SEARCH

SEARCH BY CITATION

Keywords:

  • Arctic tundra vegetation;
  • chamber measurements;
  • eddy covariance;
  • net ecosystem carbon exchange;
  • plant functional types

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information

[1] We characterized the tundra vegetation at three eddy covariance towers located along a toposequence in northern Alaska and studied seasonal variations in plot-level CO2 fluxes among the dominant vegetation types with chambers during the summer and with the gradient-diffusion technique during the winter. We performed footprint analyses to determine the source areas contributing to the tower fluxes and scaled plot-level to eddy-covariance CO2 data based on the proportion of vegetation types occurring within the footprints. At peak growing season, both gross ecosystem exchange and ecosystem respiration were greater in moist acidic tussock tundra and wet sedge tundra than in dry heath tundra. This resulted in relatively similar values of net ecosystem exchange as measured by chambers in July in tussock tundra across all topographic positions and wet sedge tundra (−2.4 to −4.2 μmol CO2/m2/s) but low values in dry heath tundra (−0.4 μmol CO2/m2/s). Winter respiration was highest for tussock tundra in December, but there were no significant differences among vegetation types in February and April. Net and gross ecosystem exchange scaled up from summer chamber measurements compared well to tower data (r2 = 0.84 and r2 = 0.78, respectively), especially on level terrain, whereas plot-level CO2-flux measurements in the winter did not agree well with tower data. This is one of few studies to compare plot-level and tower fluxes during both summer and winter and to demonstrate successful upscaling of carbon exchange in Arctic tundra systems under certain conditions.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information

[2] Arctic landscapes are made up of a patchwork of various plant-community types [Kade et al., 2005; Walker et al., 2005], and the magnitude of ecosystem productivity and respiration varies by vegetation type and the associated soil conditions. Understanding the influence of vegetation composition and structure on ecosystem processes is critical when predicting changes in carbon exchange due to altered climatic conditions in the future [Soegaard et al., 2000; Street et al., 2007]. For example, moist acidic tussock tundra in Arctic Alaska shows greater carbon exchange than dry heath tundra [Welker et al., 2000]. Various studies have investigated plot-level carbon exchange with chambers that provide good estimates of CO2 fluxes for specific Arctic plant-community types [e.g., Oberbauer et al., 1991; Vourlitis et al., 1993; Welker et al., 2000; Shaver et al., 2007], but these measurements are labor intensive, usually confined to discrete sampling periods and may be unsuitable for extrapolation to larger areas with patchy vegetation. Eddy covariance measurements give continuous estimates of landscape-level net ecosystem carbon exchange in Arctic systems [e.g., Weller et al., 1995; Vourlitis and Oechel, 1999; Nordstroem et al., 2001; Harazono et al., 2003; Lafleur and Humphreys, 2008], but they do not partition out the importance of different plant communities with varying structural and functional characteristics. That is, while a given type of tundra (e.g., heath, tussock or wet sedge tundra) may be characterized by a predominant surface type, in reality each tundra type exhibits a certain amount of spatial heterogeneity. This plot-scale heterogeneity may cause varying contributions to the overall landscape-scale carbon exchange measured with eddy covariance towers. Although a few studies have scaled carbon fluxes from chamber to tower data in tundra landscapes based on physiological and micrometeorological parameters [Vourlitis et al., 2000; Zamolodchikov et al., 2003; Loranty et al., 2011], we are aware of only one study that has focused on the contribution of carbon exchange from discrete vegetation types at the plot level to those of larger areas as measured with the eddy covariance technique in Arctic ecosystems [Fox et al., 2008]. Here, we ask how the carbon exchange of various Arctic tundra plant communities at the plot level compares to eddy-covariance data during both the summer and winter season. The comparison of winter fluxes offers a unique aspect in our study, as eddy covariance data during Arctic winters are very sparse [Euskirchen et al., 2012].

[3] In order to document the changes in CO2, water and energy fluxes of Arctic systems due to high-latitude warming, we established three eddy covariance towers along a toposequence from ridge-top heath tundra to mid-slope moist acidic tussock tundra to valley-bottom wet sedge tundra in the Imnavait Creek Watershed in the Low Arctic in 2007 [Euskirchen et al., 2012]. The goals of the present study are to:

[4] (a) Examine the seasonal variations in CO2 flux among discrete tundra plant communities within the footprint of the towers. We determined CO2 flux measurements of gross ecosystem exchange (GEE), ecosystem respiration (ER) and net ecosystem exchange (NEE) at the plot level during several measurement campaigns over a full annual cycle, including collection of flux data during the winter season. We hypothesized that the summer GEE and ER would show greater absolute values in the wet sedge and tussock tundra than in the dry heath tundra, based on the lack of vegetation on the frost-disturbed heath soils, while we expected less variation in flux among vegetation types during the winter months.

[5] (b) Relate the plot-level CO2 flux measurements to the eddy covariance data at heterogeneous landscape levels with the help of a footprint model that determines the source areas of the fluxes. We hypothesized that plot-level and tower estimates would agree better for NEE than for ER, since ER estimates cannot be measured directly by eddy covariance towers and are difficult to derive during Arctic summer nights with almost constant light. GEE estimates should only be as accurate as ER estimates, but since GEE represents a larger value than ER, the percent error should be smaller.

2. Methods

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information

2.1. Study Area

[6] The study site is located on a west-facing hillslope in the Imnavait Creek catchment of the Upper Kuparuk River region in the southern foothills north of the Alaskan Brooks Range (68°37′N, 149°18′W, elevation about 920 m asl), where broad valleys and rolling hills dominate the landscape. The ridge of the study hillslope is approximately 100 m wide and the valley bottom is approximately 200 m wide, with a slope angle of 5 to 10°. The study site is part of the Arctic Foothills in the zone of continuous permafrost [Péwé, 1975], where mean annual precipitation ranges from 140 to 270 mm, with about 40% falling as snow [Zhang et al., 1996]. At the Imnavait Creek catchment, the annual air temperature averaged −7.4°C between 1985 and 1993 [Stieglitz et al., 2000] and the January and July air temperatures averaged −20.2°C and 9.3°C between 2002 and 2008 [Kane and Hinzman, 2012]. The study site is situated on a drift of Sagavanirktok River glacial deposits [Hamilton, 2003], with peaty organic soils overlaying silt and till [Hinzman et al., 1991]. Bedrock outcrops and small water tracts, differences in elevation, and snow gradients allow for a relatively diverse mix of vegetation types [Walker et al., 1989]. The study site is part of the bioclimatic subzone E, which is classified as a low-shrub zone [Walker, 2000; Walker et al., 2005].

2.2. Vegetation Characterization

[7] We mapped the vegetation types within an 80-m radius of each of three eddy covariance towers that had instruments mounted at 2 m. The towers were placed along a west-facing slope on a topographic gradient. In order to characterize the vegetation in the footprint of each tower, we laid out a grid consisting of eight 10-m wide belt transects. The grid was centered at the tower and covered the entire footprint. Using grid lines for orientation, we manually mapped vegetation composition of every meter within the grid, observing up to 5 m at a time. We discerned the boundaries of the plant communities (based on the vegetation classification by Walker et al. [2005]) by eye, recorded the perimeters of the plant communities on paper and then digitized the maps. The character of the vegetation occurring within the footprint of the flux towers changed drastically along the topographic gradient (Figure 1). The ridge top of the study area was dominated by a mix of dry heath tundra with signs of active frost heaving and stable tussock tundra with localized spots of moist sedge meadows. The west-facing mid-slope site was covered by typical moist acidic tussock tundra (Sphagno-Eriophoretum vaginati [Walker et al., 1994]) interspersed with thickly vegetated hummocks. The vegetation at the valley bottom was made up of shrubby tussock tundra growing on elevated polygonal ground, wet sedge tundra with thick moss mats dominated by several Sphagnum species, and water-logged sedge tundra lacking moss mats. A small beaded stream draining the watershed ran along the valley bottom.

image

Figure 1. Vegetation types found within an 80-m radius of the eddy covariance towers located at the ridge top, mid-slope and valley bottom of the Imnavait Creek Watershed. The percent cover of the vegetation types and the plot locations are indicated for each tower footprint.

Download figure to PowerPoint

[8] The dry heath tundra supported less vegetation cover than the other vegetation types and was dominated by evergreen dwarf shrubs such as Cassiope tetragona, Dryas integrifolia and Vaccinium vitis-idaea, deciduous dwarf shrubs such as Betula nana, Salix phlebophylla and Vaccinium uliginosum and several fruticose lichen species (Figure 2). Bare spots due to active frost heaving were covered by a suite of crustose lichen species. In contrast, the tussock tundra was well vegetated, and evergreen shrubs such as Cassiope tetragona, Empetrum nigrum, Ledum palustre ssp. decumbens and Vaccinium vitis-idaea, deciduous shrubs such as Betula nana and Salix pulchra, the tussock-forming sedge Eriophorum vaginatum and thick moss layers consisting of Hylocomium splendens and Sphagnum sp. dominated the plant community. The sedge tundra was dominated by a suite of hydrophilic Carex species along with Andromeda polifolia, Betula nana, several Salix species and a 10-cm thick Sphagnum carpet.

image

Figure 2. Cover estimates (in % with standard errors) of plant functional types in the study plots in mid-July 2009. Cover estimates can sum to greater than 100% because multiple vegetation layers were assessed.

Download figure to PowerPoint

2.3. Environmental Parameters

[9] At each tower, we selected study areas that were representative of the major plant community types and similar in vegetation and environmental characteristics. Within these predetermined areas, we randomly selected three 0.5-m2 study plots per major plant community type for a total of fifteen sample plots. At the ridge-top tower, we chose plots in dry heath and tussock tundra, at the mid-slope tower, we selected plots in tussock tundra only, and at the valley-bottom tower, we established plots in mossy wet sedge and tussock tundra. At each of the fifteen study plots, we characterized the vegetation by estimating the cover of major plant functional types including evergreen shrubs, deciduous shrubs, graminoids, forbs, mosses and lichens [Chapin et al., 1996]. We also determined the Normalized Difference Vegetation Index (NDVI) in mid July 2009 with a Unispec SC Spectral Analyzer (PP Systems, Amesbury, Massachusetts) and calculated NDVI as (ρNIRρRed)/(ρNIR + ρRed), where ρRed and ρNIR are the reflectance values averaged over 560–680 nm and 725–1000 nm, respectively. We measured the thickness of the soil organic horizon and recorded soil temperature at 5 cm depth, volumetric soil moisture of the upper 10 cm and the depth of the thawed active layer in early June, mid-July and late August 2009 directly adjacent to the study plots. Volumetric soil-moisture content was measured with a Hydrosense probe (Campbell Scientific Inc., Logan, Utah). We also recorded ground-surface temperatures at the base of the snowpack and air temperatures in early December 2009, mid February 2010 and mid April 2010.

2.4. Plot-Level CO2-Flux Measurements

[10] We used chamber-based methods to measure ER and the light response of NEE and calculated GEE at each study plot during three measurement campaigns in June, July and August 2009. We also measured ER during three measurement campaigns in December 2009, February 2010 and April 2010. For the three snow-free measurement periods (1–4 June, 15–18 July, 20–25 August), we measured midday CO2 concentrations by connecting a clear Plexiglas chamber (0.7 × 0.7 × 0.25 m) to a LI-6400 portable infrared gas analyzer in closed-path configuration (Li-Cor Inc., Lincoln, Nebraska) and fitting the chamber to a portable rectangular base with an airtight polyethylene skirt that was sealed to the ground with a heavy metal chain [Saleska et al., 1999]. Two small fans mixed the air within the chamber, and the LI-6400 recorded internal CO2 concentrations along with humidity, temperature, barometric pressure and photosynthetic active radiation (PAR) every two seconds over a 40-s period. For each data set, only periods with stable PAR values were used to calculate net CO2 flux as described by Street et al. [2007]. At each study plot, we took two NEE measurements each under full sunlight, three levels of successive shading and complete darkness. PAR within the chamber was measured at each of the different light levels. Shading was provided with layers of fiberglass window screen material (approximately 1.5 mm mesh), and each successive layer of shading reduced the ambient light intensity by approximately 50%. To obtain complete darkness for the ER measurements, we covered the chamber with an opaque blanket. The chamber was ventilated between measurements. From these data, we constructed a light-response curve for each plot at each time point by interpolating between measured light intensities. We calculated net CO2 flux as NEE = (ρ*V/A)*(dC/dt), where ρ is air density (mol/m3), V is the chamber volume (m3), dC/dt is the rate of change in CO2 concentration (μmol/mol/s) and A is the surface area of the chamber (m2) [Saleska et al., 1999; Shaver et al., 2007; Street et al., 2007]. For this study, we report NEE values at 600 μmol photons/m2/s, because we achieved this light level consistently in the field, and we did not wish to extrapolate beyond the measured values of PAR. GEE was calculated as the difference between NEE and ER. We use negative GEE and NEE values to indicate carbon uptake by the vegetation, according to the micrometeorological sign convention.

[11] During the winter months, we estimated diffusional CO2 flux through the snowpack to the atmosphere based on Fick's Law of Diffusion as described by Musselman et al. [2005]. We determined CO2 concentrations at the surface and the base of the snowpack with an LI-820 CO2 analyzer (Li-Cor Inc., Lincoln, Nebraska) attached to a sturdy, hollow metal probe with a perforated tip that housed 3.2 mm polyethylene tubing [Sullivan, 2010]. We carefully inserted the probe into the snowpack to avoid disturbance. We recorded CO2 concentrations and average snowpack temperature and density on 2 December 2009, 18 February 2010 and 16 April 2010. We then employed the diffusion-gradient technique [e.g., Fahnestock et al., 1998; Schindlbacher et al., 2007; Sullivan, 2010] to estimate respiration.

2.5. Eddy-Covariance CO2-Flux Measurements

[12] The eddy covariance towers were outfitted with LI-7500 open-path infrared gas analyzers (Li-Cor Inc., Lincoln, Nebraska) and CSAT 3-D sonic anemometers (Campbell Scientific Instruments, Logan, Utah), which were mounted 20 cm apart to minimize flow distortion and flux loss. The instruments were connected to digital data logging systems, and data were post-processed into 30-min intervals [Euskirchen et al., 2012]. We applied the correction for instrument heating during cold temperatures as recommended by Burba et al. [2008]. NEE based on eddy covariance data can be partitioned into GEE and ER using ‘nighttime’ measurements (PPFD < 50 μmol/m2/s) fit to the equation ER = R0 * Q10 Ta/10, where Ta is air temperature, R0 is a scale parameter and Q10 is the temperature sensitivity coefficient of ER. We estimated R0 and Q10 each day using a 29-day moving window and least squares method [Ueyama et al., 2009] and calculated GEE as GEE = NEE − ER. We calculated ER and GEE during the growing season from June through August. To compare the chamber and tower data, we averaged the tower data over the time spans that corresponded to the midday sampling periods of the plot-level data (11:00 to 17:00) and compared these values to the plot-level flux estimates at the average PAR levels recorded by the towers. As the mid-slope tower was not connected to a year-round power source, our comparisons between chamber and tower data were restricted to the ridge-top and valley bottom sites in the winter.

[13] Although the vegetation types showed a relatively even distribution pattern within the mapped areas around the towers and the plot-level measurements were only taken on days under calm conditions with average wind speeds less than 3 m/s, we performed footprint analyses using the flux source-area model described by Kormann and Meixner [2001] to take into account wind direction and spatial heterogeneity of the tundra landscape. For each measurement campaign, we used the ‘footprint’ procedure in EdiRe (EdiRe, University of Edinburgh, 2011) to calculate the time-average source areas from which 95% of the CO2 fluxes at each tower were derived over the same time period that chamber measurements were taken. We superimposed the source-area ellipses over the vegetation maps in ArcMap (ESRI ArcMap, 2010) and calculated the percentage area of the major plant-community types that contributed to the tower fluxes. We then scaled the plot-level CO2-flux data to the eddy covariance measurements by summing the contribution of the dominant vegetation types weighted according to the percentage vegetation cover mapped within each tower footprint.

2.6. Data Analysis

[14] Data were analyzed using SAS (SAS Institute Inc., 2004). When including data from all measurement campaigns for each variable, the data met the assumptions of normality and homogeneity of variance. We analyzed GEE, ER and NEE with a mixed-model repeated measures design, treating vegetation types found at each site as a random effect and time as a fixed effect. Time and the interaction between time and vegetation type at each site had significant effects on the response pattern for all variables at α = 0.05, so we performed individual data analyses for each measurement campaign separately. As the smaller data sets did not meet the assumptions of normality, we used Kruskal-Wallis one-way analyses of variance to test for differences among vegetation types found at each site. The statistical power was sufficient to detect significant effects, and we used the Games-Howell test, which also does not assume a normal population, to estimate differences among vegetation types found at each site when the Kruskal-Wallis results indicated significant effects. We generated multiple linear regression models to see whether environmental characteristics or plant functional type composition had a significant effect on each of the response variables at α = 0.05 and used the minimum Akaike's Information Criterion to select the best models [Anderson et al., 2000]. Significant environmental variables were then used in individual analyses of covariance (ANCOVA) to determine whether the covariates reduced the significance of vegetation types at each site.

3. Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information

3.1. Environmental Parameters

[15] The dry heath tundra supported less biomass than the other vegetation types and showed the lowest NDVI values of all study plots at peak growing season (0.51 ± 0.03, Table 1). Due to the thin vegetation mat, the soils had shallow organic horizons (2.3 ± 0.33 cm) with relatively warm upper-soil temperatures (12.8 ± 0.5°C at 5 cm depth in mid-July) and deep active layers (67.7 ± 2.5 cm). In contrast, the tussock tundra was well vegetated, with average NDVI values ranging from 0.63 at the ridge top to 0.75 at the valley bottom, where willows were abundant. Relatively thick, unsaturated soil organic horizons (11.7–18.7 cm) insulated the soil and resulted in relatively cool upper-soil temperatures (10.1–10.5°C at 5 cm depth in mid-July) and shallow thaw depths (35.7–43.1 cm). The sedge tundra exhibited high NDVI values of 0.71 ± 0.02 at peak growing season. The soil moisture in this vegetation type (89.4 ± 5.3% by volume) was about twice as high as at the other plant communities and was associated with the accumulation of thick organic horizons (34.0 ± 2.6 cm), possibly due to lower decomposition rates at this site. This saturated substrate had warm soils in the summer (14.3 ± 1.0°C) and deep thaw depths (67.3 ± 4.2 cm). In early December, ground-surface temperatures at the base of the snowpack and snow depths ranged from −3.1 ± 0.5°C and 23.7 ± 2.2 cm at the heath to −1.3 ± 0.6°C and up to 33.6 ± 3.7 cm at the tussock plots (Table 2). In mid-April, ground-surface temperatures ranged from −4.7 ± 0.6°C at the ridge-top tussock plots to −6.0 ± 0.7°C at the valley-bottom tussock plots, and maximum snow depth was shallowest at the heath plots (40.0 ± 3.1 cm) and deepest at the surrounding ridge-top tussock tundra (54.2 ± 3.9 cm).

Table 1. Mean and Standard Error of Selected Environmental Characteristics Measured at the Various Vegetation Types at Each Sitea
Environmental VariableHeathTussock (Top)Tussock (Mid-Slope)Tussock (Bottom)Sedge
  • a

    Values for the tussock community are reported for the ridge-top, mid-slope, and valley-bottom sites. NDVI, soil moisture, summer soil temperature at 5 cm depth and thickness of the soil organic mat were determined during the peak-growing season in mid-July 2009. Maximum thaw depth was recorded in early September 2009.

NDVI0.51 ± 0.030.63 ± 0.020.68 ± 0.010.75 ± 0.020.71 ± 0.02
Depth of SOM (cm)2.3 ± 0.315.3 ± 1.711.7 ± 1.218.7 ± 2.734.0 ± 2.6
Soil moisture (vol. %)39.1 ± 5.752.1 ± 5.249.0 ± 9.645.2 ± 5.189.4 ± 5.3
Soil temperature (°C)11.8 ± 0.510.5 ± 1.510.2 ± 1.610.1 ± 1.213.3 ± 1.0
Thaw depth (cm)67.7 ± 2.537.7 ± 5.043.1 ± 0.635.7 ± 3.567.3 ± 4.2
Table 2. Mean and Standard Error of Winter Ground-Surface Temperature at the Base of the Snowpack and Snow Depth Measured at the Various Vegetation Types at Each Site
DateHeathTussock (Top)Tussock (Mid-Slope)Tussock (Bottom)Sedge
Ground Temperature (°C)
02 December 2009−3.1 ± 0.5−1.4 ± 0.4−1.3 ± 0.6−1.4 ± 0.4−2.6 ± 0.4
18 February 2010−5.5 ± 0.4−4.8 ± 0.7−5.0 ± 0.6−6.1 ± 0.6−4.9 ± 0.3
16 April 2010−5.2 ± 0.5−4.7 ± 0.6−5.6 ± 0.7−6.0 ± 0.7−4.9 ± 0.8
 
Snow Depth (cm)
02 December 200923.7 ± 2.231.1 ± 3.133.6 ± 3.728.5 ± 2.225.8 ± 2.1
18 February 201037.6 ± 3.246.2 ± 2.239.4 ± 3.232.0 ± 4.041.7 ± 3.1
16 April 201040.0 ± 3.154.2 ± 3.944.3 ± 3.340.9 ± 3.151.3 ± 3.8

3.2. Plot-Level CO2 Fluxes

[16] The ANCOVAs showed that only thaw depth had a significant covariate effect on ER in July 2009 (F = 21.4, p = 0.001). The vegetation types found at each site had a significant effect on plot-level GEE, ER and NEE (p < 0.1) during all measurement campaigns except in February and April (Table 3). During the growing season, GEE at 600 μmol photons/m2/s was significantly more negative at the mid-slope and valley-bottom sites than at the ridge-top dry heath community (Figure 3), indicating greater carbon uptake by the vegetation. The mid-slope tussock and valley-bottom wet sedge tundra reached GEE values of −9.6 ± 0.1 and −9.5 ± 0.2 μmol CO2/m2/s in mid July, respectively, but only −3.1 ± 0.2 and −5.0 ± 1.5 μmol CO2/m2/s in late August. Similarly, ER during the growing season was consistently highest in the wet sedge tundra (6.3 ± 0.3 μmol CO2/m2/s in mid July) and lowest in the dry heath and ridge-top tussock tundra (2.9 ± 0.5 and 2.5 ± 0.5 μmol CO2/m2/s in mid July, respectively). This resulted in relatively similar NEE values across vegetation types, except for low absolute peak-season NEE at the dry heath tundra (−0.4 ± 0.2 μmol CO2/m2/s). The tussock tundra at all three locations along the toposequence was the only community type to show net carbon uptake in early June, ranging from −0.2 ± 0.1 to −0.3 ± 0.1 μmol CO2/m2/s, and generally maintained greater (more negative) net uptake values throughout the growing season than the other two vegetation types.

Table 3. Results of Kruskal-Wallis One-Way Analyses of Variance for Gross Ecosystem Exchange, Net Ecosystem Exchange and Ecosystem Respiration Comparing Different Vegetation Types at Each Site (Independent Variable) for Each Seasonal Time Pointa
Response VariableDFχ2-Value
  • a

    Double asterisk (**) indicates p < 0.01; single asterisk (*) indicates p < 0.05; dagger symbol (†) indicates p < 0.1; “ns” indicates non-significant.

Gross Ecosystem Exchange
June 200949.90*
July 2009412.63**
August 200948.13†
 
Net Ecosystem Exchange
June 2009411.47*
July 200949.83*
August 200947.80†
 
Ecosystem Respiration
June 2009412.10*
July 2009411.17*
August 200948.97†
December 2009412.37*
February 201043.83 ns
April 201044.90 ns
image

Figure 3. Means and standard errors for (a) gross ecosystem exchange, (b) ecosystem respiration and (c) net ecosystem exchange of the various vegetation types at each site measured during early, mid- and late summer. Values for (a) and (c) are estimates at 600 μmol photons/m2/s. Letters denote differences between the vegetation types within each summer month as indicated by the non-parametric Games-Howell test.

Download figure to PowerPoint

[17] Multiple regressions showed that peak-season GEE had a strong negative correlation with the presence of graminoids and deciduous shrubs (Table 4), where high GEE, expressed as a negative value, corresponded with high cover of graminoids and deciduous shrubs. NDVI explained much of the variation in peak-season carbon uptake, while thaw depth and the thickness of the organic mat had a strong positive correlation with ER.

Table 4. Results of Multiple Linear Regressions of Plant Functional-Type Cover and Environmental Variables on Gross Ecosystem Exchange (GEE), Net Ecosystem Exchange (NEE) and Ecosystem Respiration (ER) Using Akaike's Information Criterion for Model Selectiona
Response VariableModel ParametersAdjusted Model r2
  • a

    Symbol [+] indicates positive relation to response variable; symbol [−] indicates negative relation to response variable. Significance level of parameter estimate: triple asterisk (***) indicates p < 0.001; double asterisk (**) indicates < 0.01; single asterisk (*) indicates < 0.05; and a dagger symbol (†) indicates < 0.1.

Plant Functional Types
GEE  
June 2009Decid. shrubs [−]†0.06
July 2009Lichens [+]***, forbs [+]**, graminoids [−]†0.83
August 2009Graminoids [−]**, forbs [+]†0.38
NEE  
June 2009Decid. shrubs [−]†, graminoids [+]†0.47
July 2009Decid. shrubs [−]***, graminoids [−]***0.83
August 2009Lichens [+]†0.09
ER  
June 2009Graminoids [+]**0.48
July 2009Lichens [−]*, decid. shrubs [−]†,forbs [−]†, graminoids [−]†0.47
August 2009Graminoids [+]*, forbs [−]†0.24
 
Environmental Variables
GEE  
June 2009Soil temp. [+]†, NDVI [−]†0.18
July 2009NDVI [−]**, thaw depth [−]*, soil moist. [+]†, SOM [−]†0.77
August 2009SOM [−]**0.48
NEE  
June 2009Soil moist. [+]***, NDVI [−]**, soil temp. [+]**0.68
July 2009NDVI [−]***0.71
August 2009Thaw depth [+]0.04
ER  
June 2009NDVI [−]***, soil moist. [+]***, soil temp. [+]**, SOM [+]*0.92
July 2009Thaw depth [+]**, NDVI [+]*, soil moist. [−]†0.66
August 2009SOM [+]**0.5

[18] During the winter months, we consistently detected ER when measuring CO2 concentrations at the base of the snowpack. In early December, the tussock tundra plots showed the highest CO2 efflux rates ranging from 0.09 ± 0.01 to 0.15 ± 0.01 μmol CO2/m2/s, while efflux was lowest at the dry heath tundra (0.02 ± 0.01 μmol CO2/m2/s, Figure 4). During the later measurement campaigns, we found no differences in ER rates among vegetation types found at each site. Average ER dropped to 0.01 μmol CO2/m2/s in mid February and rose to 0.02 μmol CO2/m2/s in mid April.

image

Figure 4. Mean respiration rates and standard errors of the various vegetation types at each site measured at three points during the winter season. Letters denote differences between the vegetation types within each winter month as indicated by the non-parametric Games-Howell test.

Download figure to PowerPoint

3.3. Scaling to Eddy Covariance Data

[19] Averaged over the summer measurement campaigns, the source-area reaches calculated with EdiRe were largest at the ridge-top tower (2,605 ± 450 m2), intermediate at the mid-slope tower (1,437 ± 220 m2) and smallest at valley-bottom tower (856 ± 122 m2). While the predominant wind direction during the summer measurements fluctuated between north and south, the wind during the winter measurements blew from the north. Over the course of the growing season, the aerial contributions from tussock and dry heath tundra to the total flux footprint measured at the ridge-top tower were consistently 67–69% and 26–27%, respectively (Table 5). During the winter, the aerial contributions of these vegetation components were slightly more variable, ranging from 65 to 70% for tussock tundra and 23–34% for dry heath tundra. Tussock tundra was the major vegetation type at the mid-slope site and contributed 95–96% to the aerial mid-slope tower footprint during the growing season. In contrast, the aerial contributions of the dominant vegetation types to the valley-bottom footprint showed the greatest variation during the six measurement campaigns, ranging from 47 to 84% for tussock tundra and 16–42% for mossy wet-sedge tundra.

Table 5. Aerial Contributions (in m2) of the Various Plant-Community Types to the Eddy-Flux Source-Area Reaches Measured at Each Tower During the Six Measurement Campaignsa
DateBare Soil/ Frost BoilHummockDry HeathMoist Sedge MeadowMoist Acidic Tussock TundraWet Sedge Tundra (No Moss)Wet Sedge Tundra (Moss)Total
  • a

    The source-area reaches contributed 95% to the CO2 fluxes measured at each tower. The mid-slope tower did not collect data during the winter months.

Ridge Top
June 200924655315492231
July 20092219212623343502
August 200923612614442085
December 200938722297963989093
February 20102056842720352951
April 201030923617922751
 
Mid-Slope
June 2009156616351716
July 2009135115181579
August 093499521004
 
Valley Bottom
June 20097995149953
July 200940756149612
August 200926711032271003
December 20099711783991548
February 2010116028110352476
April 201023631725613096

[20] Over the course of the summer, plot-level CO2 flux data showed generally higher carbon uptake and respiration than eddy covariance data (Figure 5). The two different methods of estimating carbon flux give comparable values for GEE (r2 = 0.78, y = 0.75x + 0.00) and NEE (r2 = 84, y = 0.90x + 0.23), with slopes relatively close to 1 and intercepts close to 0 indicating an approximate 1:1 relationship between chamber and tower data. We found a poor correlation between tower and chamber estimates of ER (r2 = 0.30, y = 0.30x + 0.94); however, this weak relationship was driven by the low ER value estimated by the mid-slope tower in July, and omitting this data point improved the correlation greatly (r2 = 0.89, y = 0.77x + 0.27). The differences between chamber and tower-derived CO2-flux data were greater at the mid-slope site (1.25 μmol CO2/m2/s difference in NEE between chamber and tower estimates) than at the ridge top and valley bottom (0.26 and 0.13 μmol CO2/m2/s difference in NEE between chamber and tower estimates, respectively).

image

Figure 5. Comparison of (a) gross ecosystem exchange, (b) ecosystem respiration and (c) net ecosystem exchange derived from eddy covariance towers and scaled up from plot-level measurements at the three study sites located along a toposequence. Means and standard errors are shown. Tower measurements and PAR were averaged over the time periods of plot-level measurements (11:00–17:00 LT), and plot-level values were determined from light-response curves at corresponding PAR values from the towers. Plot-level values were weighted according to the percent cover of the dominant vegetation types at each site.

Download figure to PowerPoint

[21] Scaled-up winter efflux data that were obtained from the diffusion-gradient technique at the plot level compared well to the data derived for the ridge-top and valley-bottom towers in early December 2009 (Table 6). The difference in ER between plot-level and tower estimates at the ridge top and valley bottom were 0.03 and 0.01 μmol CO2/m2/s, respectively. However, the discrepancy between the two measurement techniques grew larger in mid- and late-winter, with tower ER being 1.10 and 0.99 μmol CO2/m2/s larger at the ridge top and valley bottom, respectively, than ER values scaled up from plot-level data in April 2010.

Table 6. Comparison Between Winter ER (in μmol CO2/m2/s) Measured With Eddy Covariance Towers and Scaled up From Plot-Level Data Obtained From the Diffusion-Gradient Techniquea
DateRidge TopMid-SlopeValley Bottom
TowerPlotTowerPlotTowerPlot
  • a

    Tower ER was averaged over the time periods of plot-level measurements. Plot-level ER was scaled up by weighting fluxes from the dominant vegetation types according to the source-area fractions determined with the footprint model. Means and standard errors are shown.

December 20090.10 ± 0.030.07 ± 0.01n/a0.15 ± 0.010.10 ± 0.120.09 ± 0.01
February 20100.43 ± 0.020.01 ± <0.01n/a0.01 ± <0.010.38 ± 0.030.01 ± <0.01
April 20101.12 ± 0.050.02 ± <0.01n/a0.02 ± <0.011.01 ± 0.040.02 ± <0.01

4. Discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information

4.1. Effect of Vegetation Type and Site Factors on CO2 Flux

[22] Local landscape patterns of both plant community composition and soil variables drive predictions of overall carbon exchange in Arctic tundra. In our study, the cover of deciduous shrubs and graminoids, found on stable soils without frost disturbance, was an important compositional factor determining the difference in carbon fluxes among plant community types (Table 4). For example, we found relatively similar values of mid- and late-summer NEE for tussock and wet sedge tundra (both with a high abundance of deciduous shrubs, sedges and mosses) but drastically lower carbon uptake at the lichen-rich dry heath plots in July. Along with other studies [e.g., Welker et al., 2000; Heijmans et al., 2004], our data support the idea that plant functional types are important in explaining differences in carbon fluxes among plant communities in Arctic tundra in the summer, although we cannot distinguish between differences in photosynthetic capacity by individual species versus differences in biomass and leaf area as we did not measure these variables in the field.

[23] Vegetation type had a significant effect on ER in early winter, with tussock tundra showing two to six times greater CO2 efflux than the other vegetation types. In support, Fahnestock et al. [1998] and Grogan and Jonasson [2006] found that wintertime CO2 efflux varied significantly among Arctic and Subarctic tundra vegetation types. Interestingly, we found relatively low ER at the sedge plots when compared to the other vegetation types, although these plots had relatively high summer ER. An ice layer that formed at the soil surface of the saturated sedge plots likely impeded CO2 diffusion and may have prohibited the snow probe from accurately sampling the CO2 pocket at the base of the snowpack, leading to an underestimation of CO2 efflux. We found that ER differed more between early and late winter than among vegetation types, which supports our original hypothesis that winter ER would differ little among vegetation types. Similarly, Jones et al. [1999] report that winter CO2 efflux varied along a climate gradient in Arctic Alaska but did not differ among the various types of tundra communities. Although we did not study the source of wintertime CO2 fluxes, the decrease in efflux over the course of the winter could be caused by a decline in microbial activity due to cooling soil temperatures and a reduced transport of stored soil CO2 [Panikov et al., 2006]. The similarity of effluxes among vegetation types during mid- and late winter may potentially result from an overriding effect of soil CO2 transport processes.

4.2. Comparison between Plot-Level and Tower-Derived CO2 Fluxes

[24] Only a few studies have compared plot-level chamber measurements to eddy flux data in Arctic tundra. Our study shows a high correlation between NEE and GEE derived from chambers and towers, while chamber ER did not agree as well with the tower data. Similarly, Zamolodchikov et al. [2003] found that chamber and tower measurements in the Siberian tundra compared well for NEE and GEE but not for ER (r2 = 0.6), and Loranty et al. [2011] produced a robust model to predict NEE for tundra systems but had difficulties with ER model formulations. However, our correlations between chamber and tower ER improved drastically to r2 = 0.89 when the July data from the mid-slope site were omitted. At the mid-slope site, our peak-season tower ER average of 1.3 μmol CO2/m2/s was lower and our chamber ER average of 5.9 was substantially larger than ER values of moist tussock tundra published in other studies, which range from 2 to 3 μmol CO2/m2/s [Williams et al., 2006; Oberbauer et al., 1991]. Both the chamber and eddy-covariance technique are associated with biases, which may have contributed to the discrepancy in the scaled July ER data at the mid-slope site. The eddy-covariance technique may not give reliable data under stable atmospheric conditions and does not measure ER directly [Fox et al., 2008], making it difficult to partition NEE into GEE and ER, particularly in the Arctic with constant daylight during the growing season. In contrast, chambers can increase the internal relative humidity and temperature and diffuse the light more, thus modifying the environmental conditions the plants are exposed to [Fox et al., 2008; Kutzbach et al., 2007]. The instant artificial darkening of the chambers may result in marginally larger ER estimates due to the lag in shutdown of the plant photosynthetic machinery during instant darkening [Larcher, 2001]. However, measurement biases should have been similar at all sites, which suggests that a unique situation occurred at this site only. As we could not detect any errors in the tower and chamber data, we can only hypothesize that subsurface variations in organic matter thickness and soil moisture may have contributed to a spatial heterogeneity in ER that we did not capture with the chambers. Interestingly, although the relative contributions of mossy wet sedge and tussock tundra to the total flux measured at the valley-bottom tower varied over the course of the growing season (Table 5), the CO2 fluxes between the chamber and eddy covariance measurements were generally in good agreement (Figure 5). This suggests that the placement of the chambers adequately captured the heterogeneity in the vegetation and shows that spatial variability in vegetation types can be successfully integrated over larger spatial scales.

[25] Flux data obtained from the diffusion-gradient technique compared well with tower data in early winter, but while the plot-level effluxes declined toward mid-winter, eddy covariance data showed an increase in efflux over the course of the winter (Table 6). This discrepancy may partly be explained by the biases associated with the diffusion-gradient technique, which assumes a homogeneous snowpack and uses average snow density as a proxy for porosity [Björkman et al., 2010] and which is susceptible to enhanced CO2 transport via pressure pumping on windy days [Bowling and Massman, 2011]. However, ice layers, such as the one we encountered at the wet sedge plots, may form and impede the transport of CO2 through the snow, resulting in trapped pockets of CO2 at the base of the snowpack. Although Schindlbacher et al. [2007] and Björkman et al. [2010] anticipate that impermeable ice layers will inflate estimates of soil CO2 production and efflux, we likely underestimated CO2 efflux at the wet sedge plots as we could not penetrate the ice layer with our measurement probe to reach potentially trapped CO2. Our plot-level April efflux measurements of 0.02 μmol CO2/m2/s at the tussock plots were low but comparable to other studies that used a similar sampling technique in Alaskan tussock tundra in late spring, ranging from about 0.03 μmol CO2/m2/s [Fahnestock et al., 1998] to 0.08 μmol CO2/m2/s [Jones et al., 1999]. Thus, we are not certain at this point whether the tower or diffusion-gradient estimates approximate the “true” fluxes.

[26] Due to logistical difficulties, we were restricted to three replicates per vegetation type at each tower and performed plot-level flux measurements at six points in time over the course of the year. Other studies in Arctic tundra that successfully scaled plot-level chamber fluxes to eddy covariance towers are based on only small sample sizes. For example, Fox et al. [2008] and Zamolodchikov et al. [2003] used only two and three replicate plots per vegetation type, respectively, for chamber flux measurements, which they used to scale to eddy covariance data. Although based on a relatively small sample size, our study contributes to the understanding of estimating and partitioning carbon fluxes with small chambers and the wintertime diffusion-gradient technique at the plot level versus large-scale eddy covariance towers. Excluding the aberrant ER July measurement at the mid-slope site, our data show that growing-season GEE, ER and NEE can be successfully scaled from plot-level chambers to eddy flux towers with the help of detailed vegetation maps and footprint analyses. In contrast, the diffusion-gradient technique did not compare well to the tower data during the later winter months.

5. Conclusion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information

[27] An ongoing challenge to understanding system-wide Arctic biogeochemistry is to adequately capture the effect of spatial heterogeneity in tundra systems when estimating trace-gas fluxes. One approach to this problem is to link multiple types of trace-gas flux data at varying temporal and spatial scales. Here, we demonstrated that (a) fluxes of GEE, ER and NEE differ across a topographic gradient that encompasses three distinct types of tundra vegetation, (b) using footprint models and simple scaling principles, chamber and eddy covariance estimates of GEE, ER and NEE across different tundra types can show good agreement and (c) using the same scaling principles, this relationship does not hold as well for the comparison between the diffusion-gradient technique and flux towers in the winter. Although detailed characterizations of plant community types and CO2-flux measurements at the plot level should remain important tools for ground-truthing and estimating carbon exchange at larger spatial scales, our study is one of few to demonstrate successful upscaling of carbon exchange in Arctic tundra systems.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information

[28] This work was supported by U.S. National Science Foundation grant OPP 0632264. We would like to thank D. Dubie, B. Petitpierre and the participants of the Field Course in Arctic Science for valuable assistance in the field and G. Shaver and P. Sullivan for graciously loaning flux equipment.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information
FilenameFormatSizeDescription
jgrg992-sup-0001-t01.txtplain text document1KTab-delimited Table 1.
jgrg992-sup-0002-t02.txtplain text document1KTab-delimited Table 2.
jgrg992-sup-0003-t03.txtplain text document1KTab-delimited Table 3.
jgrg992-sup-0004-t04.txtplain text document2KTab-delimited Table 4.
jgrg992-sup-0005-t05.txtplain text document1KTab-delimited Table 5.
jgrg992-sup-0006-t06.txtplain text document1KTab-delimited Table 6.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.