Responses of CO2 flux components of Alaskan Coastal Plain tundra to shifts in water table



[1] The Arctic stores close to 14% of the global soil carbon, most of which is in a poorly decomposed state as a result of water-saturated soils and low temperatures. Climate change is expected to increase soil temperature, affecting soil moisture and the carbon storage and sink potential of many Arctic ecosystems. Additionally, increased temperatures can increase thermokarst erosion and flooding in some areas. Our goal was to determine the effects that water table shifts would have on the CO2 sink potential of the Alaskan Coastal Plain tundra. To evaluate the effects of different water regimes, we used a large hydrological manipulation at Barrow, Alaska, where we maintained flooded, drained, and intermediate water levels in a naturally drained thaw lake basin over a period of three seasons: one pretreatment (2006) and two treatment (2007–2008) seasons. To assess CO2 flux components, we used 24 h chamber-based measurements done on a weekly basis. Increased water table strongly lowered ecosystem respiration (ER) by reducing soil oxygen availability. Flooding decreased gross primary productivity (GPP), most likely by submerging mosses and graminoid photosynthetic leaf area. A decrease in water table increased GPP and ER; however, the increase in root and microbial activity was greater than the increase in photosynthesis, negatively affecting net ecosystem exchange. In the short term, ER is the CO2 flux component that responds most strongly to changes in water availability. Our results suggest that drying of the Alaskan Coastal Plain tundra in the short term could double ER rates, shifting the historic role of some Arctic ecosystems from a sink to a source of CO2.

1. Introduction

[2] The Arctic is characterized by the presence of large amounts of soil organic carbon (SOC) and unique hydrologic conditions. The highly anoxic soils, low soil temperatures and microbial activity, and slow turnover of the organic matter have resulted in the slow accumulation of nearly 14% of the global soil carbon in the Arctic, most of which is in a labile state [Post et al., 1982]. Recent estimates of SOC in the North American Arctic region suggest a carbon pool of 98.2 Gt, with 19.2 Gt in the surface layer, 42.1 Gt in the subsurface active layer and 36.9 Gt in the permafrost [Ping et al., 2008]. In some Arctic areas, shallow permafrost, low evapotranspiration, and minimal groundwater movement contribute to the formation of a unique system of wetlands composed of ice wedge polygons, lakes, and rivers, where water is a dominant driver of ecosystem structure and function [Hinzman and Kane, 1992; Oberbauer et al., 1992; Ostendorf, 1996; Hinkel and Nelson, 2003]. For instance, on the Barrow Peninsula of the Coastal Plain of Alaska, nearly 50% of the surface consists of lakes and ponds [Hinkel et al., 2003].

[3] Changes in temperature and hydrological regimes in the Arctic are likely to negatively impact the processes that have favored carbon accumulation. In fact, they may accelerate the turnover of the labile carbon, particularly accumulated SOC in anoxic and frozen areas, as a result of an increase in the microbial component of the ER. Recent climate reconstruction models suggest that a long-term cooling trend in the Arctic has been reversed as result of contemporary warming associated with anthropogenic activities [Intergovernmental Panel on Climate Change, 2007; Kaufman et al., 2009]. As a result of climate change, drying of some tundra ecosystems is likely to increase because of the combined effect of an increase in temperature, particularly in the summer, and low annual precipitation (∼200 mm a year). Most of the precipitation is largely in the form of snow that is lost as runoff during the snowmelt.

[4] The ecosystem response to drying and flooding is likely to be complex, as plants (vascular and nonvascular) and CO2 flux components may respond independently to changes in temperature and water availability [Oberbauer et al., 1991, 1992; Sommerkorn, 2008; Sullivan et al., 2008]. For instance, an increase in water table is likely to decrease ecosystem respiration (ER) by reducing oxygen availability in the soil [Oberbauer et al., 1991; Bubier et al., 1998; Oechel et al., 1998]. However, an increase in water table could also decrease gross primary productivity (GPP), if vascular plants are flood-sensitive or if water levels were sufficient to submerge the moss layer or vascular leaf area. On the other hand, drying is likely to increase ER, although severe drying could also decrease ER [Oberbauer et al., 1992; Oechel et al., 1998]. With drying, GPP of the moss layer could decrease as result of desiccation, while GPP of the vascular plants could increase as result of increased soil aeration and their ability to access suprapermafrost water.

[5] Relatively smooth landscape features characterize the northern areas of the Alaska Arctic Coastal Plain. However, the presence of continuous permafrost affects the development of several microtopographic features that, in turn, interact with the permafrost and water table to influence ecosystem function and structure [Tieszen, 1978; Engstrom et al., 2005; Sommerkorn, 2008; Sullivan et al., 2008; P. C. Olivas et al., Effects of fine-scale topography on CO2 flux components of Alaskan Coastal Plain Tundra: Response to contrasting growing seasons, submitted to Arctic, Antarctic, and Alpine Research, 2010]. Microtopography describes landscape features with small differences in elevation that are a result of the freezing and cracking of the soil and water promoting the formation of a polygonized patterned ground [Tieszen, 1978]. Some common microtopographic features are the ice wedges, low-centered, high-centered polygons, and the polygon rims, which are the areas between the ice wedges and low-centered polygons. For instance, the low-centered polygons are features with a depressed center where drainage is poor. The low-centered polygons form as result of the expansion of a system ice wedges surrounding the polygon (overlaid by troughs) that, as they expand, they up thrust the ground, producing elevated areas (the rims) relative to the center of the polygon [Billings and Peterson, 1980]. The size of the low-centered polygons is variable, ranging from 5 to 12 m in diameter [Tieszen, 1978].

[6] The Alaskan Arctic Coastal Plain is dominated by several graminoid species and bryophytes [Walker et al., 2005], and in recent decades, increasing cover of deciduous shrubs [Sturm et al., 2001]. Graminoids and bryophytes are likely to respond differently to short- and long-term changes in water availability, thereby affecting ecosystem carbon uptake and loss [Riutta et al., 2007]. As a result of the lack of a vascular system, the productivity of mosses could be negatively affected by a decrease in the water table [Riutta et al., 2007]. In contrast, some graminoid species have the ability to grow in both wet and dry microsites; therefore, graminoids are likely to play an important role in the capacity of some Arctic ecosystems to maintain their carbon sink potential in the future. For instance, Carex spp. is a particularly important component of some Arctic Coastal Plain ecosystems because of its ability to grow in polygon rims, low-centered polygons and wet troughs.

[7] In the short term, lowering of the water table is likely to have a larger impact on the productivity of the Arctic than severe flooding, potentially changing the historic role of Arctic ecosystems as carbon sinks by increasing microbial activity [Oechel et al., 1993]. However, increased nutrient availability as a result of organic matter turnover could increase the productivity of vascular plants and offset carbon losses, but shifts in ecosystem structure and function are also expected to accompany lowering of the water table [Shaver et al., 2000; Riutta et al., 2007].

[8] Our main goal was to determine the complex interactions of drying and flooding on the individual CO2 flux components using a large-scale hydrological manipulation of a naturally drained thaw lake. Our hypotheses were that: (1) lowering the water table should increase ER and decrease GPP, (2) raising the water table should decrease ER and increase GPP, and (3) microtopography should affect the magnitude of the response of ER and GPP to the decrease and increase of the water table. For instance, ER in polygon rims should be less affected by an increase in water table than in low areas. Previously, experimental manipulation of the water table of tundra ecosystems has been attempted at the microcosm [Billings et al., 1984; Johnson et al., 1996] or plot scale [Oechel et al., 1998]. However, at those scales, thermal interactions between water level, vegetation properties and microtopography are likely obscured.

[9] We focused our CO2 assessments on plot-scale chamber-based measurements that allow evaluation of the role of microtopography in the response of the ecosystem to water level changes. In addition to water table, we examined the direct and indirect effects of other environmental parameters, such as thaw, soil and air temperature on the CO2 exchange. CO2 response was evaluated using gross primary productivity (GPP, or total photosynthesis), ecosystem respiration (ER, plant and soil CO2 losses), and net ecosystem CO2 exchange (NEE, balance between ER and GPP). The data collected included one control growing season prior to the manipulation (2006, 62 days) and two seasons of experimental manipulation (2007, 64 days, and 2008, 61 days).

2. Methods

2.1. Study Site

[10] The hydrological manipulation and carbon exchange measurements were conducted at the Barrow Environmental Observatory in Barrow, Alaska (71.32°N, 156.62°W). The site is a 62 ha naturally drained lakebed located near the northernmost point of the Alaskan Arctic Coastal Plain. The lakebed was 1.4 km by 0.3 km, longitudinally oriented in a north-south direction. Prevailing northwest winds and the proximity to the Arctic Ocean influence the climate in Barrow and the orientation of lakes. Annual precipitation is low (∼200 mm), with most precipitation as snow during winter and spring. August has the highest pluvial precipitation. The snow-free season varies, but usually begins during the first week of June and ends in September. The study area exhibits shallow summer thaw, with active layers less than 50 cm deep on average. The landscape is characterized by small thaw ponds and low- and high-centered polygons. These landscape features represent different stages of the thaw lake cycle where the constant freezing and thawing of soil creates small changes in relief that are sufficient to affect water regimes and vegetation cover [Billings and Peterson, 1980].

[11] Wet sedge communities dominate the lake basin, but some aquatic and dry vegetation types are also present. The wet sedge communities dominate the waterlogged areas of the low-centered polygons and low areas with poor drainage. Dominant species in these areas include Carex aquatilis Whalenb., Eriophorum scheuchzeri Hoppe, and Dupontia fisheri R. Br., interspersed with some Sphagnum spp. A thick layer of Sphagnum spp. dominates the areas with better drainage, such as the polygon rims, along with some individuals of C. aquatilis that usually present fewer and shorter leaves than those individuals in the moist and wet sites. The aquatic sites are dominated by Arctophylla fulva (Trin.) Rupr., C. aquatils and Dupontia fisheri R. Br.

2.2. Experimental Manipulation

[12] In March 2007, the lakebed was divided into three isolated sections (north, central and south) by installation of two cross-lake dikes constructed of plastic panels inserted below maximum depth of thaw [Zona et al., 2009]. During experimental treatment, water was added to the northernmost lake section, the central was drained, and the southern section left as reference (intermediate water level). The objective of the manipulation was to maintain the water table 15 cm below and above of that of the intermediate treatment in the drained and flooded sections, respectively. Although we maintained flooding and drying treatments, interannual differences in water availability during the growing seasons of the study affected the water table of the intermediate (reference) section between years. As a result, the water level objectives and realized values for the treatment sections differed among the years depending on overall water availability as a function of spring snow cover, rainfall, and evaporative demand. Lake sections will be referred to as north, central and south.

[13] The water manipulation was carried out by moving water from the central section (drained treatment) into the north (flooded treatment) immediately following snowmelt, using large pumps distributed among the low centered polygons. The differences in water table were achieved by allowing the combined effects of evapotranspiration and seasonal thaw to further lower the water tables of the drained section. In 2008, additional water was pumped from a nearby lake to help maintain the water tables in the north section.

2.3. Plot Setup and Microsite Classification

[14] A 200 m boardwalk was built within each treatment (north, central, and south), perpendicular to the long axis of the lakebed to allow site access while minimizing trampling effects on the tundra. Each boardwalk was divided into six blocks of 30 m (10 m at each end were used as a buffer). One plot was randomly selected within each block, for a total of 18 plots, six in each treatment.

[15] To further understand the effects of water table shifts on the CO2 flux components across the landscape, we classified the 18 plots into three vegetation groups following Olivas et al. (submitted manuscript, 2010). Plant cover, soil moisture conditions, and polygon position (if present), combined with a high-resolution digital elevation model (DEM, C. E. Tweedie et al., unpublished data, 2010), were used to separate the plots into: (1) wet sedge: high vascular cover and waterlogged soils, (2) intermediate areas: transition areas with better drainage, higher moss cover and lower vascular canopy than wet sedge areas, and (3) polygon rims: high moss cover, low vascular canopy and well drained. The resulting plot classification included eight wet sedge plots, six intermediate areas, and four polygon rims.

2.4. Microclimate, Water Level, and Thaw

[16] Atmospheric conditions, such as air and soil temperature, photosynthetically active radiation (PAR), and relative humidity, were recorded and stored every 30 min on a Campbell Scientific CR10X data logger located at the center of the lakebed. Temperature and relative humidity were measured with a CS500 sensor (Campbell Scientific, Logan UT, USA) and PAR with a quantum sensor (LI-COR Li-190). Soil temperature was measured at 2 cm deep using a thermocouple. Thaw depth and water table were measured twice a week over the course of the season at each plot. Thaw depth was measured using a metal probe to nearest centimeter. Water table was measured to the nearest tenth of a centimeter inside of perforated PVC wells (2.5 cm diameter) inserted into the permafrost. The top of the green moss layer was used as the surface reference.

2.5. CO2 Exchange Sampling

[17] NEE and ER were measured using an infrared gas analyzer (LI-6200, LI-COR Inc., Lincoln, NE) with enclosed chamber techniques following Oberbauer et al. [2007]. The photosynthesis system was calibrated prior to each sampling date using a NIST-relatable gas standard (CO2 in air). The cylindrical chamber was constructed from clear acrylic plastic (95% light transmission) with two internal fans for mixing. The chamber had a volume of approximately 80 L depending on the location of the water table (basal area 1574 cm2, 50.8 cm height). Pressure gradients were minimized by: (1) detaching the head sensor of the infrared gas analyzer from the chamber before placement on the plot to allow air to flow freely between inside to outside of the chamber, and (2) by using a small diameter tube connecting the inside and outside of the chamber. The base of the chamber was attached to a short section of PVC pipe coupling. During measurements, the chamber was coupled to a PVC pipe chamber base (46 cm diameter) permanently installed in each plot (Olivas et al., submitted manuscript, 2010). These bases facilitated rapid sealing of the chamber during CO2 flux measurements while minimizing plot damage and soil disturbance during the repeated samplings.

[18] Net ecosystem exchange (NEE) was estimated by the CO2 exchange rate of the enclosed ecosystem under ambient light. Ecosystem respiration (ER) was assessed by determining the CO2 exchange rate with the chamber covered by a heavy black cloth. Gross primary production (GPP) was calculated as the absolute difference between ER and NEE. Seasonal means of NEE, GPP, and ER assessments were used to determine the effect of the water treatments on the CO2 fluxes. Measurements of NEE and ER on all plots were made weekly during the growing season (June–August) over 24 h periods divided into six samplings of 4 h. Because of the long distance between lake sections, plots were divided into two sets of nine (three plots in each treatment) to complete measurements on all plots within the sample period.

2.6. NDVI Measurements

[19] Normalized difference vegetation index (NDVI) has been a valuable tool for use in scaling up ground-based measurements, such as carbon fluxes, to larger scales with airplane or satellite imagery [Hope et al., 1993; Vourlitis et al., 2000; Steltzer and Welker, 2006]. NDVI is a ratio defined by the reflectance of the red light (RED) and near infrared light (NIR) (NDVI = (NIR-RED)/(NIR+RED)). NDVI correlates well with green biomass and has the potential to distinguish between different vegetation types, but caution must be used when applying relationships derived from temporal variation to data based on spatial variability [La Puma et al., 2007].

[20] We used NDVI to evaluate changes in green biomass over the growing season and in response to water table manipulation. In 2006, we used digital images taken with an Agricultural Digital Camera (ADC Model 4, Dycam Inc., Woodland Hills, CA) to determine NDVI. For the analysis of the images we used software provided with the camera, BRIV-32. During 2007 and 2008 we measured NDVI with a single-channel reflectometer (UniSpec-SC, PP SYSTEMS, Amesbury, MA, USA). Reflectance from the 680 (RED) and 800 (NIR) bands were used to calculate NDVI. We used a Teflon panel as a reflectance standard. Under cloudy sky conditions, measurements of the Teflon panel were taken before each plot to reduce the effects of changing light conditions. Under clear sky conditions, measurements of the panel were done every twelve plots (approximately every 20 min). All measurements were taken horizontally above the vegetation surface in a 2 h window, bracketing solar noon to minimize shadow effects, and within a day of CO2 measurements.

2.7. Data Analysis

[21] To assess the effect of drying and flooding of the tundra, we first evaluated the effects of differences in water availability on CO2 fluxes within the three sections of the thaw lake (north, central and south). Seasonal means of the NDVI, CO2 flux components, water table, and thaw were compared in a two-way ANOVA (factors: site, 3 levels, and year, 3 levels). Subsequently, we evaluated the effect of microtopography on the relationship between water table and CO2 flux components. CO2 flux components were regressed against water table changes using Standard Least Squares (JMP® 7.0.2, SAS Institute). Seasonal means of the NDVI, CO2 flux components, water table, and thaw were compared in a two-way ANOVA (factors: microtopography, 3 levels, and year, 3 levels). For the pairwise comparisons, we used Least Significant Difference. Means are accompanied by plus or minus one standard error of the mean. The analysis was done using PASW Statistics GradPack 18 (SPSS Inc., 2009).

[22] Additionally, we used PATH analysis to assess the relative importance of other environmental factors on ecosystem CO2 flux. Although similar to a multiple regression, PATH analysis is a useful tool for data analysis when causal or correlational information is known a priori about the relationships among variables. Different from a multiple regression that assumes independence of the predicting variables, PATH analysis is a more appropriate tool for data evaluation when the independence of the predictors is not certain or they are expected to be dependent. Based on our ability to assess water tables at each plot, for the PATH analyses we combined the data from all plots, years and treatments (lake sections) and treated the water table as a continuous variable rather than treatment categories. This approach allowed us to assess the carbon flux responses to periods with high, low and intermediate soil moisture and different weather conditions in addition to the flooding and drying treatments.

[23] We designed our CO2 emissions model assuming water table, thaw depth, soil and air temperature, PAR (photosynthetic active radiation), and vapor pressure deficit (VPD) as potential predictors of carbon fluxes. The interrelated paths of the variables and predictors were used to evaluate the direct and indirect effects of the predictors on the carbon flux components. Final models included only those parameters that were significant predictors.

3. Results

3.1. Microenvironment

[24] The study site presented considerable variation in weather conditions among seasons. Air temperature was lowest in 2006 followed by 2008 and 2007, even though we recorded similar PAR conditions throughout the three growing seasons (Figure 1). On average, water tables in all lake sections were higher in 2006 than in 2008, and 2007 presented the lowest water tables. The south section (intermediate or reference) presented higher water tables in 2006 than in 2007, with intermediate water tables in 2008 (Table 1 and Figure 2). Water tables in 2006 (the pretreatment season) were above the soil surface for all plots for most of the growing season (Figure 2). In 2007, (treatment year 1) the average water table of the plots along the intermediate lake section (reference) was above the surface only immediately after snowmelt. Water tables in the north and central sections in 2007 showed similar patterns to those of the south section, but were shifted higher. In 2008, water tables of the central section were slightly below those of the south section. For the same year, water tables in the north section were about 5 cm higher than those in the south section until midseason.

Figure 1.

Photosynthetic active radiation (PAR, μmol m−2 s−1) and air temperature (°C) for the three growing seasons. Values represent daily averages ±1 standard error.

Figure 2.

Water table (cm), thaw (cm), and soil temperature (at 1 cm depth) for the three growing seasons. Values for water table and thaw represent the average of all plots along each lake section, and negative values represent points below the surface. Error bars represent ±1 standard error. Soil temperature was measured at one location during the three seasons.

Table 1. Seasonal Means of Gross Primary Productivity, Net Ecosystem Exchange, Ecosystem Respiration, Normalized Difference Vegetation Index, Seasonal Water Table, and Thawa
VariableLocationSeasonal Mean ± Standard Error
  • a

    GPP, gross primary productivity; NEE, net ecosystem exchange; ER, ecosystem respiration; NDVI, normalized difference vegetation index. Flux values follow ecosystem flux notation, positive values indicate CO2 uptake, and negative values mean CO2 losses. Water table and thaw were referenced to the top of the surface or at moss layer if present. N represents the overall sample size, with the same N for each lake section within years, and six plots in each lake section. Means were compared using a two-way ANOVA. ANOVA summary: Site had a significant effect on NDVI (F = 12.274, p < 0.0001), but not on NEE, GPP, nor ER. Year had a significant effect on NEE (F = 9.274, p < 0.0001), GPP (F = 4.809, p = 0.013), ER (F = 29.416, p < 0.0001), and NDVI (F = 60.13, p < 0.0001). The interaction Site*Year was not significant for any of the CO2 flux components or for NDVI. For water table and thaw, Year was significant for water table (F = 32.459, p < 0.0001), and thaw (F = 23.816, p < 0.0001). Site was not significant for either. Interaction Year*Site was significant for water table (F = 3.014, p = 0.028), but not for thaw. Multiple comparison tests were done using Least Significant Difference. Different letters represent significant difference p < 0.05 within years, and different symbols represent significant difference p < 0.05 across years.

GPP (μmol/m2 s)north0.597 ± 0.077 (a†)0.826 ± 0.077 (a‡)0.608 ± 0.077 (a†)
 central0.621 ± 0.077 (a†)0.810 ± 0.077 (a†)0.766 ± 0.077 (ab†)
 south0.667 ± 0.077 (a†)0.823 ± 0.077 (a†)0.877 ± 0.077 (b†)
NEE (μmol/m2 s)north0.193 ± 0.052 (a†)0.068 ± 0.052 (a†)0.130 ± 0.052 (ab†)
 central0.224 ± 0.052 (a†)0.033 ± 0.052 (a‡)0.032 ± 0.052 (a‡)
 south0.207 ± 0.052 (a†)−0.021 ± 0.052 (a†)0.255 ± 0.052 (b†)
ER (μmol/m2 s)north−0.395 ± 0.060 (a‡)−0.753 ± 0.060 (a†)−0.476 ± 0.060 (a‡)
 central−0.393 ± 0.060 (a‡)−0.772 ± 0.060 (a†)−0.734 ± 0.060 (b†)
 south−0.454 ± 0.060 (a‡)−0.842 ± 0.060 (a†)−0.622 ± 0.060 (ab‡)
NDVInorth0.451 ± 0.016 (a†)0.395 ± 0.016 (a‡)0.323 ± 0.016 (a*)
 central0.432 ± 0.016 (a†)0.313 ± 0.016 (b‡)0.269 ± 0.016 (b‡)
 south0.477 ± 0.016 (a†)0.355 ± 0.016 (ab‡)0.357 ± 0.016 (a‡)
Water table (cm)north2.4 ± 1.5 (a†)−5.6 ± 1.5 (a‡)0.2 ± 1.5 (a†)
 central6.0 ± 1.5 (a†)−3.4 ± 1.5 (a‡)−4.9 ± 1.5 (b‡)
 south4.1 ± 1.5 (a†)−7.6 ± 1.5 (a‡)−1.9 ± 1.5 (ab*)
Thaw (cm)north−14.9 ± 0.9 (a†)−18.2 ± 0.9 (a‡)−22.0 ± 0.9 (a*)
 central−15.8 ± 0.9 (a†)−18.7 ± 0.9 (a‡)−18.2±0.9 (b†‡)
 south−15.8 ± 0.9 (a†)−17.8 ± 0.9 (a†)−21.2 ± 0.9 (a‡)
 sample size (N)181818

[25] The depth of thaw presented similar seasonal patterns in all three years; however the ranking of the lake sections changed among years in response to the hydrological treatments (Figure 2). On average, the thaw in the north section changed from being the shallowest in 2006 to the deepest in 2008. Conversely the central section went from being one of the deepest in 2006 to be the shallowest in 2008. The south section was generally intermediate (Figure 2). Differences within years among lake sections were not significant, except in 2008 when the central section was significantly shallower than both north (p < 0.004) and south (p < 0.021, Table 1).

[26] Soil temperatures were similar among all lake sections in 2006 (Figure 2). In 2007, temperatures were slightly warmer than those of 2006, but again were similar among the sections, especially later in the season. In contrast, soil temperatures in 2008 differed among the sections in response to hydrological treatments, with highest temperatures in the north section and lowest in the central section (Figure 2).

3.2. Seasonal Patterns of Carbon Flux Components and NDVI

[27] Values of GPP, NEE, and ER reflected differences in weather and soil water levels. The ANOVA revealed that Year (main factor) had a significant effect on all CO2 flux components. Site and the interaction Year*Site effects were not significant for any of the CO2 flux components (Table 1).

[28] On average, GPP rates were higher in 2007 than in 2006 (p = 0.004, Table 1), with intermediate rates in 2008. Within seasons, the GPP rates in 2006 and 2007 between lake sections did not differ. In 2008, the north section had lower GPP rates than the south (p = 0.017). The central section did not differ from either section. Across years, the seasonal mean GPP rates tended to be higher in 2007 for the north and central sections (only in 2007, north lake section was significantly higher), followed by 2008. The south section presented the highest rates in 2008, but difference was not significant across years (Table 1 and Figure 3).

Figure 3.

Seasonal CO2 flux components and NDVI from 2006 to 2008 within the three lake sections. Sample sizes are six plots for north, central, and south lake sections. CO2 flux components are presented using the ecosystem perspective, where positive values represent CO2 uptake and negative values loss of CO2 to the atmosphere. GPP, gross primary productivity; NEE, net ecosystem exchange; ER, ecosystem respiration; NDVI, normalized difference vegetation index.

[29] In general, the seasonal NEE indicated a positive net uptake associated with high water tables. On average, NEE rates were highest in 2006 followed by 2008 and 2007, and the south section presented the highest uptake followed by the north and central (Table 1). Lake sections did not differ significantly within years except during 2008, when the CO2 uptake of the central section responded negatively to low water table having significantly lower NEE rates than the south section (p < 0.004). The north section did not differ from the other sections. Across years, the north section did not differ. The central section presented higher rates in 2006 than in 2007 (p = 0.013) and 2008 (p = 0.012). The south section presented significantly lower rates in 2007 than in 2006 (p = 0.003) and 2008 (p = 0.001, Table 1 and Figure 3).

[30] Interannual differences in NEE largely reflected the differences in ER. ER rates were higher in 2007 (more negative) than in 2006 (p < 0.0001) and 2008 (p = 0.001). Respiratory losses were also higher in 2008 than in 2006 (p < 0.0001). On average, the north section presented lower ER rates than central and south sections (differences nonsignificant, NS). Within years, ER did not differ among lake sections, except in 2008 when the north section presented lower respiratory losses than the central section (p = 0.004). Across years, the north section had higher ER rates during periods of low water tables (2007) than during wet conditions (2006). Respiratory losses in the central section were lower in 2006 than in 2007 (p < 0.0001) and 2008 (p < 0.0001). North and south sections presented the highest respiratory losses in 2007, with no difference between 2006 and 2008 (Table 1 and Figure 3).

[31] For NDVI, the ANOVA revealed that values were higher in 2006 (0.453 ± 0.009) than in 2007 (0.355 ± 0.009, p < 0.0001) and 2008 (0.316 ± 0.009, p < 0.0001). NDVI was also significantly higher in 2007 than in 2008 (p = 0.005). The main effects (Year and Site) had significant effects on NDVI, but the interaction was not significant (Table 1). Within years, lake sections did not differ in 2006. In 2007, the north section had higher NDVI than the central section (p < 0.001), and the south section did not differ from the other two sections. In 2008, the central section had lower NDVI than the north (p < 0.02) and south (p < 0.0001) sections, and the north and south sections did not differ significantly (Table 1 and Figure 3).

3.3. Microsite Responses to Changes in Water Table

[32] We divided the study plots into the most prevalent topographic features and vegetation cover types present at the site: wet sedge and low-centered polygon centers, intermediate or transition areas, and polygon rims. The ANOVA revealed that Microtopography (main factor) had a significant effect on ER, but not on NEE, nor GPP. Year (main factor) had a significant effect on NEE, and ER, but not on GPP. The interaction Microtopography*Year was not significant for any of the CO2 flux components (Table 2).

Table 2. Results of the Two-Way Analysis of Variance Analyzing the Seasonal Measurements of Gross Primary Productivity, Net Ecosystem Exchange, Ecosystem Respiration, Normalized Difference Vegetation Index, Water Table, and Thawa
  • a

    ANOVA, analysis of variance; GPP, gross primary productivity; NEE, net ecosystem exchange; ER, ecosystem respiration; NDVI, normalized difference vegetation index; df, degrees of freedom; F, critical value; P, probability. Bold indicates significant effect at p < 0.05.

Water table

[33] On average, GPP rates in all categories tended to be lower in 2006 than in 2007, with intermediate GPP rates in 2008 (Figure 4). Within years, the polygon rims presented higher GPP rates than the intermediate and the wet sedge plots in 2006, but differences were not significant. In 2007 and 2008, all microsites presented similar rates (Figure 4). Across years, the GPP rates of wet sedge tended to be higher in 2007 than in 2006 and 2008, but did not differ from either year. The polygon rim plots presented similar GPP rates across years. The rates of intermediate plots tended to be lower in 2006 than in 2007 and 2008, but differences were not significant (Figure 4).

Figure 4.

Seasonal means for GPP, NEE, ER, water table, and thaw for all microsites during 2006, 2007, and 2008. For CO2 flux components, positive values indicate uptake, and negative values indicate loss. The zero value represents the soil surface for thaw and water table. Error bars represent ±1 standard error. N was the same for all years, with 18 plots: 8 wet sedge, 6 intermediate, and 4 polygon rims. Two-way ANOVA for (a) GPP, (b) NEE, and (c) ER. Two-way ANOVA for (d) water table and (e) thaw. Multiple comparison was done using Least Significant Difference, columns under different horizontal lines represent significant difference p < 0.05 within years, and different letters represent significant difference p < 0.05 between years.

[34] On average, all cover types showed higher growing season NEE during the wet conditions in 2006 (Figure 4). Generally, NEE rates of wet sedge and intermediate plots were similar and slightly higher than those of the polygon rims plots. Within years, NEE rates between cover types did not differ for any year. However, dry conditions in 2007 led to lower CO2 accumulation in all cover types especially in the polygon rim plots, but differences were not significant. Across years, the wet sedge plots presented higher uptake (NEE) in 2006 than in 2007 (p = 0.039), with NEE rates in 2008 not differing from either year. The intermediate and polygon rim plots presented the same pattern, where NEE rates were higher in 2006 than in 2007.

[35] On average, ER rates were higher in 2007 than in 2006 (p < 0.0001) and 2008 (p = 0.001), ER rates were also higher in 2008 than in 2006 (p < 0.0001). Generally, polygon rims presented higher respiratory losses than the wet sedge (p = 0.001) and intermediate plots (p = 0.001). Within years, lower water tables at the polygon rim plots in 2006 resulted in higher ER rates compared to those of the wet sedge (p = 0.003) and intermediate plots (p = 0.001, Figure 4), with no differences between wet sedge and intermediate plots. In 2007, vegetation cover types did not differ, regardless of much lower water tables in the polygon rim plots. Similarly, in 2008, vegetation cover types did not differ. Across years, wet sedge and intermediate plots presented similar linear responses to water table shifts (Table 3). Wet sedge presented higher ER rates in 2007 than in 2006 (p < 0.0001), and 2008 (p = 0.009). Intermediate plots presented lower ER rates in 2006 than in 2007 (p < 0.0001) and 2008 (p = 0.001), with no difference between 2007 and 2008. The polygon rim plots presented lower rates in 2006 than in 2007 (p = 0.022), with rates in 2008 not differing from either year (Figure 4).

Table 3. Linear Correlations (Adjusted R2) Between CO2 Flux Components and Water Tablea
CO2 Flux ComponentCover Type200620072008All Years
  • a

    Correlations are significant to p < 0.001, unless specified within parentheses. N represents the overall sample sizes, with eight plots for wet sedge, six for intermediate, and four for the polygon rim (PR). GPP, gross primary productivity; NEE, net ecosystem exchange; ER, ecosystem respiration.

GPPwet sedgeNS0.610.370.41
 polygon rimNSNS0.32NS
NEEwet sedgeNS0.12NSNS
 polygon rimNSNS0.320.04 (0.04)
ERwet sedge0.15 (0.024)0.640.310.53
 polygon rimNS0.22NS0.13
Nwet sedge455973177
 polygon rim22303680

3.4. Environmental Controls on Carbon Flux Components

[36] We evaluated the relationship between CO2 flux components and water table by performing correlation analysis. We found that as a result of the low variation in water table during 2006, correlations between weekly water table and CO2 flux components were not significant for the individual microsites except for ER in the wet sedge (Table 3). During the strong dry down in 2007, water table was a strong predictor of GPP and ER in the wet sedge (vascular-dominated microsites, Table 3). Correlations between water table and GPP, NEE and ER were intermediate in strength in 2008.

[37] The polygon rims (moss-dominated microsites) showed nonsignificant or low correlations with water table for all carbon flux components during all seasons (Table 3). Polygon rims also generally had the lowest water tables (Figure 4). Similar to the wet sedge microsites, the water table in the intermediate microsites best predicted GPP and ER during periods with very low water availability (e.g., 2007, Table 3). In general, water table was not a good predictor of weekly NEE rates for any of the microsites except for polygon rims in 2008 (Table 3).

[38] Correlations between weekly water table and CO2 flux components across all years combined revealed that: (1) water table is not a good predictor of NEE; (2) ER is the flux component most affected by changes in water availability followed by GPP; and (3) among all microsites wet sedge is the most responsive to changes in water table (Table 3).

[39] Although weekly assessments showed a weak relationship between water table and NEE, the seasonal NEE revealed a strong correspondence between mean seasonal water table and carbon accumulation; wet conditions presented the highest CO2 uptake. Wet sedge microsites were generally higher carbon sinks, except in 2006 when intermediate microsites were a slightly stronger sink. The CO2 balance of the polygon rim microsites was the lowest during all seasons; polygon rims had the lowest water levels in all three years. The 2007 dry conditions increased the carbon respiratory losses in all microsites, shifting the polygon rim into a source of CO2 for the growing season (Figure 4).

[40] For NDVI, Year and Microtopography had significant effects, while the effect of the interaction Year*Microtopography was not significant (Table 2). On average, the NDVI was higher in 2006 (0.453 ± 0.010) than in 2007 (0.362 ± 0.010, p < 0.0001), and 2008 (0.324 ± 0.010, p < 0.0001). NDVI in 2007 was higher than in 2008 (p = 0.014). Polygon rims (0.412 ± 0.012) presented on average, higher NDVI than the wet sedge (0.365 ± 0.009, p = 0.004) and intermediate plots (0.363 ± 0.010, p = 0.003). In 2006, microsites did not significantly differ. In 2007 and 2008, the polygon rims presented higher values than the wet sedge (p = 0.013, p = 0.009, respectively), and the intermediate plots (p = 0.007, p = 0.004). The difference between wet sedge and intermediate plots was not significant for either year. Across years, the wet sedge plots presented higher values in 2006 (0.448 ± 0.015) than in 2007 (0.343 ± 0.015, p < 0.0001) and 2008 (0.305 ± 0.015, p < 0.0001), with no difference between the last two seasons. Similarly, intermediate plots presented higher values in 2006 (0.449 ± 0.021) than in 2007 (0.410 ± 0.021, p < 0.0001) and 2008 (0.376 ± 0.021, p < 0.0001), with no difference between 2007 and 2008. Polygon rim plots were higher in 2006 (0.463 ± 0.017) than in 2008 (0.292 ± 0.017, p = 0.019), with measurements in 2007 (0.332 ± 0.017) not differing from either season.

[41] To visualize the spatial and temporal relationships between water table and carbon flux components, we combined data from all years and sections (Figure 5). We found that the response to changes in water table for GPP and ER were different. Although the highest GPP rates were associated with water tables lower than −5 cm, low rates of GPP were also observed under similar water table conditions. On average, the GPP rates were higher when water table was at or just below the surface (e.g., south section in 2008); however, the response was not as strong as in ER rates. ER rates were strongly affected by the presence of water above the surface. The highest ER rates were observed when the water table was lower than −5 cm (e.g., south and north sections in 2007, Table 1 and Figure 5). The combined response of GPP and ER to water table shifts resulted in NEE values that represent a strong sink when water is near the surface and a strong source when water is near or lower than −10 cm (e.g., south section in 2007, Table 1 and Figure 5).

Figure 5.

Seasonal pattern of CO2 flux components for growing seasons of 2006, 2007, and 2008 (18 plots). Fluxes use the ecosystem notation where positive values represent CO2 uptake and negative values represent CO2 loss from the system. Values of GPP, NEE, and ER on the same date represent diurnal patterns in response to PAR and temperature.

[42] Seasonal responses of the GPP and ER to water table were consistent throughout the growing seasons, and were particularly strong for ER. Very low GPP rates observed at the beginning of the growing season occurred when leaf areas were low and were not likely to respond to water table. Low ER rates were generally associated with high water tables (Figure 5).

[43] Water table correlations with the CO2 flux components across all seasons revealed that ER is the component most strongly controlled by water (R2 = 0.42, p < 0.001), followed by GPP (R2 = 0.24, p < 0.001). The correlation between NEE and water table was not significant.

[44] PATH analysis revealed that the predictors that significantly affect flux rates differed for each CO2 flux component. ER rates have water table and air temperature as the strongest predictors followed by PAR, VPD and thaw (an indicator of seasonality). NEE rates were most affected by thaw followed by air temperature, PAR and water table. GPP rates were most strongly affected by thaw followed by PAR and water table (Figure 6).

Figure 6.

PATH diagrams illustrate the standardized correlations (R) between predictors and CO2 flux components (unidirectional arrows) and the correlations between predictors (bidirectional arrows). Data used for the analysis include measurements for all growing seasons and sites, N = 399 (north, central, south).

4. Discussion

[45] Significant shifts in thaw depth between 2006 and 2008 and among lake sections in response to the hydrological treatments suggest that draining or flooding, combined with warmer temperatures, have the potential to increase and decrease, respectively, the active layer depth in the long term [Jorgenson et al., 2006; Zhou et al., 2009]. However, as a result of the effects of water addition and removal on depth of thaw, manipulation of the depth of water table with respect to the soil surface is not straightforward. The top of the permafrost sets the base of the water table, and because of the strong effects of water level on soil thermal conductivity and albedo, the addition and removal of water to raise and lower water tables, respectively, could act counter to the objective of raising and lowering water table relative to the soil surface. Water addition increases transfer of heat into deeper soil layers and water above the surface increases solar absorption, resulting in greater depth to the frozen layer, lowering the base of the water table with respect to the surface. Water removal increases albedo and lowers soil thermal conductivity, decreasing the thaw depth and increasing the base of the water table with respect to the surface. For instance, in 2008, the water addition and water removal resulted in the depth of thaw of the water addition treatment close to ∼4 cm below that of the water removal treatment (p = 0.004, Table 1).

4.1. Water Table Controls of Temporal and Spatial CO2 Fluxes

[46] The study site experienced seasons with very distinct weather conditions during the three years of the study (Figure 1). To our advantage we were able to assess the response of the carbon flux components to water table during a wet year (2006), a dry and warm year (2007), and a year with intermediate soil moisture and temperature (2008). These large differences resulted in very different backgroundwater levels against which the water manipulation treatments were applied.

[47] Across all years, water table was an important controller of the CO2 respiratory losses. These results follow previous findings in field and laboratory experiments on the effects of water table on CO2 flux components, especially on ER [Billings et al., 1982; Oberbauer et al., 1991, 1992; Ostendorf, 1996]. A change in NEE can be a result of multiple conditions, such as an increase in GPP, decrease in ER, or differential in the magnitude of the response of GPP and ER to altered water table. In general, weekly assessments of net ecosystem exchange did not show a significant response to water table fluctuations (Table 3) because changes in water table had effects of similar magnitude on GPP and ER resulting in little or no significant variation in net exchange rates [Sulman et al., 2009].

[48] Overall, although the highest ER rates were associated with low water tables (more than 20 cm below the surface), similar rates were also observed with water tables above −10 cm (Figure 5). This result suggests that even though soil moisture conditions were sufficient to support microbial activity at very low water tables, an important portion of the total ER comes from the topsoil layers [Sommerkorn, 2008].

[49] Among all microsites, ER was the carbon flux component that had the strongest response to changes in water table. Polygon rim plots presented the highest seasonal ER rates during all years; however, the correlation with water table in this microsite was also the poorest because of a small variation in water table with respect to ER. Nevertheless, the significant correlation (R2 = 0.22, p < 0.001) between water table and ER from polygon rims in 2007 suggests that under very dry conditions, water table exerts more control on ER in these microsites than during wet or intermediate soil moisture conditions (Table 3 and Figure 4). Conversely, the strongest response to changes in water table was observed in the wet microsites (wet sedge and intermediate, Figure 4). ER in the wet sedge (wettest microsites) presented the strongest correlations during all years, and although small, the ER in this microsite presented a significant correlation with water table even during periods of very high water tables (Table 3). Our results suggest that in these microsites, small fluctuations in water table can translate into important changes in ER, and water at or near the surface represents a strong control on ER [Sommerkorn, 2008].

[50] The magnitude of the effect of water table was not the same for all CO2 flux components during all seasons. We found that during wet years, water table was not a strong predictor of any of the CO2 flux components, while in dry years water table was a strong predictor (Table 3); however, on average, low GPP and ER rates were associated with high water table conditions. The control of water table on ER and GPP can be attributed to the combined effects of increasing resistance to diffusion of O2 through the water column (above and below the surface) and soil pores, the rapid depletion of available O2 in the soil as a result of microbial activity [Gebauer et al., 1996], and increased resistance to diffusion of CO2 out of the soil (D. Zona et al., Microtopographic controls on ecosystem respiration in the Arctic tundra, submitted to Journal of Geophysical Research, 2010). GPP was affected by water tables above the surface, but the response was not as strong as in ER. For GPP, severe flooding and oxygen deprivation reduces the ability of roots to take up water and in extreme cases submerges the leaf area, decreasing photosynthesis, especially that of mosses, as a result of the inability of the leaves to exchange CO2 with the atmosphere [Oberbauer et al., 1991; Elberling et al., 2008]. However, the presence of anoxia-tolerant species, such as Carex spp., can dampen the effects of loss of root function on GPP [Peterson et al., 1984].

[51] Oxidation-reduction potential measurements at the study site determined that soil just below the water table is anoxic (Zona et al., submitted manuscript, 2010), thus suggesting that oxygen available in the soil might leak from roots. Although we did not assess oxidation around roots, roots can be an important oxygen supply for the microbial community in anaerobic soils [Conrad, 1996]. ER represents a combination of the microbial activity from oxygenated zones plus root and foliage respiration [Billings et al., 1984; Sommerkorn, 2008]. Our results suggest that the fraction of ER most affected by the fluctuation in water table is the microbial component because of the anoxic conditions found in the study site under high water tables. The ability of the water table to control root respiration in flooding tolerant vascular plants might be indirect and a function of the available leaf area above the water surface capable of exchanging O2 with the atmosphere and transporting it to the roots via aerenchyma. For instance, Carex aquatilis, the dominant vascular species in the study site, has shown no significant growth responses to changes in water table [Peterson et al., 1984] suggesting that their ability to maintain physiological processes such as root respiration are not affected by flooding conditions.

[52] Differences in the vegetation cover among the microtopographic features should affect the response of the GPP to water table fluctuation. However, GPP did not differ between microtopographic features, except when water table was above the surface (e.g., wet sedge and intermediate microsites in 2006, Figure 4). This was an unexpected result, especially the small change of GPP rates in the polygon rims (NS) regardless of a significant decrease in water table (e.g., 2007, Figure 4). Since these areas are moss-dominated we expected that the lack of a vascular system would negatively affect water uptake and eventually photosynthesis; however, this was not observed. In general the polygon rims presented higher NVDI values than the other microsites, suggesting that the moss layer in the polygon rims was photosyntheticaly active [La Puma et al., 2007]. This result suggests that sufficient moisture was still present in these microsites to support moss photosynthesis, or the few vascular plants present in these areas experienced an increase in the photosynthetic rates that offset a decrease in moss photosynthesis.

[53] NDVI was able to detect differences in microtopography, especially between polygon rims and wet sedge plots. However, NDVI also changed in response to water table differences between years and sites (Table 1), suggesting that both microtopography (Table 2), and water table had a significant effect on the NDVI measurements. As a result, the interpretation of NDVI values is not straightforward. Since NDVI measurements are taken perpendicular to the ground, and in our study site the leaves of the vascular plants (mostly graminoids) have very steep angles, NDVI measurement might have captured primarily the greenness of the mosses rather than that of the vascular plants, failing to detect a change in the productivity of the graminoids. The highest NDVI values were observed in 2006, when the GPP values were the lowest and water tables the highest. Perhaps, these high NDVI measurements can be explained by an increase of the greenness of the mosses regardless of low GPP rates, suggesting that an increase of the graminoid productivity might not be detected by the NDVI measurements. Therefore, in tundra areas with poor drainage, combined with high moss and graminoid cover, the NDVI might not be a strong predictor of GPP.

4.2. Other Environmental Factors

[54] Water table alone explained 42% of the variation in ER (p < 0.001), 24% of seasonal variation of GPP the (p < 0.001), and had no relationship with NEE (Figure 4). PATH analysis revealed that in addition to water table, air temperature and PAR are important predictors of the CO2 flux components. Although on a daily basis, PAR is expected to drive the variation of GPP rates, air and soil temperature, water table, and thaw also play an important role in the seasonal variation of GPP (Figure 6).

5. Conclusions

[55] We determined that water table interacts differently with GPP, NEE and ER. We conclude that: (1) low water tables increased GPP but also ER, negatively affecting NEE because the response of ER was larger than that of GPP; (2) high water tables reduced GPP and ER, but the effect on the ER was larger increasing therefore the seasonal uptake; and (3) microtopography had a significant effect on ER, but not on GPP. However, the difference in strength of the correlations between water table and GPP among the different microsites suggests that microtopography position affects the response of GPP to water table, especially in the wet sedge microsites.

[56] Water addition and removal had complex effects on depth of water table with respect to the surface and therefore effects on ER. Although the polygon rims had the highest ER rates during all years, they also showed the lowest correlations with water table. Nevertheless, the increase in ER as a result of warm and dry conditions was sufficient to turn the polygon rim microsites from sinks to sources of CO2 during the growing season. On the other hand, wet sedge and intermediate areas were strongly affected by small changes in water table, significantly increasing the ER rates as result of water table decrease. Taking into account that wet sedge and intermediate microsites cover a large area in the lakebed compared to polygon rims, and that anthropogenic warming will likely accelerate drying of the tundra [Kaufman et al., 2009], the seasonal emission rates from these microsites have the potential to double in magnitude as it was observed between 2006 and 2007.

[57] The prohibitive cost of such a large hydrological manipulation limited our study to a single lake, restricting our ability to scale up our results to the whole Coastal Plain. However, wet sedge and polygon rim microsites represent similar plant communities across the area, suggesting that the response of the microsites in other areas is likely to be similar.

[58] Even though the focus of our research was to assess the controls of water table on CO2 flux components, we acknowledge the important effects that other environmental factors have on the CO2 fluxes, especially temperature. Additionally, the intertwined effects that changes in water table and/or temperature have on each other and the influence on other environmental factors, such as depth of thaw, adds a level of the complexity to the system that should be explored in further detail.


[59] This research was conducted on the Barrow Environmental Observatory, a private reserve owned by the Ukpeagvik Inupiat Corporation (UIC). This work is based in part on funding from the National Science Foundation Biocomplexity in the Environment—Coupled Biogeochemical Cycles program (award 0421588) with logistics funded by the Office of Polar Programs. We are especially grateful to the Barrow Scientific Consortium (BASC), Chico Perales and CPS, UIC, the North Slope Borough, Florida International University, and the ARCUS PolarTREC program for logistical support. We also thank the two anonymous reviewers for their contribution to improve this manuscript.