizably lower in spring com-pared to summer (p lt 005) however there was no differencebetween summer and autumn (p gt 005 Supporting Informa-tion Table S5 and Fig S3bc) The DOC showed an inverse pat-tern with spring lt summer = autumn (p lt 005 SupportingInformation Fig S3a) CH4 concentrations and CO2 and CH4

fluxes to atmosphere followed the order ldquospring gt summerrdquo(p lt 005) but there was no difference (p gt 005) between sum-mer and autumn (Supporting Information Fig S3dndashf) Therewere no seasonal differences (p gt 005) in DIC DOC CO2 andCH4 concentrations and fluxes for depressions (lt 10 m2 area)Interannual variations in DOC CO2 CH4 concentrations andCO2 and CH4 fluxes (Supporting Information Fig S4) did notexhibit any link to potential drivers like annual andor sea-sonal temperature or precipitation for the specific year ofobservation (Supporting Information Table S2) In fact despitesizeable contrast in mean June and July temperature and pre-cipitation between 2015 and 2018 (p le 001 MannndashWhitneyU-test) no difference in DOC DIC CO2 and CH4 was evidentfor these years (p gt 005)

Using Spearman rank order correlation and multiple regres-sion approach we examined the impact of water temperatureand cumulative precipitation and temperature during thesummer prior to the sampling campaign on the GHG DOCand DIC concentrations and GHG fluxes to atmosphere(Supporting Information Table S6a) Water temperature neg-atively impacted CO2 and positively impacted DOC(RSpearman = minus020 and 022 respectively at p lt 005) TheSpearman rank order correlation yielded positive depen-dence of CH4 and CO2 on cumulative air temperature priorto sampling (RSpearman = 020ndash028) as well as on the meanair temperature during June and July (RSpearman = 026ndash030in lakes RSpearman = 043ndash046 in small thaw ponds)(Supporting Information Table S6a) Precipitation exerted a

Zabelina et al C emission from thermokarst lakes

7

negative impact on GHG concentration and fluxes to atmo-sphere (RSpearman = minus023 to minus055) but a positive impact onspecific UV absorbance (SUVA RSpearman = 024ndash038) Over-all the impact of both temperature and precipitation wasmuch more pronounced for small thaw ponds (lt 500 m2)compared to thermokarst lakes (gt 500 m2) The multipleregression approach demonstrated that the GHG concentra-tion variability is partially determined by waterbody areaand cumulative climatic parameters such as air temperatureand precipitation (Supporting Information Table S6b) How-ever the multiple regression was not capable of explainingmore than 20ndash40 of the overall variability in CO2 CH4DOC and DIC concentrations

The variability of C concentration and emissions acrosssites was assessed for Naryan-Mar Shapkino and Khorey-Versites in July 2016 This comparison was conducted separatelyfor thaw ponds with surface areas lt 50 m2 and 50ndash500 m2 andfor lakes with surface areas gt 1000 m2 (Supporting Informa-tion Fig S5) The difference between sites was assessed via thenonparametric MannndashWhitney U-test (Supporting Informa-tion Table S7) All sites were similar in DOC CO2 CH4 con-centrations and C emissions to atmosphere except lakes inShapkino which had three times higher DIC and CO2 concen-trations than those of Naryan-Mar and Khorey-Ver This isconsistent with the shallower position of mineral soils onclays and fluvioglacial deposits at the Shapkino site wheresome carbonate shells have been reported (Andreicheva 2007)

In order to assess possible control of hydrochemical param-eters of waterbodies on CO2 and CH4 concentrations we cal-culated Spearman rank correlation coefficients (SupportingInformation Table S8) The concentrations of CO2 and CH4

negatively correlated with pH and O2 and positively corre-lated with DIC DOC and Total Dissolved Solids (TDS) CO2

concentration positively correlated with PO3minus4 Ptot NOminus

3 NtotSi DIC DOC Fe Co Ni and negatively correlated with NH+

4 CH4 concentration positively correlated with Ntot Si DICDOC Fe Co and Ni and negatively correlated with NH+

4 and

SO2minus4 (Supporting Information Table S8)

Regional-scale estimation of C emissions from BZTterritory (based on satellite imagery)

For the full BZT territory (195000 km2) the total numberof inventoried lakes ranging from 2 m2 to 5 times 107 m2 in sizewas 896 times 106 with a total water area of 7800 km2 (Table 1and Supporting Information Fig S6ab) Lakes lt 5000 m2 pro-vided about 10 of overall lake water area in BZT While smalllakes (lt 2000 m2) provided 97 of the total lake numberstheir area contributed to only about 6 of total area coverageThermokarst lakes between 5000 and 108 m2 in areaaccounted for only 15 of the lake count but provided gt 90of total lake area Note that from an optical remote sensingview point there is no difference between thermokarst andnonthermokarst lakes because their reflections yield similar T

able

1Lake

characteristic

sC

pools

andGHG

flux

esinventory[m

edianan

d(IQR)]forlakesof

EasternEu

rope

an(Bolshezem

elskaya)

Tund

ra(195

000

km2)

Themax

imal

contrib

utionto

both

CO

2an

dCH4po

olisprov

ided

bylakesof

gt10

5m

2surfaceareaLakeslarger

than

105m

2prov

idegt43

of

totalC

H4am

ount

whe

reas

thesharefrom

smalllakes

(lt10

00m

2)islt20

N

otethat

smalllakes

contrib

uted

less

toov

erallC

O2an

dDOClevelsthan

toCH4level

Lake

size

range(m

2)

Dep

th(m

)La

kenum

ber

Area

(m2)

Volume

(m3)

C-СO

2(g

)C-СН4(g

)DIC

(g[С

])DOC(g

[С])

C-CO

2

flux

(gСdminus1)

C-CH4

flux

(gСdminus1)

lt10

010

482

times10

6235

times10

7235

times10

6818

times10

6

(201times

107)

460

times10

4

(425times

105)

267

times10

6

(207times

106)

133

times10

8

(726times

107)

720

times10

7

(996times

107)

334

times10

5

(264times

106)

10ndash10

2029

303

times10

6915

times10

7265

times10

7156

times10

8

(228times

108)

514

times10

6

(100times

107)

539

times10

7

(691times

107)

105

times10

9

(125times

109)

448

times10

8

(403times

108)

149

times10

7

(183times

107)

102ndash10

3050

783

times10

5231

times10

8115

times10

8292

times10

8

(365times

108)

127

times10

7

(453times

107)

117

times10

8

(128times

108)

414

times10

9

(304times

109)

450

times10

8

(435times

108)

200

times10

7

(529times

107)

103ndash10

4067

246

times10

5829

times10

8556

times10

8801

times10

8

(445times

108)

106

times10

7

(138times

107)

499

times10

8

(341times

108)

107

times10

10

(109times

1010)

963

times10

8

(979times

108)

137

times10

7

(118times

107)

104ndash10

5077

653

times10

4213

times10

9164

times10

9186

times10

9

(275times

109)

216

times10

7

(180times

107)

110

times10

9

(272times

109)

226

times10

10

(738times

109)

282

times10

9

(252times

109)

210

times10

7

(244times

107)

gt105

083

130

times10

4457

times10

9379

times10

9566

times10

9

(543times

109)

386

times10

7

(903times

107)

453

times10

9

(279times

109)

518

times10

10

(177times

1010)

545

times10

9

(620times

109)

334

times10

7

(651times

107)

Total

896

times10

6784

times10

9613

times10

9878

times10

9887

times10

7631

times10

9903

times10

10

102

times10

10

103

times10

8

Zabelina et al C emission from thermokarst lakes

8

signals for Landsat-8 OLI However according to publishedexpert estimations the majority of lakes in the BZT are ofthermokarst origin (Kravtzova and Bystrova 2009) this originis also confirmed by their generally isometric shape as com-pared to elongated lakes with glacial origins Both the lakeabundance and area distribution obtained for the permafrost-affected part of BZT are in fairly good agreement (SupportingInformation Fig S7) with those reported for permafrost-affected parts of western Siberia (Polishchuk et al 2017 2018)and Swedish territory (Cael and Seekell 2016) In the BZT thepower law extrapolation (Pareto law) strongly overestimatesthe number of small lakes (lt 1000 m2) compared to empiricaldata (Supporting Information Fig S7) which is in agreementwith other regions of the world

The maximal contribution to overall GHG emissions fromthe territory of 195000 km2 (38 Tg yrminus1) was provided bylakes gt 100000 m2 whereas contributions from small thawponds (lt 1000 m2) were below 20 (Table 1 Fig 3) The dis-persity of the areal GHG fluxes to atmosphere is quite highachieving one order of magnitude in small thaw ponds and afactor of 2ndash5 for thermokarst lakes (Fig 2) These uncertaintiesvisible as IQR in Supporting Information Figs S2ndashS5 S8 do notoriginate from analytical procedures and rather reflect variabil-ity between sites and seasons It is important to note that thecontribution of small thaw ponds to overall C balance in BZT isbelow the actual uncertainties for total pools and emissions

DiscussionC pools fluxes to atmosphere and possible mechanisms

The highest DOC and CO2 concentrations and CO2 fluxesto atmosphere were observed in small (lt 1000 m2) thawponds Overall CO2 and DOC concentrations and CO2 emis-sion decreased by an order of magnitude when lake sizeincreased by gt 3 orders of magnitude Assuming that the sizegradient of thermokarst lakes in permafrost peatlands reflectsthe age gradient (Shirokova et al 2013 Manasypov et al 2014)high CO2 flux to atmosphere from small waterbodies

corresponding to the beginning of permafrost thaw and lakeformation is likely largely sustained by relatively high input ofallochthonous DOM DOC concentrations in studiedwaterbodies were quite high (median 60ndash20 mg Lminus1 for differ-ent size bins and decreasing with lake size) and exceededvalues observed in soil pore waters of wetlands in the southernpart of the studied territory (20ndash30 mg Lminus1 Avagyanet al 2014) Also given the likely short residence time ofwater in these small thaw ponds and depressions (Manasypovet al 2015) this DOC input from the peat soil would bequickly replenished as well as rapidly bio- and photodegradedas was recently suggested for small waterbodies of easternEuropean tundra (Shirokova et al 2019) The very low qualityof DOC that remains after fast initial degradation of DOM inwaterbodies of the BZT is confirmed by the high values ofSUVA254 (from 3 to 6 L mgminus1 mminus1) which was independent ofthe season and size of the waterbody (Supporting InformationFig S9) Note that Fe concentrations in studied lakes (500ndash-1000 μg Lminus1 depending on the lake size) are too low to inter-fere with SUVA values (ie Poulin et al 2014)

In large thermokarst lakes both sediment source andsuprapermafrost input of contemporary DOM likely supportCO2 fluxes to atmosphere Allochthonous (terrestrial) organiccarbon from peat and vegetation leachates is delivered intolarge lakes via shallow subsurface flow in summer and autumnand via surface overflow during spring (Manasypov et al 2015Ala-Aho et al 2017) High concentrations of DOC and somemacro- and micronutrients (P N Si Fe Co Ni) positively cor-relating with pCO2 in small thaw ponds are due to intensiveabrasion of peat at the shore and concomitant C and elementdelivery to the waterbodies via suprapermafrost flow(Shirokova et al 2013 Raudina et al 2018) Upon maturationof lakes and decrease in ratio of the circumference length tolake surfacevolume delivery of terrestrial DOM per surfacearea of lake or lake volume is decreasing This increases the pHto above 4ndash45 thereby favoring phytoplankton and periphy-ton development as it is known in lakes of frozen peatlands ofwestern Siberia (Pavlova et al 2016) This may result indecreased CO2 concentration and fluxes to atmosphere forthese lakes Furthermore lake growth leads to certain steady-state conditionsmdashwhen delivery of DOM matches itsprocessing in the water column and burial in the sedimentsmdashas was demonstrated in western Siberia (Manasypovet al 2015) Thus after some threshold lake area (103ndash104 m2)DOC and CO2 concentrations remain constant with furtherincrease in the lake area

However the lack of information on C content and thebiodegradability of BZT lake sediments require in-depth stud-ies of C processing in benthic ecosystems of the BZT In partic-ular in order to distinguish the origin of CO2 and CH4 inthermokarst lake waters radiocarbon dating of dissolved CO2

and dissolved vs ebullited CH4 (ie Walter et al 2006Matveev et al 2016 Elder et al 2019) would be useful how-ever such research is beyond the scope of this study

Fig 3 Partial contributions of different size range lakes to overall CO2

(light gray columns) and CH4 (black columns) emissions for lakes in theBZT (200000 km2) To calculate these contributions we used medianconcentrations and fluxes to atmosphere (averaged over all sites and allseasons Supporting Information Fig S2) and lake size distribution fromGeoEye-1 and Landsat data (Supporting Information Fig S6)

Zabelina et al C emission from thermokarst lakes

9

Regional yields and upscale of fluxesThe total carbon (CO2 + CH4) flux from the water surface

of BZT lakes ranged from 100 to 400 mmol C mminus2 dminus1 or 12to 48 g C mminus2 dminus1 (with median values corresponding to lakesof different size bins) This value should be considered withcaution given a factor of 25 lower values obtained by cham-ber measurements compared to those calculated by wind-based model Given the scarcity of direct chamber measure-ments the regional estimation is possible only based on calcu-lated fluxes Nevertheless these numbers are comparable to Cfluxes measured for western Siberian lakes (17 g C mminus2 dminus1range 06ndash53 g C mminus2 dminus1 Serikova et al 2019) and withfluxes from small (4ndash150 m2) thaw ponds of NorthernEuropean peatlands (34 g C-CO2 mminus2 dminus1 01 g C-CH4 mminus2 dminus1 Kuhn et al 2018) as summarized in SupportingInformation Table S9 At the same time C fluxes in the BZTare higher than those of lakes in Arctic Alaska (Kling et al 1992Elder et al 2018) thermokarst lakes in the Vorkuta region(Heikkinen et al 2004 Marushchak et al 2013) palsapeatlands of Canada (Matveev et al 2016) and runnel pondsin continuous permafrost zone of the Canadian High Arctic(Negandhi et al 2013) It should be noted that recent model-ing of CO2 emissions from Arctic (yedoma thermokarst andglacial) lakes across the globe recommended values between0036 and 008 g C-CO2 mminus2 dminus1 (Tan et al 2017) which is10ndash100 times lower than CO2 fluxes measured in BZT There-fore we caution against use of simulated values for all perma-frost regions especially peatlands Western Siberia lowland(WSL) NE European tundra and presumably other peat-bearing lowlands of Northern Eurasia might turn out to beprominent exceptions to these simulations thus they mayrequire detailed spatial and seasonal analysis of C storage andemissions as their contribution to the global Arctic C cyclemay be strongly underestimated For example even thoughthe BZT represents 1 of overall permafrost-affected land areaits contribution alone can be comparable to the whole Arcticas estimated by Tan et al (2017)

Upscaling the results to the full territory of BZT(195443 km2) yields present day CO2 + CH4 emission of 38( 065) Tg C yrminus1 (median IQR) This value is higher thanthe total annual CO2 emissions from the North Slope ofAlaska (095 Tg C yrminus1) across a similar territory (231157 km2Elder et al 2018) Also C emission flux to atmosphere fromBZT lakes is 30 higher than the annual DOC + DIC exportfor the Pechora River (Swatershed = 324000 km2) to the ArcticOcean (11 times 106 t DIC yrminus1 17 times 106 t DOC yrminus1 Gordeevet al 1996) When compared to the same land (watershed)surface area of this region the C emission exceeds the lateralC export by a factor of 22 C emission from the BZT lakes rep-resents 30 of that reported for the 1300000 km2 of WSL(12 26 Tg C yrminus1 Serikova et al 2019) a territory of similarwetlands and peatbogs Normalized to total land area the spe-cific aquatic C yields to the atmosphere are however twotimes higher in the BZT than in the WSL (195 and

(92 20) times 10minus6 Tg kmminus2 yrminus1 respectively) It is importantto note that the mean annual air temperature is 2ndash6C higherin the BZT compared to the WSL Precipitation for bothregions is relatively similar This suggests more active terres-trial C processing in inland waters of permafrost peatlands inEuropean tundra compared to western Siberia Because thepermafrost in BZT is essentially discontinuous to sporadic andthe warming is more pronounced than in the WSL (Vasilievet al 2020) lakes of BZT could be expected to experience stron-ger and faster response to warming and permafrost thaw com-pared to those in western Siberia Upscaling the average Cemissions of European tundra and western Siberia peatlands([14 5] times 10minus6 Tg kmminus2 yrminus1) to the overall area of permafrost-affected peatlands (2864000 km2 Smith et al 2007) yields avalue of 40 14 Tg yrminus1 However this value should be consid-ered with caution given the significant spatial heterogeneity ofthermokarst lakes in permafrost landscapes (Vonk et al 2019)Another issue is distinguishing between the GHG pattern ofthermokarst and nonthermokarst ecosystems including thewinter period Here Synthetic Aperture Radar (ie Engramet al 2013) can be extremely useful

A critical issue is the emission from thaw ponds andthermokarst lakes compared to the overall C balance of perma-frost wetlands None of the BZT catchment has been subjectedto extensive C exchange studies (ie by chamber and eddycovariance techniques coupled with aquatic export and emis-sion estimation) in ways which it was performed for otherpermafrost-affected regions such as Northern Sweden (Lundinet al 2016 Kuhn et al 2018) or Alaska (Bogard et al 2019)The C balance in Eastern European tundra (BZT) is poorly con-strained due to strong seasonal and spatial variations that arenot captured in existing measurements (ie Heikkinenet al 2002 Marushchak et al 2013) The upscaled value forterrestrial C uptake for this territory is minus72 Tg yrminus1 (Heikkinenet al 2004) which strongly contrasts with the mean annualterrestrial C balance for the eastern part of the BZT(minus08 113 Tg yrminus1 Zamolodchikov and Karelin 2001)Because the magnitude of these values is comparable to net Cemissions from the waterbodies estimated in this study(+38 Tg C yrminus1) lakes may play a crucial role in overall C bal-ance of the European permafrost peatlands

The ongoing climate warming and permafrost thaw in fro-zen Siberian wetlands leads to the drainage of largethermokarst lakes (Smith et al 2005) and appearance ofnumerous new small thaw ponds (Polishchuk et al 2014Bryksina and Polishchuk 2015) Our results suggest that thismay lead to a decrease in overall regional C emissions frominland water surfaces At the same time this change of regimeis not universal across the Eurasian Arctic (Polishchuket al 2020) loss of lakes in discontinuoussporadic permafrostzones may be compensated by increases in the number oflakes in regions of continuous permafrost (Smith et al 2005)Moreover interannual fluctuations may well mask the netchanges in surface hydrology (Karlsson et al 2014) Although

Zabelina et al C emission from thermokarst lakes

10

remote sensing-based lake dynamics are not available for theBZT by analogy with western Siberia and Alaska which arealso located in the discontinuous permafrost zone one mayexpect that permafrost thaw in NE Europe will reduce the totalarea of lakes Such lake drainage occurs via soil disturbanceand increase of hydrological connectivity due to permafrostthaw and active layer thickness increase phenomena observedboth in Alaska (Hinkel et al 2003 Riordan et al 2006) andwestern Siberia (Kirpotin et al 2008) Massive drainage of largelakes may have an unexpected impact on C balance becauseair-exposed sediments have high CO2 effluxes that greatlyexceed those of surface waters (Martinsen et al 2019) Thishighlights the need for systematic assessment of C emissionsin large lakes including freshly drained lake basins

ConclusionsThis multiseasonal study of thermokarst lakes in the perma-

frost peatlands of NE European tundra (BZT) demonstrated thehighest DOC and CO2 concentrations and CO2 emissions indepressions and small thaw ponds (lt 1000 m2 water surface area)However CH4 concentrations and fluxes to atmosphere exhibiteda local maximum in ponds of 10ndash1000 m2 surface area The CO2

and CH4 concentrations and fluxes to atmosphere were highestin spring and the lowest in autumn External spatial and temporalfactors exhibited the following order of control on GHG parame-ters lake size gt season gt year Overall identification of these fac-tors allowed straightforward upscaling of field results obtainedfrom several test sites in the BZT to the full territory ofpermafrost-affected NE European tundra zone The estimated lakewater surface emission flux to atmosphere from the BZT territoryis two times higher than the lateral export of C (DOC + DIC)documented for the Pechora River to the Arctic Ocean Howeverdue to temporal and spatial variations of C fluxes uncertaintieson area GHG emissions are at least one order of magnitude forsmall thaw ponds and a factor of 3ndash5 for thermokarst lakes Possi-ble decrease in lake water area of Northern Eurasia due to lakedrainage induced by permafrost thaw may decrease overall Cemissions from the surface of lentic waters in this territory

ReferencesAbnizova A J Siemens M J Langer and J Boike 2012 Small

ponds with major impact The relevance of ponds and lakes inpermafrost landscapes to carbon dioxide emissions Global Bio-geochem Cycles 26 GB2041 doi1010292011GB004237

Ala-Aho P and others 2017 Using stable isotopes to assess sur-face water source dynamics and hydrological connectivity ina high-latitude wetland and permafrost influenced landscapeJ Hydrol 556 279ndash293 doi101016jjhydrol201711024

Alin S R M F F L Rasera C I Salimon J E Richey G WHoltgrieve A V Krusche and A Snidvongs 2011 Physicalcontrols on carbon dioxide transfer velocity and flux inlow-gradient river systems and implications for regional

carbon budgets J Geophys Res 116 G01009 doi1010292010JG001398

Andreicheva L N 2007 Pleistocene deposits in r Shapkinabasin (Bolshezemelskaya tundra) Lithol Miner Resour 193ndash110 doi101134S0024490207010099

Arsenault J J Talbot and T R Moore 2018 Environmentalcontrols of C N and P biogeochemistry in peatland pondsSci Total Environ 631ndash632 714ndash722 doi101016jscitotenv201803064

Audry S O S Pokrovsky L S Shirokova S N Kirpotin andB Dupreacute 2011 Organic matter mineralization and traceelement post-depositional redistribution in Western Siberiathermokarst lake sediments Biogeosciences 8 3341ndash3358doi105194bg-8-3341-2011

Avagyan A B R K Runkle J Hartmann and L Kutzbach2014 Spatial variations in pore-water biogeochemistrygreatly exceed temporal changes during baseflow condi-tions in a boreal river valley mire complex NorthwestRussia Wetlands 34 1171ndash1182 doi101007s13157-014-0576-4

Bartosiewicz M I Laurion F Clayer and R Maranger 2016Heat-wave effects on oxygen nutrients and phytoplanktoncan alter global warming potential of gases emitted from asmall shallow lake Environ Sci Technol 50 6267ndash6275doi101021acsest5b06312

Bastviken D I Sundgren S Natchimuthu H Reyier and MGaringlfalk 2015 Technical note Cost-efficient approaches tomeasure carbon dioxide (CO2) fluxes and concentrations interrestrial and aquatic environments using mini loggers Bio-geosciences 12 3849ndash3859 doi105194bg-12-3849-2015

Bogard M J and others 2019 Negligible cycling of terrestrialcarbon in many lakes of the arid circumpolar landscapeNat Geosci 12 180ndash185 doi101038s41561-019-0299-5

Bouchard F and others 2017 Paleolimnology ofthermokarst lakes A window into permafrost landscapeevolution Arct Sci 3 91ndash117 doi101139AS-2016-0022

Brittain J F and others 2009 Arctic rivers (chapter 9)p 337ndash379 In K Tockner U Uehlinger and C T Robinson[eds] Rivers of Europe Academic Press httpswwwelseviercombooksrivers-of-europetockner978-0-12-369449-2

Bryksina N A and Y M Polishchuk 2013 Research ofremote measurement accuracy of lake areas using spaceimages Geoinformatika 1 64ndash68 Available from httpselibraryrudownloadelibrary_18882698_14046761pdf

Bryksina N A and Y M Polishchuk 2015 Analysis of changes inthe number of thermokarst lakes in permafrost of Western Sibe-ria on the basis of satellite images Kriosfera Zemli 19 114ndash120httpwwwizdatgeorupdfkrio2015-2114pdf

Cael B B and D A Seekell 2016 The size-distribution ofEarthrsquos lakes Sci Rep 6 29633 doi101038srep29633

Casper P O Chan A Furtado and D Adams 2003 Meth-ane in an acidic bog lake The influence of peat in thecatchment on the biogeochemistry of methane Aquat Sci65 36ndash46 doi101007s000270300003

Zabelina et al C emission from thermokarst lakes

11

negative impact on GHG concentration and fluxes to atmo-sphere (RSpearman = minus023 to minus055) but a positive impact onspecific UV absorbance (SUVA RSpearman = 024ndash038) Over-all the impact of both temperature and precipitation wasmuch more pronounced for small thaw ponds (lt 500 m2)compared to thermokarst lakes (gt 500 m2) The multipleregression approach demonstrated that the GHG concentra-tion variability is partially determined by waterbody areaand cumulative climatic parameters such as air temperatureand precipitation (Supporting Information Table S6b) How-ever the multiple regression was not capable of explainingmore than 20ndash40 of the overall variability in CO2 CH4DOC and DIC concentrations

The variability of C concentration and emissions acrosssites was assessed for Naryan-Mar Shapkino and Khorey-Versites in July 2016 This comparison was conducted separatelyfor thaw ponds with surface areas lt 50 m2 and 50ndash500 m2 andfor lakes with surface areas gt 1000 m2 (Supporting Informa-tion Fig S5) The difference between sites was assessed via thenonparametric MannndashWhitney U-test (Supporting Informa-tion Table S7) All sites were similar in DOC CO2 CH4 con-centrations and C emissions to atmosphere except lakes inShapkino which had three times higher DIC and CO2 concen-trations than those of Naryan-Mar and Khorey-Ver This isconsistent with the shallower position of mineral soils onclays and fluvioglacial deposits at the Shapkino site wheresome carbonate shells have been reported (Andreicheva 2007)

In order to assess possible control of hydrochemical param-eters of waterbodies on CO2 and CH4 concentrations we cal-culated Spearman rank correlation coefficients (SupportingInformation Table S8) The concentrations of CO2 and CH4

negatively correlated with pH and O2 and positively corre-lated with DIC DOC and Total Dissolved Solids (TDS) CO2

concentration positively correlated with PO3minus4 Ptot NOminus

3 NtotSi DIC DOC Fe Co Ni and negatively correlated with NH+

4 CH4 concentration positively correlated with Ntot Si DICDOC Fe Co and Ni and negatively correlated with NH+

4 and

SO2minus4 (Supporting Information Table S8)

Regional-scale estimation of C emissions from BZTterritory (based on satellite imagery)

For the full BZT territory (195000 km2) the total numberof inventoried lakes ranging from 2 m2 to 5 times 107 m2 in sizewas 896 times 106 with a total water area of 7800 km2 (Table 1and Supporting Information Fig S6ab) Lakes lt 5000 m2 pro-vided about 10 of overall lake water area in BZT While smalllakes (lt 2000 m2) provided 97 of the total lake numberstheir area contributed to only about 6 of total area coverageThermokarst lakes between 5000 and 108 m2 in areaaccounted for only 15 of the lake count but provided gt 90of total lake area Note that from an optical remote sensingview point there is no difference between thermokarst andnonthermokarst lakes because their reflections yield similar T

able

1Lake

characteristic

sC

pools

andGHG

flux

esinventory[m

edianan

d(IQR)]forlakesof

EasternEu

rope

an(Bolshezem

elskaya)

Tund

ra(195

000

km2)

Themax

imal

contrib

utionto

both

CO

2an

dCH4po

olisprov

ided

bylakesof

gt10

5m

2surfaceareaLakeslarger

than

105m

2prov

idegt43

of

totalC

H4am

ount

whe

reas

thesharefrom

smalllakes

(lt10

00m

2)islt20

N

otethat

smalllakes

contrib

uted

less

toov

erallC

O2an

dDOClevelsthan

toCH4level

Lake

size

range(m

2)

Dep

th(m

)La

kenum

ber

Area

(m2)

Volume

(m3)

C-СO

2(g

)C-СН4(g

)DIC

(g[С

])DOC(g

[С])

C-CO

2

flux

(gСdminus1)

C-CH4

flux

(gСdminus1)

lt10

010

482

times10

6235

times10

7235

times10

6818

times10

6

(201times

107)

460

times10

4

(425times

105)

267

times10

6

(207times

106)

133

times10

8

(726times

107)

720

times10

7

(996times

107)

334

times10

5

(264times

106)

10ndash10

2029

303

times10

6915

times10

7265

times10

7156

times10

8

(228times

108)

514

times10

6

(100times

107)

539

times10

7

(691times

107)

105

times10

9

(125times

109)

448

times10

8

(403times

108)

149

times10

7

(183times

107)

102ndash10

3050

783

times10

5231

times10

8115

times10

8292

times10

8

(365times

108)

127

times10

7

(453times

107)

117

times10

8

(128times

108)

414

times10

9

(304times

109)

450

times10

8

(435times

108)

200

times10

7

(529times

107)

103ndash10

4067

246

times10

5829

times10

8556

times10

8801

times10

8

(445times

108)

106

times10

7

(138times

107)

499

times10

8

(341times

108)

107

times10

10

(109times

1010)

963

times10

8

(979times

108)

137

times10

7

(118times

107)

104ndash10

5077

653

times10

4213

times10

9164

times10

9186

times10

9

(275times

109)

216

times10

7

(180times

107)

110

times10

9

(272times

109)

226

times10

10

(738times

109)

282

times10

9

(252times

109)

210

times10

7

(244times

107)

gt105

083

130

times10

4457

times10

9379

times10

9566

times10

9

(543times

109)

386

times10

7

(903times

107)

453

times10

9

(279times

109)

518

times10

10

(177times

1010)

545

times10

9

(620times

109)

334

times10

7

(651times

107)

Total

896

times10

6784

times10

9613

times10

9878

times10

9887

times10

7631

times10

9903

times10

10

102

times10

10

103

times10

8

Zabelina et al C emission from thermokarst lakes

8

signals for Landsat-8 OLI However according to publishedexpert estimations the majority of lakes in the BZT are ofthermokarst origin (Kravtzova and Bystrova 2009) this originis also confirmed by their generally isometric shape as com-pared to elongated lakes with glacial origins Both the lakeabundance and area distribution obtained for the permafrost-affected part of BZT are in fairly good agreement (SupportingInformation Fig S7) with those reported for permafrost-affected parts of western Siberia (Polishchuk et al 2017 2018)and Swedish territory (Cael and Seekell 2016) In the BZT thepower law extrapolation (Pareto law) strongly overestimatesthe number of small lakes (lt 1000 m2) compared to empiricaldata (Supporting Information Fig S7) which is in agreementwith other regions of the world

The maximal contribution to overall GHG emissions fromthe territory of 195000 km2 (38 Tg yrminus1) was provided bylakes gt 100000 m2 whereas contributions from small thawponds (lt 1000 m2) were below 20 (Table 1 Fig 3) The dis-persity of the areal GHG fluxes to atmosphere is quite highachieving one order of magnitude in small thaw ponds and afactor of 2ndash5 for thermokarst lakes (Fig 2) These uncertaintiesvisible as IQR in Supporting Information Figs S2ndashS5 S8 do notoriginate from analytical procedures and rather reflect variabil-ity between sites and seasons It is important to note that thecontribution of small thaw ponds to overall C balance in BZT isbelow the actual uncertainties for total pools and emissions

DiscussionC pools fluxes to atmosphere and possible mechanisms

The highest DOC and CO2 concentrations and CO2 fluxesto atmosphere were observed in small (lt 1000 m2) thawponds Overall CO2 and DOC concentrations and CO2 emis-sion decreased by an order of magnitude when lake sizeincreased by gt 3 orders of magnitude Assuming that the sizegradient of thermokarst lakes in permafrost peatlands reflectsthe age gradient (Shirokova et al 2013 Manasypov et al 2014)high CO2 flux to atmosphere from small waterbodies

corresponding to the beginning of permafrost thaw and lakeformation is likely largely sustained by relatively high input ofallochthonous DOM DOC concentrations in studiedwaterbodies were quite high (median 60ndash20 mg Lminus1 for differ-ent size bins and decreasing with lake size) and exceededvalues observed in soil pore waters of wetlands in the southernpart of the studied territory (20ndash30 mg Lminus1 Avagyanet al 2014) Also given the likely short residence time ofwater in these small thaw ponds and depressions (Manasypovet al 2015) this DOC input from the peat soil would bequickly replenished as well as rapidly bio- and photodegradedas was recently suggested for small waterbodies of easternEuropean tundra (Shirokova et al 2019) The very low qualityof DOC that remains after fast initial degradation of DOM inwaterbodies of the BZT is confirmed by the high values ofSUVA254 (from 3 to 6 L mgminus1 mminus1) which was independent ofthe season and size of the waterbody (Supporting InformationFig S9) Note that Fe concentrations in studied lakes (500ndash-1000 μg Lminus1 depending on the lake size) are too low to inter-fere with SUVA values (ie Poulin et al 2014)

In large thermokarst lakes both sediment source andsuprapermafrost input of contemporary DOM likely supportCO2 fluxes to atmosphere Allochthonous (terrestrial) organiccarbon from peat and vegetation leachates is delivered intolarge lakes via shallow subsurface flow in summer and autumnand via surface overflow during spring (Manasypov et al 2015Ala-Aho et al 2017) High concentrations of DOC and somemacro- and micronutrients (P N Si Fe Co Ni) positively cor-relating with pCO2 in small thaw ponds are due to intensiveabrasion of peat at the shore and concomitant C and elementdelivery to the waterbodies via suprapermafrost flow(Shirokova et al 2013 Raudina et al 2018) Upon maturationof lakes and decrease in ratio of the circumference length tolake surfacevolume delivery of terrestrial DOM per surfacearea of lake or lake volume is decreasing This increases the pHto above 4ndash45 thereby favoring phytoplankton and periphy-ton development as it is known in lakes of frozen peatlands ofwestern Siberia (Pavlova et al 2016) This may result indecreased CO2 concentration and fluxes to atmosphere forthese lakes Furthermore lake growth leads to certain steady-state conditionsmdashwhen delivery of DOM matches itsprocessing in the water column and burial in the sedimentsmdashas was demonstrated in western Siberia (Manasypovet al 2015) Thus after some threshold lake area (103ndash104 m2)DOC and CO2 concentrations remain constant with furtherincrease in the lake area

However the lack of information on C content and thebiodegradability of BZT lake sediments require in-depth stud-ies of C processing in benthic ecosystems of the BZT In partic-ular in order to distinguish the origin of CO2 and CH4 inthermokarst lake waters radiocarbon dating of dissolved CO2

and dissolved vs ebullited CH4 (ie Walter et al 2006Matveev et al 2016 Elder et al 2019) would be useful how-ever such research is beyond the scope of this study

Fig 3 Partial contributions of different size range lakes to overall CO2

(light gray columns) and CH4 (black columns) emissions for lakes in theBZT (200000 km2) To calculate these contributions we used medianconcentrations and fluxes to atmosphere (averaged over all sites and allseasons Supporting Information Fig S2) and lake size distribution fromGeoEye-1 and Landsat data (Supporting Information Fig S6)

Zabelina et al C emission from thermokarst lakes

9

Regional yields and upscale of fluxesThe total carbon (CO2 + CH4) flux from the water surface

of BZT lakes ranged from 100 to 400 mmol C mminus2 dminus1 or 12to 48 g C mminus2 dminus1 (with median values corresponding to lakesof different size bins) This value should be considered withcaution given a factor of 25 lower values obtained by cham-ber measurements compared to those calculated by wind-based model Given the scarcity of direct chamber measure-ments the regional estimation is possible only based on calcu-lated fluxes Nevertheless these numbers are comparable to Cfluxes measured for western Siberian lakes (17 g C mminus2 dminus1range 06ndash53 g C mminus2 dminus1 Serikova et al 2019) and withfluxes from small (4ndash150 m2) thaw ponds of NorthernEuropean peatlands (34 g C-CO2 mminus2 dminus1 01 g C-CH4 mminus2 dminus1 Kuhn et al 2018) as summarized in SupportingInformation Table S9 At the same time C fluxes in the BZTare higher than those of lakes in Arctic Alaska (Kling et al 1992Elder et al 2018) thermokarst lakes in the Vorkuta region(Heikkinen et al 2004 Marushchak et al 2013) palsapeatlands of Canada (Matveev et al 2016) and runnel pondsin continuous permafrost zone of the Canadian High Arctic(Negandhi et al 2013) It should be noted that recent model-ing of CO2 emissions from Arctic (yedoma thermokarst andglacial) lakes across the globe recommended values between0036 and 008 g C-CO2 mminus2 dminus1 (Tan et al 2017) which is10ndash100 times lower than CO2 fluxes measured in BZT There-fore we caution against use of simulated values for all perma-frost regions especially peatlands Western Siberia lowland(WSL) NE European tundra and presumably other peat-bearing lowlands of Northern Eurasia might turn out to beprominent exceptions to these simulations thus they mayrequire detailed spatial and seasonal analysis of C storage andemissions as their contribution to the global Arctic C cyclemay be strongly underestimated For example even thoughthe BZT represents 1 of overall permafrost-affected land areaits contribution alone can be comparable to the whole Arcticas estimated by Tan et al (2017)

Upscaling the results to the full territory of BZT(195443 km2) yields present day CO2 + CH4 emission of 38( 065) Tg C yrminus1 (median IQR) This value is higher thanthe total annual CO2 emissions from the North Slope ofAlaska (095 Tg C yrminus1) across a similar territory (231157 km2Elder et al 2018) Also C emission flux to atmosphere fromBZT lakes is 30 higher than the annual DOC + DIC exportfor the Pechora River (Swatershed = 324000 km2) to the ArcticOcean (11 times 106 t DIC yrminus1 17 times 106 t DOC yrminus1 Gordeevet al 1996) When compared to the same land (watershed)surface area of this region the C emission exceeds the lateralC export by a factor of 22 C emission from the BZT lakes rep-resents 30 of that reported for the 1300000 km2 of WSL(12 26 Tg C yrminus1 Serikova et al 2019) a territory of similarwetlands and peatbogs Normalized to total land area the spe-cific aquatic C yields to the atmosphere are however twotimes higher in the BZT than in the WSL (195 and

(92 20) times 10minus6 Tg kmminus2 yrminus1 respectively) It is importantto note that the mean annual air temperature is 2ndash6C higherin the BZT compared to the WSL Precipitation for bothregions is relatively similar This suggests more active terres-trial C processing in inland waters of permafrost peatlands inEuropean tundra compared to western Siberia Because thepermafrost in BZT is essentially discontinuous to sporadic andthe warming is more pronounced than in the WSL (Vasilievet al 2020) lakes of BZT could be expected to experience stron-ger and faster response to warming and permafrost thaw com-pared to those in western Siberia Upscaling the average Cemissions of European tundra and western Siberia peatlands([14 5] times 10minus6 Tg kmminus2 yrminus1) to the overall area of permafrost-affected peatlands (2864000 km2 Smith et al 2007) yields avalue of 40 14 Tg yrminus1 However this value should be consid-ered with caution given the significant spatial heterogeneity ofthermokarst lakes in permafrost landscapes (Vonk et al 2019)Another issue is distinguishing between the GHG pattern ofthermokarst and nonthermokarst ecosystems including thewinter period Here Synthetic Aperture Radar (ie Engramet al 2013) can be extremely useful

A critical issue is the emission from thaw ponds andthermokarst lakes compared to the overall C balance of perma-frost wetlands None of the BZT catchment has been subjectedto extensive C exchange studies (ie by chamber and eddycovariance techniques coupled with aquatic export and emis-sion estimation) in ways which it was performed for otherpermafrost-affected regions such as Northern Sweden (Lundinet al 2016 Kuhn et al 2018) or Alaska (Bogard et al 2019)The C balance in Eastern European tundra (BZT) is poorly con-strained due to strong seasonal and spatial variations that arenot captured in existing measurements (ie Heikkinenet al 2002 Marushchak et al 2013) The upscaled value forterrestrial C uptake for this territory is minus72 Tg yrminus1 (Heikkinenet al 2004) which strongly contrasts with the mean annualterrestrial C balance for the eastern part of the BZT(minus08 113 Tg yrminus1 Zamolodchikov and Karelin 2001)Because the magnitude of these values is comparable to net Cemissions from the waterbodies estimated in this study(+38 Tg C yrminus1) lakes may play a crucial role in overall C bal-ance of the European permafrost peatlands

The ongoing climate warming and permafrost thaw in fro-zen Siberian wetlands leads to the drainage of largethermokarst lakes (Smith et al 2005) and appearance ofnumerous new small thaw ponds (Polishchuk et al 2014Bryksina and Polishchuk 2015) Our results suggest that thismay lead to a decrease in overall regional C emissions frominland water surfaces At the same time this change of regimeis not universal across the Eurasian Arctic (Polishchuket al 2020) loss of lakes in discontinuoussporadic permafrostzones may be compensated by increases in the number oflakes in regions of continuous permafrost (Smith et al 2005)Moreover interannual fluctuations may well mask the netchanges in surface hydrology (Karlsson et al 2014) Although

Zabelina et al C emission from thermokarst lakes

10

remote sensing-based lake dynamics are not available for theBZT by analogy with western Siberia and Alaska which arealso located in the discontinuous permafrost zone one mayexpect that permafrost thaw in NE Europe will reduce the totalarea of lakes Such lake drainage occurs via soil disturbanceand increase of hydrological connectivity due to permafrostthaw and active layer thickness increase phenomena observedboth in Alaska (Hinkel et al 2003 Riordan et al 2006) andwestern Siberia (Kirpotin et al 2008) Massive drainage of largelakes may have an unexpected impact on C balance becauseair-exposed sediments have high CO2 effluxes that greatlyexceed those of surface waters (Martinsen et al 2019) Thishighlights the need for systematic assessment of C emissionsin large lakes including freshly drained lake basins

ConclusionsThis multiseasonal study of thermokarst lakes in the perma-

frost peatlands of NE European tundra (BZT) demonstrated thehighest DOC and CO2 concentrations and CO2 emissions indepressions and small thaw ponds (lt 1000 m2 water surface area)However CH4 concentrations and fluxes to atmosphere exhibiteda local maximum in ponds of 10ndash1000 m2 surface area The CO2

and CH4 concentrations and fluxes to atmosphere were highestin spring and the lowest in autumn External spatial and temporalfactors exhibited the following order of control on GHG parame-ters lake size gt season gt year Overall identification of these fac-tors allowed straightforward upscaling of field results obtainedfrom several test sites in the BZT to the full territory ofpermafrost-affected NE European tundra zone The estimated lakewater surface emission flux to atmosphere from the BZT territoryis two times higher than the lateral export of C (DOC + DIC)documented for the Pechora River to the Arctic Ocean Howeverdue to temporal and spatial variations of C fluxes uncertaintieson area GHG emissions are at least one order of magnitude forsmall thaw ponds and a factor of 3ndash5 for thermokarst lakes Possi-ble decrease in lake water area of Northern Eurasia due to lakedrainage induced by permafrost thaw may decrease overall Cemissions from the surface of lentic waters in this territory

ReferencesAbnizova A J Siemens M J Langer and J Boike 2012 Small

ponds with major impact The relevance of ponds and lakes inpermafrost landscapes to carbon dioxide emissions Global Bio-geochem Cycles 26 GB2041 doi1010292011GB004237

Ala-Aho P and others 2017 Using stable isotopes to assess sur-face water source dynamics and hydrological connectivity ina high-latitude wetland and permafrost influenced landscapeJ Hydrol 556 279ndash293 doi101016jjhydrol201711024

Alin S R M F F L Rasera C I Salimon J E Richey G WHoltgrieve A V Krusche and A Snidvongs 2011 Physicalcontrols on carbon dioxide transfer velocity and flux inlow-gradient river systems and implications for regional

carbon budgets J Geophys Res 116 G01009 doi1010292010JG001398

Andreicheva L N 2007 Pleistocene deposits in r Shapkinabasin (Bolshezemelskaya tundra) Lithol Miner Resour 193ndash110 doi101134S0024490207010099

Arsenault J J Talbot and T R Moore 2018 Environmentalcontrols of C N and P biogeochemistry in peatland pondsSci Total Environ 631ndash632 714ndash722 doi101016jscitotenv201803064

Audry S O S Pokrovsky L S Shirokova S N Kirpotin andB Dupreacute 2011 Organic matter mineralization and traceelement post-depositional redistribution in Western Siberiathermokarst lake sediments Biogeosciences 8 3341ndash3358doi105194bg-8-3341-2011

Avagyan A B R K Runkle J Hartmann and L Kutzbach2014 Spatial variations in pore-water biogeochemistrygreatly exceed temporal changes during baseflow condi-tions in a boreal river valley mire complex NorthwestRussia Wetlands 34 1171ndash1182 doi101007s13157-014-0576-4

Bartosiewicz M I Laurion F Clayer and R Maranger 2016Heat-wave effects on oxygen nutrients and phytoplanktoncan alter global warming potential of gases emitted from asmall shallow lake Environ Sci Technol 50 6267ndash6275doi101021acsest5b06312

Bastviken D I Sundgren S Natchimuthu H Reyier and MGaringlfalk 2015 Technical note Cost-efficient approaches tomeasure carbon dioxide (CO2) fluxes and concentrations interrestrial and aquatic environments using mini loggers Bio-geosciences 12 3849ndash3859 doi105194bg-12-3849-2015

Bogard M J and others 2019 Negligible cycling of terrestrialcarbon in many lakes of the arid circumpolar landscapeNat Geosci 12 180ndash185 doi101038s41561-019-0299-5

Bouchard F and others 2017 Paleolimnology ofthermokarst lakes A window into permafrost landscapeevolution Arct Sci 3 91ndash117 doi101139AS-2016-0022

Brittain J F and others 2009 Arctic rivers (chapter 9)p 337ndash379 In K Tockner U Uehlinger and C T Robinson[eds] Rivers of Europe Academic Press httpswwwelseviercombooksrivers-of-europetockner978-0-12-369449-2

Bryksina N A and Y M Polishchuk 2013 Research ofremote measurement accuracy of lake areas using spaceimages Geoinformatika 1 64ndash68 Available from httpselibraryrudownloadelibrary_18882698_14046761pdf

Bryksina N A and Y M Polishchuk 2015 Analysis of changes inthe number of thermokarst lakes in permafrost of Western Sibe-ria on the basis of satellite images Kriosfera Zemli 19 114ndash120httpwwwizdatgeorupdfkrio2015-2114pdf

Cael B B and D A Seekell 2016 The size-distribution ofEarthrsquos lakes Sci Rep 6 29633 doi101038srep29633

Casper P O Chan A Furtado and D Adams 2003 Meth-ane in an acidic bog lake The influence of peat in thecatchment on the biogeochemistry of methane Aquat Sci65 36ndash46 doi101007s000270300003

Zabelina et al C emission from thermokarst lakes

11

signals for Landsat-8 OLI However according to publishedexpert estimations the majority of lakes in the BZT are ofthermokarst origin (Kravtzova and Bystrova 2009) this originis also confirmed by their generally isometric shape as com-pared to elongated lakes with glacial origins Both the lakeabundance and area distribution obtained for the permafrost-affected part of BZT are in fairly good agreement (SupportingInformation Fig S7) with those reported for permafrost-affected parts of western Siberia (Polishchuk et al 2017 2018)and Swedish territory (Cael and Seekell 2016) In the BZT thepower law extrapolation (Pareto law) strongly overestimatesthe number of small lakes (lt 1000 m2) compared to empiricaldata (Supporting Information Fig S7) which is in agreementwith other regions of the world

The maximal contribution to overall GHG emissions fromthe territory of 195000 km2 (38 Tg yrminus1) was provided bylakes gt 100000 m2 whereas contributions from small thawponds (lt 1000 m2) were below 20 (Table 1 Fig 3) The dis-persity of the areal GHG fluxes to atmosphere is quite highachieving one order of magnitude in small thaw ponds and afactor of 2ndash5 for thermokarst lakes (Fig 2) These uncertaintiesvisible as IQR in Supporting Information Figs S2ndashS5 S8 do notoriginate from analytical procedures and rather reflect variabil-ity between sites and seasons It is important to note that thecontribution of small thaw ponds to overall C balance in BZT isbelow the actual uncertainties for total pools and emissions

DiscussionC pools fluxes to atmosphere and possible mechanisms

The highest DOC and CO2 concentrations and CO2 fluxesto atmosphere were observed in small (lt 1000 m2) thawponds Overall CO2 and DOC concentrations and CO2 emis-sion decreased by an order of magnitude when lake sizeincreased by gt 3 orders of magnitude Assuming that the sizegradient of thermokarst lakes in permafrost peatlands reflectsthe age gradient (Shirokova et al 2013 Manasypov et al 2014)high CO2 flux to atmosphere from small waterbodies

corresponding to the beginning of permafrost thaw and lakeformation is likely largely sustained by relatively high input ofallochthonous DOM DOC concentrations in studiedwaterbodies were quite high (median 60ndash20 mg Lminus1 for differ-ent size bins and decreasing with lake size) and exceededvalues observed in soil pore waters of wetlands in the southernpart of the studied territory (20ndash30 mg Lminus1 Avagyanet al 2014) Also given the likely short residence time ofwater in these small thaw ponds and depressions (Manasypovet al 2015) this DOC input from the peat soil would bequickly replenished as well as rapidly bio- and photodegradedas was recently suggested for small waterbodies of easternEuropean tundra (Shirokova et al 2019) The very low qualityof DOC that remains after fast initial degradation of DOM inwaterbodies of the BZT is confirmed by the high values ofSUVA254 (from 3 to 6 L mgminus1 mminus1) which was independent ofthe season and size of the waterbody (Supporting InformationFig S9) Note that Fe concentrations in studied lakes (500ndash-1000 μg Lminus1 depending on the lake size) are too low to inter-fere with SUVA values (ie Poulin et al 2014)

In large thermokarst lakes both sediment source andsuprapermafrost input of contemporary DOM likely supportCO2 fluxes to atmosphere Allochthonous (terrestrial) organiccarbon from peat and vegetation leachates is delivered intolarge lakes via shallow subsurface flow in summer and autumnand via surface overflow during spring (Manasypov et al 2015Ala-Aho et al 2017) High concentrations of DOC and somemacro- and micronutrients (P N Si Fe Co Ni) positively cor-relating with pCO2 in small thaw ponds are due to intensiveabrasion of peat at the shore and concomitant C and elementdelivery to the waterbodies via suprapermafrost flow(Shirokova et al 2013 Raudina et al 2018) Upon maturationof lakes and decrease in ratio of the circumference length tolake surfacevolume delivery of terrestrial DOM per surfacearea of lake or lake volume is decreasing This increases the pHto above 4ndash45 thereby favoring phytoplankton and periphy-ton development as it is known in lakes of frozen peatlands ofwestern Siberia (Pavlova et al 2016) This may result indecreased CO2 concentration and fluxes to atmosphere forthese lakes Furthermore lake growth leads to certain steady-state conditionsmdashwhen delivery of DOM matches itsprocessing in the water column and burial in the sedimentsmdashas was demonstrated in western Siberia (Manasypovet al 2015) Thus after some threshold lake area (103ndash104 m2)DOC and CO2 concentrations remain constant with furtherincrease in the lake area

However the lack of information on C content and thebiodegradability of BZT lake sediments require in-depth stud-ies of C processing in benthic ecosystems of the BZT In partic-ular in order to distinguish the origin of CO2 and CH4 inthermokarst lake waters radiocarbon dating of dissolved CO2

and dissolved vs ebullited CH4 (ie Walter et al 2006Matveev et al 2016 Elder et al 2019) would be useful how-ever such research is beyond the scope of this study

Fig 3 Partial contributions of different size range lakes to overall CO2

(light gray columns) and CH4 (black columns) emissions for lakes in theBZT (200000 km2) To calculate these contributions we used medianconcentrations and fluxes to atmosphere (averaged over all sites and allseasons Supporting Information Fig S2) and lake size distribution fromGeoEye-1 and Landsat data (Supporting Information Fig S6)

Zabelina et al C emission from thermokarst lakes

9

Regional yields and upscale of fluxesThe total carbon (CO2 + CH4) flux from the water surface

of BZT lakes ranged from 100 to 400 mmol C mminus2 dminus1 or 12to 48 g C mminus2 dminus1 (with median values corresponding to lakesof different size bins) This value should be considered withcaution given a factor of 25 lower values obtained by cham-ber measurements compared to those calculated by wind-based model Given the scarcity of direct chamber measure-ments the regional estimation is possible only based on calcu-lated fluxes Nevertheless these numbers are comparable to Cfluxes measured for western Siberian lakes (17 g C mminus2 dminus1range 06ndash53 g C mminus2 dminus1 Serikova et al 2019) and withfluxes from small (4ndash150 m2) thaw ponds of NorthernEuropean peatlands (34 g C-CO2 mminus2 dminus1 01 g C-CH4 mminus2 dminus1 Kuhn et al 2018) as summarized in SupportingInformation Table S9 At the same time C fluxes in the BZTare higher than those of lakes in Arctic Alaska (Kling et al 1992Elder et al 2018) thermokarst lakes in the Vorkuta region(Heikkinen et al 2004 Marushchak et al 2013) palsapeatlands of Canada (Matveev et al 2016) and runnel pondsin continuous permafrost zone of the Canadian High Arctic(Negandhi et al 2013) It should be noted that recent model-ing of CO2 emissions from Arctic (yedoma thermokarst andglacial) lakes across the globe recommended values between0036 and 008 g C-CO2 mminus2 dminus1 (Tan et al 2017) which is10ndash100 times lower than CO2 fluxes measured in BZT There-fore we caution against use of simulated values for all perma-frost regions especially peatlands Western Siberia lowland(WSL) NE European tundra and presumably other peat-bearing lowlands of Northern Eurasia might turn out to beprominent exceptions to these simulations thus they mayrequire detailed spatial and seasonal analysis of C storage andemissions as their contribution to the global Arctic C cyclemay be strongly underestimated For example even thoughthe BZT represents 1 of overall permafrost-affected land areaits contribution alone can be comparable to the whole Arcticas estimated by Tan et al (2017)

Upscaling the results to the full territory of BZT(195443 km2) yields present day CO2 + CH4 emission of 38( 065) Tg C yrminus1 (median IQR) This value is higher thanthe total annual CO2 emissions from the North Slope ofAlaska (095 Tg C yrminus1) across a similar territory (231157 km2Elder et al 2018) Also C emission flux to atmosphere fromBZT lakes is 30 higher than the annual DOC + DIC exportfor the Pechora River (Swatershed = 324000 km2) to the ArcticOcean (11 times 106 t DIC yrminus1 17 times 106 t DOC yrminus1 Gordeevet al 1996) When compared to the same land (watershed)surface area of this region the C emission exceeds the lateralC export by a factor of 22 C emission from the BZT lakes rep-resents 30 of that reported for the 1300000 km2 of WSL(12 26 Tg C yrminus1 Serikova et al 2019) a territory of similarwetlands and peatbogs Normalized to total land area the spe-cific aquatic C yields to the atmosphere are however twotimes higher in the BZT than in the WSL (195 and

(92 20) times 10minus6 Tg kmminus2 yrminus1 respectively) It is importantto note that the mean annual air temperature is 2ndash6C higherin the BZT compared to the WSL Precipitation for bothregions is relatively similar This suggests more active terres-trial C processing in inland waters of permafrost peatlands inEuropean tundra compared to western Siberia Because thepermafrost in BZT is essentially discontinuous to sporadic andthe warming is more pronounced than in the WSL (Vasilievet al 2020) lakes of BZT could be expected to experience stron-ger and faster response to warming and permafrost thaw com-pared to those in western Siberia Upscaling the average Cemissions of European tundra and western Siberia peatlands([14 5] times 10minus6 Tg kmminus2 yrminus1) to the overall area of permafrost-affected peatlands (2864000 km2 Smith et al 2007) yields avalue of 40 14 Tg yrminus1 However this value should be consid-ered with caution given the significant spatial heterogeneity ofthermokarst lakes in permafrost landscapes (Vonk et al 2019)Another issue is distinguishing between the GHG pattern ofthermokarst and nonthermokarst ecosystems including thewinter period Here Synthetic Aperture Radar (ie Engramet al 2013) can be extremely useful

A critical issue is the emission from thaw ponds andthermokarst lakes compared to the overall C balance of perma-frost wetlands None of the BZT catchment has been subjectedto extensive C exchange studies (ie by chamber and eddycovariance techniques coupled with aquatic export and emis-sion estimation) in ways which it was performed for otherpermafrost-affected regions such as Northern Sweden (Lundinet al 2016 Kuhn et al 2018) or Alaska (Bogard et al 2019)The C balance in Eastern European tundra (BZT) is poorly con-strained due to strong seasonal and spatial variations that arenot captured in existing measurements (ie Heikkinenet al 2002 Marushchak et al 2013) The upscaled value forterrestrial C uptake for this territory is minus72 Tg yrminus1 (Heikkinenet al 2004) which strongly contrasts with the mean annualterrestrial C balance for the eastern part of the BZT(minus08 113 Tg yrminus1 Zamolodchikov and Karelin 2001)Because the magnitude of these values is comparable to net Cemissions from the waterbodies estimated in this study(+38 Tg C yrminus1) lakes may play a crucial role in overall C bal-ance of the European permafrost peatlands

The ongoing climate warming and permafrost thaw in fro-zen Siberian wetlands leads to the drainage of largethermokarst lakes (Smith et al 2005) and appearance ofnumerous new small thaw ponds (Polishchuk et al 2014Bryksina and Polishchuk 2015) Our results suggest that thismay lead to a decrease in overall regional C emissions frominland water surfaces At the same time this change of regimeis not universal across the Eurasian Arctic (Polishchuket al 2020) loss of lakes in discontinuoussporadic permafrostzones may be compensated by increases in the number oflakes in regions of continuous permafrost (Smith et al 2005)Moreover interannual fluctuations may well mask the netchanges in surface hydrology (Karlsson et al 2014) Although

Zabelina et al C emission from thermokarst lakes

10

remote sensing-based lake dynamics are not available for theBZT by analogy with western Siberia and Alaska which arealso located in the discontinuous permafrost zone one mayexpect that permafrost thaw in NE Europe will reduce the totalarea of lakes Such lake drainage occurs via soil disturbanceand increase of hydrological connectivity due to permafrostthaw and active layer thickness increase phenomena observedboth in Alaska (Hinkel et al 2003 Riordan et al 2006) andwestern Siberia (Kirpotin et al 2008) Massive drainage of largelakes may have an unexpected impact on C balance becauseair-exposed sediments have high CO2 effluxes that greatlyexceed those of surface waters (Martinsen et al 2019) Thishighlights the need for systematic assessment of C emissionsin large lakes including freshly drained lake basins

ConclusionsThis multiseasonal study of thermokarst lakes in the perma-

frost peatlands of NE European tundra (BZT) demonstrated thehighest DOC and CO2 concentrations and CO2 emissions indepressions and small thaw ponds (lt 1000 m2 water surface area)However CH4 concentrations and fluxes to atmosphere exhibiteda local maximum in ponds of 10ndash1000 m2 surface area The CO2

and CH4 concentrations and fluxes to atmosphere were highestin spring and the lowest in autumn External spatial and temporalfactors exhibited the following order of control on GHG parame-ters lake size gt season gt year Overall identification of these fac-tors allowed straightforward upscaling of field results obtainedfrom several test sites in the BZT to the full territory ofpermafrost-affected NE European tundra zone The estimated lakewater surface emission flux to atmosphere from the BZT territoryis two times higher than the lateral export of C (DOC + DIC)documented for the Pechora River to the Arctic Ocean Howeverdue to temporal and spatial variations of C fluxes uncertaintieson area GHG emissions are at least one order of magnitude forsmall thaw ponds and a factor of 3ndash5 for thermokarst lakes Possi-ble decrease in lake water area of Northern Eurasia due to lakedrainage induced by permafrost thaw may decrease overall Cemissions from the surface of lentic waters in this territory

ReferencesAbnizova A J Siemens M J Langer and J Boike 2012 Small

ponds with major impact The relevance of ponds and lakes inpermafrost landscapes to carbon dioxide emissions Global Bio-geochem Cycles 26 GB2041 doi1010292011GB004237

Ala-Aho P and others 2017 Using stable isotopes to assess sur-face water source dynamics and hydrological connectivity ina high-latitude wetland and permafrost influenced landscapeJ Hydrol 556 279ndash293 doi101016jjhydrol201711024

Alin S R M F F L Rasera C I Salimon J E Richey G WHoltgrieve A V Krusche and A Snidvongs 2011 Physicalcontrols on carbon dioxide transfer velocity and flux inlow-gradient river systems and implications for regional

carbon budgets J Geophys Res 116 G01009 doi1010292010JG001398

Andreicheva L N 2007 Pleistocene deposits in r Shapkinabasin (Bolshezemelskaya tundra) Lithol Miner Resour 193ndash110 doi101134S0024490207010099

Arsenault J J Talbot and T R Moore 2018 Environmentalcontrols of C N and P biogeochemistry in peatland pondsSci Total Environ 631ndash632 714ndash722 doi101016jscitotenv201803064

Audry S O S Pokrovsky L S Shirokova S N Kirpotin andB Dupreacute 2011 Organic matter mineralization and traceelement post-depositional redistribution in Western Siberiathermokarst lake sediments Biogeosciences 8 3341ndash3358doi105194bg-8-3341-2011

Avagyan A B R K Runkle J Hartmann and L Kutzbach2014 Spatial variations in pore-water biogeochemistrygreatly exceed temporal changes during baseflow condi-tions in a boreal river valley mire complex NorthwestRussia Wetlands 34 1171ndash1182 doi101007s13157-014-0576-4

Bartosiewicz M I Laurion F Clayer and R Maranger 2016Heat-wave effects on oxygen nutrients and phytoplanktoncan alter global warming potential of gases emitted from asmall shallow lake Environ Sci Technol 50 6267ndash6275doi101021acsest5b06312

Bastviken D I Sundgren S Natchimuthu H Reyier and MGaringlfalk 2015 Technical note Cost-efficient approaches tomeasure carbon dioxide (CO2) fluxes and concentrations interrestrial and aquatic environments using mini loggers Bio-geosciences 12 3849ndash3859 doi105194bg-12-3849-2015

Bogard M J and others 2019 Negligible cycling of terrestrialcarbon in many lakes of the arid circumpolar landscapeNat Geosci 12 180ndash185 doi101038s41561-019-0299-5

Bouchard F and others 2017 Paleolimnology ofthermokarst lakes A window into permafrost landscapeevolution Arct Sci 3 91ndash117 doi101139AS-2016-0022

Brittain J F and others 2009 Arctic rivers (chapter 9)p 337ndash379 In K Tockner U Uehlinger and C T Robinson[eds] Rivers of Europe Academic Press httpswwwelseviercombooksrivers-of-europetockner978-0-12-369449-2

Bryksina N A and Y M Polishchuk 2013 Research ofremote measurement accuracy of lake areas using spaceimages Geoinformatika 1 64ndash68 Available from httpselibraryrudownloadelibrary_18882698_14046761pdf

Bryksina N A and Y M Polishchuk 2015 Analysis of changes inthe number of thermokarst lakes in permafrost of Western Sibe-ria on the basis of satellite images Kriosfera Zemli 19 114ndash120httpwwwizdatgeorupdfkrio2015-2114pdf

Cael B B and D A Seekell 2016 The size-distribution ofEarthrsquos lakes Sci Rep 6 29633 doi101038srep29633

Casper P O Chan A Furtado and D Adams 2003 Meth-ane in an acidic bog lake The influence of peat in thecatchment on the biogeochemistry of methane Aquat Sci65 36ndash46 doi101007s000270300003

Zabelina et al C emission from thermokarst lakes

11

Regional yields and upscale of fluxesThe total carbon (CO2 + CH4) flux from the water surface

of BZT lakes ranged from 100 to 400 mmol C mminus2 dminus1 or 12to 48 g C mminus2 dminus1 (with median values corresponding to lakesof different size bins) This value should be considered withcaution given a factor of 25 lower values obtained by cham-ber measurements compared to those calculated by wind-based model Given the scarcity of direct chamber measure-ments the regional estimation is possible only based on calcu-lated fluxes Nevertheless these numbers are comparable to Cfluxes measured for western Siberian lakes (17 g C mminus2 dminus1range 06ndash53 g C mminus2 dminus1 Serikova et al 2019) and withfluxes from small (4ndash150 m2) thaw ponds of NorthernEuropean peatlands (34 g C-CO2 mminus2 dminus1 01 g C-CH4 mminus2 dminus1 Kuhn et al 2018) as summarized in SupportingInformation Table S9 At the same time C fluxes in the BZTare higher than those of lakes in Arctic Alaska (Kling et al 1992Elder et al 2018) thermokarst lakes in the Vorkuta region(Heikkinen et al 2004 Marushchak et al 2013) palsapeatlands of Canada (Matveev et al 2016) and runnel pondsin continuous permafrost zone of the Canadian High Arctic(Negandhi et al 2013) It should be noted that recent model-ing of CO2 emissions from Arctic (yedoma thermokarst andglacial) lakes across the globe recommended values between0036 and 008 g C-CO2 mminus2 dminus1 (Tan et al 2017) which is10ndash100 times lower than CO2 fluxes measured in BZT There-fore we caution against use of simulated values for all perma-frost regions especially peatlands Western Siberia lowland(WSL) NE European tundra and presumably other peat-bearing lowlands of Northern Eurasia might turn out to beprominent exceptions to these simulations thus they mayrequire detailed spatial and seasonal analysis of C storage andemissions as their contribution to the global Arctic C cyclemay be strongly underestimated For example even thoughthe BZT represents 1 of overall permafrost-affected land areaits contribution alone can be comparable to the whole Arcticas estimated by Tan et al (2017)

Upscaling the results to the full territory of BZT(195443 km2) yields present day CO2 + CH4 emission of 38( 065) Tg C yrminus1 (median IQR) This value is higher thanthe total annual CO2 emissions from the North Slope ofAlaska (095 Tg C yrminus1) across a similar territory (231157 km2Elder et al 2018) Also C emission flux to atmosphere fromBZT lakes is 30 higher than the annual DOC + DIC exportfor the Pechora River (Swatershed = 324000 km2) to the ArcticOcean (11 times 106 t DIC yrminus1 17 times 106 t DOC yrminus1 Gordeevet al 1996) When compared to the same land (watershed)surface area of this region the C emission exceeds the lateralC export by a factor of 22 C emission from the BZT lakes rep-resents 30 of that reported for the 1300000 km2 of WSL(12 26 Tg C yrminus1 Serikova et al 2019) a territory of similarwetlands and peatbogs Normalized to total land area the spe-cific aquatic C yields to the atmosphere are however twotimes higher in the BZT than in the WSL (195 and

(92 20) times 10minus6 Tg kmminus2 yrminus1 respectively) It is importantto note that the mean annual air temperature is 2ndash6C higherin the BZT compared to the WSL Precipitation for bothregions is relatively similar This suggests more active terres-trial C processing in inland waters of permafrost peatlands inEuropean tundra compared to western Siberia Because thepermafrost in BZT is essentially discontinuous to sporadic andthe warming is more pronounced than in the WSL (Vasilievet al 2020) lakes of BZT could be expected to experience stron-ger and faster response to warming and permafrost thaw com-pared to those in western Siberia Upscaling the average Cemissions of European tundra and western Siberia peatlands([14 5] times 10minus6 Tg kmminus2 yrminus1) to the overall area of permafrost-affected peatlands (2864000 km2 Smith et al 2007) yields avalue of 40 14 Tg yrminus1 However this value should be consid-ered with caution given the significant spatial heterogeneity ofthermokarst lakes in permafrost landscapes (Vonk et al 2019)Another issue is distinguishing between the GHG pattern ofthermokarst and nonthermokarst ecosystems including thewinter period Here Synthetic Aperture Radar (ie Engramet al 2013) can be extremely useful

A critical issue is the emission from thaw ponds andthermokarst lakes compared to the overall C balance of perma-frost wetlands None of the BZT catchment has been subjectedto extensive C exchange studies (ie by chamber and eddycovariance techniques coupled with aquatic export and emis-sion estimation) in ways which it was performed for otherpermafrost-affected regions such as Northern Sweden (Lundinet al 2016 Kuhn et al 2018) or Alaska (Bogard et al 2019)The C balance in Eastern European tundra (BZT) is poorly con-strained due to strong seasonal and spatial variations that arenot captured in existing measurements (ie Heikkinenet al 2002 Marushchak et al 2013) The upscaled value forterrestrial C uptake for this territory is minus72 Tg yrminus1 (Heikkinenet al 2004) which strongly contrasts with the mean annualterrestrial C balance for the eastern part of the BZT(minus08 113 Tg yrminus1 Zamolodchikov and Karelin 2001)Because the magnitude of these values is comparable to net Cemissions from the waterbodies estimated in this study(+38 Tg C yrminus1) lakes may play a crucial role in overall C bal-ance of the European permafrost peatlands

The ongoing climate warming and permafrost thaw in fro-zen Siberian wetlands leads to the drainage of largethermokarst lakes (Smith et al 2005) and appearance ofnumerous new small thaw ponds (Polishchuk et al 2014Bryksina and Polishchuk 2015) Our results suggest that thismay lead to a decrease in overall regional C emissions frominland water surfaces At the same time this change of regimeis not universal across the Eurasian Arctic (Polishchuket al 2020) loss of lakes in discontinuoussporadic permafrostzones may be compensated by increases in the number oflakes in regions of continuous permafrost (Smith et al 2005)Moreover interannual fluctuations may well mask the netchanges in surface hydrology (Karlsson et al 2014) Although

Zabelina et al C emission from thermokarst lakes

10

remote sensing-based lake dynamics are not available for theBZT by analogy with western Siberia and Alaska which arealso located in the discontinuous permafrost zone one mayexpect that permafrost thaw in NE Europe will reduce the totalarea of lakes Such lake drainage occurs via soil disturbanceand increase of hydrological connectivity due to permafrostthaw and active layer thickness increase phenomena observedboth in Alaska (Hinkel et al 2003 Riordan et al 2006) andwestern Siberia (Kirpotin et al 2008) Massive drainage of largelakes may have an unexpected impact on C balance becauseair-exposed sediments have high CO2 effluxes that greatlyexceed those of surface waters (Martinsen et al 2019) Thishighlights the need for systematic assessment of C emissionsin large lakes including freshly drained lake basins

ConclusionsThis multiseasonal study of thermokarst lakes in the perma-

frost peatlands of NE European tundra (BZT) demonstrated thehighest DOC and CO2 concentrations and CO2 emissions indepressions and small thaw ponds (lt 1000 m2 water surface area)However CH4 concentrations and fluxes to atmosphere exhibiteda local maximum in ponds of 10ndash1000 m2 surface area The CO2

and CH4 concentrations and fluxes to atmosphere were highestin spring and the lowest in autumn External spatial and temporalfactors exhibited the following order of control on GHG parame-ters lake size gt season gt year Overall identification of these fac-tors allowed straightforward upscaling of field results obtainedfrom several test sites in the BZT to the full territory ofpermafrost-affected NE European tundra zone The estimated lakewater surface emission flux to atmosphere from the BZT territoryis two times higher than the lateral export of C (DOC + DIC)documented for the Pechora River to the Arctic Ocean Howeverdue to temporal and spatial variations of C fluxes uncertaintieson area GHG emissions are at least one order of magnitude forsmall thaw ponds and a factor of 3ndash5 for thermokarst lakes Possi-ble decrease in lake water area of Northern Eurasia due to lakedrainage induced by permafrost thaw may decrease overall Cemissions from the surface of lentic waters in this territory

ReferencesAbnizova A J Siemens M J Langer and J Boike 2012 Small

ponds with major impact The relevance of ponds and lakes inpermafrost landscapes to carbon dioxide emissions Global Bio-geochem Cycles 26 GB2041 doi1010292011GB004237

Ala-Aho P and others 2017 Using stable isotopes to assess sur-face water source dynamics and hydrological connectivity ina high-latitude wetland and permafrost influenced landscapeJ Hydrol 556 279ndash293 doi101016jjhydrol201711024

Alin S R M F F L Rasera C I Salimon J E Richey G WHoltgrieve A V Krusche and A Snidvongs 2011 Physicalcontrols on carbon dioxide transfer velocity and flux inlow-gradient river systems and implications for regional

carbon budgets J Geophys Res 116 G01009 doi1010292010JG001398

Andreicheva L N 2007 Pleistocene deposits in r Shapkinabasin (Bolshezemelskaya tundra) Lithol Miner Resour 193ndash110 doi101134S0024490207010099

Arsenault J J Talbot and T R Moore 2018 Environmentalcontrols of C N and P biogeochemistry in peatland pondsSci Total Environ 631ndash632 714ndash722 doi101016jscitotenv201803064

Audry S O S Pokrovsky L S Shirokova S N Kirpotin andB Dupreacute 2011 Organic matter mineralization and traceelement post-depositional redistribution in Western Siberiathermokarst lake sediments Biogeosciences 8 3341ndash3358doi105194bg-8-3341-2011

Avagyan A B R K Runkle J Hartmann and L Kutzbach2014 Spatial variations in pore-water biogeochemistrygreatly exceed temporal changes during baseflow condi-tions in a boreal river valley mire complex NorthwestRussia Wetlands 34 1171ndash1182 doi101007s13157-014-0576-4

Bartosiewicz M I Laurion F Clayer and R Maranger 2016Heat-wave effects on oxygen nutrients and phytoplanktoncan alter global warming potential of gases emitted from asmall shallow lake Environ Sci Technol 50 6267ndash6275doi101021acsest5b06312

Bastviken D I Sundgren S Natchimuthu H Reyier and MGaringlfalk 2015 Technical note Cost-efficient approaches tomeasure carbon dioxide (CO2) fluxes and concentrations interrestrial and aquatic environments using mini loggers Bio-geosciences 12 3849ndash3859 doi105194bg-12-3849-2015

Bogard M J and others 2019 Negligible cycling of terrestrialcarbon in many lakes of the arid circumpolar landscapeNat Geosci 12 180ndash185 doi101038s41561-019-0299-5

Bouchard F and others 2017 Paleolimnology ofthermokarst lakes A window into permafrost landscapeevolution Arct Sci 3 91ndash117 doi101139AS-2016-0022

Brittain J F and others 2009 Arctic rivers (chapter 9)p 337ndash379 In K Tockner U Uehlinger and C T Robinson[eds] Rivers of Europe Academic Press httpswwwelseviercombooksrivers-of-europetockner978-0-12-369449-2

Bryksina N A and Y M Polishchuk 2013 Research ofremote measurement accuracy of lake areas using spaceimages Geoinformatika 1 64ndash68 Available from httpselibraryrudownloadelibrary_18882698_14046761pdf

Bryksina N A and Y M Polishchuk 2015 Analysis of changes inthe number of thermokarst lakes in permafrost of Western Sibe-ria on the basis of satellite images Kriosfera Zemli 19 114ndash120httpwwwizdatgeorupdfkrio2015-2114pdf

Cael B B and D A Seekell 2016 The size-distribution ofEarthrsquos lakes Sci Rep 6 29633 doi101038srep29633

Casper P O Chan A Furtado and D Adams 2003 Meth-ane in an acidic bog lake The influence of peat in thecatchment on the biogeochemistry of methane Aquat Sci65 36ndash46 doi101007s000270300003

Zabelina et al C emission from thermokarst lakes

11

remote sensing-based lake dynamics are not available for theBZT by analogy with western Siberia and Alaska which arealso located in the discontinuous permafrost zone one mayexpect that permafrost thaw in NE Europe will reduce the totalarea of lakes Such lake drainage occurs via soil disturbanceand increase of hydrological connectivity due to permafrostthaw and active layer thickness increase phenomena observedboth in Alaska (Hinkel et al 2003 Riordan et al 2006) andwestern Siberia (Kirpotin et al 2008) Massive drainage of largelakes may have an unexpected impact on C balance becauseair-exposed sediments have high CO2 effluxes that greatlyexceed those of surface waters (Martinsen et al 2019) Thishighlights the need for systematic assessment of C emissionsin large lakes including freshly drained lake basins

ConclusionsThis multiseasonal study of thermokarst lakes in the perma-

frost peatlands of NE European tundra (BZT) demonstrated thehighest DOC and CO2 concentrations and CO2 emissions indepressions and small thaw ponds (lt 1000 m2 water surface area)However CH4 concentrations and fluxes to atmosphere exhibiteda local maximum in ponds of 10ndash1000 m2 surface area The CO2

and CH4 concentrations and fluxes to atmosphere were highestin spring and the lowest in autumn External spatial and temporalfactors exhibited the following order of control on GHG parame-ters lake size gt season gt year Overall identification of these fac-tors allowed straightforward upscaling of field results obtainedfrom several test sites in the BZT to the full territory ofpermafrost-affected NE European tundra zone The estimated lakewater surface emission flux to atmosphere from the BZT territoryis two times higher than the lateral export of C (DOC + DIC)documented for the Pechora River to the Arctic Ocean Howeverdue to temporal and spatial variations of C fluxes uncertaintieson area GHG emissions are at least one order of magnitude forsmall thaw ponds and a factor of 3ndash5 for thermokarst lakes Possi-ble decrease in lake water area of Northern Eurasia due to lakedrainage induced by permafrost thaw may decrease overall Cemissions from the surface of lentic waters in this territory

ReferencesAbnizova A J Siemens M J Langer and J Boike 2012 Small

ponds with major impact The relevance of ponds and lakes inpermafrost landscapes to carbon dioxide emissions Global Bio-geochem Cycles 26 GB2041 doi1010292011GB004237

Ala-Aho P and others 2017 Using stable isotopes to assess sur-face water source dynamics and hydrological connectivity ina high-latitude wetland and permafrost influenced landscapeJ Hydrol 556 279ndash293 doi101016jjhydrol201711024

Alin S R M F F L Rasera C I Salimon J E Richey G WHoltgrieve A V Krusche and A Snidvongs 2011 Physicalcontrols on carbon dioxide transfer velocity and flux inlow-gradient river systems and implications for regional

carbon budgets J Geophys Res 116 G01009 doi1010292010JG001398

Andreicheva L N 2007 Pleistocene deposits in r Shapkinabasin (Bolshezemelskaya tundra) Lithol Miner Resour 193ndash110 doi101134S0024490207010099

Arsenault J J Talbot and T R Moore 2018 Environmentalcontrols of C N and P biogeochemistry in peatland pondsSci Total Environ 631ndash632 714ndash722 doi101016jscitotenv201803064

Audry S O S Pokrovsky L S Shirokova S N Kirpotin andB Dupreacute 2011 Organic matter mineralization and traceelement post-depositional redistribution in Western Siberiathermokarst lake sediments Biogeosciences 8 3341ndash3358doi105194bg-8-3341-2011

Avagyan A B R K Runkle J Hartmann and L Kutzbach2014 Spatial variations in pore-water biogeochemistrygreatly exceed temporal changes during baseflow condi-tions in a boreal river valley mire complex NorthwestRussia Wetlands 34 1171ndash1182 doi101007s13157-014-0576-4

Bartosiewicz M I Laurion F Clayer and R Maranger 2016Heat-wave effects on oxygen nutrients and phytoplanktoncan alter global warming potential of gases emitted from asmall shallow lake Environ Sci Technol 50 6267ndash6275doi101021acsest5b06312

Bastviken D I Sundgren S Natchimuthu H Reyier and MGaringlfalk 2015 Technical note Cost-efficient approaches tomeasure carbon dioxide (CO2) fluxes and concentrations interrestrial and aquatic environments using mini loggers Bio-geosciences 12 3849ndash3859 doi105194bg-12-3849-2015

Bogard M J and others 2019 Negligible cycling of terrestrialcarbon in many lakes of the arid circumpolar landscapeNat Geosci 12 180ndash185 doi101038s41561-019-0299-5

Bouchard F and others 2017 Paleolimnology ofthermokarst lakes A window into permafrost landscapeevolution Arct Sci 3 91ndash117 doi101139AS-2016-0022

Brittain J F and others 2009 Arctic rivers (chapter 9)p 337ndash379 In K Tockner U Uehlinger and C T Robinson[eds] Rivers of Europe Academic Press httpswwwelseviercombooksrivers-of-europetockner978-0-12-369449-2

Bryksina N A and Y M Polishchuk 2013 Research ofremote measurement accuracy of lake areas using spaceimages Geoinformatika 1 64ndash68 Available from httpselibraryrudownloadelibrary_18882698_14046761pdf

Bryksina N A and Y M Polishchuk 2015 Analysis of changes inthe number of thermokarst lakes in permafrost of Western Sibe-ria on the basis of satellite images Kriosfera Zemli 19 114ndash120httpwwwizdatgeorupdfkrio2015-2114pdf

Cael B B and D A Seekell 2016 The size-distribution ofEarthrsquos lakes Sci Rep 6 29633 doi101038srep29633

Casper P O Chan A Furtado and D Adams 2003 Meth-ane in an acidic bog lake The influence of peat in thecatchment on the biogeochemistry of methane Aquat Sci65 36ndash46 doi101007s000270300003

Zabelina et al C emission from thermokarst lakes

11

Chupakov A V and others 2017 Allochthonous andautochthonous carbon in deep organic-rich and organic-poor lakes of the European Russian subarctic Boreal Envi-ron Res 22 213ndash230 Available from httpwwwborenvnetBERpdfsber22ber22-213-230pdf

Cole J J and N F Caraco 1998 Atmospheric exchange ofcarbon dioxide in a low-wind oligotrophic lake measuredby the addition of SF6 Limnol Oceanogr 43 647ndash656doi104319lo19984340647

Davydova N N V K Kuznetsov I V Delyusina 1994 Fea-tures of the ecosystem structure of the lakes of the FarNorth (on the example of the lakes of the Bol-shezemelskaya tundra) p 260 In V G Drabkova and I STrifonova [eds] Saint Petersburg Science Publ House1994

de Wit G A R-M Couture L Jackson-Blake M N Futter SValinia and K Austnes 2018 Pipes or chimneys For car-bon cycling in small boreal lakes precipitation mattersmost Limnol Oceanogr Lett 3 275ndash284 doi101002lol210077

Drake T W R M Holmes A V Zhulidov T GurtovayaP A Raymond J W McClelland and R G M Spencer2019 Multidecadal climate-induced changes in Arctic tun-dra lake geochemistry and geomorphology Limnol Ocean-ogr 64 S179ndashS191 doi101002lno11015

Elder C D and others 2018 Greenhouse gas emissions fromdiverse Arctic Alaskan lakes are dominated by young carbonNat Clim Chang 8 166ndash171 doi101038s41558-017-0066-9

Elder C D and others 2019 Seasonal sources of whole-lakeCH4 and CO2 emissions from interior Alaskan thermokarstlakes J Geophys Res Biogeosci 124 1209ndash1229 doi1010292018JG004735

Engram M K Walter Anthony F J Meyer and G Grosse2013 Synthetic aperture radar (SAR) backscatter responsefrom methane ebullition bubbles trapped by thermokarst lakeice Can J Remote Sens 38 667ndash682 doi105589m12-054

Erkkilauml K-M and others 2018 Methane and carbon dioxidefluxes over a lake Comparison between eddy covariancefloating chambers and boundary layer method Bio-geosciences 15 429ndash445 doi105194bg-15-429-2018

Goldina L P 1972 Geography of the lakes of the Bol-shezemelskaya tundra Izd-vo Nauka

Gordeev V V J M Martin I S Sidorov and M V Sidorova1996 A reassessment of the Eurasian river input of watersediment major elements and nutrients to the ArcticOcean Am J Sci 296 664ndash691 doi102475ajs2966664

Guerin F G Abril D Serccedila C Delon S Richard R Delmas ATremblay and L Varfalvy 2007 Gas transfer velocities ofCO2 and CH4 in a tropical reservoir and its river downstreamJ Mar Syst 66 161ndash172 doi101016jjmarsys200603019

Heikkinen J E P V Elsakov and P J Martikainen 2002 Carbondioxide and methane dynamics and annual carbon balance intundra wetland in NE Europe Russia Global BiogeochemCycles 16 62-1ndash62-15 doi1010292002GB001930

Heikkinen J E P T Virtanen J T Huttunen V Elsakov andP J Martikainen 2004 Carbon balance in East Europeantundra Global Biogeochem Cycles 18 GB1023 doi1010292003GB002054

Hinkel K M W R Eisner J G Bockheim F E Nelson K MPeterson and X Dai 2003 Spatial extent age and carbonstocks in drained thaw lake basins on the Barrow PeninsulaAlaska Arct Antarct Alp Res 35 291ndash300 doi1016571523-0430(2003)035[0291SEAACS]20CO2

Iglovsky S A 2016 GPR researches of the frozen deposits inthe area of thermokarst lakes of Bolshezemelskaya tundrap 185ndash191 In V I Pavlenko [ed] Natural resources andintegrated development of coastal regions of the Arcticzone Collection of scientific papers Publ House FCIARcticRAS Available from httpselibraryruitemaspid=26848189

Juutinen S M Rantakari P Kortelainen J T Huttunen TLarmola J Alm J Silvola and P J Martikainen 2009Methane dynamics in different boreal lake types Bio-geosciences 6 209ndash223 doi105194bg-6-209-2009

Karlsson J R Giesler J Persson and E J Lundin 2013 Highemission of carbon dioxide and methane during ice thawin high latitude lakes Geophys Res Lett 40 1ndash5 doi101002grl50152

Karlsson J M S W Lyon and G Destouni 2014 Temporalbehavior of lake size-distribution in a thawing permafrostlandscape in Nothwestern Siberia Remote Sens 6621ndash636 doi103390rs6010621

Kirpotin S Y Polishchuk E Zakharova L Shirokova OPokrovsky M Kolmakova and B Dupreacute 2008 One of thepossible mechanisms of thermokarst lake drainage in West-Siberian North Int J Environ Stud 65 631ndash635 doi10108000207230802525208

Kirpotin S N and others 2011 West Siberian palsapeatlands Distribution typology cyclic development pre-sent day climate-driven changes seasonal hydrology andimpact on CO2 cycle Int J Environ Stud 68 603ndash623doi101080002072332011593901

Kling G W G W Kipphut and M C Miller 1992 Theflux of CO2 and CH4 from lakes and rivers in arcticAlaska Hydrobiologia 240 23ndash36 doi101007BF00013449

Kornienko S G 2014 Assessment accuracy of measurementof the water body area in the permafrost using differentspatial resolution satellite imagery Kriosf Zemli 18 86ndash93httpwwwizdatgeorupdfkrio2014-486pdf

Kosten S and others 2010 Climate - dependent CO2 emis-sions from lakes Global Biogeochem Cycles 24 GB2007doi1010292009GB003618

Kravtzova V I and A G Bystrova 2009 Study ofthermokarst lake size in different regions of Russia over last30 years Earth Cryosphere 13 12ndash26 in Russian httpwwwizdatgeorupdfkrio2009-216pdf

Kuhn M E J Lundin R Giesler M Johansson and JKarlsson 2018 Emissions from thaw ponds largely offset

Zabelina et al C emission from thermokarst lakes

12

the carbon sink of northern permafrost wetlands Sci Rep8 9535 doi101038s41598-018-27770-x

Langer M S Westermann K Walter Anthony KWischnewski and J Boike 2015 Frozen ponds Productionand storage of methane during the Arctic winter in a low-land tundra landscape in northern Siberia Lena River deltaBiogeosciences 12 977ndash990 doi105194bg-12-977-2015

Laurion I W F Vincent S MacIntyre L Retamal C DupontP Francus and R Pienitz 2010 Variability in greenhousegas emissions from permafrost thaw ponds Limnol Ocean-ogr 55 115ndash133 doi104319lo20105510115

Liss P S and P G Slater 1974 Flux of gases across the air-sea interface Nature 247 181ndash184 doi101038247181a0

Lundin E J R Giesler A Persson M S Thompson and JKarlsson 2013 Integrating carbon emissions from lakesand streams in a subarctic catchment J Geophys ResBiogeosci 118 1200ndash1207 doi101002jgrg20092

Lundin E J J Klaminder R Giesler A Persson D OlefeldtM Heliasz T R Christensen and J Karlsson 2016 Is thesubarctic landscape still a carbon sink Evidence from adetailed catchment balance Geophys Res Lett 431988ndash1995 doi1010022015GL066970

Maksimova L N and E N Ospennikov 2012 Evolution ofmire systems and permafrost of Bolshezemelskaya tundrain Holocene Earthrsquos Cryosphere 16 53ndash61 in Russianhttpwwwizdatgeorupdfkrio2012-353pdf

Manasypov R M O S Pokrovsky S N Kirpotin and L SShirokova 2014 Thermokarst lake waters across permafrostzones of Western Siberia Cryosphere 8 1177ndash1193 doi105194tc-8-1177-2014

Manasypov R M and others 2015 Seasonal dynamics oforganic carbon and metals in thermokarst lakes from thediscontinuous permafrost zone of western Siberia Bio-geosciences 12 3009ndash3028 doi105194bg-12-3009-2015

Martinsen K T T Kragh and K Sand-Jensen 2019 Carbondioxide fluxes of air-exposed sediments and desiccatingponds Biogeochemistry 144 165ndash180 doi101007s10533-019-00579-0

Marushchak M E and others 2013 Carbon dioxide balanceof subarctic tundra from plot to regional scales Bio-geosciences 10 437ndash452 doi105194bg-10-437-2013

Matthews J A S-O Dahl M S Berrisford and A Nesje 1997Cyclic development and thermokarstic degradation of palsasin the mid-alpine zone at Leirpullan Dovrefjell SouthernNorway Permafr Periglac Process 8 107ndash122 doi101002(SICI)1099-1530(199701)81lt107AID-PPP237gt30CO2-Z

Matveev A 2019 Variable effects of climate change on carbonbalance in northern ecosystems IOP Conf Ser Earth Envi-ron Sci 226 012023 doi1010881755-13152261012023

Matveev A I Laurion B N Deshpande N Bhiry and W FVincent 2016 High methane emissions from thermokarstlakes in subarctic peatlands Limnol Oceanogr 61S150ndashS164 doi101002lno10311

Matveev A I Laurion and W F Vincent 2018 Methane andcarbon dioxide emissions from thermokarst lakes on mineralsoils Arct Sci 4 584ndash604 doi101139as-2017-0047

Negandhi K I Laurion M J Whiticar P E Galand X Xuand C Lovejoy 2013 Small thaw ponds An unaccountedsource of methane in the Canadian High Arctic PLoS One8 e78204 doi101371journalpone0078204

Pavlova O A O S Pokrovsky R M Manasypov L SShirokova and S N Vorobyev 2016 Seasonal dynamics ofphytoplankton in acidic and humic environment in shal-low thaw ponds of western Siberia discontinuous perma-frost zone Ann Limnol Int J Limnol 52 47ndash60 doi101051limn2016006

Pokrovsky O S L S Shirokova S N Kirpotin S Audry JViers and B Dupreacute 2011 Effect of permafrost thawing onorganic carbon and trace element colloidal speciation inthe thermokarst lakes of western Siberia Biogeosciences 8565ndash583 doi105194bg-8-565-2011

Pokrovsky O S L S Shirokova S N Kirpotin S PKulizhsky and S N Vorobiev 2013 Impact of westernSiberia heat wave 2012 on greenhouse gases and trace metalconcentration in thaw lakes of discontinuous permafrostzone Biogeosciences 10 5349ndash5365 doi105194bg-10-5349-2013

Polishchuk Y S Kirpotin and N Bryksina 2014 Remotestudy of thermokarst lakes dynamics in West-Siberian per-mafrost p 173ndash204 In O S Pokrovsky [ed] PermafrostDistribution composition and impacts on infrastructure andecosystems Nova Science Publishers httpsnovapublisherscomshoppermafrost-distribution-composition-and-impacts-on-infrastructure-and-ecosystems

Polishchuk Y M A N Bogdanov V Y Polischuk R MManasypov L S Shirokova S N Kirpotin and O SPokrovsky 2017 Size-distribution surface coverage watercarbon and metal storage of thermokarst lakes (gt 05 ha) inpermafrost zone of the Western Siberia Lowland Water 9228 doi103390w9030228

Polishchuk Y M A N Bogdanov I N Muratov V YPolishchuk A Lim R M Manasypov L S Shirokova andO S Pokrovsky 2018 Minor contribution of small thawponds to the pools of carbon and methane in the inlandwaters of the permafrost-affected part of the Western Sibe-rian Lowland Environ Res Lett 13 04500 doi1010881748-9326aab046

Polishchuk Y M I N Muratov and V Y Polishchuk 2020Remote research of spatiotemporal dynamics ofthermokarst lakes fields in Siberian permafrost p 425 InO S Pokrovsky S N Kirpotin and A I Malov [eds] TheArctic Current issues and challenges Nova Science Pub-lishers httpsnovapublisherscomshopthe-arctic-current-issues-and-challenges

Poulin B A J N Ryan and G R Aiken 2014 Effects of ironon optical properties of dissolved organic matter EnvironSci Technol 48 10098ndash10106 doi101021es502670

Zabelina et al C emission from thermokarst lakes

13

Prokushkin A S and others 2011 Sources and export fluxesof dissolved carbon in rivers draining larch-dominatedbasins of the Central Siberian Plateau Environ Res Lett 6045212 doi1010881748-932664045212

Raudina T and others 2018 Permafrost thaw and climatewarming may decrease the CO2 carbon and metal concen-tration in peat soil waters of the Western Siberia LowlandSci Total Environ 634 1004ndash1023 doi101016jscitotenv201804059

Rautio M F Dufresne I Laurion S Bonilla W Vincent andK Christoffersen 2011 Shallow freshwater ecosystems ofthe circumpolar Arctic Ecoscience 18 204ndash222 doi10298018-3-3463

Repo M E J T Huttunen A V Naumov A V ChichulinE D Lapshina W Bleuten and P J Martikainen 2007Release of CO2 and CH4 from small wetland lakes in west-ern Siberia Tellus B 59 788ndash796 doi101111j1600-0889200700301x

Riordan B D Verbyla and A D McGuire 2006 Shrinkingponds in subarctic Alaska based on 1950ndash2002 remotelysensed images J Geophys Res 111 G04002 doi1010292005JG000150

Sabrekov A F B R K Runkle M V Glagolev I ETerentieva V M Stepanenko O R Kotsyurbenko S SMaksyutov and O S Pokrovsky 2017 Variability in meth-ane emissions from West Siberiarsquos shallow boreal lakes Bio-geosciences 14 3715ndash3742 doi105194bg-14-3715-2017

Sepulveda-Jauregui A K W Anthony K Martinez - Crus SGreene and F Thalasso 2015 Methane and carbon diox-ide emissions from 40 lakes along a north-south latitudinaltransect in Alaska Biogeosciences 12 3197ndash3223 doi105194bg-12-3197-2015

Serikova S and others 2018 High riverine CO2 emissions atthe permafrost boundary of Western Siberia Nat Geosci11 825ndash829 doi101038s41561-018-0218-1

Serikova S O S Pokrovsky H Laudon I V KrickovA G Lim R M Manasypov and J Karlsson 2019 Highcarbon emissions from thermokarst lakes of WesternSiberia Nat Commun 10 1552 doi101038s41467-019-09592-1

Shirokova L S O S Pokrovsky S N Kirpotin C DesmukhB G Pokrovsky S Audry and J Viers 2013 Biogeochemis-try of organic carbon CO2 CH4 and trace elements inthermokarst water bodies in discontinuous permafrostzones of Western Siberia Biogeochemistry 113 573ndash593doi101007s10533-012-9790-4

Shirokova L S and others 2019 Humic surface waters of fro-zen peat bogs (permafrost zone) are highly resistant to bio-and photodegradation Biogeosciences 16 2511ndash2526 doi105194bg-16-2511-2019

Sjoumlberg Y G Hugelius and P Kuhry 2013 Thermokarst lakemorphometry and erosion features in two peat plateauareas of Northeast European Russa Permafr Periglac Pro-cess 24 75ndash81 doi101002ppp1762

Smith L C Y Sheng G M MacDonald and L D Hinzman2005 Disappearing arctic lakes Science 308 1429 doi101126science1108142

Smith L C Y Sheng and G M MacDonald 2007 A firstpan-Arctic assessment of the influence of glaciation perma-frost topography and peatlands on northern hemispherelake distribution Permafr Periglac Process 18 201ndash208doi101002ppp581

Stackpoole S M D E Butman D W Clow K L VerdinB V Gaglioti H Genet and R G Striegl 2017 Inlandwaters and their role in the carbon cycle of Alaska EcolAppl 27 1403ndash1420 doi101002eap1552

Tan Z and Q Zhuang 2015 Arctic lakes are continuousmethane sources to the atmosphere under warming condi-tions Environ Res Lett 10 054016 doi1010881748-9326105054016

Tan Z Q Zhuang N J Shurpali M E Marushchak C BiasiW Eugster and K Walter Anthony 2017 Modeling CO2

emissions from Arctic lakes Model development and site-level study J Adv Model Earth Syst 9 2190ndash2213 doi1010022017MS001028

Tank S E R H H Esslein and L F W L Lesack 2009Northern delta lakes as summertime CO2 absorbers withinthe arctic landscape Ecosystems 12 144ndash157 doi101007s10021-008-9213-5

Tranvik L J and others 2009 Lakes and reservoirs as reg-ulators of carbon cycling and climate Limnol Ocean-ogr 54 2298ndash2314 doi104319lo2009546_part_22298

Vachon D and Y T Prairie 2013 The ecosystem size andshape dependence of gas transfer velocity versus windspeed relationships in lakes Can J Fish Aquat Sci 701757ndash1764 doi101139cjfas-2013-0241

van Huissteden J C Berrittella F J W Parmentier Y MiT C Maximov and A J Dolman 2011 Methane emis-sions from permafrost thaw lakes limited by lake drain-age Nat Clim Chang 1 119ndash123 doi101038nclimate1101

Vasiliev A V D S Drozdov A G Gravis G V MalkovaK E Nyland and D A Streletskiy 2020 Permafrost degra-dation in the Western Russian Arctic Environ Res Lett15 045001 doi1010881748-9326ab6f12

Vonk J E and others 2015 Reviews and syntheses Effectsof permafrost thaw on Arctic aquatic ecosystems Bio-geosciences 12 7129ndash7167 doi105194bg-12-7129-2015

Vonk J E S E Tank and M A Walvoord 2019 Integrat-ing hydrology and biogeochemistry across frozen land-scapes Nat Commun 10 5377 doi101038s41467-019-13361-5

Walter K M S A Zimov J P Chanton D Verbyla and F SChapin III 2006 Methane bubbling from Siberian thawlakes as a positive feedback to climate warming Nature443 71ndash75 doi101038nature05040

Zabelina et al C emission from thermokarst lakes

14

Walter K M L C Smith and F S Chapin III 2007 Methanebubbling from northern lakes Present and future contribu-tions to the global methane budget Philos Trans A MathPhys Eng Sci 365 1657ndash1676 doi101098rsta20072036

Walter Anthony K M and others 2014 A shift ofthermokarst lakes from carbon sources to sinks during theHolocene epoch Nature 511 452ndash456 doi101038nature13560

Wik M R K Varner K Walter Anthony S MacIntyre andD Bastviken 2016 Climate-sensitive northern lakes andponds are critical components of methane release NatGeosci 9 99ndash105 doi101038ngeo2578

Zamolodchikov D and D Karelin 2001 An empirical modelof carbon fluxes in Russian tundra Glob Chang Biol 7147ndash161 doi101046j1365-2486200100380x

AcknowledgmentsThis work was supported by the RNF (RSCF) grant 15-17-10009 ldquoEvo-

lution of thermokarst lake ecosystems in the context of climate changerdquo(sampling interpretation 50) State Task AAAA-A18-118012390200-5RFBR grant 18-05-70087 ldquoArctic Resourcesrdquo 17-05-00348 19-07-0028218-45-860002 18-45-703001 and 18-47-700001 and the SwedishResearch Council (grant 2016-05275) Chris Benker is thanked for thor-ough English editing

Conflict of InterestNone declared

Submitted 05 November 2019

Revised 06 March 2020

Accepted 25 June 2020

Associate editor Suzanne Tank

Zabelina et al C emission from thermokarst lakes

15