Partitioning CO2 net ecosystem exchange fluxes on the microsite scale in the Lena River Delta, Siberia

Arctic tundra ecosystems are currently facing rates of amplified climate change. This is critical as these ecosystems store significant amounts of carbon in their soils, which can be mineralized to CO2 and CH4 and released to the atmosphere. To understand how the CO2 net ecosystem exchange (NEE) fluxes will react to changing climatic conditions, it is necessary to understand the individual responses of the physiological processes 15 contributing to CO2 NEE. Therefore, this study aimed: i) to partition NEE fluxes at the soil-plant-atmosphere interface in an arctic tundra ecosystem; and ii) to identify the main environmental drivers of these fluxes. Hereby, the NEE fluxes were partitioned into gross primary productivity (GPP) and ecosystem respiration (Reco) and further into autotrophic (RA) and heterotrophic respiration (RH). The study examined flux data collected during the growing season in 2015 using closed chamber measurements in a polygonal tundra landscape in the Lena River 20 Delta, northeastern Siberia. The measured fluxes on the microscale (1 m – 10 m) were used to model the NEE, GPP, Reco, RH, RA and net ecosystem production (NPP) over the growing season. Here, for the first time, the differing response of in situ measured RA and RH fluxes from permafrost-affected soils to hydrological conditions have been examined. It was shown that low RA fluxes are associated to a high water table, most likely due to the submersion of mosses, while an effect of water table fluctuations on RH fluxes was not observed. Furthermore, 25 this work found the polygonal tundra in the Lena River Delta to be a sink for atmospheric CO2 during the growing season. Spatial heterogeneity was apparent with the net CO2 uptake at a wet, depressed polygon center being more than twice as high as that measured at a drier polygon rim. In addition to higher GPP fluxes, the differences in NEE between the two microsites were caused by lower Reco fluxes at the center compared to the rim. Here, the contrasting hydrological conditions caused the CO2 flux differences between the microsites, where high water 30 levels lad to lower decomposition rates due to anoxic conditions.


Introduction
More than 1,000 Petagrams of organic carbon (OC) are stored in the upper 3 m of northern permafrost-affected soils (Hugelius et al., 2014).Given the large amount of OC stored in these soils, the response of the arctic carbon (C) cycle to a changing climate is of global importance (McGuire et al., 2009).Over thousands of years, C has 35 been sequestered in permafrost-affected soils and sediments due to cold conditions and poor drainage resulting in water saturation and slow organic matter decomposition.Currently, arctic ecosystems are facing amplified warming (Chapin et al., 2005), which will lead to the longer and deeper thawing of permafrost-affected soils Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-311Manuscript under review for journal Biogeosciences Discussion started: 30 August 2018 c Author(s) 2018.CC BY 4.0 License.(Romanovsky et al., 2010).On the one hand, the microbial decomposition of liberated permafrost organic matter releases carbon dioxide (CO2) and methane (CH4) (Grosse et al., 2011;Knoblauch et al., 2018;Knoblauch et al., 40 2013;Schuur et al., 2009;Zimov et al., 2006a).On the other hand, higher temperatures increase the assimilation of CO2 by tundra vegetation due to a longer growing period and increased nutrient availability in the deeper layers of thawed soils (Jia et al., 2009;Schuur et al., 2007).
Although the CO2 budget of the arctic tundra has been the topic of several studies (e.g.Kittler et al., 2016;Kutzbach et al., 2007b;Marushchak et al., 2013;Merbold et al., 2009;Oechel et al., 2000) the extent of their source or sink 45 strength has not been well established.Based on flux measurements, atmospheric inversion and process models, McGuire et al. (2012) report that the arctic tundra still acts as a sink for atmospheric CO2 on an annual basis with a total CO2 uptake of 110 Teragram (Tg) C yr -1 .However, this value is associated with a high uncertainty ranging from 291 Tg C yr -1 uptake to 91 Tg C yr -1 loss.In contrast, in a meta-analysis of available flux observations in the arctic tundra Belshe et al. (2013) report that these ecosystems act as source for atmospheric CO2 (462 Tg C yr -1 ).50 However, the vast majority of the flux measurements taken into account by Belshe et al. (2013) have been conducted in North America although the Russian arctic comprises an area of 3 million km 2 (CAVM- Team, 2003) which is a large amount of the global tundra region.These current uncertainties in the arctic CO2 budget clearly show the need for intensified CO2 flux observations in tundra ecosystems, in particular on the microsite scale (1-10 m) in northern Siberia (Virkkala et al., 2017) as only a few studies are available from these ecosystems (e.g.55 Corradi et al., 2005;Heikkinen et al., 2004;Kwon et al., 2016;Zamolodchikov et al., 2000).An improved understanding of CO2 dynamics in permafrost-affected soils is needed to improve estimates of future CO2 balances of the highly heterogeneous arctic tundra regions.Without furthering knowledge, estimates of the carbon balance of the circum-arctic tundra and its future reaction to changing climatic conditions remain biased.
The net ecosystem exchange (NEE) of CO2 between the land surface and the atmosphere is composed of the CO2 60 uptake by plants, termed the gross primary productivity (GPP); and the release of CO2 from soils and plants, the ecosystem respiration (Reco) (Chapin et al., 2006).The latter can further be split into autotrophic respiration by plants (RA) and heterotrophic respiration (RH) mainly by microorganisms that decompose soil organic matter.
Inorganic sinks and sources of CO2 are generally neglected because of their minor contribution to NEE (Elsgaard et al., 2012;Kuzyakov, 2006).In order to partition NEE into the underlying fluxes, measurements of GPP, Reco, 65 RA and RH are required.These individual process-based fluxes governing the CO2 balance react differently to changing climatic conditions.
The release of CO2 from soils by Reco is the largest efflux of C from terrestrial ecosystems to the atmosphere (Mahecha et al., 2010).RA can be separated into aboveground plant respiration and belowground root respiration.
RH is associated with the decomposition of soil organic matter (SOM) by heterotrophic soil organisms.It is 70 challenging to separate belowground respiration fluxes into autotrophic and heterotrophic components because roots and microorganisms are closely linked within the rhizosphere (Hanson et al., 2000).There is a wide range of methods to partition ecosystem respiration (Kuzyakov, 2006;Subke et al., 2006), each with its associated advantages and disadvantages.Root trenching for example, despite some disturbance on the plant-soil interface, can give accurate estimates of the rates of RA and RH (Diaz-Pines et al., 2010) and produces similar results as a 75 non-disturbing 14 C partitioning approach in an arctic tundra ecosystem (Biasi et al., 2014) and a partitioning approach based on 13 C (Chemidlin Prévost-Bouré et al., 2008).
To date, a few estimates on the contribution of RH to Reco from tundra ecosystems during the growing season have been published (Biasi et al., 2014;Nobrega and Grogan, 2008), with data lacking for ecosystems such as the Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-311Manuscript under review for journal Biogeosciences Discussion started: 30 August 2018 c Author(s) 2018.CC BY 4.0 License.
polygonal tundra.This is critical, as warming of the Arctic will influence Reco fluxes both directly and indirectly: 80 Warming will increase decomposition of soil organic matter (RH), but it will also cause permafrost to thaw, which will expose previously frozen SOM to microbial decomposition (Dorrepaal et al., 2009;Schuur et al., 2011).This could cause a substantial reduction of the carbon sink function of arctic tundra ecosystems as gross ecosystem productivity has been found to be less temperature-sensitive than Reco in these ecosystems (Dorrepaal et al., 2009;Grogan and Chapin, 2000).Furthermore, warming could reduce microbial biomass (Frey et al., 2008) and soil 85 moisture (Suseela et al., 2012), and increase RA due to increasing aboveground biomass, which can lead to a lower contribution of RH to Reco (Chen et al., 2016;Hicks Pries et al., 2015).The increase of RA and RH fluxes due to warming might be compensated by higher net primary production (Hicks Pries et al., 2013), but whether this is valid for the complete growing season and across the highly heterogeneous arctic ecosystems remains uncertain.
The current study improves the understanding of CO2 flux dynamics in permafrost-affected ecosystems of 90 northeastern Siberia by partitioning CO2 NEE into its underlying processes: photosynthesis, ecosystem respiration as well as autotrophic and heterotrophic respiration at two typical polygonal tundra microsites.Furthermore, the response of these processes to different environmental parameters such as temperature and hydrology is revealed.
Finally, a CO2 budget for a nearly complete vegetation period is determined for the two microsites using datacalibrated flux models.These models were based on the time-sensitive bulk flux partitioning model by Runkle et 95 al. (2013), which was used in different arctic ecosystems (Helbig et al., 2017;Zona et al., 2014).

Study site
The investigation area is located on Samoylov Island in the southern central Lena River Delta, northeastern Siberia (72°22'N, 126°28'E -Figure 1).The Lena river forms the largest delta in the Arctic, which can be geomorphologically divided in river terraces of different ages and flood-plain levels (Schwamborn et al., 2002).100 The delta is located in the continuous permafrost zone with permafrost extending to depths of down to 300 to 500 m (Yershov, 1998) and relatively low mean annual soil temperatures of -7.8 °C at 1.7 m depth (Boike et al., 2013).With low temperatures and low precipitation, the study site has an arctic continental climate.The mean annual air temperature between 1998 and 2011 was -12.5 °C, and mean annual rainfall is 125 mm, ranging from 52 mm to 199 mm (Boike et al., 2013).Polar day lasts from 7 May until 8 August, and polar night lasts from 15 105 November to 28 January.Snowmelt usually starts in the first half of June, and growing season usually occurs from around mid-June until mid-September.
The study site is covered by an ice-wedge polygonal tundra on a Late-Holocene river terrace with elevations from 10 to 16 m above sea level on the eastern part of Samoylov Island.The development of polygonal structures has created depressed polygon centers surrounded by elevated polygon rims with elevation differences of about 0.5 m. 110 Underlying permafrost prevents drainage in polygon centers with water-saturated soils, anoxic conditions at shallow depths, and significant amounts of soil organic carbon of around 33 kg m -2 in the uppermost meter (Zubrzycki et al., 2013).In contrast, the elevated polygon rim soils accumulated less soil organic carbon of around 19 kg m -2 (Zubrzycki et al., 2013) due to oxic conditions in the top soil.The land cover ratio of the polygonal tundra on Samoylov Island is 65% for polygon rims, 19% for polygon centers and 16% for open water bodies 115 (Muster et al., 2012).
In this study, two different microsites were investigated: i) a wet-depressed polygon center; and ii) its surrounding elevated polygon rim (72°22,442 N; 126°29.828E).These microsites were located within the footprint area of an Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-311Manuscript under review for journal Biogeosciences Discussion started: 30 August 2018 c Author(s) 2018.CC BY 4.0 License.eddy covariance (EC) system (Kutzbach et al., 2007b;Runkle et al., 2013;Wille et al., 2008).At this polygon, the maximum active layer depth (ALD) was deeper at the polygon center (40 cm) compared to the polygon rim 120 (30 cm).The soils at the polygon centers were classified as Histic or Reductaquic Cryosols with water table close to the soil surface (WRB, 2014).Polygon rim soils were characterized by cryoturbation and therefore classified as Turbic Glacic Cryosols with a water table just a few centimeters above the permafrost table.Total organic carbon (TOC) contents above 10% were found in the surface horizon (0-6 cm) above the cryoturbated horizons of the polygon rim at the study site, while high TOC contents at the polygon center were found in the entire active layer.125 The vegetation of the polygon rims were dominated by mosses (Hylocomium splendens, Polytrichum spp., Rhytidium rugosum), some small vascular plants (Dryas punctata and Astragalus frigidus) as well as lichens (Peltigera spp.) and was classified after Walker et al. (2005) as non-tussock sedge, dwarf-shrub, moss tundra.The vegetation of the polygon centers were dominated by the hydrophilic sedge Carex aquatilis and mosses (Drepanocladus revolvens, Meesia triqueta, Scorpidium scorpioides) and was classified as sedge, moss, dwarf-130 shrub wetland according to Walker et al. (2005).

Meteorological data
Meteorological variables were recorded at 30 minute intervals at the nearby EC system and adjacent meteorological station 40 m southwest of the study site.Data on relative humidity and air temperature (MP103A, 135 ROTRONIC AG, Switzerland), air pressure (RPT410F, Druck Messtechnik GmbH, Germany) and photosynthetically active radiation (PAR; wavelength: 400 -700 nanometers; QS2, Delta-T Devices Ltd., UK) as well as the incoming and reflected components of shortwave and longwave radiation, respectively (CNR 1, Kipp and Zonen, Netherlands), were collected.The radiative surface temperature (Tsurf; in Kelvin (K)) was calculated as: 140 where  ↑  is the upward infrared radiation (W m -2 ), σ is the Stefan-Boltzmann constant (W m -2 K -4 ), and the dimensionless emissivity ε was assumed to be 0.98 after Wilber et al. (1999).Furthermore, soil temperature was measured at 2 cm soil depth in intervals of 30 minutes at an adjacent polygon rim and center.

Soil sampling and vegetation indices 145
A total of six soil samples were taken from the thawed layer at the polygon rim using steel rings.At the polygon center, one soil sample was taken from the thawed layer using a spade, and separated into four soil layers based on their degradation status.Soil samples were homogenized for analysis of soil water content (mass difference between wet and dried (105 °C) soil samples), pH (CG820, Schott AG, Mainz, Germany) total C and nitrogen (N) contents (VarioMAX cube, Elementar Analysesysteme GmbH, Hanau, Germany), as well as total organic carbon 150 (TOC, liquiTOC II, Elementar Analysesysteme GmbH, Hanau, Germany) and total inorganic C (difference between total C and TOC).To analyze vegetation indices, gridded quadrats of 10 cm x 10 cm squares were placed over the collars, and a visual identification of the plant species present as well as their abundance (% surface cover) was conducted in four grid squares.

Net ecosystem exchange and ecosystem respiration 155
A total of eight PVC collars (50 cm x 50 cm), four at each microsite, were installed in July 2014 in preparation for NEE and Reco flux measurements with closed chambers the following year.The collars were equipped with a U-shaped frame filled with water to avoid gas exchange between the chamber headspace and ambient air.The chamber (50 cm x 50 cm x 50 cm) used for NEE and Reco flux measurements was made of clear plexiglas (Plexiglas SunActive GS, Evonik Industries AG, Germany).The chamber was equipped with a fan for continuous 160 mixing of headspace air (axial fan, 12V/DC, Conrad Electronic SE, Germany).Furthermore, a PAR sensor (SKP212, Skye Instruments Ltd., UK) and a temperature probe (107 Thermistor probe, Campbell Scientific Ltd., USA) were installed inside the chamber.Including the volume inside the chamber collars, the chamber enclosed a volume of 124 -143 L. For Reco measurements, the chamber was covered with an opaque box (dark chamber measurements).Boardwalks were installed at both microsites to avoid disturbance.The volumetric soil water 165 content (VWC) was measured with a GS3 sensor (Decagon Devices, Inc., USA) during each measurement directly beside the chamber collar in a depth of 5 cm.A diver (Schlumberger Ltd., USA) was installed at the polygon center to measure water table (WT) depth every 15 minutes.To prevent pressure-induced gas release during chamber closure (Christiansen et al., 2011), two holes (3 cm in diameter) at the top of the chamber were left open while placing the chamber on the collars and then closed for measurements.Soil temperatures between the surface 170 and the frozen ground in 5 cm intervals and thaw depth were measured daily at both microsites.For each chamber flux measurement, CO2 concentrations in the chamber headspace were continuously measured with an Ultra-Portable Gas Analyzer (UGGA 30-p, Los Gatos Research, USA).The chamber headspace was pumped in a closed loop system via transparent polyurethane tubes (inner diameter 4 mm, each 10 m length) through the analyzer with a flowrate of 200 mL min -1 .The CO2 concentration was logged (CR800series, Campbell Scientific 175 Ltd., USA) together with PAR as well as soil and air temperature at a frequency of 1 Hz.Each chamber measurement was restricted to 120 sec to minimize warming inside the chamber relative to the ambient environment.
Chamber measurements were conducted from 11 July until 22 September 2015.The chamber measurements were done at least every third day between 6 am and 9 pm (local time), apart from the period 2-9 August and 17-24 180 August.Two consecutive measurements were performed at each collar: first, NEE (n = 679) was measured with the transparent chamber, followed by an Reco measurement (n = 679) with the dark chamber shortly after.The four collars of one microsite were measured in sequence before moving to the other microsite.GPP fluxes were calculated from the sum of the measured Reco and NEE fluxes.

Heterotrophic respiration 185
For RH measurements the root-trenching method was applied at both microsites.By inserting PVC collars 20 cm deep into the soil, which is below the main rooting zone, lateral roots were cut off.All living plant biomass including living moss tissue inside the collars were removed carefully in 2014.To prevent re-growth, the living plant biomass was removed periodically over the measurement period.This removal causes the die-off of roots, and days after the disturbance RH equals NEE.A total of eight collars, four at each microsite, were prepared for 190 RH measurements.RH fluxes (n = 662) were measured during the same periods and with the same measurement interval as NEE and Reco measurements on unaltered plots.

Flux calculation 200
CO2 fluxes (µg CO2 m -2 s -1 ) were calculated using MATLAB® R2015a (The MathWorks Inc., Natick, MA, 2000) with a routine that uses different regression models to describe the change in the chamber headspace CO2 concentration over time and conducts statistical analysis to aid model selection (Eckhardt and Kutzbach, 2016;Kutzbach et al., 2007a).
Due to possible perturbations while placing the chamber on the collar, the first 30 seconds of each 2-minute 205 measurement period were discarded and the remaining 90 data points were used for flux calculations.The precision of the gas analyzer with 1 s signal filtering is < 0.3 ppm for CO2.Typically, the root mean square error (RMSE) of chamber measurements and the fitting of the linear and non-linear regression models did not exceed this value, and higher RMSE values indicated failed model fitting or disturbed chamber measurements.Therefore, if RMSE exceeded 0.3 ppm, the concentration-over-time curve was re-inspected.Variation of PAR during chamber 210 measurements due to shifts in cloud cover leads to irregular CO2 concentration time series and perturbation of the calculated CO2 fluxes (Schneider et al., 2011).These perturbed concentration time series show distinct autocorrelation of the residuals of the regression models and were filtered out by using a threshold for residual autocorrelation indicated by the Durbin-Watson test (Durbin and Watson, 1950).If the RMSE exceeded 0.3 ppm or showed a distinct autocorrelation, the flux curve was re-inspected to see if irregularities could be removed by 215 adjusting the time series.If irregularities could be removed by adjusting the time series, the flux curve was re-calculated and if not, the dataset was discarded.Overall, about 3% (n = 47) of the CO2 flux measurements (NEE, Reco and RH measurements) were discarded from the dataset, because they did not meet the abovementioned quality criteria.
It was shown that CO2 fluxes calculated with linear regression models can be seriously biased (Kutzbach et al., 220 2007a), while non-linear regression models significantly improve flux calculations (Pihlatie et al., 2013).However, we found that the change in CO2 concentration in the chamber was modelled best with a linear regression model, as determined by the Akaike Information Criterion corrected for small samples sizes (AICc) (Burnham and Anderson, 2004).This is in good agreement with other studies, which have shown that in some cases a linear regression model can produce a better CO2 flux estimate for a non-linear concentration-over-time curve than a 225 non-linear regression model (Görres et al., 2014;Koskinen et al., 2014).

Modelling CO2 fluxes
In this study the atmospheric sign convention is used, whereby a positive NEE defines a net release of CO2 from the soil to the atmosphere.To determine their single contribution to the NEE, different a range of numerical models were fitted to the measured CO2 fluxes to quantify seasonal GPP,Reco,and RH fluxes. 230 To calibrate the models, the used functions were fitted to the GPP, Reco, and RH fluxes.The resulting fitting parameters were used to reproduce the fluxes over the complete measurement period.Model calibration was done by applying a 15-day moving window over the measurement period moving on a daily basis.If less than eight Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-311Manuscript under review for journal Biogeosciences Discussion started: 30 August 2018 c Author(s) 2018.CC BY 4.0 License.chamber measurements were performed during these 15 days, the moving window was extended to 19 days.Subsequently, the modelled fluxes for each measurement plot were averaged for each microsite.CO2 fluxes from 235 each of the four measurement plots were used separately for model calibration and the summed fluxes were used to analyze differences between both microsites using a student's t-test.
The empirical Q10 model (van't Hoff, 1898) was fitted to the measured Reco and RH fluxes: where the fit parameter Rbase was the respiration at the reference temperature Tref (15 °C).Tref and γ (10 °C) were 240 held constant according to Mahecha et al. (2010).Q10 was a fit parameter indicating the ecosystem sensitivity of respiration to a 10 °C change in temperature.For this study a fixed Q10 value of 1.52 was used, which represents the seasonal mean value of the bulk partitioning model for the CO2 fluxes in the EC footprint area established by Runkle et al. (2013).Air temperature (Ta), surface temperature (Tsurf), and soil temperature (Tsoil) measured at a depth of 5 cm were tested as input variables.245 The model calibration was done with MATLAB® R2015a (The MathWorks Inc., Natick, MA, 2000).The model parameters were estimated by nonlinear least-squares regression fitting (nlinfit function), and the uncertainty of the parameters was determined by calculating the 95% confidence intervals using the nlparci function.The selection of the best performing temperature as input variable for the Reco and RH model was based on comparing the R 2 adj of the model runs with different temperatures as input variable.The selected input variable was chosen 250 for all measurement plots of the same microsite, although at times for a given plot a different input variable might have had a better R 2 adj.
To estimate GPP, the chamber measured Reco fluxes were subtracted from the NEE fluxes separately for each measurement plot.The rectangular hyperbola function was fitted to the calculated GPP fluxes as a function of PAR (in µmol m -2 s -1 ): 255 where the fit parameter Pmax was the maximum canopy photosynthetic potential (hypothetical maximum of Pmax at infinite PAR).The values for the initial canopy quantum efficiency α (dimensionless; initial slope of the Pmax -PAR curve at PAR = 0) were obtained from modelling the CO2 fluxes with EC data (Kutzbach et al., unpublished).
From the determined values when α was held variable, a function was formulated that account for the seasonality 260 of α with specific values for every day of the growing season using the following function: where  = 0.042,  = 209.5, = 2,  = 25.51, = 0.008 and  = day of year 2015.Afterwards, these values (variable on daily basis) were used to reproduce GPP fluxes from chamber measurements over the complete measurement period.265 Although the transmissivity of the chamber material was high (> 92%), it caused a reduction in the amount of incoming radiation reaching the surface.During the complete measurement period, the PAR values measured inside the chamber were on average 20% lower than the PAR values measured outside the chamber (data not shown).Therefore, GPP modelling was conducted in two steps.First, the GPP model was calibrated using PAR values measured inside the chamber; and secondly, the reproduction of GPP fluxes over the growing season was 270 carried out using PAR values measured outside the chamber.Without this two-step calibration the CO2 uptake rates would have been underestimated.

Meteorological data and environmental conditions
From mid-July to the end of September 2015, soil temperatures at 2 cm depth at the polygon rim showed a higher diurnal variability than at the center.The highest soil temperatures were measured in mid-July and at the beginning of August.At the end of September, the temperatures became slightly negative (Figure 2).The mean daily air 280 temperature ranged between 23 °C and -2 °C.Three warm periods were recorded, one in mid-July (up to 27 °C) and one at the beginning of August (up to 25 °C) and a third warm period was recorded at the beginning of September with temperatures of up to 20 °C.The average temperature in August 2015 (9 °C) was similar to the long-term mean temperature over the period 1998-2011 (Boike et al., 2013).Compared to the long-term mean, it was about 1°C colder during July (9 °C), whereas September was around 2 °C warmer than the reference period 285 (3 °C).The total precipitation from mid-July to end of September 2015 was 78 mm.
At the polygon rim, the thaw depth increased from the beginning of the campaign in mid-July until mid-September to reach a maximum depth of 36 cm.Maximum thaw depth was reached at the polygon center much earlier in the season (mid-July) and remained relatively constant until mid-September.From mid-September onwards the thaw depth decreased slightly until the end of September at both microsites.The water table depth at the polygon center 290 were tightly coupled to rainfall.The VWC at 5 cm soil depth was on average 30% at the polygon rim, with highest values observed after rainfall events (Figure 2).
Photosynthetically active radiation showed strong diurnal variability, with lowest PAR values during nighttime.
From mid-July until 12 August, mean nighttime (9 pm -3 am) PAR values did not drop below 5 µmol m -2 s -1 (Figure 2).Afterwards, with upcoming polar night conditions, the number of periods with PAR values below 295 5 µmol m -2 s -1 increased.Midday PAR values above 1000 µmol m -2 s -1 were measured in mid-July, at the beginning of August, and once at the end of August.Throughout September, the daily maximum PAR values were decreasing.
The total soil C content was lower at the polygon rim (2-12%) than the polygon center (10-20%) and showed a reduction with depth.At the polygon center the N content showed little variability with depth, with being around 300 0.6%, and the C/N ratio decreased from 33.1 at the surface to 16.9 at the bottom of the active layer.In contrast, in the polygon rim, the N content was considerably higher in the organic-rich layer compared to the mineral soil layer (0.5% vs. 0.1%), and the difference of the C/N ratio at different soil depths was smaller than in the center.Strongly acidic pH values were measured throughout the active layer of both microsites, with values around 5.3 at the polygon center, while in the organic-rich layer and in the mineral soil layer at the rim the pH was moderately acidic 305 with values of 5.7 and 6.0, respectively.

Carbon dioxide fluxes
The highest NEE fluxes were measured at the end of July with -97 ± 27 µg CO2 m -2 s -1 and -209 ± 17 µg CO2 m -2 s -1 at the rim and center (Figure 3), respectively.In September, both microsites acted as small net sources of atmospheric CO2.The net CO2 uptake at the center was generally higher than at the rim.The 310 Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-311Manuscript under review for journal Biogeosciences Discussion started: 30 August 2018 c Author(s) 2018.CC BY 4.0 License.
highest net CO2 release at the polygon rim was measured on 17 August with 16 ± 5 µg CO2 m -2 s -1 , and at the polygon center on 19 September with 22 ± 3 µg CO2 m -2 s -1 .The standard error of the flux calculation was about 3.5 and 2.3 µg CO2 m -2 s -1 for polygon center and rim, respectively, and decreased slightly towards the end of the season.
In contrast to the NEE fluxes, the measured Reco fluxes were on average higher at the rim compared to the center.315 The lowest ecosystem respiration fluxes at the polygon center were observed on 23 July with 10 ± 3 µg CO2 m -2 s -1 and at the polygon rim on 21 September with 17 ± 1 µg CO2 m -2 s -1 .The highest ecosystem respiration fluxes of 80 ± 11 and 88 ± 10 µg CO2 m -2 s -1 for rim and center, respectively, were measured at 9 August, when the air temperature exceeded 20 °C.
The net CO2 uptake increased from mid-July until it peaked during the vegetation maximum at the end of July and 320 beginning of August.Subsequently, NEE decreased until the end of September.This seasonality was more pronounced at the polygon center than at the polygon rim.Interestingly, towards September the net CO2 uptake at the polygon rim exhibited an increase for a period of about one week until it decreased again towards the end of September.Reco fluxes showed a similar, but less distinct seasonal pattern, and the peak of the highest Reco fluxes was in mid-August.In contrast, RH fluxes showed no seasonal trend.325 The lowest GPP fluxes were calculated for the end of September with -10 ± 3 µg CO2 m -2 s -1 and -16 ± 6 µg CO2 m -2 s -1 for the polygon center and rim, respectively (Figure 3).The highest GPP fluxes at the polygon rim were found at the end of July and beginning of August with -143 ± 33 µg CO2 m -2 s -1 as well as at the polygon center with -245 ± 19 µg CO2 m -2 s -1 .
As observed for the GPP fluxes, calculated NPP fluxes showed a distinct seasonality (Figure 3).The mean NPP fluxes were -94 ± 61 and -55 ± 26 µg CO2 m -2 s -1 at the polygon center and rim, respectively.The highest NPP 335 flux at the center was on 30 July with -222 ± 18 µg CO2 m -2 s -1 , which is three days earlier than the maximum GPP flux was determined.Similar to the highest GPP fluxes, the highest NPP flux at the rim was determined on 27 July with -115 ± 29 µg CO2 m -2 s -1 .The lowest NPP fluxes were determined in September with -10 ± 11 µg CO2 m -2 s -1 at the rim and -2 ± 6 µg CO2 m -2 s -1 at the center.
The highest releases of CO2 by RH were measured on 9 August at the polygon center and rim with 38 ± 6 and 340 51 ± 12 µg CO2 m -2 s -1 , respectively.The lowest RH fluxes of 3 ± 1 µg CO2 m -2 s -1 were measured at the center on 3 September, while at the rim lowest RH fluxes of 10 ± 3 µg CO2 m -2 s -1 were observed at the end of September.
Increased RH fluxes after periodical re-clipping of the vegetation were not observed.The comparison of RH fluxes from measurement plots that were trenched in 2014 with those that were trenched in 2015 revealed no significant differences (t-test, p > 0.05) between the years of root-trenching (data not shown).345 Interestingly, the Reco fluxes were linearly correlated with WT fluctuations from the beginning of July until the end of August (Figure 4).In contrast to this, neither a trend of higher RH fluxes during times of high WT, nor lower RH fluxes during times of low WT were observed.Instead, the RA fluxes showed a good correlation (R 2 = 0.71; p < 0.05) with WT fluctuations.

Modelled CO2 fluxes 350
The fitting parameter of the GPP model, Pmax, showed strong spatial and temporal variability (Figure 5).The α values used for the GPP model showed a high temporal variability with a mean of 1.47 ± 0.62.This value increased sharply towards the peak vegetation period at the end of July and decreased afterwards towards the end of the growing season.The Pmax values showed a strong temporal variability at the polygon center (mean: 250.7 ± 101.9 µg CO2 m -2 s -1 ).Considerable differences in Pmax were also observed between polygon rim and center.The 355 averaged Pmax at the polygon rim (135.4 ± 37.2 µg CO2 m -2 s -1 ) was substantially lower than at the polygon center (250.7 ± 101.9 µg CO2 m -2 s -1 ).As with the measured NEE fluxes, Pmax values displayed an increase at the polygon rim towards the end of September.The fitting parameter of the Reco and RH model, Rbase, also showed a strong spatial and temporal variability (Figure 5).In general, Rbase was higher at the polygon rim.The averaged Rbase values for the RH model fit differed substantially between microsites with 14.6 ± 2.1 µg CO2 m -2 s -1 at the polygon 360 center and 29.0 ± 2.9 µg µg CO2 m -2 s -1 at the polygon rim.
The best model fit of polygon center Reco fluxes (R 2 adj = 0.70) was achieved when the surface temperature was used as explanatory variable; while for the polygon rim the soil temperature showed the best fitting (R 2 adj = 0.46).In contrast to the Reco fluxes, the best model fit for polygon center RH fluxes was achieved when the air temperature was used as explanatory variable (R 2 adj = 0.55).At the polygon rim, using the soil temperature as explanatory 365 variable showed the best fitting (R 2 adj = 0.45) when modelling RH fluxes.However, differences in the goodness of the fits were small.At the polygon rim, the averaged modelled Reco fluxes were higher than at the center (Table 1), however the difference between the microsites was not statistically significant (t-test, p > 0.05).The highest Reco values were encountered at both microsites at the beginning of August (Figure 6 and Figure 7).At the polygon rim, the lowest 370 Reco fluxes where obtained at the end of September, while the lowest fluxes at the polygon center where observed on 21 July, associated with the highest water table during the campaign.
At the polygon center, the modelled RH fluxes were substantially lower than at the polygon rim, and this difference between the microsites was highly significant (t-test, p < 0.001).The lowest RH fluxes at the center were encountered on 3 September, a period of low air temperatures.The highest RH flux was observed when 375 temperatures first rose above 25 °C in July.The highest and lowest modelled RH fluxes at the polygon rim were encountered at the same time as the highest and lowest modelled Reco fluxes (8 August and 20 September).When comparing modelled data for RH and the Reco over the whole measurement period, the contribution of RH to Reco was on average 42 % at the center and 60 % at the rim.
The modelled GPP fluxes showed a distinct seasonal trend.From the mid of July until 12 August, photosynthesis 380 took place for 24 hours per day because of polar day conditions, even though the CO2 uptake was low during nighttime.Afterwards, periods when GPP was zero extended due to extended night conditions.At both microsites, the diurnal amplitude of GPP increased from the beginning of the campaign until it reached a maximum in mid-August.After this peak, the GPP fluxes at the center continuously decreased until the lowest daily maximum GPP on 21 September at the polygon center.Interestingly, the lowest daily GPP maximum at the polygon rim was 385 observed about a week earlier than at the polygon center.Later, the GPP at the polygon rim increased again.The difference in GPP fluxes between the microsites was statistically significant (p < 0.01).The R 2 adj of the GPP model was 0.82 for the polygon center and 0.45 for the polygon rim.

Integrated fluxes
Based on the modelled CO2 fluxes, the time-integrated CO2 fluxes for the period between mid-July and end of September 2015 were calculated (Figure 8).The integrated net CO2 uptake (NEE) was more than twice as high at 400 the center than at the rim.The integrated GPP flux at the polygon center was substantially higher than at the polygon rim.Interestingly, the integrated Reco flux at the rim was higher than at the center, and an almost twice as high RH flux at the rim was observed.The integrated RA fluxes at polygon center and rim were within the same range.During the vegetation period, the NPP was almost twice as high at the center compared to the rim.

Discussion 405
This study presented the first values of NEE, GPP, NPP as well as Reco, RH and RA fluxes obtained from modelling approaches for different microsites of the polygonal tundra were presented.These fluxes are of crucial importance as they show the different response that the underlying processes governing CO2 NEE have to environmental controls over the growing season, both spatially and temporally.Both the water-saturated polygon center and the non-saturated polygon rim acted as net sinks for atmospheric CO2 for the period from mid-July to end of September 410 2015.The CO2 sink strength differed substantially between the microsites, which is related to the different hydrological conditions.The integrated Reco fluxes at the rim were higher than at the center.This is remarkable as Reco fluxes are expected to rise with increasing GPP fluxes (Bubier et al., 2003), since CO2 uptake via photosynthesis is the source of RA fluxes.However, despite substantial higher GPP fluxes, the RA fluxes at the center were within the same range as those at the rim.Furthermore, higher Reco fluxes at the polygon rim compared 415 to the center were also caused by the higher rim RH fluxes due to drier soil conditions.Overall, the differences in RA and GPP fluxes between the two microsites led to NPP at the polygonal center being almost two-times higher than at the polygon rim.

CO2 fluxes from arctic tundra sites
The modelled Reco fluxes at both studied microsites were at the lower end of those reported for other arctic tundra 420 sites with similar vegetation and soil composition (Table 2).Solely, a wet tundra site in the Komi Republic, Russia (Heikkinen et al., 2004), a wet sedge site at Daring Lake, Canada (Nobrega and Grogan, 2008) and a polygon center site in Alaska (Oechel et al., 1995) showed Reco fluxes that were within the same range as in this study.The at the two studied microsites are lower than reported fluxes from most arctic study sites (Olivas et al., 2011;Ström et al., 2012;Vourlitis et al., 2000;Zamolodchikov et al., 2000).

Environmental controls on CO2 fluxes 430
The polygonal tundra on Samoylov Island had to be considered as an ecosystem with low GPP due to low vascular plant coverage with a maximum leaf coverage of 0.3 (Kutzbach et al., 2007b).Mosses, which have a high leaf coverage (> 0.9), were dominant at both microsites and have, similar to lichens, a much lower photosynthetic capacity than vascular plants (Brown et al., 1980).Furthermore, photosynthesis of vascular plants is restricted in arctic tundra ecosystems due to the low nutrient availability.Low nutrient availability is typical for most tundra 435 soils due to water saturation and low soil temperatures (Oechel et al., 1998), as these conditions ensure low microbial decomposition rates (Hobbie et al., 2002), which in turn result in a low supply of bioavailable nutrients (Beermann et al., 2015).Nitrogen stocks at the studied microsites were rather low compared to other arctic tundra sites (Harden et al., 2012).Additionally, the net radiation at the study site from June to August was with a mean of 85 W m -2 (1999-2011) lower than those reported from most other arctic tundra sites (cf.Boike et al., 2013;440 Lynch et al., 1999;Soegaard et al., 2001).These factors might explain the comparatively low GPP fluxes at the polygon rim and center compared to other arctic tundra sites.
Differences in GPP between the rim and center can be related to the vascular plant coverage.The polygon center had a much higher abundance of sedges, while the rim was moss-dominated and the sparsely spread vascular plants had shorter and fewer leaves.Therefore, the photosynthetic capacity of the center is higher than the rim, resulting 445 in higher GPP.Additionally, limited water availability due to the elevation of the polygon rim caused moisture to run off, with a drier or desiccated moss layer potentially contributing to lower GPP (Olivas et al., 2011).On the other hand, Olivas et al. (2011) found GPP fluxes to be higher at a polygon rim than at a polygon center in the Alaskan coastal plains.They related low GPP fluxes at the center to submersion of the moss layer and vascular plants.At the polygon center of the current study, the WT was frequently below the soil surface so that submersion 450 of erect vascular plants was not observed regularly, and even the moss layer was not submerged for most of the time.This difference in GPP between the Alaskan study sites (Olivas et al., 2011) and those presented in this study reveals the importance of water level and its fluctuations throughout the season.
Differences in NEE fluxes between the two microsites can also be related to their different soil conditions.The cold and water-logged conditions typical of the polygon centers, inhibited decomposition and mineralization of 455 SOM due to oxygen limitation, which caused low microbial activity (Hobbie et al., 2002;Walz et al., 2017).
Furthermore, the moisture run-off at the rim created drier conditions in the top soils at the rim, which increased soil oxygen availability and therefore enhanced RH and Reco (Oechel et al., 1998).The higher diurnal amplitude of the soil temperature, a product of the thermic buffer function of the standing water at the center, led to higher daily soil temperatures at the polygon rim compared to the center.Both the higher temperatures and oxygen supply 460 enhanced microbial decomposition and therefore higher RH fluxes at the polygon rim compared to the polygon center.Hence, low net carbon uptake at the rim occurred not only because of low GPP, but also due to higher RH fluxes compared to the center.This finding is in good agreement with Nobrega and Grogan (2008) who compared a wet sedge with a dry heath and a mesic birch site and found that the net carbon uptake at the wet sedge site was highest because of limited respiration due to water-logged conditions.465 Interestingly, measurements of CO2 fluxes at the polygon rim showed an increase of NEE throughout September, whereas at the polygon center the net CO2 uptake appeared to continuously decrease.This increase at the rim Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-311Manuscript under review for journal Biogeosciences Discussion started: 30 August 2018 c Author(s) 2018.CC BY 4.0 License.cannot be explained by rising PAR or temperature.Rather, the increase of net CO2 uptake at the rim towards the end of September may be related to the photosynthetic activity of mosses.Kutzbach et al. (2007b) considered the September at the EC footprint area as period where C uptake occurs mostly due to moss photosynthesis.During 470 this time of the growing season, mosses can still assimilate substantial amounts of CO2 because they tend to reach light saturation at lower irradiance (Harley et al., 1989).The photosynthetic activity of mosses declines rapidly when they face desiccation, because they cannot control their tissue water content (Turetsky et al., 2012).It was also shown that mosses face light stress during times of high PAR (Murray et al., 1993).This light stress causes delayed senescence and more late-season photosynthesis (Zona et al., 2011).Therefore, the photosynthetic activity 475 at the polygon rim is expected to be low during warm and dry periods such as those seen at the beginning of September 2015, and during times of high PAR.With continuous rainfall, dew formation and lower PAR in mid-September, the mosses are likely resume metabolic active, which led to increasing net CO2 uptake at the rim.

Partitioning respiration fluxes in arctic tundra ecosystems
To date only a handful of studies have estimated growing season RH fluxes from arctic tundra ecosystems (Biasi 480 et al., 2014;Nobrega and Grogan, 2008).Surprisingly, the differences between these estimates of RH fluxes rather low.Differences in RH fluxes measured with the trenching method may be caused by differences in the time between trenching and start of the measurement.Nobrega and Grogan (2008) started the RH measurements one day after clipping, while measurements in this study as well as in the study of Biasi et al. (2014) started about one year after treatment.Therefore, even though the partitioning approach for seasonal estimates of RH fluxes was 485 similar for all studies, any comparison must be made with caution.The few RH flux estimates reported from other arctic tundra sites were higher than the RH values from the Lena River Delta (0.5 ± 0.1 and 0.3 ± 0.02 g C m -2 d -1 at polygon rim and center, respectively).Higher RH fluxes throughout the growing season (0.8-1.8 g C m -2 d -1 ) were measured at a mesic birch and a dry heath site at Daring Lake in Canada (Nobrega and Grogan, 2008) and at a bare peat site (1.0 g C m -2 d -1 ) in the subarctic tundra at Seida, Russia (Biasi et al., 2014).Both sites contained 490 substantially higher amounts of SOC in the organic-rich layer than the soil at the polygon rim and were wellaerated compared to the soil at the polygon center, which both most likely caused a higher organic matter decomposition rate and could explain the higher RH fluxes than at the polygonal tundra microsites.Similar RH fluxes to those reported in our study were measured at a wet sedge site in Daring Lake (0.4 g C m -2 d -1 ) (Nobrega and Grogan, 2008), where soil and environmental conditions like WT, ALD, soil temperature, vegetation and SOC 495 were similar to the Samoylov sites and at vegetated peat sites in Seida (0.4-0.6 g C m -2 d -1 ) (Biasi et al., 2014).Despite these differences, the averaged contributions of RH to Reco of 42% at the center and 60% at the rim are in good agreement with those observed at Seida (37 -64%) and Daring Lake (44 -64%).Similar contributions were determined from an arctic tussock tundra site, where RH makes up approximately 40% of growing season Reco (Segal and Sullivan, 2014) and from a moist acidic tussock tundra site (Hicks Pries et al., 2013).In contrast to 500 these results, Dorrepaal et al. (2009) determined a substantially higher contribution of RH to Reco with about 70% in a subarctic peatland.
The difference in the contribution of RH to Reco between the polygon rim and center at our study site can be related to differences in vascular plant coverage and moisture conditions between both microsites.Higher GPP at the center than at the rim caused also higher rates of RA and in turn lowered the contribution of RH to Reco.Additionally, 505 anoxic soil conditions due to standing water, like at the polygon center, were not favorable for decomposition of Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-311Manuscript under review for journal Biogeosciences Discussion started: 30 August 2018 c Author(s) 2018.CC BY 4.0 License.SOM.Furthermore, Nobrega and Grogan (2008) concluded that consistently moderate moisture conditions, as at the polygon rim, promotes fast decomposition of SOM and therefore ensures higher RH rates than at the center.
Interestingly, at the polygon center, we observed significant correlations of the WT with Reco and RA fluxes, but no correlation between RH fluxes and WT.The RA fluxes might be negatively affected by high WT due to 510 submersion of the moss layer and partwise vascular leaf area as submersion can lead to plant stress, reducing productivity and nutrient turnover (Gebauer et al., 1995).Low soil moisture contents can limit the growth and productivity of an ecosystem (Chen et al., 2015) as desiccation lowers the photosynthetic activity (Turetsky et al., 2012), and in turn lowers RA fluxes.However, if this were the case, we would expect a correlation between GPP and WT, which was not observed.Instead, it is likely that the respired CO2 is 'recycled' due to of slow diffusion 515 through the moss layer.Evidence for this process was already shown in polygonal ponds (Liebner et al., 2011).This finding is in contrast to a set of studies who explained correlations between Reco fluxes and WT fluctuations with the impact of oxygen availability on RH fluxes (Chimner and Cooper, 2003;Dorrepaal et al., 2009;Juszczak et al., 2013) or observed an impact of moisture conditions on RH fluxes across multiple peatland ecosystems (Estop-Aragonés et al., 2018), while another study has shown no effect between water table fluctuations and Reco 520 fluxes (Chivers et al., 2009).However, these findings show the importance of the soil water content for Reco fluxes and the need for partitioning approaches to understand the response of the underlying processes of Reco fluxes on changing hydrologic conditions.
In order to determine the individual impacts of hydrological conditions and temperature on the RH and RA fluxes, it would be useful to perform both warming and wetting experiments in situ.So far, a number of studies have 525 determined the temperature response of NEE, GPP, and Reco fluxes in arctic ecosystems with warming experiments (e.g.Frey et al., 2008;Natali et al., 2011;Voigt et al., 2016), however, much less research has focused on the response of RA and RH fluxes to increased temperatures (Hicks Pries et al., 2015).Wetting experiments in arctic tundra ecosystems to determine the individual response of RA and RH fluxes to changing hydrological conditions are lacking so far, despite their importance as highlighted in this study, which is critical as climate warming will 530 likely lead to severe changes of the hydrological regimes in Siberian tundra regions (Merbold et al., 2009;Zimov et al., 2006b).

Conclusion
The contributions of GPP, Reco, RH and RA to CO2 NEE fluxes in a drained (rim) and water-saturated (center) microsite in the arctic polygonal tundra of northeast Siberia have been quantified in this study.Both investigated 535 microsites acted as CO2 sinks during the measurement period mid-July to end of September 2015.The polygon center acted as a considerably stronger CO2 sink than the polygon rim.The main drivers behind these differences in CO2 fluxes at the microsites were the higher GPP at the polygon center low RH and RA fluxes.The substantial differences identified in NEE between the two investigated microsites highlight the importance of microscale measurements for reliable estimates of CO2 surface-atmosphere fluxes from arctic tundra sites and the important 540 To test if RH fluxes are related to artefacts from the root-trenching approach, four additional PVC collars (two per microsite) were installed in 2015 following the sampling and preparation protocol of 2014.A total of 302 RH flux Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-311Manuscript under review for journal Biogeosciences Discussion started: 30 August 2018 c Author(s) 2018.CC BY 4.0 License.measurements were made on these newly installed plots.The differences of the mean RH fluxes of each single plot 195 that was trenched in 2014 and 2015 were analyzed using a student's t-test.RA fluxes at the unaltered sites were calculated by subtracting the mean of the measured RH fluxes from the four replicate plots at the trenched sites from the mean of the measured Reco fluxes at the unaltered sites from the same day.The calculated RA fluxes summed with GPP to estimate the net primary productivity (NPP) fluxes.
Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-311Manuscript under review for journal Biogeosciences Discussion started: 30 August 2018 c Author(s) 2018.CC BY 4.0 License.The NEE fluxes were calculated as the sum of the modelled GPP and Reco fluxes.The RA fluxes were calculated as the difference of the modelled Reco and RH fluxes.Furthermore, NPP was calculated from the sum of RA and GPP fluxes.275 4 Results Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-311Manuscript under review for journal Biogeosciences Discussion started: 30 August 2018 c Author(s) 2018.CC BY 4.0 License.
The modelled GPP and Reco fluxes were used to calculate NEE fluxes.The highest net CO2 uptake at the rim was encountered on 23 July and on 16 August at the polygon center, while the highest net CO2 release was measured 390 Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-311Manuscript under review for journal Biogeosciences Discussion started: 30 August 2018 c Author(s) 2018.CC BY 4.0 License.at both microsites on 17 August.The diurnal amplitude of NEE oscillation was greatest between the end of July and mid-August.The modelled Reco and RH fluxes were used to calculate RA fluxes.Therefore, the lowest and highest RA fluxes were similar to the Reco and RH fluxes.The modelled RA fluxes correlated substantially with WT fluctuations with lowest RA fluxes during times of lowest WT.The NPP fluxes were calculated from the sum of GPP and RA fluxes.At both microsites the highest NPP was encountered in mid-August, while the lowest daily 395 net primary productivity was in mid-September.At the polygon center the NPP showed a distinct seasonal trend, while at the polygon rim this trend less clear.
comparison of partitioned CO2 fluxes from different arctic tundra sites highlighted the importance of individual GPP and Reco fluxes to explain NEE fluxes.The comparatively low Reco fluxes and moderate GPP fluxes reported 425 from this study led to relatively high NEE fluxes at the polygon center, compared to other tundra sites.Furthermore, the highest net CO2 uptake fluxes were reported from wet and sedge-dominated sites.The GPP fluxes Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-311Manuscript under review for journal Biogeosciences Discussion started: 30 August 2018 c Author(s) 2018.CC BY 4.0 License.
role of soil moisture conditions on CO2 fluxes.Hereby, it was shown that RA and RH fluxes respond differently depending on hydrological conditions, with low RA fluxes during times of high water tables.Therefore, it is recommended that future studies determining partitioned CO2 fluxes from arctic tundra ecosystems should focus on the role of hydrological conditions as a driver of these fluxes to obtain a more in-depth insight into this relationship.545 Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-311Manuscript under review for journal Biogeosciences Discussion started: 30 August 2018 c Author(s) 2018.CC BY 4.0 License.

Figure 2 -
Figure 2 -Meteorological conditions from mid of July to end September.Panel (a) soil temperatures measured at 2 cm 825

Figure 3 -
Figure 3 -Chamber measured NEE, Reco and RH fluxes as well as calculated GPP, NPP and RA fluxes.The error bars denote the standard deviation of the four replicate measurements at each microsite.Panel (a) fluxes of NEE (n = 83), Reco (n = 85) and RH (n = 85) at the polygon center; (b) calculated fluxes of GPP (n = 83), NPP (n = 83) and RA (n = 85) at the polygon center; panel (c) measured fluxes of NEE (n = 83), Reco (n = 85) and RH (n = 85) at the polygon rim; (d) 835

Figure 4 -
Figure 4 -Relationships between water table fluctuations and (a) Reco fluxes, (b) RH fluxes, (c) RA fluxes and (d) GPP fluxes during the period July-August at the polygon center.Negative values on the x-axis indicate a water table below840

Figure 5 -
Figure 5 -Fitting parameters of the fitted CO2 flux models.The values are given with the standard deviation of the model results from the single measurement plots (light grey error bars) and the confidence intervals (95%) of the fitting parameters (dark grey error bars).845

Figure 6 -
Figure 6 -Modelled and measured CO2 fluxes at the polygon center.Measured fluxes are available for NEE, Reco and RH.NEE model fluxes were calculated from modelled GPP minus modelled Reco, RA model fluxes from modelled Reco minus modelled RH and NPP model fluxes from modelled GPP minus modelled RA.Note the different scales of the yaxes.850

Figure 7 -
Figure 7 -Modelled and measured CO2 fluxes at the polygon rim.Measured fluxes are available for NEE, Reco and RH.NEE model fluxes were calculated from modelled GPP minus modelled Reco, RA model fluxes from modelled Reco minus modelled RH and NPP model fluxes from modelled GPP minus modelled RA.Note the different scales of the y-axes.

Figure 8 -
Figure 8 -Integrated CO2 fluxes at the polygon rim and center.The values were calculated from the model results and are given in g C m -2 .In total, both microsites acted as a net CO2 sink during the growing season.NEE= net ecosystem exchange; GPP= gross primary productivity; Reco= ecosystem respiration; RH= heterotrophic respiration; RA= autotrophic respiration and NPP= net primary productivity.