2.1. Site Description
 The study was conducted approximately 14 km west of Healy, Alaska at Eight Mile Lake (63° 52′ 42″N, 149°15′ 12″W), which is in the northern foothills of the Alaska Range near Denali National Park and Preserve. The mean annual temperature (1976–2005) was −1.0°C in Healy, with extreme differences between the coldest month (December, −16°C) and warmest month (July, 15°C). The total annual average precipitation was 378 mm (National Climate Data Center, NOAA). During this study (spring 2004 to winter 2006–2007), summer precipitation and air temperature were near average in 2006, but temperatures were warmer, and precipitation less, in 2004 and 2005 compared to the long-term average (Table 1). The study site was located at 700 m elevation on a gently sloping (∼4%), north facing glacial terminal moraine that dates to the Early Pleistocene [Wahrhaftig, 1958]. An organic horizon, 0.45–0.65 m thick, covered cryoturbated mineral soil that was a mixture of glacial till (small stones and cobbles) and windblown loess. Soil organic C pools to 1 m depth averaged between 55 and 69 kg C m−2 across all three sites (C. Hicks, unpublished data, 2008). Permafrost was found within 1 m of the soil surface, and therefore the soils were classified in the soil order Gelisol [Soil Survey Staff, 1999]. Permafrost temperatures have been monitored in a 30 m deep borehole in the study area since 1985 (maximum temperature range = −0.7 to −1.2°C at 10 m) [Osterkamp and Romanovsky, 1999]. During this time frame, researchers recorded rapidly increasing deep permafrost temperatures (by ∼0.6°C at 10 m) from 1990 until 1998, followed by a slight cooling (by ∼0.2°C) between 1998 and 2004 [Osterkamp and Jorgenson, 2006].
Table 1. Average Temperature and Precipitation Estimates for a Long-Term Monitoring Station and the Yearly Climatic Values for the Onsite Weather Stationa
|Average Air Temperatureb|
|Average Soil Temperaturee|
 We selected three sites within 200 m of the borehole for our CO2 exchange study. The sites represented differing degrees of permafrost thaw, thermokarst disturbance, and vegetation change. The sites included a 1) minimally disturbed moist acidic tundra, where the vegetation was dominated by the tussock-forming sedge Eriophorum vaginatum (site hereafter called “Minimal Thaw”), 2) a site adjacent to the borehole where shallow thermokarst depressions (<0.5 m deep) had increased in occurrence since 1990 (“Moderate Thaw”), and where tussock-forming sedges dominated productivity both inside and outside thermokarst, and 3) an area where deeper thermokarst depressions (0.5–1.0 m) were found (“Extensive Thaw”), and where, based on aerial photographs from 1951, thermokarst depressions had likely been present for at least 5 decades from present. Vegetation productivity at Extensive Thaw was dominated by shrubs (Vaccinium ulignosum) and Sphagnum spp. mosses [Schuur et al., 2007]. Jorgenson and Osterkamp  developed a thermokarst classification scheme that incorporates hydrology, soil texture, soil ice content and morphology, and karst size. Out of 16 categories in this classification system, the thermokarst at the Moderate and Extensive Thaw sites were designated as “thaw slumps.”
2.2. Climate Measurements
 From May 2004 to May 2007, a Campbell Scientific (Logan, UT) CR10X data logger recorded soil profile temperatures (10, 20, 30, and 40 cm from surface, n = 3) hourly at each site. For all temperature measurements, constantan-copper thermocouples and a reference thermistor were used. Water table depths were measured weekly in wells installed at each site (n = 3). The wells were 15 cm in diameter and extended from the surface to between 0.50 and 1.2 m into the soil profile. The depth of well placement resulted in the bottom 10–30 cm of pipe extending into the permafrost. For water table height, we measured the distance from the soil surface to the top of the water table. To determine the depth of water perched on the frozen soil, the depth from the soil surface to frozen soil layer was also measured immediately adjacent to the pipe. The wells were installed so that one occurred at the uppermost, the middle slope, and the deepest part of a thermokarst depression for a given site. We extrapolated water table depth from the wells to each chamber using a topographic survey [Lee et al., 2009].
 An Onset HOBO (Bourne, MA) weather station was used to measure air temperature, photosynthetically active radiation (PAR, band length 470–940 nm), relative humidity, precipitation, and wind speed and direction at 15 s intervals. Incoming solar radiation measurements from a nearby (∼4 km from site) weather station maintained by Denali National Park and Preserve were used to interpolate any gaps in our PAR using a solar radiation/PAR conversion factor derived for our site. The temperature and humidity sensor were 1.5 m above the ground surface, while the PAR, precipitation, and wind speed sensors were 2 m above the ground surface.
 The difference between photosynthesis and respiration equals the net exchange of CO2. Each of these components of ecosystem CO2 exchange was either directly measured (NEE, Reco) or indirectly estimated (GPP). We used closed chambers rather than an eddy covariance tower because in our study area, permafrost thaw and thermokarst depressions occurred at a smaller spatial scale than what could be resolved with an eddy covariance footprint [Vourlitis et al., 2000]. At each site, NEE was estimated at six locations spaced 8 m apart along a 40 m transect. In the summer of 2003, two square plastic bases (0.49 m2) were cut into the soil organic layer to a depth of ∼5 cm at each location. Two bases were initially used (2004–2005) so that vegetation harvests could take place within one, if needed. However, in 2006 the measurements were exclusively made on one chamber base because the fluxes from the paired bases were highly correlated with one another. The bases were subsequently used to seat a clear acrylic chamber that was 40 cm high, a height that fully accommodated the tallest plants. In 2004 and 2005, the two chamber bases in each location were treated as one replicate, for n = 6 replicates per site. At each site, the chamber bases spanned the range of microtopography created by thermokarst depressions. During measurements, a clear chamber was firmly fixed to the base, but not sealed airtight thereby minimizing the pressure differentials that can affect chamber flux measurements [Lund et al., 1999]. In 2004, we only used a single chamber that was moved among chamber bases and sites; in 2005 and 2006 we used both an automated chamber system along with periodic manual measurements. Side-by-side comparisons showed no difference between the two methods for flux calculation. For the autochamber system, Reco was measured only at night (PAR < 5 μmole m−2 s−1). For the static chamber, Reco was also measured during the day time by placing an opaque cloth over the top of the chamber to stop photosynthetic uptake.
 For all static and autochamber measurements, air was circulated between the chamber and an infrared gas analyzer (LI-820, LICOR Corp., Lincoln, Nebraska) at 1 L min−1 for 1.5 min and the CO2 concentration measured at 2 s intervals. For the static measurements, CO2 concentrations were recorded to a Palm (Sunnyvale, CA) Tungsten C portable computing device using the software program Online (Conklin Systems, Eaton Rapids, MI). For the automated system, the LI-820 data was recorded on a Campbell Scientific CR10X at 5 s intervals. Two small fans mixed the chamber air, while air temperature and relative humidity were monitored inside the chamber with a HOBO sensor. For the static chamber measurements, PAR was measured inside the chamber with a LICOR quantum sensor attached to a LI-1400 data logger. For the automated chamber, PAR was recorded at a weather station 50–400 m from the sites. A correction factor (19% reduction) was developed for the weather station PAR to account for the effect of reduced light transmission by the Plexiglass, and light interception by the chamber support structures. In conjunction with flux measurements, the thickness of unfrozen ground or the active layer thickness (ALT) was measured once per week at each chamber base.
 To describe daily and weekly environmental and gas exchange variability, in 2004 we conducted static chamber measurements 5 times per week at 5 different times of day (pre-0600 LT, 0900 to noon, noon to 1500 LT, 1700 LT to 2000 LT, and after 2300 LT). Variability in incoming PAR was also created by artificially reducing PAR to 1/2 and 1/4 of ambient conditions using mesh screens placed entirely over the chamber during the midmorning and late afternoon measurements. Static chamber measurements were continued in 2005 and 2006, but in these years measurements were only conducted three times of day (approximately 0900 LT to noon, 1400 LT to 1800 LT, and after 1900 LT) because the autochamber system was collecting gas exchange measurements every 1.2 h at each individual chamber base. Burrows et al.  compared manual and autochamber measurements and found when a warmer temperature bias in the manual chambers was accounted for, the two methods returned similar results for annual NEE. Our manual and automated chambers only differed by 1.2°C in 2005 and 2006 for daytime NEE, which was likely why we found no difference in NEE estimated with the two techniques. The automated chamber system was moved every 7–12 days among the three sites, and in general, measured for at least 3 days in a given location. These measurement regimes generally began immediately after snowmelt (∼2–5 May) and continued until a lasting snowfall (∼25–30 September).
 After the first snowfall, C exchange measurements were continued on the snow surface (October 2004 to March 2005), and in snow pits dug to the soil surface (November 2005 to April 2007). Measurements using the snow surface approach were only made when the wind was calm (<0.2 m/s) because wind can evacuate CO2 from the snow profile and result in artificially low or highly erratic efflux rates. A rectangular chamber (0.0794 m2) on a 0.5 m pole was first pressed into the snow surface to create an imprint, then lifted ∼1 m to flush the chamber, and after 20 s placed again in the imprint. The chamber CO2 concentration was measured for 6 min, and snow depth was measured afterwards. We used the snow pit method in 2005–2006 because it was less sensitive to wind conditions. The method followed that of Grogan and Jonasson , where we dug snow pits that ranged between 0.15 and 0.75 m in depth. At the bottom of the pit, we left ∼5 cm of residual snow on the ground surface. Using the same approach as the surface measurements, the residual snow was used to create a seal with the edge of the chamber. In addition, the outside edge of the chamber was covered with a thin snow layer. Based on repeated measurements of efflux at 3 min intervals, we determined that after the snow pit was excavated, another 13–15 min was needed for CO2 efflux to approach a near constant rate [Grogan and Jonasson, 2006]. At each site, six locations were measured adjacent to the chamber base locations (5–15 m downslope). We avoided measuring directly at the established bases in the winter because we were concerned that disturbing the snow cover would alter the soil thermal regime.
2.3. Estimates of Annual CO2 Exchange
 Seasonal and annual estimates of CO2 balance were estimated using two methods: gap filling with response functions to environmental factors, and by interpolating mean estimates between time points. For the period from June through August, a rectangular hyperbola equation was used to describe the relationships between PAR and NEE [Thornley and Johnson, 1990]. For Reco, we used nighttime and dark chamber measurements to develop exponential relationships between Reco and air (June–August) or soil temperature (winter, and June–August 2006). During spring (May) and fall (September), integrated estimates of NEE and Reco were developed by interpolating fluxes between time points. Reco and NEE at these times of year could not be explained with response functions, likely because plant phenology and soil conditions were changing rapidly. We defined the total growing season as being the period of May–September, but we note that snow often covered the site for the first few days of May and the last days of September. GPP was estimated as the difference between the integrated NEE and Reco values.
 We filtered flux measurements for estimates that may have been biased by unusual environmental conditions created by the chamber, or when background conditions created erratic fluxes. When chamber air temperature exceeded 30°C, the collected data point was removed because this temperature was greater than the maximum air temperature measured at the weather station. For the remaining data, chamber air temperature warmed <1°C for 72% of all daytime NEE measurements with a maximum warming of 5.4°C. In addition, automated chamber measurements that occurred during rain, or when wind speeds exceeded >5 m s−1, were also removed because the fluxes were generally erratic. No data were removed based on RH changes inside the chamber. In total, 14% of measurements were eliminated with the data filters. After screening fluxes for environmental conditions, the total number of individual NEE and Reco measurements used for making integrated growing season estimates were 2,986 in 2004, 5,148 in 2005, and 7,152 in 2006.
2.4. Statistical Analysis
 The equation parameters for the rectangular hyperbola PAR-NEE relationships, and the exponential Reco-temperature relationships were developed using the nonlinear equation feature in SAS statistical software version 9.1. The integrated May, September, growing season and annual GPP, Reco and NEE were compared using a two-way analysis of variance (mixed model) with site and year as fixed effects. Tukey's HSD post hoc test was used to correct significance levels for multiple comparisons. Soil environmental variables were only statistically compared for those measurements that occurred in each pair of chamber bases (soil temperature at 10 cm, surface moisture, ALT) during the growing season, and for depth to water table. Multiple least squares linear regression was used to determine if site and a combination of soil environmental variables (ALT, average surface soil moisture, soil temperature, and depth to water table) could describe integrated growing season C exchange across sites. Selection for the maximum coefficient of variation (adjusted R2) and Mallow's CP statistic were used for model selection [Sokal and Rohlf, 1995]. When a minimum CP statistic and maximum R2 are selection criteria, the result is the best fit model with the minimum number of parameters.