Controls Over Tundra Non-Growing Season Carbon Loss


Material Information

Controls Over Tundra Non-Growing Season Carbon Loss
Physical Description:
1 online resource (63 p.)
Webb, Elizabeth E
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Master's ( M.S.)
Degree Grantor:
University of Florida
Degree Disciplines:
Botany, Biology
Committee Chair:
Committee Co-Chair:
Committee Members:


Subjects / Keywords:
carbon -- permafrost -- winter
Biology -- Dissertations, Academic -- UF
Botany thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation


Permafrost soils store roughly 1672 Pg of carbon (C), twice the amount currently in the atmosphere. But as high latitudes warm, this temperature-protected C reservoir will become vulnerable to higher rates of decomposition. In recent decades, air temperatures in the high latitudes have warmed more than other region globally, particularly during the winter. Over the coming century, the arctic winter is also expected to experience the most warming of any region or season, yet it is notably understudied. Though warming has been shown to increase plant productivity during the growing season, these seasonal C gains may be offset on an annual basis by C losses during the non-growing season (NGS). Here we present NGS CO2 flux data from the Carbon in Permafrost Experimental Heating Research (CiPEHR) project, a tundra ecosystem warming experiment in interior Alaska. Our goals were to compare methods of measuring winter CO2 flux, determine the environmental controls of winter CO2 flux, account for subnivean photosynthesis and late fall plant C uptake in our estimate of NGS CO2 exchange, and quantify the effect of warming on total NGS CO2 flux
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility:
by Elizabeth E Webb.
Thesis (M.S.)--University of Florida, 2014.

Record Information

Source Institution:
Rights Management:
Applicable rights reserved.
lcc - LD1780 2014
System ID:

This item is only available as the following downloads:

Full Text




2014 Elizabeth E. Webb


3 ACKNOWLEDGMENTS I thank Peter Ganzlin, John Krapek Tom Lane, John Wood, and the researchers and technicians of the Bonanza Creek LTER for their assistance with fieldwork and logistics; the members of the UF Ecosystem Dynamics Research Lab for their help vetting ideas and with the writing process; Kiva Ok en for statistical support; David Risk for the allowing me to use the Flux Lab flux generator and, along with Nick Nickerson, aid in on plot chamber flux calculations; Rosvel Bracho for processing the eddy covariance data; the members of my M.S. committee: Sue Natali, and Tim Martin for their support and constructive criticism and my major advisor, Ted Schuur for his guidance and feedback through every step of this endeavor. Funding for this project was provided by the National Science Foundation CAREER p rogram, the Bonanza Creek LTER program, Denali National Park and Preserve Vital Signs Monitoring Program, and the Department of Energy Terrestrial Ecosystem Processes Program.


4 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 3 LIST OF TABLES ................................ ................................ ................................ ............ 6 LIST OF FIGURES ................................ ................................ ................................ .......... 7 ABSTRACT ................................ ................................ ................................ ..................... 9 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 10 2 MATERIALS AND METHODS ................................ ................................ ................ 15 Site Description ................................ ................................ ................................ ....... 15 Experimental Design ................................ ................................ ............................... 15 Ecosystem C Fluxes ................................ ................................ ............................... 16 Field Measurements ................................ ................................ ............................... 16 On plot Chambe r Measurements ................................ ................................ ..... 16 Snow Pit Measurements ................................ ................................ ................... 17 Eddy Covariance ................................ ................................ .............................. 18 Soda Lime ................................ ................................ ................................ ........ 18 Fall Chamber Flux ................................ ................................ ............................ 19 Radiocarbon ................................ ................................ ................................ ..... 19 Field Environmental Measurements ................................ ................................ ....... 20 Statistical Analysis ................................ ................................ ................................ .. 21 Seasonal Definitions ................................ ................................ ......................... 21 Analysis of the Environmental Drivers of Winter CO 2 Flux ............................... 22 Modeling Wintertime R eco ................................ ................................ ................. 23 Modeling Fall NEE ................................ ................................ ............................ 24 Modeling Spring NEE ................................ ................................ ....................... 24 Total NGS C Balance ................................ ................................ ....................... 25 3 RESULTS ................................ ................................ ................................ ............... 28 Method Comparison ................................ ................................ ............................... 28 Drivers of Winter CO 2 Flux ................................ ................................ ...................... 29 Season al Variation ................................ ................................ ................................ .. 31 Non Growing Season C Balance ................................ ................................ ............ 31 4 DISCUSSION ................................ ................................ ................................ ......... 41 Comparison of Measurement Methods ................................ ................................ ... 41 Drivers of Winter CO 2 Flux ................................ ................................ ...................... 42


5 Non Growing Season C Balance ................................ ................................ ............ 4 8 APPENDIX : SUPPLEMENTARY MATERIAL ................................ ................................ 51 On plot Chamber Measurement Flux Calculation ................................ ................... 51 Winter CO 2 Flux Models ................................ ................................ ......................... 52 LIST OF REFERENCES ................................ ................................ ............................... 55 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 63


6 LIST OF TABLES Table page 2 1 Average soil temperature, snow depth, air temperature, and atmospheric pressure measured at CiPEHR during the October April of 2008 2013 ............. 26 2 2 Mean estimate and standard error of parameters for the control and warming fall NEE and R eco 2 s 1 ) models. ................................ ............................ 26 3 1 Mean estimate and standard error of parameters for the on plot, snow pit, and eddy covariance biological winter CO 2 2 s 1 ) models. .............. 34 3 2 Total non growing season CO 2 loss estimates (g CO 2 C m 2 ) for the five years of soil temperature and PAR data available at CiPEHR. .......................... 34 3 3 Mean estimate and standard error of parameters from the full on plot, snow pit, and EC winter CO 2 2 m 2 s 1 ) mo dels. ................................ ..... 35 3 4 Winter season (period of no photosynthesis) length a nd start and end dates for the five years measured. ................................ ................................ ............... 36


7 LIST OF FIGURES Figure page 2 1 Location of CiPEHR, eddy covariance tower, and the Eight Mile Lake regi on within t he context of Alaska ................................ ................................ ................ 27 2 2 Photographs of the four methods of measuring winter CO 2 flux ......................... 27 3 1 Winter 2012 2013 CO 2 loss measured using the soda lime technique and modeled using the snow pit, on plot, and eddy covariance biological models.. .. 36 3 2 Winter CO 2 flux as measured by the three met hods used to create flux models ................................ ................................ ................................ ................ 37 3 3 r representative years using a five day moving window for EC fluxes ................................ ................................ ......................... 38 3 4 Winter (snow pit and on plot) and summer R eco fluxes (control and wa rming, May September 2008) p lotted against soil temperature ................................ ..... 38 3 5 eco measured at control and warmed plots at CiPEHR during Ap ril and August of 2012 and 2 013 ................................ ........... 39 3 6 Total NGS CO 2 loss for the control and warming treatment partitioned by season (fal l, winter, spring) for the five years of temperature a nd PAR data available at CiPEHR ................................ ................................ ........................... 39 3 7 Total NGS CO 2 loss as a function of October April average soil temperature .... 40 3 8 Perc ent by which assuming only R eco occurs during the NGS over estimates NGS CO 2 loss. ................................ ................................ ................................ .... 40 A 1 Representative example of the CO 2 concentration vs. time curve for the on plot chamber measurements ................................ ................................ .............. 54 A 2 Comparison between the full and biological models used to predict wintertime CO 2 loss ................................ ................................ ............................ 54


8 LIST OF ABBREVIATIONS C Carbon CiPEHR Carbon in permafrost experimental heating research CO 2 Carbon dioxide EC Eddy covariance DOS Day of season GPP Gross primary productivity NEE Net ecosystem exchange NGS Non growing season PAR Photosynthetically active radiation R eco Ecosystem respiration


9 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science CONTROLS OVER TUNDRA NON GROWING SEASON CARBON LOSS By Elizabeth E. Webb May 2014 Chair: Edward A.G. Schuur Major: Botany Permafrost soils store roughly 1672 Pg of carbon (C), twice the amount currently atmosphere. But as high latitudes warm, this temperature protected C reservoir will become vulnerable to higher rates of decomposition. In recent decades, air temperatures in the high latitudes have warmed more than other region globally, particularly du ring the winter. Over the coming century, the arctic winter is also expected to experience the most warming of any region or season, yet it is notably understudied. Though warming has been shown to increase plant productivity during the growing season, t hese seasonal C gains may be offset on an annual basis by C losses during the non growing season (NGS). Here we present NGS CO 2 flux data from the Carbon in Permafrost Experimental Heating Research (CiPEHR) project, a tundra ecosystem warming experiment i n interior Alaska. Our goals were to compare methods of measuring winter CO 2 flux, determine the environmental controls of winter CO 2 flux, account for subnivean photosynthesis and late fall plant C uptake in our estimate of NGS CO 2 exchange, and quantify the effect of warming on total NGS CO 2 flux.


10 CHAPTER 1 INTRODUCTION Cold and wet conditions in the high latitudes have limited microbial decomposition of soil carbon (C), and as a result these soils have been accumulating C over the past several millennia (Harden et al. 1992; Trumbore & Harden, 1997; Hobbie et al. 2000; Hicks Pries et al. 2011) Permafrost soil belowground organic C pool and store twice as much C as is currently in the atmosphere (Tarnocai et al. 2009) However, as a result of climate change, temperatures in high latitudes have increased and are project ed to rise more than any other region globally (Christensen et al. 2013a) As permafrost thaws, the magnitude and direction of future permafrost soil C storage are uncertain (Shaver et al. 1992; Oechel et al. 1993, 2000; Schuur et al. 2008, 2013; Koven et al. 2011; Sistla et al. 2013) It is well documented that warming stimulates plant growth as well as ecosystem respiration (R eco ) in tundra e cosystems (Rustad et al. 2001; Oberbauer et al. 2007; Ueyama et al. 2013) In some cases, experimental warming has c aused R eco to increase more strongly than plant growth during the growing season (Oberbauer et al. 2007; Biasi et al. 2008) thus increasing the release of C to the atmosphere. Other warming experiments and observational studies on the tundra show that increased C uptake by plants during the growing season outweighs the concurrent increase in R eco causing the tundra to act as a growing season C sink (Huemmrich et al. 2010; Natali et al. 2011; Trucco et al. 2012; Ueyama et al. 2013) Whether these seasonal C gains are offset during the non growing season (NGS), however, is dependent on the magnitude of winter R eco It is particularly important that we understand winter C


11 dynamics because future warming is predicted to be greatest during winter, increasing on average 4.8 C by 2100 as compared to 2.2 C during the summer months over the same time period (Christensen et al. 2013a) Neverthel ess, there are few field measurements conducted during this season and winter C fluxes remain a key unknown in estimating the annual C balance of the tundra (Jones et al. 1999; Euskirchen et al. 2012; Belshe et al. 2013) In a review of wintertime CO 2 fluxes from arctic tundra, Bjrkman et al. (2010) found estimates of cumulative winter C loss ranged from 0.19 to 210 g CO 2 C m 2 yr 1 This wide range is due in part to poo rly understood environmental drivers of winter R eco As such, the gap filling methods used to extrapolate over months of missing data are poorly developed. Soil temperature (Oechel et al. 1997; Elberling, 2007; Sullivan et al. 2008, Natali et al. in press) vegetation type (Jones et al. 1999; Grogan, 2012) and snow depth (Fahnestock et al. 1998; Nobrega & Grogan, 2007) h ave all been linked to field measurements of winter CO 2 flux, but studies often lack sufficient data to create gap filling models. Furthermore, it is largely thought that soil moisture also plays an important role in winter R eco on the tundra (Fahnestock et al. 1998; Vogel et al. 2009; Grogan, 2012) but its effect has only been documented in laboratory settings (Panikov et al. 2006; Tilston et al. 2010) The broad range of winter C loss estimates is also attributable to the discrepancies in how winter is defined; winter CO 2 flux measurements are interpolated over different time periods depending on the study. Like many others, Fahnestock et al. (1998) define winter by calendar days that generally c oincide with the snow covered season and assume that only respiration occurs during this time. While this is a


12 convenient and consistent way to partition the year, it does not allow for inter annual variation in snow cover nor does it take subnivean photo synthesis into account. Grogan and Jonasson (2006) improved the cale ndar definition by delineating winter as the continuous period during which mean diel soil temperature at 3 cm was less than 0.5C and Oechel et al. (in press) use a metric based on net radiation to define seasonality. However, photosynthesis does occur u nder the snow pack and while the soil is frozen (Tieszen, 1974; Kappen, 1993; Starr & Oberbauer, 2003) and assuming that only respiration occurs during these times may over estimate C release if plant uptake is significant. In addition to the relatively small amount of winter R eco data available, there are a variety of methods used to measure R eco, all with different assumptions. Different methods often produce di fferent estimates of CO 2 flux even at the same site. A popular method is to measure the concentration of CO 2 within the snowpack and calculate a flux (Fahnestock et al. 1 998; Jones et al. 1999; Sullivan et al. 2008; Bjrkman et al. 2010a) Due to the heterogeneity of arctic snow packs, however, ensuring that the vertical diffusion of CO 2 is accurately measured without CO 2 diversion to lateral flow or lost during sno w pack venting events can be difficult. Soda lime adsorption is also widely employed (Grogan & Chapin, 1999; Welker et al. 2000; Nobrega & Grogan, 2007; Sjgersten et al. 2008; Rogers et al. 2011) and is especial ly useful in remote locations where other methods are impractical for extended periods of time because of adverse travel and weather conditions. At sites with the requisite power requirements, eddy covariance (EC) towers are also used to gather data with limited presence of field personnel (e.g. Belshe et al. 2012; Euskirchen et al. 2012,


13 Oechel et al. 2014) Because EC towers can gather data continuously, including during conditions that are dangerous to people, the sheer amount of data generated is a powerful tool to answer high resolution questions about C flux. The EC method, however, cannot measure ove r small scales and thus cannot be used on small scale experimental plots such as in this study. Lastly, chamber measurements (e.g. Grogan & Jonasson, 2006; Elberling, 2007; Vogel et al. 2009; Morgner et al. 2010) are a good way to measure temporal variability (and associated environmental controls over R eco ) on a small scale, but are labor intensive, can only be performed in favorable weather conditions, and are sensitive to pressure changes that can affect flux estimates. Often chamber measurements are performed by removing snow and plac ing a chamber directly over the soil but this technique disturbs the local snow pack, affecting snow mediated soil insulation when sampling is repeated on the same plots. In contrast, chambers that remain under the snow for the duration of the winter are understudied but may affect the CO 2 diffusion gradient between the soil and ambient snow pack. Thus, due to the harshness of the arctic winter environment, evaluations of measurement accuracy and error are limited and there is no standardized method of me asuring winter R eco Because winter measurements and associated CO 2 flux models form a key component of the annual C balance calculations, accurately understanding the environmental controls of winter R eco defining winter, and recognizing the biases asso ciated with winter measurement methods are critical steps to ensure accurate analysis. To understand the role of winter processes under likely climate change scenarios, we measured CO 2 flux at the Carbon in Permafrost Experimental Research


14 (CiPEHR) projec t, a tundra warming experiment in interior Alaska. Our goals were four fold: (1) to identify the drivers of winter R eco (2) to define winter such that photosynthetic activity was accurately incorporated, (3) to compare methods of measuring winter flux, an d (4) to quantify the effect of warming on NGS CO 2 loss.


15 CHAPTER 2 MATERIALS AND METHODS Site Description The CiPEHR field site is located in the northern foothills of the Alaska Range the Eight Mile Lake (EML) watershed, Alaska (Figure 2 1; Schuur et al. 2009; Natali et al. 2011; Trucco et al. 2012) The vegetation at the site is moist acidic tundra at an elevation of 700m on a relatively wel l drained gentle northeast facing slope (Natali et al. 2012) Mean annual ai r temperature is 2.8C (2008 2013) and 11.3C for the October April period (Table 2 1). The EML watershed is within the zone of discontinuous permafrost, but the entire field site is underlain by permafrost; the average active layer is 57.9 1.9 cm for control plots and 65.8 3.4 cm for warmed plots (standard error, n=5 years: 2009 2013). Permafrost temperatures have been monitored in the EML region since 1985 and document regional permafrost thaw over the past three decades (Osterkamp et al. 2009) Permafrost degradation has also been documented by more extensive measurements of CO 2 fluxes, active layer depth, aboveground biomass, and radiocarbon at a nearby natural permafrost thaw gradient since 2004 (Schuur et al. 2007, 2009; Vogel et al. 2009; Lee et al. 2011; Belshe et al. 2012; Trucco et al. 2012; Hicks Pries et al. 2013) Experimental Design Soil warming at CiPEHR is achieved using snow fences (n=6) distributed throughout the landscape in three experimental blocks (Figure 2 1). The fences, which are installed in September of each year and removed in the spring, cause snow to accumulate preferentially on the warmin g (leeward) side of the fence. This increased


16 snow pack insulates the soil and the soil in the treated plots is on average 1.5 C warmer during October April (Table 2 1), with the bulk of this warming occurring in the late winter (December April) after sn ow has accumulated (Natali et al. 2011, 2012, 2014) In the early spring, the increased snow layer is removed to prevent delayed melt out. This soil warming treatment has been ongoing since the winter of 2008 2009 and the sustained warming has resulted in surface permafrost degradation ( Natali et al. 2014). Ecosystem C Fluxes Net ecosystem exchange (NEE) is the net gain or loss of C to an ecosystem over a set amount o f time. NEE comprises both R eco and plant C uptake (gross primary productivity; GPP) such that NEE = R eco GPP. R eco can be measured during dark periods when GPP is assumed to be zero and NEE can be measured at all times; GPP is inferred from the differ ence between NEE and R eco We used the convention that negative NEE values denote net C uptake by the ecosystem. Field Measurements On plot C hamber M easurements In late September 2012, 24 PVC chambers (10 cm diameter, 0.86 L average volume) were installed 10 cm into the soil of the experimental plots and remained in place fo r the entire winter (Figure 2 2 ). An open piece of tubing was installed next to the chambers to measure ambient CO 2 under the snowpack. To eliminate disturbance of the snowpa ck, this tubing, along with tubing from the chambers, extended to an off plot location. During sampling, the CO 2 concentration in the snowpack was determined using a Li 820 infrared gas analyzer (LI COR Biosciences, Lincoln, NE, USA; hereafter


17 referred to as IRGA). The CO 2 concentration within the chamber was then measured and scrubbed down to the ambient snowpack concentration using soda lime, at which point flux measurement began. Air was circulated between the chamber and the IRGA at 1L min 1 and the C O 2 concentration was recorded every 2 seconds. The flux calculation is described in detail in the appendix. All chamber measurements (n=152 on control and warming plots) were performed in low wind (< 8 mps) conditions. Snow P it M easurements Because the Ci PEHR experimental design requires an undisturbed snowpack to maintain the warming treatment, we established separate plots in the fall of 2009 at a nearby off experiment location (~100 m away) to measure R eco using the snow pit method (e.g. Elberling, 2007; Morgner et al. 2010) At the snow pit site, we installed forty PVC bases (25.4 cm diameter, 10 cm into soil, 12.8 L average chamber volume) in sets of four. Each set of four was adjacent to an individual soil temperature and soil moisture probe (n=10). A snow pit was dug to uncover the base prior to each sampling effort. This snow removal pri or to sampling eliminates issues of diffusivity through the snowpack that must be accounted for when measuring directly on top of the snow. Snow was left in the base and snow density was measured to account for this change in volume. The snow pits ranged from 3 to 100 cm in depth, depending on the plot and time of year. We waited 20 minutes after removing snow to allow the CO 2 to equilibrate between the soil and snow before placing the chamber on the base (Vogel et al. 2009) Once the chamber was placed on the base, we circulated the air through an IRGA and the chamber at 1L min 1 recording the CO 2 concentration every 1 s econd for 2 minutes (Figure 2 2 ). After sampling, the pit was re filled with snow. The set of four bases were re measured sequentially over the winter so that each sampling effort did not use the


18 previously uncovered bases; this reduced the impact of snow removal related disturbance t o the plots. Snow pit measurements (n=826) were taken in 2009, 2011, 2012, and 2013. Eddy C ovariance CO 2 exchange during the snow covered period was also measured using eddy covariance (EC) at a location approximately 1 km west of CiPEHR fr om 2008 though 2013 (Figure 2 2 ). The EC system consisted of a Campbell Scientific CSAT3 (Campbell Sci Inc. Logan, UT, USA) sonic anemometer and a Li COR Li 7500 open path CO 2 /H 2 O gas analyzer (LI COR Biosciences, Lincoln, NE, USA) mounted on a 3.5 m tower. Data were r ecorded at a frequency of 10 Hz on a Campbell Scientific CR5000 data logger and fluxes were Reynolds averaged over 30 min time periods (Reynolds, 1985) CO 2 flux es were corrected for variations in air density due to fluctuation in water vapor and heat fluxes (Webb et al. 1980) and for fluctuations caused by surface heat exch ange from the open path sensor during wintertime conditions (Burba et al. 2008) The tower footprint was greater than 400 m in all directions and consisted of vegetation similar to that found at CiPEHR. Belshe et al. (2012 ) provide more details of EC flux processing at this site. Soda L ime Winter R eco at CiPEHR was also measured by soda lime adsorption (e.g.Grogan & Chapi n, 1999b; Nobrega & Grogan, 2007; Rogers et al. 2011) Prior to placing soda lime at the CiPEHR plots, we weighed 300g of 4 8 mesh soda lime into 1L mason jars and dried for 24 hours at 100 C. Immediately after removing from the drying oven, mason ja rs were capped and stored until deployment. Bottomless 5 gallon buckets were inserted 15cm into the ground on the control and warming plots (n=12 per


19 treatment). In late September, we added 100 150 ml of water to each Mason jar to increase the adsorption capacity of the soda lime and placed one open jar inside each bucket, which was immediately capped for the entirety of the winter (Figure 2 2). We also deployed 6 mason jars into capped buckets with the bottom still intact as blanks such that the soda li me was not exposed to the soil or the atmosphere. In late April (2010 2012) or May (2013), the mason jars were collected from the field and immediately capped. The soda lime was then dried for 240 hours at 100 C and 24 hours at 75 C. R eco was determined by the change in soda lime dry weight, corrected for water loss associated with adsorption (Grogan, 1998) Fall C hamber F lux To capture CO 2 flux dynamics in the fall before photosynthesis had stopped for the year, sta tic chamber measurements were made on the CiPEHR plots in the fall of 2009, 2011, 2012, and 2013. Fluxes were measured on the warming and control plots by circulating air through a plexi glass chamber and an IRGA. At each plot, both light and dark (achie ved with chamber cover) fluxes were performed to estimate NEE and R eco. Fluxes from all three years (n=588) were combined into one data set for light determined that photosy nthesis had stopped before static flux measurements began (no significant difference between light and dark fluxes, mixed effects ANOVA, p>0.4), so light measurements were not included in analysis. Radiocarbon To determine how the relative contribution of old C to R eco changed by season, 14 13 C were measured from the on plot chambers


20 installed at CiPEHR (n=6 per treatment) in April of 2012 and 2013. These R eco isotopes were also meas ured at CiPEHR in August of the same years from 12 permanent PVC collars fixed 8 cm in the soil. Detailed methods of radiocarbon field collection and lab processing are available in Schuur et al. (2009) 14 C values were corrected for residual atmospheric CO 2 13 C values in a 2 pool mixing model as described by Schuur and Trumbore (2006 ). These corrected values were then normalized by the atmospheric 14 C value for 2012 to compare between years. Field Environmental Measurements In 2012 2013, snow depth was measured on the control and winter warming side of one snow fence every three hours using a Campbell Scientific SR50A sonic ranging senor (Campbell Sci In. Logan, UT, USA). Snow depth at the snow pit plots was measured before sampling using a 1 m ruler. Soil moisture and temperature sensors were located were within 1 m of each on plot and snow pit flux plot. Volumetric water content (hereafter refe rred to as soil moisture) was measured as an integrated value from the soil surface to 20 cm depth using site calibrated Campbell Scientific CS616 water content reflectometer probes. Soil temperature was measured using constantan copper thermocouples at 5 10, 20, and 40 cm. Soil moisture and temperature were recorded every thirty minutes using Campbell Scientific CR1000 data loggers (Campbell Sci Inc. Logan, UT, USA). An Onset HOBO (Onset Computer Corporation, Bourne, MA, USA) weather station located app roximately 100 m from the experimental plots recorded PAR, and air temperature and pressure every two minutes. Active layer depth was measured using a metal depth probe in September at each winter flux location.


21 Statistical Analysis Seasonal D efinitions We defined winter as the period when the photosynthetic flux is not significantly different from zero. This is different from the point at which NEE is negative (net C uptake), as photosynthesis may still be occurring despite a net loss of C when R eco los ses are greater than photosynthetic gain. Because there was limited data from clear chamber measurements directly on the CiPEHR plots, we used spring and fall EC data to define the threshold when photosynthesis stopped in the fall and started in the sprin g. To determine when photosynthesis was active, we used a one sided t test to test when d ata were not continuous in all years, we compared the population of light fluxes to the population of dark fluxes within a five day moving window and then applied the Bonferroni correction to account for multiple comparisons. In years with insufficient EC data, the most reasonable estimates of seasonal change were used based on the average of all other years or on available data. In the fall of 2010, our data showed some C uptake until DOY 323 but after that point there is a gap in the data. All statisti cal analyses were conducted using R version 2.15.1 (R Core Development Team, 2012) At the CiPEHR experiment, the growing season has previously been defined as the period between May 1 st and September 30 th ( Natali et al. 2011, Natali et al. 2014) By defining the growing season by the same number of calendar days every year, it is possible to determine whether differences in C fluxes across years are due to changes in ecosystem processes (such as greater summertime R eco ) rather than changes in


22 s eason length (such as a later onset of a fall). Here we define the weeks after September 30 th but before photosynthesis had reached the minimum threshold for the year as fall The months before April 30 th when photosynthesis was above the minimum, thresho ld we call spring. The period between September 30 th and April 30 th which comprises fall, winter, and spring, we call the non growing season (NGS). Analysis of the E nvironmental D rivers of W inter CO 2 F lux To determine the drivers of winter CO 2 flux, we constructed a separate model for each of the three flux methods with temporal resolution: on plot chamber measurements, snow pit chamber measurements, and EC. In all cases, the response variable (ecosystem CO 2 flux) was log transformed. Variance inflati on factors were used to determine which explanatory variables were collinear and should not be included in the analysis (values with a variance inflation factor over three were not used, as suggested by Zuur et al. 2009 ) Explanatory variables considered were: wind speed, atmospheric pressure, air temperature, soil temperature, and day of season (DOS). Snow depth, active layer depth, and treatment (on plot only) were also included in the initial snow pit and on plot mod els, but snow depth was too highly correlated with the remaining variables in the snow pit model, so it was excluded from further analysis. The first DOS was defined as October 1 st in all years and the season extended until April 30 th For the two chambe r measurements, we used mixed effects models using the lme command in the nlme package in R (Pinherio et al. 2013) with multiple environmental variables as the fixed effects and experimental design (nested: block, snow fence, plot for on plot measurements and nested: plot, subplot for the snow pit method) as the random effe cts. For the EC model, we used the lm function in the base


23 package in R and there were no random effects. The best model for each measurement method was selected using the lowest Akaike information criterion (AIC) and the residuals of all final models we re checked for normality and homogeneity of variances. Modeling W intertime R eco We determined CO 2 loss during the winter using a non linear multiple regression model (separate models were created for the on plot, snow pit, and EC methods). Because our goal was to model biological CO 2 production (R eco ) and not CO 2 flux, which results from both biological and physical processes, we chose a model based only on the biological drivers of winter R eco We described winter CO 2 loss as follows: R eco e + random effects. Analysis of all methods, including full and biological mod els, is provided in the appendix. We estimated CO 2 loss during all winters (2008 2009 through 2012 2013) including those during which we had no flux measurements (2008 2009; 2010 2011) by using the model to predict CO 2 loss based on measured environmental variables at CiPEHR. Carbon loss was predicted at each of the 24 CiPEHR plots and each of these plot predictions was averaged by treatment per experimental fence and then by treatment to calculate total amount of C lost from the experiment during the winte r. Standard error was calculated using snow fence (n=6) as the level of replication.


24 Modeling F all NEE The CO 2 balance during the fall was determined using response functions from measured R eco and NEE. R eco was modeled separately for the control and w arming treatment using the nlme function in the R nlme package (Pinherio et al. 2013) using the equation: R eco e random effects: experimental bl ock, fence, and plot. NEE was also modeled separately for the control and warming treatment using the nlme function and the equation: max max )]+R d max is the maximum photosynthesis at light saturation, PAR is photosynthetically active radiation at the time of measurement, and R d is dark ecosystem respiration. Again, R d was allowed to vary by the nested random effects: experimental block, fence, and plot. Parameter v alues for these functions are reported in T able 2 2. Modeling S pring NEE We used the EC data to determine the spring photosynthetic uptake as a GPP during the first two we eks of May). Because we were unable to measure NEE at CiPEHR in the spring (we could only measure R eco of the plots (n=12 per treatment) at CiPEHR such that such that spring NEE= R eco (winter modeled, above) ge measured GPP from the first two weeks in May). Average measured GPP during the first two weeks of May was calculated on an


25 individual plot basis based on data collected from automated chambers deployed during the growing season (Natali et al. 2011). In 2013, there were no autochamber measurements from 2013, so flux values from 2012 were used. In the spring of 2010 d 2012. Total NGS C B alance The NGS CO 2 balance was calculated by adding the NEE estimates from spring and fall to the winter R eco estimate. As with the winter CO 2 balance, error was calculated using snow fence as the level of replication and describes the spatial error of this estimate.


26 Table 2 1. Average soil temperature, snow depth, air temperature, and atmospheric pressure measured at CiPEHR during the October April of 2008 2013. Soil temperature is the average of 5, 10, 20, and 40 cm depths. In all years, the treatment effect on soil temperature was significant (p<0.001). Spring snow depth was measured before the shoveling effort each spring (see methods). Snow depth was measured in mid March in 2009, early April in 2013, and mid April in 2010 2012. Values in parenthesis are the spatial standard error of the estimate. Non growing season Average soil temperature (C) Sprin g snow depth (cm) Average air temperature (C) Average atmospheric pressure (atm) Control Warming Treatment difference Control Warming 2008 2009 2.8 (0.3) 1.2 (0.2) 1.6 (0.3) 39 (3) 130 (1) 12.92 0.92* 2009 2010 4.2 (0.2) 2.9 (0.2) 1.2 (0.3) 17 (2) 75 (1) 9.24 0.92* 2010 2011 3.8 (0.2) 1.5 (0.2) 2.3 (0.3) 23 (1) 103 (3) 11.07 0.92* 2011 2012 2.6 (0.4) 1.0 (0.1) 1.5 (0.3) 55 (3) 118 (2) 10.68 0.91 2012 2013 1.5 (0.3) 0.9 (0.2) 0.6 (0.5) 66 (1) 119 (2) 12.83 0.92 Indicates missing data from the on site weather station during this year. Values are filled from nearby weather station. Table 2 2. Mean estimate and standard error of parameters for the control and warming fall NEE and R eco 2 s 1 ) models. Model Coefficient Treatment Mean estimate SE Reco Basal respiration ( Control 0.41 0.03 Warming 0.57 0.08 Rate of respiration change ( Control 0.18 0.06 Warming 0.43 0.06 NEE Maximum photosynthesis ( F max) Control 1.46 1.79 Warming 3.88 8.28 Initial linear slope ( Control 1.18E 03 6.75E 04 Warming 1.57E 03 7.90E 04 Dark ecosystem respiration ( R d ) Control 0.35 0.06 Warming 0.46 0.09


27 Figure 2 1. Location of CiPEHR, eddy covariance tower, and the Eight Mile Lake region within the context of Alaska. A) A n aerial photograph of CiPEHR during the growing season. Note the three experimental blocks. B) Within each experimental block are two snow fences Photographs courtesy of Elizabeth E. Webb Figure 2 2. Photographs of the four methods of measuring winter CO 2 flux. A ) Snow pit B ) On plot C ) Soda lime D ) Eddy covariance. Photographs courtesy of Elizabeth E. Webb.


28 CHAPTER 3 RESULTS Method Comparison To compare between methods, we modeled cumulative R eco during the winter of 2012 2013 based on each of the three (EC, snow pit, on plot) measurement derived models and the biologically relevant variables (day of season and soil temperature) meas ured at the CiPEHR control plots during this time period (Table 3 1). The mean of all measurement methods was 107.5 g CO 2 C m 2 winter 1 There was more than a two fold difference between methods (Figure 3 1), with EC estimating the highest amount of CO 2 (45% higher than the mean of all methods). The on plot chamber measurements showed the lowest amount of CO 2 loss, estimating 39% lower than the mean of all methods. Soda lime, which measures the cumulative amount of CO 2 adsorbed and is not modeled, was higher than the two chamber measurements, which were most similar to each other. The trend observed in method differences for measuring winter CO 2 flux was preserved in NGS C loss estimates; NGS estimates made with the chamber models were lower than for soda lime adsorption (which was measured, not modeled) and EC (Table 3 2). Yet while EC estimated the highest NGS C loss, it showed the lowest difference (9%) between the control and warming treatment when the model was applied to both sets of measured te mperatures. In contrast, soda lime showed the greatest difference between control and warming (average 36% increase). The on plot method was next highest, at an average increase of 32%, and the snow pit showed a 14% increase between the control and warm ing treatments over five years.


29 The range of estimated cumulative CO 2 loss between measurement methods was a direct result of differences in measured fluxes, as all models were run on the same soil temperatures and season length for comparison. The on pl ot chambers were only sampled during 2012 2013, the winter with the warmest soil temperatures (Table 2 1), and as such we do not have measurements below 4.5 C (Figure 3 2). Despite sampling exclusively during the warmest winter, the on plot chambers mea sured the lowest average flux of the three methods ( 0.34 vs. CO 2 m 2 s 1 for the snow pit and EC methods, respectively) The on plot measurements also had the least variability between fluxes, even when compared with the other two methods at soil temperatures greater than 4.5 C. However, the medians of the on plot and snow pit measurements were very close (0.28 vs 0.32 CO 2 m 2 s 1 ) whereas the median of the EC measurements was much larger (0.55 CO 2 m 2 s 1 ). In addition to the highest average flux, the EC also had the lowest and highest measurements, which is likely a result of the high temporal coverage of this dataset compared to the other methods. When measurements taken on the same day under similar environmental conditions (low wind, equivalent soil temperatures) were compared, the snow pit and EC were not significantly different (p=0.1; mixed effects repeated measures ANOVA). It is likely, however, that the reduced number of measurements observed during these times (snow pit n=196; EC n=164) was not sufficient to detect a difference and a type II error occurred. The on plot and EC measurements were s ignificantly different (p=0.001) when compared during the same day. Drivers of Winter CO 2 Flux The three models derived separately from snow pit, on plot, and EC measurements quantify the relationship between CO 2 flux and possible covariates.


30 These model s identify the drivers of winter CO 2 flux as soil temperature, DOS, atmospheric pressure, air temperature, and the interaction between pressure and air temperature. With the exception of DOS in the on plot model and soil temperature in the EC model, all f actors were significant (p<0.05) in all three models ( Table 3 3). It is well documented that soil temperature is an important driver in soil respiration and thus we kept soil temperature in the EC model. It is likely that the relationship between soil te mperature and flux is weak in the EC model because EC calculates flux at the landscape level, whereas soil temperature was measured at one point near the tower. In all models, wind speed, snow depth, active layer depth, and experimental treatment were not significant (p>0.05). The relationship between CO 2 flux and DOS was negative, meaning that holding all other covariates constant, CO 2 flux decreased throughout the winter. The relationship between CO 2 flux and soil temperatures was positive; in all mode ls higher soil temperatures increased CO 2 flux, all other covariates being equal (Table 3 1; Table 3 3). Of all the soil depths measured (5, 10, 20, 40 cm, and the average of these depths), we found soil temperature at 10 cm to be the best predictor of wi nter R eco Air temperature, atmospheric pressure, and the interaction between the two were also significant drivers of winter CO 2 flux, likely because both impact the process of gas diffusion from the soil to the atmosphere. All gas diffusion is driven i n part by atmospheric pressure, and CO 2 movement is also governed by air temperature via advection. These two effects were magnified at low air temperatures and low atmospheric pressures.


31 Seasonal Variation The onset of winter (end of detectable plant C u ptake) varied as much as 55 days over the five years measured (Figure 3 3, Table 3 4) whereas the end of winter (start of plant C uptake) was much more stable, varying only by 11 days despite the drastically different snow conditions during the five winter s (Table 2 1). This means the period of the year during which there is no photosynthesis varied as much as 49 days, and this variation was driven almost entirely by when photosynthesis stops in the late fall. We were unable to attribute the end or beginn ing of photosynthesis to environmental variables such as a temperature or moisture thresholds. As determined by EC measurements, photosynthetic C uptake in the spring was 7, 13, and 30% of the C uptake in mid May in 2009, 2010, and 2011, respectively. This 2 uptake under the snowpack on the CiPEHR plots. The temperature sensitivity of R eco also varied between the growing season and the NGS (Figure 3 4). Wintertime C fluxes were systematically over estimated by the temperature response curve fit to both growing season and NGS R eco fluxes, indicating that Q 10 is not constant throughout the y 14 C signature of R eco changed between the growing season and the end of winter; it was significantly lower in April than in August (Figure 3 14 C values, however, were not affected by year (p=0.6) or by treatment ( p=0.9). Non Growing Season C Balance While we calculated NGS CO 2 loss estimates using all four methods (Table 3 2), we proceeded with the snow pit method for NGS CO 2 loss analysis because it estimated a moderate amount of CO 2 loss (of all the methods teste d, it did not predict the smallest or largest amount of CO 2 released; Table 3 2), had a sufficient number of


32 measurements (n=826) to capture yearly and seasonal variation, and has been well tested in the existing literature. Despite frigid temperatures, t he sheer length of the winter season meant that CO 2 loss during the NGS was driven mostly by the winter (81%) with small contributions from the fall (10%) and the spring (8%; values in parenthesis are averaged over treatment and years; Figure 3 6). In all years, the warming treatment lost more CO 2 than the control in the fall (p=0.009; one sided paired t test) and the winter (p=0.002; paired one sided t test) whereas in the spring, the warming treatment did not show a difference in C loss (p=0.2;paired one sided t test), as higher R eco was offset by greater plant uptake. The warming treatment lost, on average 26% more CO 2 in the fall and 15% more CO 2 in the winter. Total NGS CO 2 loss was between 68 and 93 g CO 2 C m 2 season 1 for the control and 82 and 99 g CO 2 C m 2 season 1 for the warming treatment (Figure 3 6; Table 3 2). Experimental warming increased NGS C loss (p=0.002; paired one sided t test) on average by 14%, but this percent increase varied from 7 to 24%, depending on the year. Differences in inter annual NGS C loss were primarily due to variation in soil temperatures (Figure 3 7; p= 0.001; mixed effects ANOVA), though years with the longest winters showed the greatest C loss over the entire NGS (p=0.1; mixed effects ANOVA). Overall, NGS CO 2 loss was exponentially related to soil temperature (Figure 3 7). We next modeled NGS CO 2 loss assuming R eco was the only ecosystem CO 2 flux only from October April (i.e. there is no photosynthesis during the snow covered period) and compared this estimate to our method of incorporating GPP during the spring and


33 fall. On average, the R eco only method estimated 12% more C loss than the GPP sensitive method (Figure 3 8). However, this difference was more pronounced for the control than for the warming treat ment (p=0.007; repeated measures mixed effects ANOVA). This suggests that, based on our method of modeling fall and spring GPP, warming stimulated R eco more than GPP during the NGS.


34 Table 3 1. Mean estimate and standard error of parameters for the on plot, snow pit, and eddy covariance biological 2 s 1) models. Day of season was not significant in the on plot model. Method Coefficient Soil temperature at 10cm Mean estimate SE Mean estimate SE M ean estimate SE On plot 0.53 0.092 0.250 0.053 Snow pit 0.67 0.041 0.094 0.022 0.0027 4.9E 04 Eddy covariance 1.05 0.066 0.056 0.026 7.6E 04 7.5E 04 Table 3 2. Total non growing season CO 2 loss estimates (g CO 2 C m 2 ) for the 5 years of soil temperature and PAR data available at CiPEHR. The on plot, snow pit, and eddy covariance estimates are modeled based on fall and method specific winter, and spring parameters. The soda lime was measured directly on the pl ots and the estimate was scaled to the same number of days (October 1 April 30) in all years. Values in parenthesis are the spatial standard error of the estimate. Method Year 2008 2009 2009 2010 2010 2011 2011 2012 2012 2013 Control Warmin g Control Warming Control Warmin g Control Warmin g Control Warmin g On plot 66 (3) 80 (3) 47 (1) 60 (1) 48 (1) 73 (2) 54 (4) 77 (2) 78 (4) 91 (4) Snow pit 89 (2) 97 (1) 71 (1) 82 (0.3) 68 (1) 84 (1) 74 (2) 86 (1) 93 (2) 99 (2) Eddy covariance 182 (2) 193 (2) 152 (1) 166 (0.4) 136 (1) 158 (1) 157 (3) 173 (1) 186 (2) 194 (2) Soda lime 136 (31) 159 (35) 120 (24) 205 (28) 148 (19) 176 (22) Indicates that the warming treatment released significantly (p<0.05; one sided t test) more CO 2 than the control for this method in this year.


35 Table 3 3. Mean estimate and standard error of parameters from the full on plot, snow pit, and EC winter CO 2 flux ( CO 2 m 2 s 1 ) models. Normalized estimates are a way to compare the relative importance of variables but should not be used in prediction. The numbers in parenthesis represent the rank order of variable importance within each model. While there is debate in the statistics community about the validity of p values from para meter estimates of mixed effects models, we report p values here, as is common in ecology, as a way to measure signal strength on a common scale. The EC model is a multiple linear regression with fixed effects only and thus has a p value whereas the on pl ot and snow removal models are mixed effects models and do not have p values. R 2 values for the mixed effects models were calculated after Nakagawa and Schielzeth 2013. Results are shown for the best fit models only. Coefficient Method On plot Snow pit Eddy covariance Intercept Mean estimate 24.19 0.48 10.98 SE 7.74 3.81 4.70 P value 0.002 0.901 0.020 Normalized estimate 1.80 1.81 0.70 Atmospheric pressure Mean estimate 23.69 0.10 12.49 SE 8.40 4.12 5.17 P value 0.006 0.981 0.016 Normalized estimate 0.07 0.10 0.03 Air temperature Mean estimate 3.68 1.73 1.30 SE 0.84 0.35 0.34 P value <0.001 <0.001 <0.001 Normalized estimate 0.11 0.10 0.03 Day of season Mean estimate 0.006 0.002 SE 0.00 0.00 P value <0.001 0.012 Normalized estimate 0.48 0.12 (2) Soil temperature (10 cm) Mean estimate 0.19 0.09 0.01 SE 0.06 0.02 0.03 P value 0.004 <0.001 0.686 Normalized estimate 0.21 (2) 0.21 0.02 (3) Atmospheric temperature Mean estimate 3.96 1.87 1.42 x Air temperature SE 0.91 0.38 0.38 P value < 0.001 <0.001 <0.001 Normalized estimate 0.26 (1) 0.15 0.12 (1) Model R 2 Fixed effects only 0.21 0.36 0.03 Fixed and random effects 0.46 0.52 NA Model p value NA NA <0.001


36 Table 3 4. Winter season (period of no photosynthesis) length and start and end dates for the 5 years measured. Winter Season length (days) Start End DOY Date DOY Date 2008 2009 183 270 29 Sep 88* 25 Mar 2009 2010 158 295* 14 Oct 88* 26 Mar 2010 2011 134 325* 21 Nov 94 30 Mar 2011 2012 148 300 29 Oct 83 23 Mar 2012 2013 168 283 11 Oct 86 26 Mar Indicates there was not enough EC data from this year to determine start or end date and instead the most reasonable estimate based on available data was used. Figure 3 1. Winter 2012 2013 CO 2 loss measured using the soda lime technique and modeled using the snow pit, on plot, and eddy covariance (EC) biological models. All models were applied to soil temperature measured at the CiPEHR control plots. Since the soda lime was left on the plots fo r the entire NGS but here we show a comparison of winter CO 2 loss, the soda lime measurement was scaled by the number of days during the winter. Error bars represent standard error of spatial variation. All estimates were significantly different from each hoc test) except the snow pit and on plot methods.


37 Figure 3 2. Winter CO 2 flux as measured by the three methods used to create flux models. A) Eddy covariance (EC) B ) Snow pit C ) On plot. Note the difference in y axis scales. D) A ll methods plotted on the same y axis; some EC measurements have been cut off for comparison.


38 Figure 3 3. da y moving window for EC fluxes. Dashed lines represent start and end of photosynthesis for each year as determined by a one sided t test. Figure 3 4. Winter (snow pit and on plot) and summer R eco fluxes (control and warming, May September 2008) plotte d against soil temperature. Eddy data was excluded from this figure to compare chamber measurements between seasons. The dash ed line represents the best fit exponential curve through all the data points. Information about methods of collecting summer flux measurements is provided in Natali et al. 2011.)


39 Figure 3 5. eco measured at control and warmed plots at CiPEHR during April and August of 2012 and 2013. Delta 14 C varied by season (p=0.01) but not by year (p=0.5) or treatment (p= 0.9). Smaller values reflect longer residence time of C, indicating an increased contribution of old C to R eco Figure 3 6. Total NGS CO 2 loss for the control (C) and warming (W) treatment partitioned by season (fall, winter, spring) for the five years of temperature and PAR data available at CiPEHR. Total NGS CO 2 loss was calculated using the snow pit biological model for winter and spring respiration. Standard error bars represent the spatial error for the total NGS estimate. In all years, warming released significantly more CO 2 during the NGS than control (p<0.02).


40 Figure 3 7. Total NGS CO 2 loss as a function of October April average s oil temperature (10 cm). Each point represents the modeled CO 2 loss of the control or warming treatment during one NGS. The solid line is the best fit exponential confidence inter parameters are significant (p<0.003). The Q 10 of this function is 2.5. Figure 3 8. Percent by which assuming only R eco occurs during the NGS over estimates NGS CO2 loss. The percent increase grew over time (p<0.001; repeated measures mixed effects ANOVA) and was more pronounced in the control than in the warming treatment (p=0.007). Standard error bars represent the spatial error of the estimate.


41 C HAPTER 4 DISCUSSION Comparison of Measurement Methods While winter respiration plays a pivotal role in the annual C balance in the arctic, there is large uncertainty in winter C budgets because of methodological limitations of measuring fluxes in arctic winter field conditions. Our study, which aimed to reduce this uncer tainty through a methods comparison, found that for a single field site, there was a four fold range in NGS C loss across the four methods employed. However, these measurements constrained NGS C flux (between 47 and 186 g CO 2 C m 2 season 1 for the contro l) at the mid to high end of the published arctic tundra winter respiration estimates (0.19 to 210 g CO 2 C m 2 season 1 over a similar time period; converted units from Bjrkman et al. 2010a) In contrast with other studies that showed the highest rates of CO 2 flux from chambers placed over the soil as oppo sed to diffusion methods or chambers placed directly on the snow (Bjrkman et al. 2010a) we found that c hamber methods measured the lowest CO 2 flux of the four methods tested (Figure 3 1). Because chamber methods must only be employed during low wind conditions (Bain et al. 2005) and because more CO 2 is released in windy conditions as a result of snow and soil venting (Bowling & Massman, 2011) extrapolating fluxes during low wind conditions to the entire winter may underestimate the amount of C released. In contrast, EC is able to capture soil and snow venting events because it measures during turbulent conditi ons. Indeed, while wind speed was not a significant control of CO 2 flux in any of our models, Eu skirchen et al. (2012) found that wind speed was an important driver of winter CO 2 flux in tundra ecosystems using EC towers. It is possible that the EC


42 method employed in this study overestimated total CO 2 lost by extrapolating measurements from high wind conditions to times of no wind. However we found that even in times of low wind (< 8 mps), the EC tower measured fluxes higher than the two chamber methods (p=0.001 for on plot and p=0.1 for snow pit). In this study, the soda lime adsorption method s howed higher cumulative winter CO 2 loss than the two chamber measurements. Soda lime may over estimate C loss by drawing down CO 2 in the bucket headspace, thus creating a gradient where CO 2 diffuses out of the soil at a higher rate than under natural conditions (Nay et al. 1994; Grogan & Chapin, 1999; Nobrega & Grogan, 2007) Indeed, Nay et al. (1994) found that at low rates of CO 2 flux, soda lime over estimated the expected flux by 25%. Nobrega and Grogan (2011 ) noted that this effect can be decreased by installing the chambers deep in the soil (10cm), thus isolating the soil volume affected. Our buckets were installed 15cm into the ground, yet when we measur ed the CO 2 concentration of the bucket headspace immediately before removing the soda lime jars from the field in the spring of 2013, we found concentrations below ambient CO 2 levels (the average bucket was 125 ppm above the ambient concentration; data not shown). This indicated that after eight months in the field, the soda lime appeared to be drawing down the bucket headspace CO 2 concentration, potentially resulting in an overestimated CO 2 flux. Drivers of Winter CO 2 Flux Soil temperature is a primary driver of soil respiration due to the temperature sensitivity of microbial enzymes (Davidson & Janssens, 2006) and the effect of temperature on available water in solution (Clein & Schimel, 1995; Tilston et al. 2010) Additionally, it is well established that winter R e co in tundra soils results from biological activity rather than from diffusion of stored CO 2 in the soils (Zimov et al. 1996; Panikov


43 et al. 2006) The temperature sensitivity of microbial decomposition (Q 10 ; Q 10 is the change in R eco f or a 10C temperature increase) in frozen soils, however, is less well understood; the range of Q 10 values reported in the literature is between 1 and 237 for winter arctic soils (Mikan et al. 2002; Elberling & Brandt, 2003; Panikov et al. 2006; Elberling, 2007; Morgner et al. 2010; Euskirchen et al. 2012, Natali et al., 2014, Oechel et al., 2014) although most values are between 2 and 13 ( when the Q 10 was not published, we calculated the value based on the reported exponential curve parmaters; Elberling & Brandt, 2003; Panikov et al. 2006; Mor gner et al. 2010; Euskirchen et al. 2012; Natali et al. in press; Oechel et al. 2014) Our Q 10 values are on the lower end of this spread (12.2 for on plot, 2.6 for snow pit, and 1.8 for EC), which is, at least in part, due to the fact that soil temp erature is not the only explanatory variable in our (snow pit and EC) models. Other studies that quantify the relationship between C flux and soil temperature with additional explanatory variables (such as wind speed, pressure, or net radiation) also repo rt lower values (Euskirchen et al. 2012; Oechel et al. 2014) These l ower Q 10 values (~1 13) are more in the range of values published for terrestrial soils (Lenton & Huntingford, 2003; Hamdi et al. 2013) but it is understood that, in general, Q 10 increases at lower temperatures (Lloyd & Taylor, 1994; Davidson & Janssens, 2006) Elberling & B randt (2003) and Elberling (2007) found that there was no evidence for a shift in temperature sensitivity around 0C and that a single Q 10 value could be applied to the same soil at all tempera tures. In contrast, Schuur et al. (2009) argued for a biological threshold around 0C and applied different temperature response curves to measurements above and below 0C. In general this has been the approach of most


44 studies, wh ich analyze CO 2 flux data from the growing season separately from the NGS (e.g. Welker et al. 2000; Vogel et al. 2009; Euskirchen et al. 2012; Natali et al. 2014) We found that when a single Q 10 value was applied to measurements from both the growing season and the winter, the winter was systematically over estimated (Fi gure 3 4). This supports the system of partitioning the year into the growing season and the NGS. However, because growing season R eco measurements comprise both plant and soil processes whereas winter R eco is, for the most part, a result of heterotrophic respiration, we cannot decipher whether the Q 10 of soil respiration itself changes over the course of the year. In addition to soil temperature, we found that DOS is a biological driver (contributing to CO 2 production rather than diffusion out of the soil) of winter C flux, with more C released at the beginning of the season than at the end when all other conditions are similar. This phenomenon of a change in the magnitude of C flux over the winter season h as also been documented by others (Zimov et al. 1996; Oechel et al. 1997; Fahnestock et al. 1999; Morgner et al. 2010) although in some cases a spring surge in C release was also observed (Zimov et al. 1996; Oechel et al. 1997; Fahnestock et al. 1999) It is not known whether this spring surge is physical (CO 2 that was trapped in the soils is released upon thaw; Elberling & Brandt, 2003) or biological (nutrients from dead microbial biomass is released upon thaw, stimulating decompos ition from freeze tolerant microbes ; Skogland et al. 1988; Schimel & Clein, 1996) We did not observe a burst in R eco in the spring, although less than one percent of our data we re collected when soil temperatures were above 0C in the spring (Figure 3 2).


45 The heightened C loss early in the winter is most often attributed to the relatively warmer soil temperatures and greater amount of unfrozen water present in the soils at this time (Oechel et al. 1997; Fahnestock et al. 1999) However, our models indicated that DOS decreased over the course of the winter even when soil temperature was held con stant We hypothesize that the larger CO 2 release early in the season to the presence of recently senesced plant inputs, which are easily decomposable and dominate the available C pool. Later in the season, this pool decreases, leading to declines in CO 2 flux rates. Indeed, it is thought that winter R eco in the arctic is derived primarily from recent labile plant inputs as opposed to bulk soil C (Grogan et al. 2001; Grogan, 201 2) In a meta analysis of high latitude soil incubation studies, Schdel et al. (2014 ) established t hat the labile C pool of permafrost soils is exhausted after ~150 days at 5 C, after which the microbial community shifts to a reliance on more slowly decomposing C. While 5 C is warmer than the average temperature of winter tundra soils, it is likely t hat after 7 months without any new plant inputs, there is much less labile C in our soils than at the start of winter. Schimel e t al. (2004) measured nitrogen mineralization in tundra soils in the fall (Sept Nov) and during the winter (Nov March or Nov May) and found nitrogen mineralization rates were higher during the winter than in the fall, indicating microbes are more C limited by the end of the winter than in the fall. Additionally, 14 C data from our field site and from boreal soils (Winsto n et al. 1997) show a 14 C signature for R eco that is more negative at the end of the winter than it is mid summer (Figure 3 5), indicating that by the end of the winter, microbes are decomposing a greater proportion of older, and often more slowly deco mposing C (Trumbore & Czimczik, 2008; Schmidt et al. 2011) It is also possible that the


46 seasonal difference in the R eco 14 C signature is due to increased plant respiration in the summer as opposed to April. An additional hypothesis for our observed C flux decrease over the course of the winter is that in the early part of the winter, there is unfrozen water in soil pores but as the season progresses, this water freezes and is thus unavailable to microbes. Hence, there would be more C released at the beginning of the season, when the soils are freezing as opposed to at the same temperatures in the spring before the soils have th awed. This hypothesis is supported by microbial ecology; microbes need liquid water for respiration, extracellular enzyme movement, and substrate diffusion (Clein & Schimel, 1995; Mikan et al. 2002) Panikov et al. (2006) and Tilston et al. (2010) showed that in the laboratory, frozen soil C flux was related to both water content and temperature. However, field studies of moisture control have remained elusive and, at least in one study, summertime moisture levels did not affect winter flux (Elberling, 2007) We opted not to use moistur e as a covariate in our analysis due to uncertainty about the sensor calibrations during the frozen period (Pers. communication with Campbell Scientific, 2013) However, a simple regression between the raw millivolt output of our moisture probes and winter C flux showed decreased flux with increasing resistance (more moisture). Including the raw millivolt output from our moisture probes in the models did not eliminate DOS as a significant predictor of CO 2 flux. This suggests that ice may actually be restricting CO 2 diffusion out of the soil, which has also been documented in frozen soils laboratory experiments (Elberling & Brandt, 2003) In this circumstance, the effect of the frozen water is to act as a physical barrier to CO 2 diffusion but that does not exclude the possibility that the unfrozen water content is


47 simultaneously limiting biological CO 2 production. Thus, while water may limit microbial activity in frozen soils, it may also limit CO 2 diffusion out of the soil, making it challenging to isolate the effect of moisture on CO 2 flux in the field. While many studies have shown a positive correlation between snow depth and winter CO 2 flux (e.g. Fahnestock et al. 1998, 1999; Jones et al. 1999; Nobrega & Grogan, 2007; Rogers et al. 2011) snow depth was not a significant contributo r to winter C release in any of our models. This could be because studies that point to snow depth as an important driver measured C flux at plots that span a range of snow depths at a few times throughout the year. In contrast, we sampled the same plots as they accumulated snow over the winter. Because our measurements spanned many years and include data of different snow depths in the same plot, we tracked intra plot as well as inter plot response to snow depth. While (snow pit) plots with higher aver age snow depth did show more C release, snow depth was too highly correlated with other variables to be used on our modeling analysis. However, when CO 2 flux was regressed against snow depth alone, there was a negative relationship. Euskirchen et al. (2012) also found a negative relationship between snow depth and winter CO 2 flux using EC towers, which they attribute to CO 2 trapping in the snow pack. Since our chamber measurements were made directly over the soil instead of over the snowpack, it is unlikely that CO 2 trapping within the snowpack was the cause of this negative relationship. Instead, snow depth was highly correlated with DOS and s oil temperature, indicating that the negative relationship between CO 2 flux and snow depth was because there is a negative relationship with soil temperature and DOS. Hence, while snow


48 plays an important role in insulating the soil from the cold winter ai r temperatures, it is soil temperature and DOS that is driving CO 2 flux. Non Growing Season C Balance Uncovering the mechanisms underlying NGS C cycling and quantifying NGS C quantify NGS CO 2 loss, we first partitioned the NGS into the times when photosynthesis was active or dormant. This approach is more rigorous than previous definitions of winter (Grogan & Jonasson, 2006, Natali et al. 2014, Oechel et al. 2014 ) because it incorporates inter annual var iability and accounts for plant C uptake even when snow is present. Our fall and winter CO 2 loss values are estimated based on measurement derived models and environmental variables measured directly at the CiPEHR plots. Similarly, the spring R eco values are also based on a measurement derived model and soil temperature measured directly at CiPEHR whereas the spring GPP estimates were made based on a percentage, as determined by EC tower measurements, of CO 2 flux measurements made at the plots in May. We found that seasonal variation led to estimates that were 12% lower than when we modeled NGS CO 2 loss by assuming that R eco is the only ecosystem process during the NGS (Figure 3 8). This evaluation might be an underestimate, however, because both the spr ing and the fall seasons are expected to lengthen with climate change (Euskirchen et al. 2009; Christensen et al. 2013b) Next we quantified NGS C loss by modeling NEE based on environmental variables, rather than applying an average rate to the entire season or interpolating between points (e.g. Oechel et al. 1997; Welker et al. 2000; Bjrkman et al. 2010b) We used DOS and soil temperature to model R eco during the winter and spring, improving upon other models that only used soil te mperature to quantify winter C loss


49 (Elberling, 2007; Morgner et al. 20 10; Natali et al. 2014) Research conducted in March and April of 2009 at our study site showed a 28% increase in basal respiration in response to warming (Natali et al. 2014 ), however we did not observe this pattern during the years of our study. Instead we found that the warming treatment was not a significant (p>0.05) predictor of winter CO 2 flux in the on plot model, indicating that experimental warming did not influence CO 2 flux aside from its effect on temperature. In other words, our study f ound that the warming treatment released more CO 2 because of increased soil temperature rather than because of the additive effects of increased basal respiration and increased soil temperature. We attribute these contrasting results to different methodol ogical approaches. Similar to other studies (Welker et al. 2000; Nobrega & Grogan, 2007; Morgner et al. 2010; Natali et al. 2014) our results show that the warming treatment l ost more C over the NGS than the control (Figure 3 6). The soda lime method indicated that warming increased CO 2 loss by an average of 36%, but the variation was high enough such that this effect as only significant in 2011 2012 (Table 3 2). Across the o ther three methods, the warming treatment released significantly more CO 2 in all years (on average 32 % (on plot), 14% (snow pit), and 9 % (EC) more). Thus, even though soil temperature was not the only predictor of NGS CO 2 loss, our data demonstrate that as the winter climate warms, CO 2 loss during the NGS could increase exponentially (Figure 3 7). Additionally, CO 2 loss during the NGS was driven mostly by the winter (Figure 3 6) with small contributions from the fall and the spring, indicating that NGS CO 2 loss will be further magnified with climate warming, since the greatest temperature increases are expected in December February (Christensen et al. 2013a)


50 To understand the annual C balance of the tundra, the warming induced increase in NGS CO 2 loss must be put in the context of the gro wing season, where experimental warming increased R eco at the same site by 18, 24, and 46 % across the study years, 2009, 2010, and 2011( Natali et al. 2014 ). When combined with growing season NEE at the same site, our results show that the tundra was, on average, a CO 2 source in both the control and warming plots for the three years (2009, 2010, 2011) of growing season data available ( Natali et al. 2014 ). Depending on the method used to calculate NGS CO 2 loss, the control plots lost, on average, 27 (on plot, SE=17), 49 (snow pit; SE=18), and 129 (EC; SE=25) g CO 2 C m 2 yr 1 Annual CO 2 loss for the warming plots was 22 (on plot; SE=33), 38 (snow pit; SE=34), 123 (EC; SE=40) g CO 2 C m 2 yr 1 (estimates are the average and SE of annual CO 2 loss by method for the first three years of the experiment). It has long been speculated that C loss over the winter period could shift the tundra from a C sink to a source, but lack of data precluded many substantive studies. While we only have three y ears of data, our results support multiple lines of evidence from individual field studies (Oechel et al. 1993; Welker et al. 2000; Natali et al 2014) a meta analysis of field studies (Belshe et al. 2013) and modeling simulations that indicate that the tundra is shifting from the historical C sink to a C source (Koven et al. 2011; Schneider von Deimling et al. 2012) These results highlight the great importance of considering both the growing and non growing season contributions to annual CO 2 flux from arctic ecosystems and the importance of refining our understanding of climate change driven changes in soil temperature, seasonal patterns, and the mechanisms driving ecosystem C loss.


51 APPENDIX SUPPLEMENTARY MATERIAL On plot Chamber Measurement Flux Calculation The on plot chambers were sampled according to the following method: 1.) The CO 2 concentration of the snow pack was measured. 2.) CO 2 free air was run through the chamber until the chamber concentratio n matched that of the ambient snow pack. 3.) Unaltered air was circulated between the IRGA and the chamber headspace for fifteen minutes. We found that it took longer to bring the chamber headspace to the ambient snow pack CO 2 concentration than would be expected from the 1 L min 1 flow rate of the IRGA pump. We attributed this to the fact that the chamber headspace is not a closed system and so as the CO 2 concentration within the chamber decreased, CO 2 stored in the soil diffused into the chamber headspa ce. Indeed, after scrubbing stopped, we saw an initial spike in the rate of CO 2 accumulation (Figure A 1) that can be explained by rapid CO 2 diffusion from the soil to the chamber as a result of the altered diffusion gradient. We also saw this initial sh arp increase when we tested this method on a closed chamber installed in sand. Because there was no biological flux from the sand, we can be sure this initial rise in CO 2 concentration is the result of physical processes. Thus, it is important to not inc lude this high rate of CO 2 diffusion in the flux calculation. However, at some point the perturbation related diffusion into the chamber will end and CO 2 will diffuse out of the soil at the same rate as under the surrounding snowpack. When the headspace concentration increases enough, CO 2 will diffuse out of the chamber or laterally through the soil instead of vertically into the chamber. The problem we faced was knowing when we were measuring diffusion that was occurring


5 2 at the same rate as the surrounding snow pack, or when CO 2 diffusion into the chamber was increasing linearly. In other words, the point at which the first derivative of the CO 2 conce ntration vs. time curve is linear should be the point at which perturbation related diffusion stopped but before CO 2 diffusion out of the chamber started. To determine this point, we took the second derivative of the CO 2 concentration vs. time curve. We then recorded the first derivative at the point where the second derivative was zero. We tested this method of data processing against laboratory tests on a CO 2 flux generator (Martin et al. 2004) and field tests during the summer when we could remove the chamber top. Winter CO 2 Flux Models We created multiple regression (for EC) and mixed effects (for the chamber methods) models to determine the drivers o f winter CO 2 flux. We found that DOS, soil temperature, air temperature, atmospheric pressure, and the interaction between air temperature and atmospheric pressure were significant controls over winter CO 2 flux (Table 3 3). However, because our sampling was not frequent enough to capture all physical aspects of CO 2 flux such as snow pack venting events, we chose to model CO 2 production over the winter, which is dependent only on biological controls: DOS and soil temperature. We compared the outcome of t he biology only models with the full models (DOS, soil temperature, air temperature, atmospheric pressure, and the interaction between air temperature and atmospheric pressure) of each measurement technique (Figure A 2). The EC full model was consistentl y higher than the EC biological model and the difference between the two models varied between 8 and 25 %, depending on the year. For the chamber models, the pattern was not consistent; in some years the biological


53 model predicted higher CO 2 loss than the full model, but in other years this trend was reversed. The difference between the full and biological models varied between 2 and 18 % for the chamber models. The trend of the on plot method predicting the lowest CO 2 loss and the EC predicting the high est CO 2 loss was preserved in all models and years. The difference between the full and biological models is small and all values lie well within the range of published winter tundra CO 2 loss estimates (Bjrkman et al. 2010a) Additionally, the lack of consistent directionality in the difference between the full and biological models suggests that we are no t systematically biasing our calculations by predicting winter CO 2 loss with the biology only models.


54 Figure A 1. Representative example of the CO 2 concentration vs. time curve for the on plot chamber measurements. After CO 2 free air is run through the chamber, there is an initial sharp increase in slope that is the result of diffusion and not a biological flux. After this rapid flow into the chamber has stopped, there is a biological flux before the chamber concentration inc reases enough so that CO 2 diffuses out of the chamber. Figure A 2. Comparison between the full and biological models used to predict wintertime CO 2 l oss. Estimates are for the control and warming treatments in four representative years. EC model was a pplied to soil temperatures measured at CiPEHR plots.


55 LIST OF REFERENCES Bain WG, Hutyra L, Patterson DC, Bright A V, Daube BC, Munger JW, Wofsy SC (2005) Wind induced error in the measurement of soil respirati on using closed dynamic chambers. Agricultural and Forest Meteorology 131 225 232. Belshe EF, Schuur EAG, Bolker BM, Bracho R (2012) Incorporating spatial heterogeneity created by permafrost thaw into a landscape carbon estimate. Journal of Geophysical R esearch 117 1 14. Belshe EF, Schuur EAG, Bolker BM (2013) Tundra ecosystems observed to be CO 2 sources due to differential amplification of the carbon cycle. Ecology letters 1307 1315. Biasi C, Meyer H, Rusalimova O et al. (2008) Initial effects of experimental warming on carbon exchange rates, plant growth and microbial dynamics of a lichen rich dwarf shrub tundra in Siberia. Plant and Soil 307 191 205. Bjrkman MP, Morgner E, Cooper EJ, Elberling B, Klemedtsson L, Bjrk RG (2010a) Winter carbon dioxide effluxes from Arctic ecosystems: An overview and comparison of methodologies. Global Biogeochemical Cycles 24 Bjrkman MP, Morgner E, Bjrk RG, Cooper EJ, Elberling B, Klemedtsson L ( 2010b) A comparison of annual and seasonal carbon dioxide effluxes between sub Arctic Sweden and High Arctic Svalbard. Polar Research 29 75 84. Bowling DR, Massman WJ (2011) Persistent wind induced enhancement of diffusive CO 2 transport in a mountain for est snowpack. Journal of Geophysical Research 116 G04006. Burba GG, McDdermitt DK, Grelle A, Anderson DJ, Xu L (2008) Addressing the influence of instrument surface heat exchange on the measurements of CO 2 flux from open path gas analyzers. Global Change Biology 14 1854 1876. Campbell Scientific (2013) Personal Communication. Christensen JH, Kumar KK, Aldrian E et al. (2013a) Climate Phenomena and their Relevance for Future Regional Climate Change. In: Climate Change 2013: The Physical Science Basis. Co ntribution of Working Group 1 to the fifth Assessment Report of the Intergovernmental Panel on Climate Change (eds Stocker T., Qin D, Plattner G K, Tignor M, Allen SK, Boschung J, Nauels A, Xia Y, Bex V, Midgley PM). Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA.


56 Christensen JH, Kumar KK, Aldrian E et al. (2013b) Long term Climate Change: Projections, Commitments and Irreversibility. In: Climate Change 2013: The Physical Science Basis. Contribution of Working Group 1 to the fi fth Assessment Report of the Intergovernmental Panel on Climate Change (eds Stocker T., Qin D, Plattner G K, Tignor M, Allen SK, Boschung J, Nauels A, Xia Y, Bex V, Midgley PM). Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA. C lein JS, Schimel JP (1995) Microbial activity of tundra and taiga soils at sub zero Ttemperatures. Soil Biology and Biochemistry 27 1231 1234. Davidson EA, Janssens IA (2006) Temperature sensitivity of soil carbon decomposition and feedbacks to climate c hange. Nature 440 165 73. Elberling B (2007) Annual soil CO 2 effluxes in the High Arctic: The role of snow thickness and vegetation type. Soil Biology and Biochemistry 39 646 654. Elberling B, Brandt KK (2003) Uncoupling of microbial CO 2 production an d release in frozen soil and its implications for field studies of arctic C cycling. Soil Biology and Biochemistry 35 263 272. Euskirchen ES, McGuire A D, Rupp TS, Chapin FS, Walsh JE (2009) Projected changes in atmospheric heating due to changes in fire disturbance and the snow season in the western Arctic, 2003 2100. Journal of Geophysical Research 114 G04022. Euskirchen ES, Bret Harte MS, Scott GJ, Edgar C, Shaver GR (2012) Seasonal patterns of carbon dioxide and water fluxes in three representative tundra ecosystems in northern Alaska. Ecosphere 3 Fahnestock JT, Jones MH, Brooks PD, Walker DA, Welker JM (1998) Winter and early spring CO 2 efflux from tundra communities of northern Alaska. Journal of Geophysical Research 103 23 27. Fahnestock JT, Jones H, Welker M (1999) Wintertime CO 2 Implications for annual carbon budgets. Global Biogeochemical Cycles 13 77 5 779. Grogan P (1998) CO 2 during CO 2 adsorption. Ecology 79 1467 1468. of vegetation type snow depth and interannual climatic variation. Arctic, Antarctic, and Alpine Research 44 446 456. depend on season. Ecosystems 2 451 459.


57 Grogan P, Jonasson S (2006) Ecosystem CO 2 production during winter in a Swedish subarctic region: the relative importance of climate and vegetation type. Global Change Biology 12 1479 1495. Grogan P, Illeris L, Michelsen A, Jonasson S (2001) Respiration of recently fixed plant carbon domina tes mid winter ecosystem CO 2 production in sub arctic heath tundra. Climactic Change 129 142. Hamdi S, Moyano F, Sall S, Bernoux M, Chevallier T (2013) Synthesis analysis of the temperature sensitivity of soil respiration from laboratory studies in relati on to incubation methods and soil conditions. Soil Biology and Biochemistry 58 115 126. Harden JW, Sundquist ET, Stallard RF, Mark RK (1992) Dynamics of soil carbon during deglaciation of the laurentide ice sheet. Science 258 1921 1924. Hicks Pries CE, Schuur EAG, Crummer KG (2011) Holocene carbon stocks and carbon accumulation rates altered in soils undergoing permafrost thaw. Ecosystems 15 162 173. Hicks Pries CE, Schuur E a. G, Vogel JG, Natali SM (2013) Moisture drives surface decom position in thawing tundra. Journal of Geophysical Research: Biogeosciences 118 Hobbie SE, Schimel JP, Trumbore SE, Randerson JR (2000) Controls over carbon storage and turnover in high latitude soils. Global Change Biology 6 196 210. Huemmrich KF, Kin oshita G, Gamon JA, Houston S, Kwon H, Oechel WC (2010) Tundra carbon balance under varying temperature and moisture regimes. Journal of Geophysical Research 115 G00I02. Jones MH, Fahnestock JT, Welker JM (1999) Early and late winter CO 2 efflux from arct ic tundra in the Kuparuk river watershed, Alaska, U.S.A. Arctic, Antarctic, and Alpine Research Kappen L (1993) Plant activity under snow and ice, with particular reference to lichens. Arctic 46 297 302. Koven CD, Ringeval B, Friedlingstein P et al. (20 11) Permafrost carbon climate feedbacks accelerate global warming. Proceedings of the National Academy of Sciences of the United States of America 108 14769 74. Lee H, Schuur E a. G, Vogel JG, Lavoie M, Bhadra D, Staudhammer CL (2011) A spatially explici t analysis to extrapolate carbon fluxes in upland tundra where permafrost is thawing. Global Change Biology 17 1379 1393.


58 Lenton TM, Huntingford C (2003) Global terrestrial carbon storage and uncertainties in its temperature sensitivity examined with a s imple model. Global Change Biology 9 1333 1352. Lloyd J, Taylor JA (1994) On the temperature dependence of soil respiration. Functional Ecology 8 315 323. Martin JG, Bolstad P V, Norman JM (2004) A carbon dioxide flux generator for testing infrared gas analyzer based soil respiration systems. Soil Science Society of America Journal 68 514 518. Mikan CJ, Schimel JP, Doyle AP (2002) Temperature controls of m icrobial respiration in arctic tundra soils above and below freezing. Soil Biology and Biochemistry 34 1785 1795. Morgner E, Elberling B, Strebel D, Cooper EJ (2010) The importance of winter in annual ecosystem respiration in the High Arctic: effects of snow depth in two vegetation types. Polar Research 29 58 74. Nakagawa S, Schielzeth H (2013) A general and simple method for obtaining R 2 from generalized linear mixed Methods in Ecology and Evolution 4 133 142. Natali SM, Schuur EAG, Trucco C, Hicks Pries CE, Crummer KG, Baron Lopez AF (2011) Effects of experimental warming of air, soil and permafrost on carbon balance in Alaskan tundra. Global Change Biology 17 1394 1407. Natali SM, Schuur EAG, Rubin RL (2012) Increased plant productivity in Alaskan tundra as a result of experimental warming of soil and permafrost. Journal of Ecology 100 488 498. Natali SM, Schuur EAG, Webb EE, Hicks Pries CE, Crummer KG (2014) Permafros t degradation stimulates carbon loss from experimentally warmed tundra. Ecology Nay MS, Mattson KG, Bormann BT (1994) Biases of chamber wethods for measuring soil CO 2 efflux demonstrated with a laboratory apparatus. Ecology 75 2460 2463. Nobrega S, Grog an P (2007) Deeper snow enhances winter resipration from both plant associated and bulk soil carbon pools in birch hummock tundra. Ecosystems 10 419 431. Oberbauer SF, Tweedie CE, Welker JM et al. (2007) Tundra CO 2 fluxes in response to experimental warming across latitudinal and moisture gradients. Ecological Monographs 221 238.


59 Oechel WC, Hastings SJ, Vourlitis G, Jenkins M, Riechers G, Grulke N (1993) Recent change of Arctic tundra ecosystems from a net carbon d ioxide sink to a source. Nature 361 9 12. Oechel WC, Vourlitis G, Hastings SJ (1997) Cold Season CO 2 emissions from arctic soils. Global Biogeochemical Cycles 11 163 172. Oechel WC, Vourlitis G, Hastings S, Zulueta R, Hinzman L, Kane D (2000) Acclimati on of ecosystem CO2 exchange in the Alaskan Arctic in response to decadal climate warming. Nature 406 978 81. Oechel WC, Laskowski CA, Burba G, Gioli B, Kalhori, AAM (2014) Annual patterns and budget of CO 2 flux in an arctic tussock tundra ecosystem. Jou rnal of Geophysical Research. Osterkamp TE, Jorgenson MT, Schuur EAG, Shur YL, Kanevskiy MZ, Vogel JG (2009) Physical and ecological changes associated with warming permafrost and thermokarst in interior Alaska. Permafrost and Periglacial Processes 256 2 35 256. Panikov NS, Flanagan PW, Oechel WC, Mastepanov M a., Christensen TR (2006) Soil Biology and Biochemistry 38 785 794. Pinherio J, Bates D, Saikat D, Sarkar D (2013) nlme: Linear and nonlinear mixe d effects models. R package. R core R Core Development Team (2012) R: A language and environment for statistical computing. Reynolds O (1985) On the dynanmical theory of incompressible viscous fluids and the determination of the criterion. Philosophical transactions of the Royal Society of London. Series A 186 123 164. Rogers MC, Sullivan PF, Welker JM (2011) Evidence of nonlinearity in the response of net ecosystem CO 2 exchange to Iincreasing levels of winter snow depth in the high arctic of northwest Greenland. Arctic, Antarctic, and Alpine Research 43 95 106. Rustad LE, Campbel JL, Marion GM et al. (2001) A meta analysis of the response of soil respiration, net nitrogen mineralization, and aboveground plant growth to experimental ecosystem warming. Oecologia 126 543 562. Schdel C, Schuur EAG, Bracho R et al. (2014) Circumpolar assessment of permafrost C quality and its vulnerability over time using long term incubation data. Global change biology 20 641 52.


60 Schimel JP, Clein JS (1996) Microbial response to freeze thaw cycles in tundra and taiga soils. Soil Biol. Biochem 28 1061 1066. Schimel JP, Bilbrough C, Welker JM (2004) Increased snow depth affects microbial activity and nitrogen mineralization in two Arctic tundra communities. Soil Biolog y and Biochemistry Schmidt MWI, Torn MS, Abiven S et al. (2011) Persistence of soil organic matter as an ecosystem property. Nature 478 49 56. Schneider von Deimling T, Meinshausen M, Levermann A, Huber V, Frieler K, Lawrence DM, Brovkin V (2012) Estima ting the near surface permafrost carbon feedback on global warming. Biogeosciences 9 649 665. Schuur EAG, Trumbore SE (2006) Partitioning sources of soil respiration in boreal black spruce forest using radiocarbon. Global Change Biology 12 165 176. Sch uur EAG, Crummer KG, Vogel JG, Mack MC (2007) Plant species composition and productivity following permafrost thaw and thermokarst in Alaskan tundra. Ecosystems 10 280 292. Schuur EAG, Bockheim J, Canadell JG et al. (2008) Vulnerability of permafrost car bon to climate change: implications for the global carbon cycle. BioScience 58 701. Schuur EAG, Vogel JG, Crummer KG, Lee H, Sickman JO, Osterkamp TE (2009) The effect of permafrost thaw on old carbon release and net carbon exchange from tundra. Nature 459 556 9. Schuur EAG, Abbott BW, Bowden WB et al. (2013) Expert assessment of vulnerability of permafrost carbon to climate change. Climatic Change 119 359 374. Shaver GR, Billings WD, Chapin FS, Giblin AE, Nadelhoffer KJ, Oechel WC, Rastetter EB (1992 ) Global change and the carbon Bbalance of arctic ecosystems. BioScience 42 433 441. Sistla SA, Moore JC, Simpson RT, Gough L, Shaver GR, Schimel JP (2013) Long term warming restructures Arctic tundra without changing net soil carbon storage. Nature 1 5 Sjgersten S, van der Wal R, Woodin SJ (2008) Habitat type determines herbivory controls over CO 2 fluxes in a warmer Arctic. Ecology 89 2103 16. Skogland T, Lomeland S, Goksoyr J (1988) Respiratory burst after freezing and thawing of soil: experiments with soil bacteria. Soil Biology and Biochemistry 20 Starr G, Oberbauer SF (2003) Photosynthesis of arctic evergreens under snow: implications for tundra ecosystem carbon balance. Ecology 84 1415 1420.


61 Sullivan PF, Welker JM, Arens SJT, Sveinbjrnsson B (2008) Continuous estimates of CO 2 efflux from arctic and boreal soils during the snow covered season in Alaska. Journal of Geophysical Research 113 Tarnocai C, Canadell JG, Schuur E a. G, Kuhry P, Mazhitova G, Zimov S (2009) Soil organic carbon pools in the northern circumpolar permafrost region. Global Biogeochemical Cycles 23 Tieszen LL (1974) Photosynthetic competence of the subnivean vegetation of an arctic tundra. Arctic and Alpine Research 6 253 256. Tilston EL, Sparrman T, quist MG (2010) Unfrozen water content moderates temperature dependence of sub zero microbial respiration. Soil Biology and Biochemistry 42 1396 1407. Trucco C, Schuur EAG, Natali SM, Belshe EF, Bracho R, Vogel J (2012) Seven year trends of CO 2 exchange in a tundra ecosystem affected by long term permafrost thaw. Journal of Geophysical Research 117 1 12. Trumbore SE, Czimczik CI (2008) An uncertain future for soil carbon. Science (New York, N.Y.) 321 1455 6. Trumbore SE, Harden JW (1997) Accumulation and turnover of carbon in organic and mineral soils of the BOREAS northern study area. Journal of Geophys ical Research 102 817 28. Ueyama M, Iwata H, Harazono Y, Euskirchen ES, Oechel WC, Zona D (2013) Growing season and spatial variations of carbon fluxes of Arctic and boreal ecosystems in Alaska ( USA ). Ecological Applications 23 1798 1816. Vogel J, Sc huur EA, Trucco C, Lee H (2009) Response of CO 2 exchange in a tussock tundra ecosystem to permafrost thaw and thermokarst development. Journal of Geophysical Research 114 Webb E., Pearman G., Leuning R (1980) Correction of flux measurements for density e ffects due to heat and water vapour transfer. Quarterly Journal of the Royal Meteorological Society 106 85 100. Welker JM, Fahnestock JT, Jones MH (2000) Annual CO 2 flux in dry and moist arctic tundra: field responses to increases in summer temperatures and winter snow depth. Climactic Change 44 139 150. 2 exchange in three Canadian High Arctic ecosystems: response to long term experimental warming. Global Change Biology 10 1981 1995 Winston G., Sundquist E., Stephens B., Trumbore SE (1997) Winter CO 2 fluxes in a boreal forest. Journal of Geophysical Research 102


62 Zimov SA, Davidov SP, Voropaev VY, Prosiannikov S., Semiletov IP, Chapin MC, Chapin FS (1996) Siberian CO2 efflux in win ter as a CO2 source and cause of seasonality in atmospheric CO2. Climactic Change 33 111 120. Zuur AF, Ieno EN, Walker NJ, Saveliev AA, Smith GM (2009) Mixed Effects Models and Extensions in Ecology with R


63 BIOGRAPHICAL SKETCH Elizabeth E. Webb earned a Bachelor of Arts from Carleton C ollege in 2009 with a major in geology and a concentration in environmental and technology s tudies. Academically, Elizabeth is interested in anthropogenic impacts on global biogeochemical cycling and local ecosyste the University of Florida, Elizabeth also participated in science outreach through PolarTREC (Teach ers and Researchers Exploring and Collaborating) and worked with middle school and high school teachers in the field to communicate high latitude science to students in the classroom.