Impacts of active retrogressive thaw slumps on vegetation, soil, and net ecosystem exchange of carbon dioxide in the Canadian High Arctic

Retrogressive thaw slumps (RTS) are permafrost disturbances common on the Fosheim Peninsula, Ellesmere Island, Canada. During the 2013 growing season, three different RTS were studied to investigate the impact on vegetation composition, soil, and growing season net ecosystem exchange (NEE) of carbon dioxide (CO 2 ) by comparing to the adjacent undisturbed tundra. Eddy covariance (EC) and static chamber measurements were used to determine NEE and ecosystem respiration ( R e ), respectively. Vegetation cover was significantly lower in all active disturbances, relative to the surrounding tundra and this affected the overall impact of disturbance on CO 2 fluxes. Disturbances were characterized by greater R e compared to surrounding undisturbed tundra. Over the mid-growing season (34 days), EC NEE measurements indicated there was greater net CO 2 uptake in undisturbed vs disturbed tundra. At one site, the undisturbed tundra was a weak net sink (-0.05±0.02 g C m -2 day -1 ) while the disturbed tundra acted as a weak net source (+0.07±0.04 g C m -2 day -1 ). At the other site, the NEE of the undisturbed tundra was Page 0.20±0.03 g C m -2 day -1 (sink) while the disturbed tundra still sequestered CO 2 , but less than the undisturbed tundra (NEE = -0.05±0.04 g C m -2 day -1 ). Two of the RTS exhibited average soil temperatures that were greater compared to the surrounding undisturbed tundra. In one case the opposite effect was observed. All RTS exhibited elevated soil moisture (+14 %) and nutrient availability (specifically nitrogen) relative to the undisturbed tundra. We conclude that RTS, although limited in space, have a profound environmental impacts by reducing vegetation coverage, increasing wet soil conditions, and alter NEE during the growing season in the High Arctic.

D r a f t D r a f t 2 0.20±0.03 g C m -2 day -1 (sink) while the disturbed tundra still sequestered CO 2 , but less than the undisturbed tundra (NEE = -0.05±0.04 g C m -2 day -1 ). Two of the RTS exhibited average soil temperatures that were greater compared to the surrounding undisturbed tundra. In one case the opposite effect was observed. All RTS exhibited elevated soil

Introduction
Permafrost disturbances are expected to increase in frequency and magnitude with predicted climate change (ACIA 2005;Vincent 2011;Kokelj and Jorgenson 2013;Segal et al. 2016). In the High Arctic, common permafrost disturbances include active layer detachment slides (ALDS) and retrogressive thaw slumps (RTS). RTS occur when ground ice is exposed (including after an ALDS has occurred), soil and vegetation are removed upslope and the thawed material moves downslope as the ground ice melts (Lantuit and Pollard 2008). RTS remain active until ground ice is depleted or further thaw is prevented by falling blocks of soil and vegetation, which act as insulation (Burn and Friele 1989).
Ecosystem responses to these permafrost disturbances have focused on hydrological impacts (Kokelj and Lewkowicz 1998;Kokelj and Lewkowicz 1999; D r a f t Lamoureux and Lafrenière 2009). In aquatic ecosystems, physical and chemical changes of sediment and water following slumping impact invertebrates, macrophytes, and diatoms, altering composition and abundance (Chin et al. 2016;Mesquita et al. 2010;Moquin et al. 2015;Thienpont et al. 2013). The ecological impacts of ALDS have been analyzed at few sites in the High Arctic and changes to the physical environment following a disturbance event have been shown to influence vegetation recovery (Bosquet 2011;Cannone et al. 2010;Cassidy 2011;Desforges 2001). The impacts of RTS on vegetation development have focused on Low Arctic ecosystems (Bartleman et al. 2001;Burn and Friele 1989;Cray and Pollard 2015;Lantz et al. 2009). Thaw slumps modify vegetation communities with changes persisting for centuries (Cray and Pollard 2015).
Due to the harsh conditions in the High Arctic (low temperatures, short growing season), recovery is predicted to occur more slowly than in the Low Arctic (Svoboda and Henry 1987). The removal of vegetation in the disturbance can reduce albedo by up to 50% (Babb and Bliss 1974). A reduction in albedo and removal of the soil organic layer can increase soil temperatures and deepen the active layer, however the active layer may also increase due to the disturbance itself (Bliss and Wein 1972;Auerbach et al. 1997;Lantz et al. 2009). An increase in the active layer provides plant roots with more volume in the soil (Bliss and Wein 1972). In addition, the deeper active layer and warmer soil temperatures could result in greater nutrient availability through increased rates of decomposition and mineralization (Lantz et al. 2009). Plants that have deep roots, including species such as Calamagrostis canadensis (bluejoint grass) and Eriophorum angustifolium (tall cotton-grass) are common invaders in some disturbed tundra sites due D r a f t 4 to their ability to take advantage of extra soil volume and ability to disperse (Chapin and Shaver 1981).
Carbon dioxide (CO 2 ) fluxes between the surface and the atmosphere can be quantified by measuring net ecosystem exchange (NEE), the difference between CO 2 emissions due to ecosystem respiration (R e ) minus gross primary production (GPP). NEE of ecosystems is usually quantified by eddy covariance (EC) (Emmerton et al. 2015;Humphreys and Lafleur 2011;Lafleur et al. 2012) or clear closed chamber systems (Welker et al. 2004). NEE is influenced by temperature, moisture and light levels amongst other factors and differs among plant community types (Baldocchi 2008).
During the growing season, tundra ecosystems have generally been found to be CO 2 sinks, but can easily shift to CO 2 sources due to changes in temperature, moisture, and the water table (Oberbauer et al. 2007;Vourlitis et al. 2000). Chamber studies have found small but steady losses of CO 2 during the winter (Welker et al. 2004). At Alexandra Fiord, Ellesmere Island, experimental warming using open top chambers impacted NEE differently based on soil moisture, due to differences in respiration between wet and dry sites (Welker et al. 2004). Across a latitudinal gradient, warming tended to increase respiration, with the greatest increases found in dry ecosystems (Oberbauer et al. 2007).
Previous studies examining Arctic NEE have found large inter-annual variability within and among sites. This variability has been substantial enough to shift the system during the growing season from a CO 2 sink to CO 2 source (Griffis and Rouse 2001;Kwon et al. 2006;Merbold et al. 2009). However, there are few measurements of NEE from Arctic tundra sites, especially from the High Arctic, which combined with the inherent variability and the low flux strength in Arctic tundra ecosystems makes it difficult to D r a f t 5 determine relative NEE across the Arctic and their contribution to regional and global fluxes (Lafleur et al. 2012). Cassidy et al. (2016) simultaneously quantified growing season NEE over an active retrogressive thaw slump (RTS) and its surrounding undisturbed tundra at a High Arctic site. The RTS acted as a net source of carbon while surrounding undisturbed tundra acted as a net sink.
Vast amounts of carbon are stored in permafrost soils, and Hugelius et al. (2013) estimated these soils contain 50% of worldwide below ground organic carbon, with the greatest stores found in yedoma and low Arctic soils. These estimates are likely an underestimation (by up to a factor of two) due to measurement difficulties and uncertainty regarding carbon storage in cryoturbated soils (Hugelius et al. 2013). Schuur et al. (2008 found organic carbon that is unfrozen can be released to the atmosphere through microbial respiration. Thus, permafrost disturbance could result in the release of previously frozen carbon from this massive storage reservoir.
In this study, we assessed how the changes that result from permafrost disturbances in the High Arctic influence spatial heterogeneity in CO 2 exchange. We focused on changes in vegetation, the unique soil characteristics, and the fluxes of CO 2 that are associated with thaw slumps at different stages of recovery. The research objective was to determine the variability of vegetation composition and cover, thaw depth, soil thermal, moisture and nutrient characteristics, and CO 2 fluxes among three RTS that varied in time since their formation and adjacent undisturbed tundra.
D r a f t 6 2 Materials and methods

Study area
Research was conducted on the Fosheim Peninsula, western Ellesmere Island, Nunavut (79° 58' 21" N; 84° 17' 41" W, 100 m asl). Active layer detachment slides and thaw slumps are common in the area (Kokelj and Lewkowicz 1998). Multiple sites with RTS were identified during summer 2012 and monitored during summer 2013. The dominant vegetation in the area is characterized as dwarf-shrub-graminoid tundra that are found on uniform, weakly alkaline to neutral cryosols (Edlund et al. 1989). The geological substrate is sandstone of the Eureka Sound group (Bell 1996) with marine silts and sands varying in thickness above the bedrock (Robinson and Pollard 1998). The upper limit of marine inundation at the end of the last glaciation is approximately 140 m above sea level (Bell 1996), and limited vegetation is found above this level. This region is an area of ice rich permafrost. Due to the nature of the permafrost and increased summer temperatures and precipitation over the last 20 years, there has been an increase in the occurrence of permafrost disturbances (Lewkowicz and Harris 2005).

Site selection and description
Three RTS were selected for detailed measurement and analysis, based on vegetation characteristics and the nature of the disturbances (Table 1). Air photo analysis using D r a f t 7 photos taken in 1949 show these sites were not visible on the landscape and indicate that these disturbances formed after 1949. Therefore, these disturbances could be 50+ years old; however, numerous disturbances formed during the particularly warm summers of 1988 and 1998, so they may have occurred during these seasons (Lewkowicz and Harris 2005).
Two of these disturbances were situated on east facing slopes, and are referred to as sites RTS-1 and RTS-2 ( Figure 1) and were selected as they are both part of a larger string of disturbances. Eddy covariance (EC) flux towers were established at RTS-1 and RTS-2 on the periphery of the disturbance (Figure 2; Figure 3). Both EC flux towers were located on the northern edge of the RTS; the area north of both towers was undisturbed tundra, while that south was influenced by the RTS and in part by the other disturbances in the vicinity. were located in each zone (recovered and undisturbed), at a distance of at least 3 m from the current active region and 3 m from the recovered headwall, respectively. Plots were located at distances of 5 m along the transects.

Vegetation sampling
Vegetation was sampled at each plot along transects using a 50 cm x 50 cm quadrat separated into 5 cm squares. Vegetation composition and abundance (cover) were estimated inside five quadrats in each plot: one at the centre of the plot, and four times surrounding the plot centre. These quadrats were haphazardly located by blindly throwing a small trowel and placing the quadrat where the trowel fell with the tip of the trowel end in the SE corner of the quadrat. Percent cover of each vascular plant species and total moss and lichen cover was visually estimated for each species in each quadrat. The same observer made all visual estimates.
D r a f t

Soil characteristics
Thaw depth was measured using a thin metal probe, which was inserted into the ground until the depth of refusal (permafrost) was reached. Thaw depths were measured at the end of the season on 27 July 2013 at each site. At each slump, thaw depth was measured 18 times while 24 measurements were made outside the slump in the undisturbed tundra. with 12 cm rods. Soil moisture was recorded at three locations within each plot along each transect and three locations between plots at 10 day intervals throughout the season (N d =18, N c =18). Additional measurements were taken at both flux towers and 10 m upslope and downslope of the tower (N=9).
Surface albedo was estimated using a handheld pyranometer (LI200X, Campbell Scientific Inc., Logan, UT, USA), mounted at a height of 1 m to allow for instrumentation to be leveled for all measurements. Measurements were made upwards D r a f t and downwards during cloud free conditions on 10 July 2013 (between 10:00 -14:00) at each plot along the transects.
Soil nutrient availability was measured using ion exchange membranes (PRS Probes, Western AG, Saskatoon, SK, Canada). Four cation and four anion membranes (each 17.5 cm 2 membrane was mounted in a plastic holder 15 cm x 3 cm x 0.5 cm) were inserted 10 cm into the soil in the centre of each plot. These were deployed on 2 July 2013 and retrieved 26 July 2013 for a 24 day sampling period. At RTS-1 and RTS-2, probes were placed along the transects for a total of N d = 6 and N c =9 sites. At RTS-3, probes were placed in the centre of each plot (N d = 6 and N c =6). Samples were processed during September 2013 at Western AG in Saskatoon, Canada, using an automated flow injection analysis system and inductively-coupled plasma spectrometry.

Portable CO 2 efflux chamber system
On 30 June 2013, 28 opaque PVC collars (10 cm diameter, Area = 78.5 cm 2 , Depth = 6 cm) were installed at sites RTS-1 (N=8), RTS-2 (N=8), and RTS-3 (N=12), for a total of 14 collars in both disturbed and undisturbed tundra. The collars were inserted 4 cm into the soil to minimally disturb soil and vegetation and they extended 2 cm above the ground surface. Collars were spaced across transects at each plot ( Figure 3). Due to the variability of vegetation cover, collars were placed on both vegetated and unvegetated tundra within the disturbed and undisturbed tundra.
A non-steady state portable chamber system similar to Jassal et al. (2005) was used to measure ecosystem respiration from each collar using an opaque chamber ( Figure   3 a -b). The measurement head was a PVC chamber with a volume of 1.4 x 10 -3 m 3 (height: 15.6 cm, diameter: 10.7 cm). The chamber head was placed on each collar for 2-D r a f t 11 minute intervals. A foam gasket was used to seal the connection between collar and chamber head. A pump (flow rate: 600 cm 3 min -1 ) circulated air from the chamber head into a portable battery operated infrared gas analyzer (IRGA) (LI-840, LI-COR Inc., Lincoln, USA) and back into the chamber through a closed circuit. The IRGA determined CO 2 mixing ratios ([CO 2 ] in ppm) and water vapour concentrations at a temporal resolution of 1 Hz during each run.
Respiration (R e ) was calculated from ∆[CO 2 ]/∆t (linear regression over 2 min, discarding the first 10 seconds), using Eq. 1: where ρ is the molar density of air (mol m -3 ) calculated from measured air temperature, ‫ܦ‬ ഥ is dilution considering [H 2 O], ∆[CO 2dry ]/∆t is the rate of change of CO 2 mixing ratio over time (µmol mol -1 s -1 ), and V and A are chamber volume and area, respectively.
Measurements were made on two days during the season (17 and 27 July 2013) and were taken between 10:00 and 16:00 on each day to minimize diurnal changes in light and temperature.

Eddy covariance measurement of NEE
We used an eddy covariance (EC) tower and wind directional partitioning to measure NEE from disturbed and undisturbed tundra. Two towers were located at RTS-1 and RTS-2, and each tower was established 2 m from the edge of the disturbance (and 50 m and 60 m away from the headwall, respectively). Both EC systems were mounted on m. Ultrasonic anemometers were sampled at 60 Hz and data output was stored at 10 Hz.
The IRGA was sampled at 10 Hz, and all data were stored on a data logger (CR1000, Campbell Scientific Inc., Edmonton, Canada). Both towers were placed with the IRGA and sonic anemometer parallel to the slump edge to avoid any flow distortion effects from preferred wind sectors associated with flow past the sensors and the head wall of the slump (data were removed in sectors with flow distortion). Partitioning based on wind direction allowed measurement of fluxes from disturbed and undisturbed tundra. Friction velocity ‫ݑ(‬ * ) thresholds of 0.10 m s -1 were applied to remove data under low-turbulence conditions. Both IRGAs were tiled 30° from the vertical to minimize issues with sensor heating and reduce pooling of moisture on the windows. All IRGAs (EC tower and portable chamber) were calibrated prior to the field season using a two-point calibration in the lab against standards from the Greenhouse Gas Measurement Laboratory (GGML), Meteorological Service of Canada (using a zero gas and span gas of known mixing ratio).

Flux data processing
Molar fluxes of CO 2 (F c in µmol m -2 s -1 ) were computed in EddyPro (V5.1.1, LI-COR Inc.) with a missing sampling allowance of 30%. F c was calculated over a 30 minute averaging interval using double rotation for tilt correction, block average detrending, time lag detection, and Webb-Pearman-Leuning corrections (Webb et al., 1980). Data quality controls based on the flagging system proposed by Mauder and Foken (2004) were used D r a f t 13 and data categorized as level 2 were discarded. Corrections were applied for both low and high-pass filtering effects according to Moncrieff el al. (1997) and Moncrieff et al. (2004). At low temperatures measurements from upright (vertically mounted) open-path LI-7500 CO 2 sensors have been shown to overestimate uptake rates of CO 2 (Burba et al., 2008). To minimize this error, LI-7500 head was mounted tilted from the vertical by 30º.
As measurements were made during the warmest month of the year, with mean temperatures of 8°C and a range of 0.5°C to 17°C, and with tilted IRGA installation, heating issues have been shown to be minimal under most conditions (see Cassidy et al. 2016 supplement for a detailed justification for not performing corrections on flux data).
However, as potential issues have been associated with this sensor when fluxes are measured at lower temperatures, caution should be taken when transferring this methodology to shoulder seasons, or when interpreting values measured at lower temperatures.
When winds were parallel to the edge of the disturbance, the source terrain could not be separated as disturbed or undisturbed, calculated NEE (F c ) values were not reliable, and fluxes from these sectors were discarded (46% of data). Fluxes with a difference greater than 5 standard deviations from the daily average of the 30 min values (of the same day) were removed. We averaged half hour NEE data into hourly fluxes. If one of the half hour values was not available and from the same segment (disturbed/undisturbed), the hourly value was then based on the single 30 min flux measurement. We filled remaining hourly gaps using the following methods: a) gaps of less than 2 hours from the same segment were filled using linear interpolation; and b) gaps greater than 2 hours were filled using aggregate averaging over a rolling five-day D r a f t 14 window selecting the same time of day and same segment. The dataset was comprised of 90% original data and 10% gap filled data, as 173 of the 1723 data points were modeled.

Statistical analysis
All statistical analysis was completed using the R programming language (Version 3.1.2 (R Core Team, 2013) to analyze differences between sites (RTS-1, RTS-2 and RTS-3) and the impact of disturbance (two zones: disturbed and undisturbed). To determine differences in community composition of vegetation among sites and between disturbed and undisturbed tundra (zone), we used non-metric multidimensional scaling, a multivariate ordination technique based on Bray Curtis distance matrices derived from percent cover data. A two-dimensional ordination displayed the least stress and was repeated 100 times to reach the best solution (Legendre and Legendre 1998). ANOSIM was used to test for differences among groups using the vegan package (Oksanen et al. 2012).
Indicator species analysis was used to determine species characteristic of each zone and site. Indicator species analysis calculates an indicator value ‫ܸܫ(‬ ) for species i in group j based on relative abundance (specificity: ‫ܣ‬ ) and relative frequency (fidelity: where ‫̅ݔ‬ is the mean cover of species i within group j, ∑ ‫ݔ‬ ప ഥ Total cover was calculated as the sum of percent cover of live green material in each plot. Differences in environmental variables (total vegetation cover, soil moisture, thaw depth, soil temperatures) and CO 2 fluxes were tested using two-way ANOVA (site x zone) with Bonferroni correction. Data were transformed to meet normality assumptions, when necessary. Post hoc Tukey tests were used to conduct pairwise comparisons. Soil nutrient availability data were analyzed with a non-parametric Kruskal-Wallis test to determine the effect of site (RTS-1, RTS-2, and RTS-3) and zone (disturbed, undisturbed).

Micrometeorological conditions
Minimal differences in micrometeorological conditions were found between the two flux D r a f t

Vegetation
Significant differences were found in plant community composition among the sites ( Figure 5). ANOSIM indicated differences in composition based on site and disturbance status (ANOSIM R=0.4312; p=0.001, based on 999 permutations). Pairwise comparisons at each site also indicate compositional differences (Table 2).
Indicator species analysis (Table 3) supported the differences in composition among the sites and between zones (disturbed and undisturbed tundra) as shown using NMDS and ANOSIM. Undisturbed tundra was characterized as dwarf shrub graminoid tundra, with shrubs Salix arctica (Arctic willow) and Dryas integrifolia (mountain avens) found at these sites. Disturbed areas were dominated by rhizomatous grasses and sedges.
At RTS-1, undisturbed tundra was characterized by Dryas integrifolia and Carex nardina, however, in adjacent disturbed tundra vegetation cover was characterized by grasses, Poa glauca (glaucous bluegrass) and Alopecurus magellanicus (alpine foxtail).
At RTS-2, undisturbed tundra was characterized by Salix arctica and moss, however no unique vegetation was characteristic within the disturbance at RTS-2, which was largely unvegetated. At RTS-3, undisturbed tundra was characterized by Salix arctica, Dryas integrifolia, Puccinellia spp. (alkali grass), and lichen. The stabilized slump at RTS-3 was largely colonized by Carex aquatilis (aquatic sedge) and Alopecurus magellanicus.
Common species were found at multiple sites, including Dryas integrifolia at RTS-1 and RTS-3 and Salix arctica at RTS-2 and RTS-3, however differences were found in the overall community composition. Disturbed tundra was characterized by species able to recover and recolonize quickly and tolerate site conditions present in retrogressive thaw slumps.

D r a f t
Total cover (Figure 6a) was not significantly different between disturbed and undisturbed tundra (F (1,878) = 2.52, p=0.12); however, differences in cover were found between all sites (F (2,10534) =15.08, p< 0.001). The interaction between zone (disturbed and undisturbed tundra) and site (RTS-1, RTS-2, RTS-3) was significant (F (2,9574) =13.71, p<0.001), as total cover was lower within the disturbance at RTS-1 and RTS-2. However, this difference in cover was only significant at RTS-2. While at RTS-3, the reverse pattern was found, with significantly greater cover characterizing disturbed terrain. The lack of overall significance in the difference in total cover between disturbed and undisturbed tundra is based on stabilization of RTS-3 and the active nature of the RTS-1 and RTS-2.

D r a f t
Soil temperatures at the 5-cm-depth over the growing season were impacted by disturbance (Table 4; F (1,14286) = 251.9, p<0.001) as greater soil temperatures were found within disturbed soils at RTS-1 and RTS-2 (9.3°C±0.1 and 8.6°C±0.1, respectively) when compared to adjacent undisturbed soils (7.2°C±0.1 and 8.3°C±0.1, respectively). At RTS-3, lower temperatures were found in the disturbed (6.5°C±0.1) than undisturbed soils (8.2°C±0.1). This divergence in the direction of soil temperatures in active and stabilized slumps is shown in Figure 7. Soils temperatures were significantly different among sites F (2,14286) =140.7, p<0.001), with the coolest soil temperatures found at RTS-3 and the overall warmest temperatures at RTS-2.

Soil moisture regime
Soil moisture was greater (38.5±0.9 %) in the disturbed tundra than the undisturbed terrain (24.0±0.6 %) at all three sites (Figure 6c; F (1,370) =218.5, p<0.01). There was a significant interaction between site and disturbance (F (2,370) =19.8, p<0.01) as soil moisture measured within disturbed soils at RTS-3 was greater than in RTS-2; however RTS-1 was not significantly different than RTS-2 or RTS-3. Undisturbed soils in RTS-1 and RTS-2 had similar soil moisture levels, and these undisturbed soil moisture values were greater than those in undisturbed tundra at RTS-3.
The average (±SE) albedo measured on 10 July 2013 at RTS-1 was 0.18±0.00 in undisturbed tundra and 0.15±0.01 in the disturbance. At RTS-2, undisturbed tundra had a slightly greater albedo (0.19±0.01) than at RTS-1, while albedo in the disturbed portion of this site (RTS-2) (0.21±0.01) was notably greater than at RTS-1. With all measurements combined, average albedo was very similar between the undisturbed D r a f t tundra (0.18±0.004) and the disturbed tundra (0.18±0.01). Albedo was not measured at RTS-3.

Nutrient availability
Summary values of nutrient availabilities as measured using IEMs are presented in Table   5. Total nitrogen was greater in disturbed soils, and differences in total N were driven by NO 3 -N, which was also higher in the disturbances. Availability of NH 4 did not differ between disturbed and undisturbed tundra ( higher concentrations of Cu + were found in disturbed soils at RTS-2 and RTS-3 but not RTS-1.

Ecosystem respiration
Ecosystem respiration (R e ) measured over the growing season (Figure 6d found at RTS-2) compared to undisturbed areas (0.96 µmol m -2 s -1 ). When respiration was compared between all sites and zones, differences in respiration were found between sites (F (2,44) =4.49, p=0.02). Correlations among ecosystem respiration and ecosystem variables are presented in Table 7.

Net ecosystem exchange (NEE)
When EC measured NEE was separated by wind direction (Table 8) At each site, the relationship between air temperature and NEE was also explored.
Linear regressions were calculated to predict NEE based on temperature and disturbance RTS-1 and RTS-2, respectively). As air temperatures were different between situations that were sampled as disturbed and undisturbed tundra at each site, a conditional sampling by weighting bin averaged measurements based on air temperature was conducted ( Figure 9). Results indicated a significant effect of location, as fluxes differed between RTS-1 and RTS-2 (F (1,34) =6.31, p<0.05), however, CO 2 fluxes from disturbed tundra were not statistically significantly different from fluxes from undisturbed tundra (F (1,34) =1.84, p=0.18).
D r a f t

4 Discussion and conclusion
Landscape mosaics were evident on the Fosheim Peninsula as a result of permafrost disturbances. Site characteristics that result from morphological modifications following retrogressive thaw slumping are dependent on initial site characteristics. These permafrost disturbances increase spatial heterogeneity at the fine and landscape scale. At our study location, three disturbances were compared to their respective undisturbed tundra in the surrounding area. Each site was characterized by different combinations of plant species, possibly due to the initial site conditions. We found vegetation differences across the three disturbances, which were likely the result of differences in soil moisture and are supported by a positive correlation between these variables: the newly disturbed tundra in RTS-1 was characterized by the grass Poa glauca while the recovered tundra at RTS-3 had wet soil conditions and was dominated by Carex aquatilis. The greater soil moisture at RTS-3 was due to the re-exposure of ground ice that had begun to melt and the continued flow of meltwater, which allowed moisture dependent species to colonize, including Carex aquatilis (Chapin et al. 1992). Increases in soil moisture were also present at the other two sites, and field observations suggest that differences were influenced by the quantity of meltwater provided by ground ice thaw at the headwall in addition to flow patterns present in the disturbance. The dynamic nature of these flow patterns can greatly modify vegetation composition during recovery. Bartleman et al. (2001) found recovery trajectories in the Low Arctic (Mayo, Yukon) were altered by the depletion of ground ice and resulting lack of flow, which resulted in moisture rich areas that were dominated by Equisetum spp. (horsetail) and drier areas of the slump that were characterized by Salix spp.
D r a f t 23 Additionally, thaw depths were impacted by disturbance. Differences in thaw depth at our study sites corresponded with reduced albedo. Lower albedo at RTS-1 resulted in greater absorption of solar radiation and a deeper thaw layer, while greater albedo corresponded with a shallower thaw layer at RTS-2. We found negative correlations between vegetation cover and thaw depth likely in response to greater soil moisture in shallower active layers (and resulting in a slight negative correlation between soil moisture and thaw depth). Thaw depths are also impacted by winter snow accumulation, soil texture, organic layer development, and soil moisture, amongst other factors. At sites impacted by vehicle disturbance, Emers et al. (1995) found both increases and decreases in active layer depth associated with disturbed tundra, with shallower depth associated with insulating vegetation and complete removal of vegetation resulting in deeper depths. We found increased soil temperatures within disturbed soils at RTS-1 and RTS-2. However, at RTS-3 greater soil temperatures were found in the undisturbed terrain, which were likely a result of the increased shading by the greater vegetation cover in the recovering disturbance at this site. Less vegetation was found on the undisturbed surrounding terrain; thus, albedo was likely greater in the disturbance.
The ground thermal regime can be altered by these permafrost disturbances for more than a century (Burn and Friele 1989), which contributes to the vegetation differences in and out of the disturbances.
Plant nutrient availability may be elevated by permafrost thaw as soluble materials in the frozen ground may be released with ground ice melt associated with disturbance. On the Fosheim Peninsula, salt efflorescence accumulations have been found at sites located below the glacial marine limit, and are largely related to disturbance as dissolved solids that were previously trapped in frozen sediments were released and redistributed downslope (Kokelj and Lewkowicz 1999). Efflorescences were found within scar floors of the RTS sites and downstream from disturbances. Large concentrations of Na + and salts in the active layer and runoff could negatively affect plant growth and revegetation of disturbed terrain, increasing the duration of modified drainage and enhancing erosion and may continue to alter the terrestrial system for 30 years or more (Kokelj and Lewkowicz 1999).
Ecosystem respiration (R e ) was greater in disturbed soils than in surrounding tundra, and these patterns were consistent with measurements from other permafrost disturbances (Beamish et al. 2014;Cassidy et al. 2016). However, we found differential responses of D r a f t R e based on site, as disturbed tundra was characterized by greater R e at sites RTS-2 and RTS-3, while undisturbed and disturbed tundra at RTS-1 had similar R e rates. Disturbed tundra that has undergone some revegetation, such as the recovered site at RTS-3, may also have larger GPP, which would offset increases in R e at these sites. R e was found to be positively correlated with thaw depth and weakly negatively correlated with soil moisture.
Loss of old carbon from permafrost has been associated with disturbance. Schuur et al. (2008) found a positive relationship between ecosystem respiration and old carbon release associated with permafrost thawing. On the Fosheim Peninsula, old carbon may be released due to permafrost disturbances, which may be responsible for increases in respiration in disturbed areas relative to undisturbed tundra at some sites; however, these increases were not significant at all sites.
Our NEE measurements are comparable to those at other high Arctic locations (specifically Lake Hazen, Ellesmere Island and Cape Bounty, Melville Island), with uptake values that ranged between 0.2 -2.2 g C m -2 day -1 (Lafleur et al. 2012). Despite the small magnitude of these fluxes, the impact of disturbance was evident. Although both disturbed and undisturbed tundra at RTS-1 sequestered a minimal quantity of CO 2 , the undisturbed tundra sequestered nearly four times the CO 2 of the disturbed tundra. In addition, at RTS-2 the undisturbed terrain sequestered CO 2 throughout the season whereas the disturbed tundra emitted CO 2 to the atmosphere. The NEE of the undisturbed tundra at RTS-1 was substantially greater than at RTS-2, which was likely due to greater vegetation cover at this site. This indicates that the initial conditions of undisturbed tundra, such as vegetation composition and cover, influence the resulting carbon D r a f t 26 exchange of disturbed tundra through vegetation colonization and recovery within the disturbance.
When meteorological conditions were compared during times when fluxes were measured from disturbed and undisturbed wind sectors over the measurement period, significant differences were found (including air temperature and vapour pressure deficit). As we were unable to measure CO 2 fluxes from both disturbed and undisturbed tundra simultaneously, our flux comparisons are dependent on wind directions. The magnitude of the estimated impact of disturbance may be influenced by the seasonality of CO 2 fluxes. Cooler temperatures and greater wind speeds characterized times when fluxes were measured from disturbed tundra at both RTS-1 and RTS-2. As such, fluxes from these areas were found to be smaller than those measured during times when wind came from the undisturbed tundra sector, which were characterized by warmer temperatures during our measurement period. We used conditional sampling to address biases of meteorological conditions associated with fluxes from disturbed and undisturbed tundra. We posit disturbance had some effect on NEE, which is supported by our chamber measurements and the findings of Cassidy et al. (2016).
NEE measured at High Arctic sites are generally smaller than those measured at Low Arctic sites (Lafleur et al. 2012). If disturbances occur in areas that sequester the most CO 2 in a landscape the effect on the landscape CO 2 balance will be greater than in areas with lower initial sequestration. In the Low Arctic, very large slumps could have large impacts on the CO 2 balance of landscapes as this region is carbon rich and has higher fluxes (Lafleur et al. 2012). The lower CO 2 fluxes measured and the smaller amounts of carbon in the soils of the High Arctic indicate that the impacts of disturbances here will D r a f t 27 not have the same magnitude of effects on regional carbon balance or the atmosphere as in the Low Arctic. Understanding the spatial variation in thermokarst will be critical to understanding the magnitude of their effects on the climate system.
Based on the data presented here we draw the following conclusions: 1) initial site conditions influence the impacts of permafrost disturbance on ecosystem structure and function; 2) differences between the three sites for many variables were greater than the impacts of disturbance within one site; and 3) RTS impact NEE.
Heterogeneity in conditions and responses was amplified by disturbance, resulting in unique combinations of vegetation communities, site characteristics, and carbon fluxes.
The establishment of flux towers at two locations allowed us to determine the variability in the impacts of permafrost disturbances on CO 2 fluxes simultaneously throughout the season and examine the role of heterogeneity on NEE. However, as fluxes were dependent on wind direction, only either disturbed or undisturbed tundra could be measured at any time and site.
As the magnitude and frequency of disturbance is predicted to increase with increasing temperatures and precipitation, it becomes important to determine the impacts of disturbance both at the plot-scale but also at the landscape scale. Understanding both scales will help us predict the potential ecosystem changes that may result from modified disturbance regimes in the Arctic.    Table 4. Summary of soil characteristics by site and treatment (mean ± SE) Table 5. Summary of soil nutrient availability at three study locations (mean ± SE), RTS-1, RTS-2, and RTS-3 in disturbed (d) and undisturbed (c) tundra. All concentrations are µg cm 2 x 24 days. Table 6. Results of soil nutrient availability analysis with significant differences between zones and among sites shown in bold          D r a f t Tables   Table 1  Site  Type     D r a f t