Bulk partitioning the growing season net ecosystem exchange of CO 2 in Siberian tundra reveals the seasonality of its carbon sequestration strength

This paper evaluates the relative contribution of light and temperature on net ecosystem CO 2 uptake during the 2006 growing season in a polygonal tundra ecosystem in the Lena River Delta in Northern Siberia (72 2 ′ N, 12630 E). The occurrence and frequency of warm periods may be an important determinant of the magnitude of the ecosystem’s carbon sink function, as they drive temperatureinduced changes in respiration. Hot spells during the early portion of the growing season, when the photosynthetic apparatus of vascular plants is not fully developed, are shown to be more influential in creating positive mid-day surface-toatmosphere net ecosystem CO 2 exchange fluxes than those occurring later in the season. In this work we also develop and present a multi-step bulk flux partition model to better account for tundra plant physiology and the specific light conditions of the arctic region. These conditions preclude the successful use of traditional partition methods that derive a respiration–temperature relationship from all nighttime data or from other bulk approaches that are insensitive to temperature or light stress. Nighttime growing season measurements are rare during the arctic summer, however, so the new method allows for temporal variation in the parameters describing both ecosystem respiration and gross uptake by fitting both processes at the same time. Much of the apparent temperature sensitivity of respiration seen in the traditional partition method is revealed in the new method to reflect seasonal changes in basal respiration rates. Understanding and quantifying the flux partition is an essential precursor to describing links between assimilation and respiration at different timescales, as it allows a more confident evaluation of measured net exchange over a broad r range of environmental conditions. The growing season CO 2 sink estimated by this study is similar to those reported previously for this site, and is substantial enough to withstand the long, low-level respiratory CO2 release during the rest of the year to maintain the site’s CO2 sink function on an annual basis.

of the ecosystem's carbon sink function, as they drive temperature-induced changes in respiration. Hot spells during the early portion of the growing season are shown to be more influential in creating mid-day surface-to-atmosphere net ecosystem CO 2 exchange fluxes than those occurring later in the season. In this work we also develop and present a bulk flux partition model to better account for tundra plant physiology 10 and the specific light conditions of the arctic region that preclude the successful use of traditional partition methods that derive a respiration-temperature relationship from all night-time data. Night-time, growing season measurements are rare during the arctic summer, however, so the new method allows for temporal variation in the parameters describing both ecosystem respiration and gross uptake by fitting both processes at 15 the same time. Much of the apparent temperature sensitivity of respiration seen in the traditional partition method is revealed in the new method to reflect seasonal changes in basal respiration rates. Understanding and quantifying the flux partition is an essential precursor to describing links between assimilation and respiration at different time scales, as it allows a more confident evaluation of measured net exchange over

Introduction
Amplified arctic warming (Serreze and Barry, 2011;Serreze and Francis, 2006) has created a widespread interest in the CO 2 exchange fluxes of tundra ecosystems (McGuire et al., 2009), and there exist a number of unresolved questions regarding the seasonality of key controls on the processes governing this land-atmosphere flux. 5 While all ecosystems respond to the timing and magnitude of hot periods, the short Arctic growing season is particularly sensitive to synoptic weather conditions, including the number and extent of warm weather events characterized by warm, dry winds from the continental south (Johannessen et al., 2004). In this study, we examine how these few, brief warm periods (with air temperatures exceeding 20 • C and approaching peaks 10 near 30 • C for three to five days) have the potential to alter the CO 2 balance of tundra ecosystems.
The net ecosystem exchange (NEE) of CO 2 between the land surface and atmosphere is commonly partitioned between gross primary productivity (P gross ) and ecosystem respiration (R eco ) as a means to boost understanding of the underlying environ- 15 mental processes driving this flux term (Reichstein et al., 2005) as well as to fill measurement gaps in flux time series (Falge et al., 2001). In the simplest of such methods, the R eco term is often modelled as a function of temperature from night-time conditions where P gross is assumed to be negligible (Falge et al., 2002). This method has been widely applied at arctic sites, despite the seasonal near-absence of truly dark (Mahecha et al., 2010;Yvon-Durocher et al., 2012), including descriptions of increases in decomposition with temperature for sedge litter (Thormann et al., 2004). Considerable research has also demonstrated the temperature dependence of the photosynthetic apparatus (e.g. Berry and Björkman, 1980;Medlyn et al., 2002). Particularly in some moss-sedge environments, high-temperature stress on photosynthetic perfor-10 mance has been quantified in studies at the closed chamber level (i.e. 60 cm × 60 cm squares) (Riutta et al., 2007) and in leaf-or shoot-level measurements (Williams and Flanagan, 1998). As such these effects are often incorporated into soil-vegetationatmosphere transfer models that upscale arctic leaf-scale fluxes to the fetch of an eddy covariance tower (Williams and Rastetter, 1999), though not always (Shaver et al., 15 2007). However, photosynthetic deactivation driven by higher temperature is often not considered important on seasonal timescales due to tundra mosses' relatively high optimal temperatures and ability to adapt quickly (Furness and Grime, 1982;Oechel, 1976;Riutta et al., 2007;Sveinbjörnsson and Oechel, 1983;Zona et al., 2011). Similarly, the tussock tundra species Eriophorum vaginatum has been found to have only a minimal physiological response to temperature (Tissue and Oechel, 1987).
Photosynthesis is a light-sensitive process governed at first order on the ecosystem level by incoming light levels (Haxeltine and Prentice, 1996). Plants can also adapt over time scales of days to changes in photoperiod and other light conditions (Bauerle et al., 2012). In addition to this adaptation, certain Sphagnum moss species face photo-25 inhibition or light stress, in part as a response to low tissue nitrogen levels, at photosynthetically active radiation (PAR) levels less than 800 µmol m −2 s −1 (Murray et al., 1993). Light stress in tundra moss species has also been shown to be greater early in the season, with subsurface morphological adaptations to this stress sustaining more 13716 Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | late-season photosynthesis and delayed senescence (Zona et al., 2011). Similar morphological adaptations have been seen in open-bog boreal Sphagnum species (Hájek et al., 2009). A partition method should then be receptive to the possibility of light and temperature sensitivities that change through the season. These stresses (or process amplifiers) may be apparent through an examination of the day-time residuals of a light-5 response model.
The specific objectives of this study are to: 1. Test the proposed bulk flux-partition method against more traditional procedures for a low arctic tundra ecosystem; 2. Determine the growing season NEE flux and temporal changes in the relative 10 strength of its two partitioned components; 3. Quantify the short-term net CO 2 flux during hot spells in a tundra ecosystem; 4. Examine the effect of the timing of hot periods in the ecosystem's growing season carbon balance.
2 Site description 15 The study site is located on Samoylov Island (5 km 2 ), 120 km south of the Arctic Ocean in the Southern Central Lena River Delta (72 • 22 N, 126 • 30 E), and is considered representative of the region's modern delta areas that include a Late-Holocene river terrace and different active floodplain levels, and cover about 65 % of the total delta area. Over the past fourteen years a variety of investigations has been performed at the site, Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | the terrace features wet polygonal tundra, whose development has created regular microrelief with typical elevation differences of around 0.5 m between depressed polygon centres and elevated polygon rims (Kutzbach, 2006). These landscape units contain a large pool of accumulated organic matter -greater than 25 kg m −2 soil organic carbon in the top 1 m (Zubrzycki et al., 2012) -facing slow decomposition rates due to low 5 annual temperatures and chemical recalcitrance (Höfle et al., 2012). The site has a true arctic continental climate with very low temperatures and low precipitation. Mean annual conditions at the site's meteorological station have been determined from 1999 to 2005, and include mean air temperature of −14.7 • C and mean summer precipitation of 137 mm, ranging from 72 mm to 208 mm in this period 10 (Boike et al., 2008). Frequent cyclonic activity in the area causes rapidly changing weather conditions throughout the growing season by advection of cold and moist air from the Arctic Ocean or warm and dry air from continental Siberia, respectively. Polar day lasts from 7 May to 8 August, and polar night lasts from 15 November to 28 January. Typically, snowmelt and river break up start in the first half of June, and the growing 15 season lasts from mid-June through mid-September. The delta's continuous permafrost reaches depths of 500 to 600 m (Grigoriev, 1960) and is characterized by relatively low temperatures with the top-of-permafrost (1.7 m) temperature on Samoylov being approximately −7.8 • C from 2006-2011 .
The wet polygon centres and their edges are dominated by hydrophytic sedges such 20 as Carex aquatilis, Carex chordorrhiza, and Carex rariflora as well as mosses (e.g. Drepanocladus revolvens, Meesia triquetra, and Aulacomnium turgidum) Kutzbach et al., 2004). Mesophytic dwarf shrubs such as Dryas octopetala and Salix glauca, forbs (Astragalus frigidus), and mosses (Hylocomium splendens, Timmia austriaca) dominate the polygon rims. Surface classification of high-resolution aerial 25 photographs taken in the eddy footprint region of the island in 2003 shows that elevated and dryer polygon rims cover approximately 60 % of the area surrounding the study site, while depressed and wet polygon centres and troughs cover 40 % of the area (G. Grosse, personal communication, 2005).
struments Ltd, UK) measured wind velocity components in three dimensions and sonic temperature at 20 Hz frequency at a height of 4 m. Sample air was drawn at a rate of 20 l min −1 from the air intake 15 cm below the anemometer measurement point, through a polyethylene/aluminium composite-wall tube of 5 m length and 6.375 mm inner diameter (Dekabon ® 1300) to the closed-path gas analyzer. All analog signals were syn-10 chronously digitized at 20 Hz and logged on a laptop PC running EdiSol software (J. Massheder, University of Edinburgh, UK). The system was powered by a diesel generator located 100 m southwest from the tower in the least frequent wind direction. Continuous operation was ensured by an uninterruptible power supply. This site is supported by an adjacent meteorological station that collected data on 15 relative humidity and air temperature (MP103A, ROTRONIC AG, Switzerland), air pressure (RPT410F, Druck Messtechnik GmbH, Germany), photosynthetically active radiation (PAR; QS2, Delta-T Devices Ltd., UK) and the incoming and reflected components of shortwave and long wave radiation, respectively (CNR 1, Kipp and Zonen, Netherlands). Surface temperature (T s ) was determined from the outgoing long-wave radiation 20 measurement with the Stefan-Boltzmann law and an assumed emissivity of 0.98. Precipitation and soil temperature data were recorded at a long-term monitoring station 700 m south of the eddy covariance tower (Boike et al., 2008). The site's fetch is relatively flat and homogeneous despite the microtopographic variation in the polygonal surface (Kutzbach et al., 2007;Sachs et al., 2008;Wille et al., 2008). 25 Our flux-data processing and correction routine is presented and summarized in Table 1, and includes data screens based on stationarity, instrument performance, and integral turbulence characteristics (Foken and Wichura, 1996)

Data analysis and flux partitioning
Several partition models are used to separate the net CO 2 flux (NEE) into gross 5 primary productivity (the atmosphere-surface flux, P gross ) and ecosystem respiration (the surface-atmosphere flux, R eco ). In general, the PAR-sensitive portion of measured NEE is assigned to P gross and at least part of the temperature-sensitive portion of NEE is assigned to R eco . The NEE fluxes are first partitioned in a "traditional model" (as in Kutzbach et al., 2007) by assuming fluxes during low-light periods 10 (PAR < 20 µmol m −2 s −1 ) are fully composed of ecosystem respiration, are an exponential function of surface temperature, and are not a function of PAR. These flux measurements are pooled for a single best-fit relationship between R eco and T s using the empirical Q 10 model (van't Hoff, 1898): (1) 15 where, as in Mahecha et al. (2010), T ref = 15 • C and γ = 10 • C are independent parameters, Q 10,1 is a best-fit parameter indicating sensitivity of ecosystem respiration to surface temperature, and R base,1 is a best-fit parameter indicating basal respiration at the reference temperature T ref . R eco,1 is the modelled respiration flux. This relationship is comparable to the exponential relationship The P gross portion of the flux is then estimated from the difference between measured NEE and modelled R eco,1 , and is modelled as a function of PAR using the rectangular hyperbola function: The fit parameters α x and P max,1 represent, respectively, the initial canopy quantum ef-5 ficiency (that is, the initial slope of the P gross -PAR curve at PAR = 0) and the maximum canopy photosynthetic potential, which is the hypothetical maximum of P gross at infinite PAR; P gross,1 is the modelled CO 2 uptake using this approach. Both α x and P max,1 are assumed to have positive values, necessitating the negative sign on the equation's right-hand side. This model contains the explicit assumption that the gross productivity 10 flux is not influenced by light stress or temperature effects. This model is parameterized for each 7-day interval during the measurement period. A different two-step "bulk model" proposed here is developed to allow its governing parameters to change over the measurement period and to enable a portion of the low-light flux to be assigned to P gross . This method fits both a parabolic light curve 15 and a temperature response to the NEE flux measured when PAR < 500 µmol m −2 s −1 , thus below the range where light stress on P gross are expected to occur. The model is a best fit of the parameters α y , P max,2 , R base,2 , and Q 10,2 , for 7-day intervals during the measurement period using the following function: (3) 20 The NEE time series is then partitioned in 7-day intervals using R base,2 and Q 10,2 to estimate R eco,2 and setting the residual flux to P gross,2 . This derived P gross flux term is then modelled by re-fitting a light-response curve across the whole range of incoming PAR using a new pair of positive-valued parameters, α z and P max,3 : Parameters in both sets of models are found via unconstrained nonlinear regression to minimize the mean-square-error of the residuals (i.e. the nlinfit function in Matlab Release R2011b, The Mathworks, Inc.). The parameter 95 % confidence intervals are 5 determined using the Jacobian matrix computed in the nonlinear fitting (i.e. using Matlab's nlparci function); these are used to generate an error propagation estimate on modelled fluxes. This error propagation method assumes a normally distributed random error of 20 % on measured NEE fluxes, and uses the model parameter uncertainty estimates in a first-order partial derivative of the respective model equations (i.e. Eqs. 2-4) for gap-filled points in the time series. The final error estimate for the cumulative fluxes is then defined according to the following equation, relating individual flux errors to the cumulative value in quadrature: where u F cumul is the cumulative flux uncertainty, d int is the interval length (i.e. 30 min), n 15 is the number of intervals, and u 2 F −i is the estimated uncertainty for each (i th) 30-min flux (either measured or modelled).

Sensitivity to the timing of warm spells
The sensitivity of the ecosystem's mid-day sink-source function to warm spell timing is investigated using the two partition models to determine the threshold temperature 20 (T s,TH ) at which mid-day CO 2 flux turns positive (i.e. from surface to atmosphere). This modelling is performed using the daily maximum PAR value (PAR max ) measured for 13722 Introduction  fluxes are consistently negative between 28 June (DOY 179) and 2 August (DOY 214), and following a two-day net CO 2 release are again negative until 30 August (DOY 242). Cumulative NEE include a large sustained uptake from 12 July (DOY 193) to 2 August (DOY 214) (also shown in Fig. 6). The two-day net CO 2 release in early August coincides with a major storm event with more than 35 mm rainfall, daily maximum PAR 20 levels below 500 µmol m −2 s −1 , and wind speeds regularly exceeding 8 m s −1 .
The partition models generate different parameterizations for the CO 2 fluxes (Fig. 2). The exponential relationship of R eco to surface temperature used in the traditional partition model had parameters R base,1 (56.4 ± 3.1 µg m −2 s −1 ) and Q 10 with n = 448 measurement intervals contributing to this model. This model explained low-light NEE fluxes better than a trial model fit to air temperature rather than surface temperature (r 2 = 0.37; r.m.s.e. = 9.0 µg m −2 s −1 ). When compared to the traditional method (Fig. 2), the weekly fitted parameters of the new bulk partition method tends to 5 have a lower temperature response (Q 10,2 ; mean 1.52 ± 0.33) and more variation in the basal respiration rate (R base,2 ; mean 54.1 ± 13.3 µg m −2 s −1 ). Additionally, the weekly P max,1 estimates of the traditional method tend to be higher than those of the bulk method, while the light response parameter α tends to be higher in the bulk method.
Best-fit values of α z in the bulk method vary between 0.16 and 0.95 µg (CO 2 ) µmol The bulk model slightly reduces the r.m.s.e. relative to the traditional model (on average, by 5 % each week). The traditional partitioning model, relative to the bulk model, creates residuals that are more often correlated to temperature (9 vs. 2 weeks, respectively) (Fig. 2, lower left panel). In each of these cases, the models significantly 15 over-predict CO 2 fluxes at higher temperatures (i.e. the NEE magnitude is higher, and the model predicts greater respiratory or reduced uptake fluxes). The bulk partitioning model (i.e. Eq. 3) has been fit to week-long time slices of NEE, PAR and T s measurements. One example of this fit is for the growing season period containing the first of the hot spells, 9-16 July, which demonstrates how such a hot pe-20 riod may induce positive CO 2 fluxes during mid-day, high-PAR conditions theoretically optimal for photosynthetic CO 2 uptake (Fig. 3). Applying the new bulk partition method reveals relatively high basal respiration rates with moderate temperature dependence. This period precedes the ecosystem's maximum light response, and so the higher temperatures drive more respiratory fluxes than the light and temperature together drive 25 uptake. The data from this period, shown in Fig. 3, also highlight the strong response to increases in both light and temperature even in the lowest light region. This response is evident from the relatively high α 2 value (1.03 ± 0.35 µg µmol −1 ) reducing net fluxes at low light. The resultant partitioned fluxes from the bulk method are shown in Fig. 4. This partition highlights several environmental processes. First, there is a slow ramping up of P max,3 , which peaks at 240 ± 11 µg m −2 s −1 during the first week in August (this process is also evident in Fig. 2; the traditional partition estimates 334 ± 23 µg m −2 s −1 for P max,1 during this period) and remains rather high with a steady decline following the pattern 5 of vegetation senescence until freezing a month later. Second, there is little evidence of either light stress or a temperature response within the light curve (this finding is also demonstrated statistically through the residuals analysis in Fig. 2). Third, due to the synchronicity of surface temperature and light in the diurnal cycle, the respiratory fluxes tend to peak at the highest light levels. 10 The ecosystem's positive mid-day NEE in response to higher temperatures (as highlighted in Fig. 3) is explored in more detail by finding the threshold temperature at which the NEE turns positive for the PAR time series of this growing season (Fig. 5). A model is derived from each partition method, and shows when the ecosystem is susceptible to this mid-day efflux of CO 2 . The mid-season increase in photosynthetic capacity (deter-15 mined through the P max parameters) increases the threshold temperature so much as to make this response very unlikely. Later-season heat spells are less likely to generate a net positive CO 2 flux, whereas earlier heat spells are able to generate this mid-day positive NEE flux (i.e. occuring when the surface temperature exceeds the threshold). The bulk partition method generates higher temperature thresholds than the traditional 20 partition method, reflecting its assignment of some of the low-light NEE flux to uptake rather than to dampened respiratory activity. More generally, the relatively lower threshold periods in this figure (such as following 6 August or 20 August) represent a response to synoptic weather conditions, when cloudier (lower PAR) conditions lower photosynthetic uptake and so require less respiration to offset P gross . The mid-day net 25 CO 2 release events driven by early season high temperatures differ from cumulative, but not mid-day, CO 2 releases in the period 3-5 August, when low radiation, high precipitation, and high wind speeds dampen photosynthesis enough to create net positive fluxes. Introduction The net outcome of this work is an NEE flux time series that is gap-filled and partitioned using the two methods; these cumulative seasonal fluxes are shown in Fig. 6. Both methods suggest an unambiguous sink function, with partitioned fluxes changing relatively little (< 10 %) between methods at this time scale. Using the traditional partition method for gap-filling yields a greater sink function (a larger magnitude of NEE) by 5 partitioning less of the NEE flux to respiration than in the bulk partition case. All three fluxes show an inflection near day 190 (i.e. 9 July), after which the fluxes (in particular, the P gross flux) increase substantially until day 215 (i.e. 3 August), when the rate of increase slows. The ecosystem's carbon sink function is nearly finished at day 240 (i.e. 28 August), when the upward and downward flux terms are in balance. Partitioning after day 246 (i.e. 3 September) was discarded due to the sustained positive fluxes in this period and negative correlation with temperaure. This period behaves differently, ecologically, due to its freezing temperatures and onset of snowfall.

Discussion
There are challenges in partitioning R eco and P gross during the long polar day; in partic-15 ular, separating temperature effects between two flux portions may involve an extrapolation out of the conditions used for parameterization. Understanding the net effect of these environmental flux drivers is important in determining present-day carbon balances and cycling as well as making predictive models for these landscapes. In this paper, the method used to model NEE fluxes implicitly assumes that all of the tem-20 perature sensitivity of NEE in the low-PAR range is revealed through changes in the R eco parameters, and are therefore unrelated to changes in the P gross portion of the net flux. This decision appears justified by the general lack of correlation between the bulk model's residuals and temperature, even when the full PAR range is considered.
There are a number of benefits to the bulk partitioning method considered here. First, 25 the method allows one to discover the seasonality of the respiration parameters during a season where very few measurement points support the traditional parameterization Introduction scheme. In particular, the basal respiration parameter increases during midseason when using the newer method and the temperature sensitivity parameter (Q 10 ) decreases (from 1.9 to mean 1.5). This shift implies that the traditional method would over-estimate respiratory responses to temperature, whereas actually the base respiration has a stronger phenological pattern. This phenology may follow from seasonal growth in the overall microbial biomass, increases in the thaw depth that allow a greater zone of respiration to occur, and larger mid-summer contributions from plant respiration to R eco . Such a change in interpreting the R eco flux follows recent work demonstrating that temperature sensitivity is often over-estimated at the expense of ignoring seasonal changes in base respiration (Mahecha et al., 2010). 10 A second benefit to the bulk partition method is its use of a broader range of environmental conditions to parameterize the respiration flux, thus allowing the respiration model to extrapolate into a narrower range of temperatures. For example, the maximum measured surface temperature in this study was 36.5 • C, the maximum temperature during PAR conditions less than 500 µmol m −2 s −1 (the bulk method's threshold for 15 the initial combined R eco -P gross parameterization) was 26.2 • C, and the maximum temperature under PAR conditions less than 20 µmol m −2 s −1 (the traditional threshold for night-time conditions) was 17.6 • C. Thus a portion of the uncertainty in extending nighttime temperature relationships during the much warmer daytime has been removed in this model. Yet, the new method still may allow for the discovery of ecosystem-wide de-20 activation in response to higher light and/or higher temperature (both of which, in this case, seem minimal). Finally, the new method also avoids extrapolation of R eco under dark conditions to R eco under light conditions, when leaf respiration may be inhibited by up to 20 % (Brooks and Farquhar, 1985;Wohlfahrt et al., 2005). Limits remain for both partitioning model strategies. The methods are only parame- 25 terized under existing environmental conditions, so they are unable to suggest how the ecosystem may acclimate to changes in the timing, magnitude, or duration of hot periods. Additionally, the drying of the land surface (i.e. reduced soil moisture or changes in water table) or of the air (in terms of humidity) are not considered. However, the Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | relatively few low humidity periods prevent a proper parameterization of a model representing these low-humidity effects. Moreover, their scarcity may prevent the need (or ability) for such a model in this landscape, though there may be some risk that not including these effects artificially inflates the temperature effect on respiration rather than into the photosynthetic portion of NEE (Lasslop et al., 2010). An additional challenge is that neither method accounts for variation in the contributions to P gross -e.g. mosses and vascular plants are governed by different light-response mechanisms and parameters -nor to R eco -e.g. differences across the micro-topographic zones (dryer polygon rims, inundated centres). Encouraging progress has been made elsewhere on parts of this question (Belshe et al., 2012) though a full footprint contribution model would be  et al., 2007). In that work, the authors estimated a moderate CO 2 source for the rest 15 of the year (+48 g m −2 ), implying that in both cases, the site is a CO 2 sink even on an annual basis. In the season presented here, previous work has demonstrated an ecosystem CH 4 source function of 1.93 g CH 4 m −2 using both eddy covariance and closed chamber methods (Sachs et al., 2008(Sachs et al., , 2010. Thus the vertical CO 2 flux sink strength is substantially greater than CH 4 emissions. It is also stronger than the CO 2 20 source-strength of respiration, which includes outgassing from flooded area and ponds (Abnizova et al., 2012) that cover up to 28 % of the eddy footprint, and which seem to be adequately parameterized within the respiration model used here. Lateral releases of dissolved organic and inorganic carbon are also expected to be much less than the net CO 2 sink, due to the site's flat landscape, pronounced microrelief, ponding and 25 low thaw depth. These factors together prevent lateral fluxes from the polygon centres through most of the growing season, keeping fluxes low (Helbig et al., 2012).
In addition to the season-wide results, this study has also demonstrated the importance of early-season hot periods in generating net CO 2 emissions despite seemingly Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | optimal growing conditions. In the future, a sensitivity analysis of the timing, magnitude, and duration of these events could be performed to show the importance of synoptic meteorological conditions on temporal changes to the local carbon budget. This question is especially urgent given likely consequences of changes in Arctic sea ice coverage and the resultant modifications of the balance between continentally-and Arctic

5
Ocean-derived weather systems (Deser et al., 2000). Such modelling can be coupled with laboratory, field, and greenhouse studies regarding plant and bacterial adaptations to heat, light, and other stresses.

Conclusions
This study provides a more appropriate method to partition ecosystem CO 2 fluxes be-10 tween their upward and downward components by accounting for seasonal changes in respiration and for the effect of even low levels of light in driving photosynthetic uptake. This new bulk method allows a discovery of the effect of early-season higher temperature periods in driving higher respiration (rather than reduced or de-activated uptake), with the net effect of a mid-day CO 2 source despite high light conditions. These events 15 seem more likely earlier in the growing season when plants have not yet fully matured enough to take advantage of the warm, sunny conditions. Such "hot moments" of ecosystem CO 2 emissions may change in frequency depending on changes in the region's synoptic weather conditions.
-Angle of attack correction for the response of the sonic anemometer (Nakai et al., 2006) -Double coordinate rotation in order to (i) rotate u into the mean horizontal wind and (ii) reduce mean w to zero -Determine cross-correlation sequence of 30-min interval to find the lag time which maximizes the covariance of scalar transport (i.e. w s , the covariance of the fluctuations of the molar concentration of scalar s and the vertical wind fluctuations w ) and de-lag the time series of s -Linearly de-trend the scalar time series and calculation of flux estimate -Webb-Pearman-Leuning terms (WPL) are applied to the CO 2 signal (Leuning, 2007;Webb et al., 1980) with a latent heat flux (LE) determined using the lag time to maximum covariance of CO 2 , rather than H 2 O (Ibrom et al., 2007), i.e. LE WPL -Apply frequency response corrections (Moore, 1986) for each sensor: -Sonic path-length -Sensor separation between the sonic and gas sampling -Tube attenuation -Signal high-pass filtering (linear de-trend) -Temporal averaging due to the sample lifetime in the cell -Spatial averaging at the sample intake -The CO 2 signal is additionally corrected with a first-order low-pass filter with time constant 0.3183 s for the period prior to 20 June 2006, when the LI-7000 low pass filter was set to 1 s.
-Correction of sensible heat flux (Schotanus et al., 1983) and calculation of Obhukov stability parameter -Evaluate integral turbulence characteristics and stationarity (Foken and Wichura, 1996) in 30-min period for use in filtering out inadequate flux measurements -Removal of data in a 30 • mean wind-direction window (238-268 • ) due to contaminating influence by the diesel generator used to power eddy-covariance equipment (Sachs et al., 2008) Printer-friendly Version between the NEE data and models and significance of the correlation ρ (in terms of its p 5 value) between NEE model residuals and surface temperature (T s ). The sign of this correlation 6 (when significant, with p < 0.05) is given above each time interval. For example, residuals of 7 the traditional model (in grey) are negatively correlated to temperature between 11 Jun and 9 8 July 2006; in these time periods the model would over-predict CO 2 fluxes at higher 9 temperatures. Time series of NEE parameters (right-hand panel) from the new, weekly-fit 10 bulk method (in black) and the traditional method (in grey) through the measurement period 11 are presented with parameter confidence intervals shown in dotted lines. 12

Fig. 2.
A time series of mean daily surface temperature is shown for comparative purposes. Time series of fitting statistics (left-hand panel) include root mean square error (rmse) between the NEE data and models and significance of the correlation ρ (in terms of its p-value) between NEE model residuals and surface temperature (T s ). The sign of this correlation (when significant, with p < 0.05) is given above each time interval. For example, residuals of the traditional model (in grey) are negatively correlated to temperature between 11 June and 9 July 2006; in these time periods the model would over-predict CO 2 fluxes at higher temperatures. Time series of NEE parameters (right-hand panel) from the new, weekly-fit bulk method (in black) and the traditional method (in grey) through the measurement period are presented with parameter confidence intervals shown in dotted lines.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 1 2 Figure 3. NEE flux measurements during period 9 July -16 July, 2006 (i.e., days 190-197), 3 plotted according to PAR and coloured according to surface temperature. The right-hand 4 panel provides an example of the bulk flux partition model (i.e., equation 3) in the low-PAR 5 range during this period with five of the fitted isotherms provided for comparison. Note the 6 change in the plotting range of the x-axis between the two images. 7 Fig. 3. NEE flux measurements during period 9 July- 16 July 2006 (i.e. days 190-197), plotted according to PAR and coloured according to surface temperature. The right-hand panel provides an example of the bulk flux partition model (i.e. Eq. 3) in the low-PAR range during this period with five of the fitted isotherms provided for comparison. Note the change in the plotting range of the x-axis between the two images.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 1 2 Figure 4. Time series of scatter plots showing partitioned NEE flux against PAR, with points 3 colored according to surface temperature (T s ) and generated according to the bulk method 4 described in the text (i.e., equations 3-4). The P gross,2 model's light-response parameters (P max,3 5 α ,3 ) are fit to each time slice by minimizing the root-mean-square of the residuals (rmse). 6 P gross,2 and R eco,2 have units of µg m -2 s -1 . Each subplot represents a different one-week time 7 period. 8 Fig. 4. Time series of scatter plots showing partitioned NEE flux against PAR, with points colored according to surface temperature (T s ) and generated according to the bulk method described in the text (i.e. Eqs. 3-4). The P gross,2 model's light-response parameters (P max , 3 α 3 ) are fit to each time slice by minimizing the root-mean-square of the residuals (rmse). P gross,2 and R eco,2 have units of µg m −2 s −1 . Each subplot represents a different one-week time period.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 1 2 Figure 5. Ecosystem sensitivity to surface temperature, with respect to positive mid-day CO 2 3 fluxes generated using measured PAR time series for the 2006 growing season with model 4 parameters generated as described in the text (equation 6). Surface temperatures higher than 5 the threshold would enable net positive mid-day CO 2 fluxes. 6 Fig. 5. Ecosystem sensitivity to surface temperature, with respect to positive mid-day CO 2 fluxes generated using measured PAR time series for the 2006 growing season with model parameters generated as described in the text (Eq. 6). Surface temperatures higher than the threshold would enable net positive mid-day CO 2 fluxes. Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 1 2 Figure 6. Cumulative, gap-filled CO 2 fluxes for an 84 day period (11 June -3 September, 3 2006) using different models. Note the sign-switching for P gross and NEE to ease comparison 4 with R eco . Gap-filling was necessary for 21% of the 4032 measurement intervals considered in 5 this period, and creates the differences between the two methods' cumulative NEE fluxes. 6 Error propagation assumes 20% randomly distributed error on the flux measurements and half 7 the 95% confidence intervals on the flux partition parameters; these are combined in 8 quadrature (equation 5). 9 10 Fig. 6. Cumulative, gap-filled CO 2 fluxes for an 84 day period (11 June-3 September 2006) using different models. Note the sign-switching for P gross and NEE to ease comparison with R eco . Gap-filling was necessary for 21 % of the 4032 measurement intervals considered in this period, and creates the differences between the two methods' cumulative NEE fluxes. Error propagation assumes 20 % randomly distributed error on the flux measurements and half the 95 % confidence intervals on the flux partition parameters; these are combined in quadrature (Eq. 5).