Introduction
Arctic ecosystems have long acted as carbon sinks due to their consistent low
air temperatures (Ta) and the presence of permafrost that both
inhibit the mineralization of soil carbon. Although Arctic net primary production and
standing biomass are smaller than those of adjacent climate zones (Saugier et
al., 2001), the low decomposition rates of Arctic ecosystems have resulted in
an accumulation over 1000 Pg of belowground organic carbon in the upper 3 m
of the soil in permafrost regions (Hugelius et al., 2014; Schuur et al.,
2015). However, the tendency for Arctic ecosystems to take up more CO2
on average than they release may be changing due to global climate change,
which has given rise to shifts in Ta and precipitation patterns.
While the photosynthetic rates and standing biomass in the Arctic have become
larger (Epstein et al., 2012; Jia, 2003; Myneni et al., 1997; Xu et al.,
2013), the rate of organic carbon decomposition has also increased
(Bond-Lamberty and Thomson, 2010), which could potentially accelerate
CO2 cycle processes. Further, it is not only a matter of how fast
CO2 circulates between the atmosphere and the upper soil layers but
also what will happen to the massive amount of stored carbon (Schuur et al.,
2009). Thus, understanding how CO2 flux patterns of Arctic ecosystems
change as a consequence of climate change, as well as how this affects the
fate of permafrost carbon, is of great importance (Abbott et al., 2016; Koven
et al., 2011; Schuur et al., 2008, 2015).
Gross primary production (GPP) is chiefly determined by the length of the
growing season (Baptist and Choler, 2008; White et al., 1999; Xia et al.,
2015) and leaf area index (LAI; Barr et al., 2004); secondary influences are
water and nutrient availability, as well as local climate conditions such as
Ta and radiation (Chapin et al., 2012a). As each plant species
responds differently to changes in the aforementioned factors controlling
GPP, and as successional changes in vegetation species distribution may take
place under a changing climate, the total amount of carbon assimilated
(net primary production, NPP), and plant respiration
(autotrophic respiration, Ra) may undergo changes. The
rate of organic matter decomposition (heterotrophic respiration,
Rh) increases under warmer and more aerobic conditions, and is
also influenced by the quality of available organic matter. If any of these
conditions are modified due to climate change, the rate of Rh may
also change in response.
Warming Ta have been observed in the Arctic (Serreze et al.,
2000), and disproportionately warmer conditions are forecast in response to
climate change (Collins et al., 2013; Kirtman et al., 2013; Overland et al.,
2014). As noted in the preceding paragraph, rates of both photosynthesis and
organic matter decomposition have increased with warmer Ta
(Belshe et al., 2013; Bond-Lamberty and Thomson, 2010; Epstein et al., 2012;
Jia, 2003; Myneni et al., 1997; Xu et al., 2013), and these trends are
predicted to continue. This has the potential to change Arctic terrestrial
ecosystems from a carbon sink to a source, with accelerated organic carbon
decomposition as a dominant process (Koven et al., 2011; Schaefer et al.,
2011). Schuur et al. (2015) predict that, under the current climate warming
trajectory, ∼ 5–15 % of the permafrost carbon pool may be
released into the atmosphere by 2100.
An increase in Ta can have an immediate impact on soil hydrology,
potentially adding complexity to the drivers of CO2 fluxes and the
permafrost carbon pool. In permafrost regions, land surface warming is
usually followed by topographical changes, and thus the formation of
small-scale local hydrological conditions; wetter microsites can form due to
subsiding ground following permafrost thaw (Jorgenson et al., 2006; O'Donnell
et al., 2011), while adjacent areas become drier as water drains laterally to
subsided areas. These phenomena are particularly pronounced when increased
Ta thaws ice-rich permafrost, such as ice wedges and ice lenses
(Liljedahl et al., 2016). In some Arctic regions, changing precipitation
patterns can aggravate or offset this situation. Precipitation in the Arctic
has been generally increasing over the last 5 decades (Kattsov and Walsh,
2000), but patterns are fluctuating across both time and space (Curtis et
al., 1998; Stafford et al., 2000); at times, the surface water balance has
also been found to be decreasing (Oechel et al., 2000). Although, overall,
greater precipitation is expected in the Arctic as the result of intensified
hydrological cycles under climate change, the net effect may significantly
vary by region (Bintanja and Selten, 2014; Huntington, 2006; Kirtman et al.,
2013). Different areas in the Arctic may therefore become either wetter or
drier through the combined effects of atmospheric warming and permafrost
thaw, as well as varying rates of precipitation.
Several studies have investigated the effects of drainage on CO2 fluxes
in the Arctic (Table 1). Field water table depth (WTD) manipulation
experiments and comparison studies with varying WTD have generally shown
decreased net CO2 uptake or increased net CO2 emission at lower
water levels, primarily due to increased ecosystem respiration
(Reco; Christensen et al., 2000; Huemmrich et al., 2010; Kim, 2015;
McEwing et al., 2015; Oechel et al., 1998; Olivas et al., 2010; Zona et al.,
2011); in most cases, GPP increased as well. Although some studies have shown
slightly increased net CO2 uptake when the increase in GPP is larger
than the increase in Reco under drier conditions (Natali et al.,
2015), the magnitude of the increase in Reco is usually larger than
that of GPP (Christensen et al., 2000; Huemmrich et al., 2010; Kim, 2015;
McEwing et al., 2015; Oechel et al., 1998; Olivas et al., 2010; Zona et al.,
2011). The between-site variability of changes in net ecosystem exchange
(NEE) presented in Table 1 can be attributed to differences in the
observation period, vegetation type, as well as the intensity and duration of
WTD changes in the specific studies. Microcosm experiments have also shown
inconsistent results, with a decrease in water level resulting in either
decreased (Johnson et al., 1996) or increased (Peterson et al., 1984) net
CO2 fluxes. These findings exemplify how the net effect of changes in
WTD arise from interactions between multiple factors, and can vary strongly
depending on vegetation and soil types (Billings et al., 1982). Therefore,
although previous studies have shown that WTD reduction affects GPP and
Reco rates, the direction and significance of changes in net
CO2 cycling have been found to differ from ecosystem to ecosystem.
CO2 flux changes (g C-CO2 m-2 day-1) in
response to a WTD decrease, expressed as either
fluxcontrol–fluxlower-WTD or
fluxhigher-WTD–fluxlower-WTD. Negative net CO2 flux
rates represent a net increase in terrestrial CO2 uptake; positive
changes denote a decrease in net CO2 uptake by the terrestrial ecosystem
or an increase in terrestrial CO2 emissions to the atmosphere. The
ranges of these changes are from different years, soil types, and study
sites. Numbers in parentheses represent percent change compared to the
original (control, WTD condition) flux.
Sites
WTD change
Net CO2 flux change
Reference
Coastal plain
Drawdown
+0.59 (+160 %)
Huemmrich et al. (2010)
3 cm lower
+0.23 (+87 %)
Olivas et al. (2010)2
Up to 3.6 cm lower1
+1.17 (+63 %)
Christensen et al. (2000)3
7–7.5cm lower
+0.36 to +0.4 (+450 to +500 %)
Oechel et al. (1998)
8.5 cm lower
+2.99 (+365 %)
Kim (2015)4
11.9 cm lower1
+0.41 (+67 %)
Zona et al. (2011)4
20 cm lower1
+0.72 (+37 %)
McEwing et al. (2015)
Floodplain
20–35 cm lower
-0.06 (-47 %)
Merbold et al. (2009)5
Moist tundra
2.5 cm lower
-0.02 (-3 %)
Natali et al. (2015)6
Laboratory
Saturated vs.
-2.63 to -1.41 (-1716 to -344 %)
Johnson et al. (1996)
field capacity
5 cm lower
-0.61 to +0.96 (-59 to +72 %)
Billings et al. (1982)
10 cm lower
+2.21 (+184 %)
Peterson et al. (1984)
1 WTD difference from natural variation instead of
manipulation. 2 Only data from 2008 were used, as this was the only time
when the WTD of the drained area was lower than that of the control area.
3 Only from grassland data. 4 Only Reco was
considered (no GPP). 5 Only data from 2003 and 2005
were used, as these were the only years when climate conditions were similar.
6 Only data from 2013 were used, as this was the only time when the WTD
of the drained area was lower than that of the control area.
Most of the existing field observation and incubation studies (Table 1) have
focused on the short-term effects of changes in WTD, with a few exceptions
that included permafrost thaw history (Johnston et al., 2014; Schuur et al.,
2009). A further limitation is that most of these studies have been carried
out in North America, despite the fact that permafrost regions in Eurasia not
only cover about twice the area but also contain twice the amount of carbon
as compared to North America (Tarnocai et al., 2009). Drying manipulation
experiments in the Eurasian Arctic with timescales of decades or more will
therefore greatly contribute to understanding drainage effects on CO2
fluxes in Arctic ecosystems. In addition to these growing-season CO2
fluxes, several studies have highlighted significant contributions of
non-growing-season CO2 emissions to the annual CO2 budget in the
Arctic (Coyne and Kelley, 1971; Kelley et al., 1968; Panikov and Dedysh,
2000; Webb et al., 2016; Zimov et al., 1993, 1996). Because of the insulation
provided by snow, soil temperatures (Tsoil) remain warmer compared
to Ta, and biological processes may continue throughout the
non-growing season (Kelley et al., 1968; Webb et al., 2016; Zimov et al.,
1993, 1996). Non-growing-season fluxes are also affected by state changes
from water to ice (Mastepanov et al., 2013). However, no studies have yet
compared non-growing-season CO2 fluxes between wet and dry sites.
A schematic showing how a decade-long drainage installation affects
a floodplain ecosystem and CO2 fluxes. Drainage of a floodplain
ecosystem alters Tsoil through changing heat capacity and
thermal conductivity, with increased Tsoil in shallow layers,
decreased Tsoil in deeper layers, and shallower TDs, as well as
decreasing the abundance of wetland grasses while increasing the abundance
of shrubs. These modifications will subsequently affect CO2 fluxes by
changing the rates of GPP and possibly increasing
Reco, which consists of Ra and Rh.
![]()
Aerial photograph of the site, including schematics of the drained
(bottom left) and the control (top right) transects. Names of plots are
written with numbers and two core plots for more frequent flux measurements
are highlighted in yellow in each transect. Fluxes, vegetation community
structures (created using a non-destructive method), WTDs, and
TDs were measured in all plots, and Tsoil were measured
in even-numbered plots only.
![]()
As a continuation of hydrological manipulation initiated a decade ago in
northeastern Siberia (Merbold et al., 2009), the present study investigates
how 10 years of drainage have affected ecosystem structure and CO2
fluxes. By directly comparing CO2 fluxes of a pristine area to those
from the drained area, our results go beyond a mere description of the
immediate disturbance effects, and clearly point out differences between the
properties of pristine and drained ecosystems. These differences highlight
how the disturbed area has adapted to persistently drier conditions. Our
investigation is focused on shifts in Tsoil, thaw depths
(TDs), and vegetation community
structure, as well as how these changes then influence net CO2 exchange
and its component fluxes, GPP and Reco (Fig. 1). In addition to the
growing season, phenomena during the non-growing season will be also
described; this represents the first drying manipulation experiment of this
nature that extends beyond the growing season.
Methodology
Site description
The study site is located in a Kolyma River floodplain near Chersky,
northeastern Siberia (also written as Cherskii or Cherskiy). The dominant
vegetation species are tussock-forming Carex appendiculata and
lugens, and Eriophorum angustifolium. An organic peat layer
(15–20 cm deep) has accumulated on top of alluvial material soils (composed
of silty clay), although some organic peat materials can be found within
alluvial layers due to cryoturbation.
Based on the record filtered by the Berkeley Earth project
(berkeleyearth.org, station ID 169921) for the period 1960–2013, mean
monthly Ta at the Chersky weather station ranged between
-33 ∘C in January and +13 ∘C in July, and the annual
mean Ta was -11 ∘C. World Meteorological Organization
(WMO) records for the period 1950–1999 indicate a total annual precipitation
of 197 mm, with about half of this falling as rain in summer. Snowmelt at
the site and in the surrounding river basin usually results in a spring
flood. This flooding brings an increased water level of up to 50 cm above
the soil surface in late May or early June, followed by a gradual decrease in
the water level starting in early July. After the flood waters have receded,
the primary water source is precipitation.
A drainage ring with a ∼ 200 m diameter and minimum depth of 50 cm
was constructed in fall 2004 (Merbold et al., 2009), to drain water into the
nearest river channel (Ambolikha). As a result, WTD in this drained area is
lowered by 20 cm on average and by up to 30 cm in the growing season
compared to control areas (Merbold et al., 2009). While the spatial range of
drainage effects varies by soil topography, high-resolution land cover
classification (WorldView with 2 m resolution; Richards and Xiuping, 1999)
has indicated a high abundance of vegetation groups dominant in dry areas to
only within 200 m on both sides of the ditch (Burjack et al., unpublished
data); we can therefore limit the drainage effect to this maximum distance.
Starting summer 2013, we measured ecosystem properties and CO2 fluxes at
two sites in parallel (Fig. 2): the drained area affected by the ditch since
2004 (68∘36′47′′ N, 161∘20′29′′ E), and a control
area (68∘37′00′′ N, 161∘20′59′′ E) approximately
600 m away from the drained area that is not affected by the drainage ditch.
Despite some short-term diurnal fluctuations of up to a few centimeters following
evapotranspiration, as well as precipitation events and water supply from
thawing permafrost, distinct differences in WTD between these treatment areas
persist over the growing season. Each transect of 10 plots in the drained
and control areas (henceforth referred to as drained and control transects,
respectively) was selected using a stratified systematic sampling method.
First, we selected 10 approximate positions with 25 m intervals along the
boardwalks or transects; we then selected the final plots by considering
representative vegetation groups of the selected positions, and by selecting
specimens small enough to fit within flux chambers (Table 2 and Fig. 2). All
plots were located within ca. 2 m of boardwalks to minimize disturbances.
WTD and vegetation characteristics of plots. Vegetation groups were
created by taking into account only Carex sp., E. angustifolium, and shrubs when the relative abundance of each species
exceeded 10 %. The relative abundances of consisting plant species (mean
±SD) are separately presented. Average WTD was calculated by pooling all
WTD measurements from both years (mean ±SD), except the period where the
whole area was flooded from snowmelt. When the average WTD of the growing
season was larger than -10 cm, plots were classified as wet groups.
Transect
Plot ID
Vegetation group
Vegetation
Average
WTD
Group abbr.
no.
abundance (%)
WTD (cm)
group
Drained
0
EriophorumShrub
90, 10
4.6 ± 2.2
Wet
Drained_wet
1, 2, 4
CarexEriophorum
31 ± 23, 64 ± 21
-14.1 ± 8.4
Dry
Drained_dry
3, 5, 6, 7, 8, 9
Carex
82 ± 30
-19.2 ± 6.1
Dry
Drained_dry
Control
0
CarexShrub
85, 15
-1.3 ± 2.3
Wet
Control_wet
1, 3, 6, 7, 8, 9
Eriophorum
79 ± 33
4.3 ± 2.4
Wet
Control_wet
2
EriophorumShrub
80, 20
3.9 ± 2.1
Wet
Conrol_wet
4, 5
CarexShrub
71 ± 12, 27 ± 15
-18.5 ± 4.1
Dry
Control_dry
We conducted three field campaigns. The first was 3 weeks, starting on
20 July 2013 (representing the mid-growing season); the second was 4 weeks,
starting on 1 November 2013 (representing the non-growing fall season); and
the third was 10 weeks, starting on 15 June 2014 (representing the growing
season). The non-growing season was defined as the time period when the
average daily Ta was below 0 ∘C. Although WTD of the
drained transect was generally lower by 20 cm than that of the control
transect after the spring flood in both years, heterogeneous soil topography
rendered varying WTD within each transect: one plot in the drained transect
had an average WTD close to that of wet plots in the control transect, and
two plots in the control transect had an average WTD close to that of dry
plots in the drained transect. Since our objective was to analyze how a
decade-long drainage disturbance affects CO2 fluxes and its links to
environmental parameters, we categorized 20 plots into four groups –
drained(D)_wet, drained_dry, control(C)_wet, and control_dry – according
to transect and WTD category (Table 2). Plots were classified as “dry” when
the average WTD of the growing season was lower than -10 cm. In 2013, all
20 plots were observed with equal frequency to investigate spatial
variability among plots; in 2014, four core plots (i.e., one plot from each
group; Table 2) were more frequently observed to highlight temporal
variability over the growing season (Table 2 and Fig. 2). Due to different
lengths of the observation periods between the two years, we divided data
from 2014 into three subseasons to distinguish seasonal variability: (2014.1)
15 June–5 July, (2014.2) 6 July–26 July, and (2014.3) 27 July–20 August.
Subseason 2014.3 and the 2013 field campaign covered similar periods, based
on an analysis of plant phenology with the normalized difference vegetation
index (NDVI), and both periods included peak growing season (i.e., when the
NDVI of the site was the highest).
CO<inline-formula><mml:math display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> flux measurements
At each plot, a 60 × 60 cm2 polyvinyl chloride (PVC) collar
was inserted 15 cm into the ground in late June 2013, 3 weeks before the
first flux measurements. No noticeable plant damage was identified around the
collars after installation, and 3 weeks was expected to provide enough
buffer time for any stabilization needed in the event of minor belowground
damage (Högberg et al., 2001; Parkin and Venterea, 2010). To take the
flux measurements, a transparent chamber (60 cm on each side, made of
4 mm thick plexiglass) was placed on the collar. The chamber had an opening
valve on the top to avoid pressure effects when we placed the chamber onto
the collars. Sensors for Ta, air humidity, air pressure
(Pa), and photosynthetically active radiation (PAR) were attached
to one side of the chamber and all parameters were measured in parallel with
fluxes. These sensors – along with three small fans on a vertical pole
attached in one of the corners, for the purpose of mixing the air inside –
were placed such that their shadows would not bias incoming solar radiation.
CO2 flux was measured with a non-steady-state flow-through (i.e., closed
dynamic) method using an Ultra-Portable Greenhouse Gas Analyzer (UGGA, Los
Gatos Research, USA), and all data were recorded at 1 Hz with a CR1000 data
logger (Campbell, USA).
We restricted each flux measurement to a maximum of 2 min to minimize
saturation effects (i.e., warming and pressurized effects) within the
chamber. In the event of strong incoming radiation, which can cause
Ta to increase more than 1 ∘C per minute, we placed ice
packs on the collar rims inside the chamber to keep Ta constant
while measuring fluxes. The number of ice packs was adjusted by observing the
Ta changes at 1 Hz frequency. In addition to measuring NEE using
the transparent chamber, in summer we also measured Reco by
covering the chamber with a tarp that blocked incoming radiation. In the
non-growing season we did not find significant differences between NEE and
Reco, probably due to the role of low temperatures, low solar
radiation, and snow cover in limiting photosynthesis; we therefore measured
NEE only with the transparent chamber.
To calculate the CO2 flux from the observed changes in CO2
concentrations ([CO2]) within the sampling time of 2 min, median values
of the [CO2] slopes were computed by selecting multiple time windows
based on a bootstrapping approach, and fluxes
(mg C-CO2 m-2 s-1) were calculated by taking into account
Ta, Pa, and the volume and area of the chamber
(Rochette and Hutchinson, 2005). Flux rates that fell outside of the range of
seasonal mean ±3σ (i.e., standard deviation) were removed as
outliers. GPP and Reco are expressed in positive values, indicating
the amount of CO2 assimilated and respired, respectively. Negative
values for NEE denote net CO2 uptake by the terrestrial ecosystem, while
positive values denote net CO2 emission to the atmosphere.
WTD, TD, and <inline-formula><mml:math display="inline"><mml:mrow><mml:msub><mml:mi>T</mml:mi><mml:mtext>soil</mml:mtext></mml:msub></mml:mrow></mml:math></inline-formula>
WTD was measured during each flux measurement using perforated PVC pipes with
a 25 mm diameter, which were installed at each plot. WTD was measured
relative to soil surface, with values larger than 0 cm denoting water
standing above the soil surface. TD was estimated by pushing a measuring pole
into the ground. At every second plot Tsoil probes were installed
at 5, 15, 25, and 35 cm (Th3-s, UMS, Germany), and data were recorded while
fluxes were measured.
To investigate the effect of Tsoil on Rh, we measured
respiration rates of soils at 0–15 and 15–30 cm depths by aerobically
incubating soils at 15 ∘C in the laboratory (N=6 for each
depth). Respiration rates were corrected for bulk density and average
growing-season Tsoil at each 0–15 and 15–30 cm depth of both the
wet and dry plots by assuming a Q10 value of 2 as the mean for tundra
ecosystems (Zhou et al., 2009). The relative Rh rates
between the wet and the dry plots were subsequently compared, and were linked
to changes in Tsoil.
Vegetation community structure
Changes in vegetation community structure between 2003 (before the drainage
ditch was installed) and 2013 (9 years after the drainage ditch was
installed) were examined using historical data collected in 2003 through the
Terrestrial Carbon Observation System Siberia project (TCOS Siberia; Corradi
et al., 2005). Vegetation community structure was then identified in 2013
along the same transect as in 2003 (which had not been drained in 2003, but
was drained in 2013), as well as in the control transect (newly selected in
2013). Identification was carried out using the same harvest method in all
transects. All living vegetation inside a 1 × 1 m2 quadrat (N=4 per transect) was harvested. Collected vegetation was sorted by
species, completely dried at 40 ∘C, and then weighed (g dry biomass
m-2). Relative abundance of each species (%) was calculated based on
the dry biomass to avoid potential biases linked to the water content of
plants.
To correlate abundances of plant species with CO2 fluxes without
destroying plots for further flux observations, we applied a non-destructive
point-intercept method using a 60 × 60 cm2 quadrat that was
divided into 10 × 10 cm2 subgrids in 2014. After creating
this grid, we recorded the plant species that a laser pointer hit when
pointed downward at each subgrid intersection, and calculated the percentage
of each species' cover. This analysis was performed within each collar, so
that vegetation community structure of each plot could be linked directly to
CO2 fluxes. As plots were selected using a stratified method (see
Sect. 2.1), this analysis was also performed at a spot 10 m away from each
plot, to confirm that the vegetation community structure of each plot
accurately represented the transects.
Data analysis and interpolation
Interpolation of growing-season CO<inline-formula><mml:math display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> fluxes
To compare flux variability among plots induced by temporal discrepancies in
sampling, and to visualize the implications of these differences for net
growing-season CO2 uptake, CO2 fluxes for each vegetation and WTD
group were interpolated throughout the growing-season observation period. To
simulate CO2 flux rates we adapted a satellite-data-driven CO2 flux
model, the Polar Vegetation Photosynthesis and Respiration Model (PolarVPRM),
which calculates high-latitude NEE by subtracting GPP from Reco
(Luus and Lin, 2015):
GPP=λ×Tscale×Wscale×FAPARPAV×11+PARPAR0×PARTscale=(Ta-Tmin)×(Ta-Tmax)Ta-Tmin×Ta-Tmax-(Ta-Topt)2Wscale=a×WTDWTDmax-WTDmin+b(0<a<1,a+b=1),
where λ is a parameter representing maximum light use
efficiency at low light levels, and PAR0 represents the half-saturation
value of PAR. Tscale and Wscale are scaling variables
ranging between 0 and 1 that reflect the influence of Ta and
water availability, respectively, on photosynthesis. The set of three
parameters required for calculating Tscale, i.e., Tmin,
Tmax, and Topt were set to 0, 40, and 20 ∘C
according to literature recommendations to avoid the parameter instability
that would arise from empirically fitting these parameters, due to the strong
positive correlations between Ta and PAR (Mahadevan et al.,
2008). FAPARPAV is the fraction of PAR absorbed by the vegetation,
and is calculated using the Moderate Resolution Imaging Spectroradiometer
(MODIS) Enhanced Vegetation Index (EVI).
Site-level meteorological observations of Ta and PAR were used as
inputs for PolarVPRM; these observations were taken from sensors installed in
the chamber system (for calibration) and from nearby meteorological towers
(for temporal interpolation; Fig. 2). The influence of water availability on
photosynthesis (Wscale) was calculated based on WTD determined next
to each plot at the time of flux measurement, with an optimized scaling
factor (a, b) to obtain the best fits between GPPmodeled and
GPPobserved.
Both parameters (λ and PAR0) were fitted empirically
in R (R Core Team, 2013). PAR0 was obtained from the curve fit between
GPP and PAR measured with flux observations using the nonlinear least squares
curve fitting in R (R Core Team, 2013); λ was calculated as
the slope of the linear regression of observed GPP, and of GPP calculated
from Eq. (1). GPP was estimated excluding PAR terms for CarexEriophorum in
2013 because no positive relationship between GPP and PAR was found (see
Sect. 3.3.2). GPP was then computed half-hourly using linearly interpolated
WTD and EVI, as well as half-hourly measured Ta and PAR from the
meteorological station.
Reco was calculated using an empirical Q10 model:
Reco=α×ek×Ta.
The two free parameters in this exponential relationship between
Reco and Ta, α and k, were empirically
calculated from chamber-based measurements of Reco and
Ta using nonlinear least squares curve fitting in R (R Core
Team, 2013). Once these coefficients were calculated, Reco was
calculated at half-hourly intervals with half-hourly-averaged Ta
from the meteorological station at each transect.
Parameter optimization and flux interpolation were carried out separately
across four core plots for the year 2014, while 10 plots from each transect
were categorized into three vegetation groups and pooled for the year 2013.
These vegetation categories took into account only Carex sp.,
E. angustifolium, and shrubs when the relative abundance of each
species exceeded 10 % (Table 2). The categorized vegetation groups of the
drained transect were EriophorumShrub, CarexEriophorum, and Carex, while those
of the control transect were CarexShrub, Eriophorum, and EriophorumShrub
(Table 2). The period of interpolation was restricted to the observation
periods within each year because WTD (Wscale) was not measured
continuously outside of this period. The discrepancies between the observed
and modeled fluxes were calculated using root mean squared error (RMSE) and
mean bias error (MBE). All data points that were used for calibration were
utilized for the error estimates due to the limited number of data points.
Uncertainty ranges of the interpolated fluxes were calculated using cross
validation by creating 2000 data subsets consisting of randomly selected data
points (bootstrapping, 80 % of the total data set). To obtain an error
range of Reco, the 2000 resulting pairs of parameters, and,
subsequently, Reco ± 2σ were computed for each
1 ∘C Ta bin. Similarly, 2000 pairs of PAR0 and
λ were estimated for binned PAR. The range of GPP was subsequently
estimated by including the rest of the terms from Eq. (1). To constrain the
uncertainty ranges of the interpolated fluxes, we took the GPP and
Reco error ranges at each point from the corresponding PAR and
Ta bin, respectively, that reflected the current condition.
Because NEE is calculated as the difference of GPP and Reco,
uncertainty ranges were also determined by adding the two error ranges of GPP
and Reco. For CarexEriophorum and EriophorumShrub groups in 2013 –
for which no positive relationship between GPP and PAR was found or the
number of data points was not enough to produce uncertainty ranges,
respectively – the bootstrapping step was skipped.
(a) Spatial variability in WTDs and TDs measured across the two transects on 10 August 2013. Plots are
indicated with squares (core plots indicate closed squares). The letters W and D
indicate the wet and dry WTD category of each plot, respectively. (b) Temporal variability in WTD and TD observed at the four core plots over the
growing season of 2014, separated by transect (columns) and WTD category
(rows).
![]()
Statistical analysis
Spatial differences in the 2013 WTD and TD between the two transects were
tested using an independent t test. A permutational multivariate analysis
of variance (PERMANOVA) was performed to compare vegetation community
structure between the drained and the control transects of 2013 and 2003.
Data from 2014 were not compared with those from 2003 due to the different
experimental methods employed. A two-way analysis of covariance (ANCOVA) was
carried out with WTD category (wet and dry) and depth as independent
variables, to compare Tsoil between WTD categories. Correlations
between WTD and TD were tested by taking values from August of each year when
TD was the deepest and the effects of WTD were strongest.
To see if vegetation groups affected the 2013 fluxes, all fluxes were
aggregated by vegetation group (see Sect. 2.5.1; Table 2) and a one-way
analysis of variance (ANOVA) was performed for each vegetation group as an
independent variable. When independent variables significantly influenced
dependent variables, Tukey's post hoc test was applied. To investigate
whether vegetation group and Tsoil significantly affected the
non-growing-season CO2 fluxes, one-way ANOVA and multiple linear
regressions were performed, respectively. A multiple linear regression
analysis was also performed to identify additional major environmental
drivers for cold-season CO2 fluxes. For multiple linear regression
analyses, significant variables were defined based on the Bayesian
information criterion (BIC); with these selected variables the best-fit
regression models were identified, based on the Akaike information criterion
(AIC). All statistical analyses were performed using R (R Core Team, 2013).
Results and discussion
WTD changes from drainage
Following flooding due to snowmelt in early June, the drainage ditch
effectively lowered WTD in the drained transect. Average differences in WTD
between the two transects were significant with a mean drop of approximately
20 cm, and a maximum difference of up to 30 cm during a 3-week period
in summer 2013 (independent t test, P<0.001, t=-4.55,
df = 17.91; Fig. 3a). Approximately the same difference in mean WTD
was observed in the middle of the 2014 growing season. However, several
significant rainfall events from late July of 2014 triggered an increase in
WTD in the dry plots, especially in the drained transect (Fig. 3b). The
amount of precipitation was similar at both transects, but WTD for some
drained_dry plots was more susceptible to increases in WTD compared to
control_dry plots; this was because the width of the area within the
drainage ring was 3 times larger than that of the elevated areas of
control_dry plots. In addition, drainage may slow when the water level rises
within the drainage ditch due to the obstruction of water flow by taller
vegetation – E. angustifolium and aquatic plants – at the end of
the growing season (Allan, 1995; Green, 2005). As a result, WTD in
drained_dry plots stayed high longer than in the control transect following
heavy rainfalls. Similar patterns were also observed in 2005, 1 year after
the drainage ditch was installed (Merbold et al., 2009). Nonetheless, WTD
difference between the wet and the dry plots showed distinct patterns. In the
long term, it can be speculated that new drainage pathways will be
established, which will lead water away more effectively after precipitation
events, and thus reduce the fluctuations in WTD we observed at our site.
Transferring our findings to a natural disturbance (e.g., the formation of a
connected system of troughs following ice-rich permafrost thaw), we expect
that water drainage will be more effective than our drainage manipulation, as
thawing permafrost following persistently warmer conditions will induce more
pronounced topographical changes (Jorgenson et al., 2006; Liljedahl et al.,
2016; O'Donnell et al., 2011).
Shifts in <inline-formula><mml:math display="inline"><mml:mrow><mml:msub><mml:mi>T</mml:mi><mml:mtext>soil</mml:mtext></mml:msub></mml:mrow></mml:math></inline-formula> and TD and their effects on CO<inline-formula><mml:math display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> fluxes
<inline-formula><mml:math display="inline"><mml:mrow><mml:msub><mml:mi>T</mml:mi><mml:mtext>soil</mml:mtext></mml:msub></mml:mrow></mml:math></inline-formula> and TD
Our two-way ANCOVA indicated that drainage resulted in both stronger diurnal
fluctuations in Tsoil in shallow layers and colder Tsoil
in deep layers, as compared to the wet plots (Table 3 and Fig. 4). This
finding highlights the important role of water content in the thermal
properties of organic soils, with the soil of shallow layers of the dry plots
tending to heat up more easily during the daytime due to the reduced heat
capacity of dry organic soil (Abu-Hamdeh, 2003; Idso et al., 1975; Lakshmi et
al., 2003; Reginato et al., 1976). At the same time, these dry organic soils
also have lower thermal conductivity, limiting downward heat transfer; as a
result, deeper layers remained colder than soil at the same depth in the wet
plots (Abu-Hamdeh, 2003). This mechanism reduced TD in the dry plots, the
effect of which became more distinct at the end of the growing season due to
the continued effects of WTD (Fig. 5). The positive correlations between WTD
and TD in August clearly show this trend (for 2013: r=0.47, P<0.05; for 2014: r=0.67, P<0.001; Fig. 5).
Tsoil profiles based on observations at 5, 15, 25, and
35 cm depths from even-numbered plots each in the drained (top) and the
control (bottom) transects. Boxplot contains median, 25 and 75 %
quartiles, and ±1.5 interquartile ranges. To minimize the impact of the
diurnal temperature cycle on this temporally discontinuous data set, the time
window for averaging was restricted to 13:00–17:00 LT. Panels from left to right
show data from 2013, as well as from three subseasons of the growing season
of 2014: (2014.1) 15 June–5 July, (2014.2) 6–26 July, and (2014.3)
27 July–20 August. Data subsets where significant differences in WTD between
the wet and the dry plots were detected are marked with asterisks (***
P<0.001, ** P<0.01, * P<0.05).
![]()
Correlations of TDs and WTDs in
mid-August 2013 and 2014. Error bars of WTD represent the minimum and the
maximum ranges of WTD of the previous 20 days. Results of correlation
analysis for each year are presented with black lines.
![]()
ANCOVA results with WTD category (wet and dry), and soil depth (cm) as the independent
variables and Tsoil (∘C) as the dependent variable. The
time periods of the entire year 2013, as well as three subseasons of 2014
– (2014.1) 15 June–5 July, (2014.2) 6–26 July, and (2014.3) 27 July–20
August – were separately analyzed. The significance of F values are
denoted with asterisks (*** P<0.001, ** P<0.01, *
P<0.05).
Transect
WTD
Depth
WTD ×
Depth
2013
Drained
12.75***
602.64***
13.38***
Control
15.38***
700.93***
2.64
2014.1
Drained
3.54
169.46***
0.02
Control
32.55***
165.35***
29.56***
2014.2
Drained
26.21***
400.48***
0.52
Control
1.24
380.91***
2.42
2014.3
Drained
101.87***
680.50***
7.55**
Control
6.49*
813.62***
4.91 *
<inline-formula><mml:math display="inline"><mml:mrow><mml:msub><mml:mi>T</mml:mi><mml:mtext>soil</mml:mtext></mml:msub></mml:mrow></mml:math></inline-formula> and TD effects on CO<inline-formula><mml:math display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> fluxes
GPP and Reco rates increased with Tsoil at 5 cm (Fig. 6)
because warmer Tsoil generally accelerates both photosynthesis
(Lawrence and Oechel, 1983; Schwarz et al., 1997) and root respiration (Boone
et al., 1998). The average Reco rates of the dry plots were
25 % higher than those of the wet plots in 2013 (independent t test, P<0.001, t=-5.70, df = 532) despite the fact that GPP
rates were found to be lower in the dry plots, meaning that lower
Ra rates would have been expected (Fig. 6). This increase in the
Reco rates can be partly explained by the increased rates of
Rh under more aerobic conditions following drainage: as anaerobic
respiration is slower and less efficient than aerobic respiration, carbon
release from both organic and mineral soils (surface and deep soil layers,
respectively) under aerobic conditions can be 4–10 times higher than under
anaerobic conditions (Lee et al., 2012). However, drier conditions altered
Tsoil regimes, and these effects further affected Rh
rates.
Links between average Tsoil at 5 cm and (a)
GPP and (b) Reco rates by transect (column). Data are from 2013
(20 July–10 August) and subseason 2014.3 (27 July–20 August); both cover
similar phenological periods. Data were grouped into temperature bins of
5 ∘C.
![]()
Modifications in Tsoil in shallow layers had greater impacts on
Rh than those in Tsoil in deep layers: the warmer
Tsoil in shallow layers of the dry plots increased Rh by
240 %, while colder Tsoil in deep layers reduced Rh
only marginally as compared to the wet plots. Combining these two contrasting
effects, Rh rates in dry plots were elevated by 95 % as a
result of the stronger effects of Tsoil in shallow layers as
compared to Tsoil in deep layers on compacted peat soils. This
increase was largely due to a greater amount of organic carbon – an
increased total carbon content by 11 % as well as more compacted soil
with an increase in bulk density by 44 % (data not shown) – in the dry
plots affected by warmer Tsoil in shallow layers. Contrary to the
accelerated respiration rates at the surface, colder Tsoil in deep
layers and reduced TDs imply that carbon currently stored in
permafrost can be preserved following drainage. In addition to these
contrasting effects, another opposing influence of the physical structures of
vegetation on Tsoil – for example, the negative relationship
between shrub abundance and TD due to shade (Blok et al., 2010) and the
positive relationship between shrub abundance and TD due to decreased albedo
(Bonfils et al., 2012) – with continuously changing vegetation communities
following drainage (see Sect. 3.3) need to be monitored over a longer period
of time to gain further insight into the net impact of secondary drainage on
carbon accumulation and CO2 fluxes.
Shifts in vegetation community structure and its effects on CO<inline-formula><mml:math display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>
fluxes
Vegetation community structure
In its natural, undisturbed state, the vegetation community of this
floodplain has historically been dominated by Eriophorum angustifolium, followed by Carex appendiculata and lugens.
This vegetation community structure was reflected in the observations made in
2003 (Corradi et al., 2005) – that is, before the drainage ditch was
constructed (Fig. 7, black) – as well as in the control transect in 2013
(Fig. 7, blue). After a decade of drainage, the abundance of E. angustifolium decreased, while shrubs (Betula exilis, and
Salix fuscescens and pulchra) and Carex
sp. became the dominant species in the drained transect (Fig. 7, red). While
no statistically significant differences were found between the vegetation
community structures in 2003 and in the control transect of 2013 (PERMANOVA,
F=1.62, P=0.19), significant differences were found between both the
2003 and the drained transect of 2013 (PERMANOVA, F=3.31, P<0.05) and between the two transects of 2013 (PERMANOVA, F=5.22, P<0.05). Although we did not experimentally compare the two
observation methods (see Sect. 2.4), a qualitative comparison of results from
2013 (i.e., harvest) and 2014 (i.e., point intercept) showed a similar
abundance of each species; this implies that these two different methods can
be used to compare vegetation community structures.
Abundances of vegetation species observed across the transects in
2003 and 2013. Numbers in parentheses are the number of replicates. Boxplot
contains median, 25 and 75 % quartiles, and ±1.5
interquartile ranges. Betul: Betula exilis, Calam:
Calamagrostis purpurascens, Carex: Carex species, Chama:
Chamaedaphne calyculata, Eriop: Eriophorum angustifolium,
Poten: Potentilla palustris, Salix: Salix species.
![]()
In the control transect, the vegetation community structures of the wet and
the dry plots were dominated by E. angustifolium and Carex
sp., respectively, but some dry plots within the drained transect showed a
vegetation transition stage (Table 2). Plots in the drained transect that
were categorized as CarexEriophorum (Table 2) showed a mixture of young
Carex sp. (without discrete tussock forms or small developing
tussocks) and short and thin E. angustifolium. The presence of this
mixture implies that these areas were formerly dominated by E. angustifolium, which is abundant in water-saturated areas, but whose
abundance decreased due to drainage. The core plots that were selected based
on drainage manipulation and WTD category represented this vegetation shift
well; control_wet and drained_wet were dominated by E. angustifolium, control_dry was dominated by Carex sp. and shrubs,
and drained_dry showed a transition stage from E. angustifolium to
Carex sp. (Table 2).
We underestimated the abundance of shrubs (B. exilis and
Salix sp.) within the collars in the drained transect as a result of
the methodological choice to exclude tall shrubs when selecting plots to
ensure that all of the vegetation could fit into the chambers when measuring
fluxes (note that these results are presented to compare CO2 fluxes by
vegetation group; see Sect. 2.4). The abundance of shrubs within the collars
of the drained transect was 2 % on average, while independently
investigated average abundance along the transect was 20 % on average.
This discrepancy will be taken into account in the following sections when
interpreting the effects of shrubs on CO2 fluxes.
Vegetation effects on CO<inline-formula><mml:math display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> fluxes
Chamber-based CO2 flux measurements during the 2013 growing season
showed similar mean and standard deviations of NEE, GPP, and Reco
rates between the two transects (Fig. 8). However, fluxes showed a large
variability across plots within each transect (each of which was
ca. 225 m), which results from one-way ANOVA indicated to be closely
linked to the dominant vegetation groups (NEE: F=24.99, P<0.001; Fig. 8a). Reco also differed by dominant vegetation group,
but this difference was not as pronounced as it was for GPP (GPP: F=11.23, P<0.001; Reco: F=3.63, P<0.01;
Fig. 8b, c).
(a) Variability of NEE among individual
plots during the 2013 growing season. Boxplot contains median, 25 and
75 % quartiles, ±1.5 interquartile ranges, as well as mean values
with cross points per plot. The black horizontal bars show the mean flux
rates averaged for the entire transect. (b) GPP
and (c) Reco rates aggregated by the vegetation
group. Significance of differences between groups, determined by one-way
ANOVA and Tukey's post hoc test, is indicated by the letters. Different
letters indicate significant differences between groups while the same
letters indicate significant similarities.
![]()
One of the vegetation effects on CO2 fluxes was that,
Eriophorum-dominated plots in both transects had higher rates of
photosynthetic uptake than Carex-dominated plots. GPP rates of
EriophorumShrub were 55 % higher than those of Carex in the drained
transect, and those of EriophorumShrub and Eriophorum were 20 % higher
than those of CarexShrub in the control transect in 2013 (Fig. 8b). In 2014,
conversely, GPP rates of CarexShrub were 5 % higher than those of
EriophorumShrub in the control transect, but this difference was
insignificant (Table 4). Thus, the decrease in E. angustifolium as a
result of drainage generally reduced carbon accumulation in the terrestrial
ecosystem. CarexEriophorum plots in the drained transect – which represent a
vegetation transition from E. angustifolium to Carex sp.
following drainage – showed the lowest GPP rates in both years, despite the
presence of E. angustifolium (Fig. 8b). In this transition stage
(which is characterized here by declining E. angustifolium) or in
early succession stages, plants assimilate less CO2 than they previously
did due to lower biomass, and can be more susceptible to disturbances (Chapin
et al., 2012b; Niinemets, 2010). The dry and warm year of 2013 was an
especially good example of this process: these plots showed slightly
decreased GPP rates along with increasing PAR (Fig. S1 in the Supplement),
implying that the combination of high PAR and high Ta caused
water stress to plants. Under the same climate conditions, CarexShrub in the
control transect – which can be considered to be the potential vegetation
communities of CarexEriophorum groups – took up significantly less CO2
than in 2014, but did not show decreasing GPP rates (Fig. S1). This implies
that when E. angustifolium is fully replaced by Carex sp.
and shrubs, the current CarexEriophorum plots may not undergo water stress as
easily as they currently do although they can be still strongly influenced by
climate. Taking into account such transition effects after 10 years of
drainage is important given that the fraction of these areas of the total
area – 3 out of 10 plots – is not small. Moreover, this finding highlights
the fact that ecosystem adaptation to new environmental conditions may take a
long time, as 10 years was evidently not sufficient for this ecosystem to be
resistant to disturbances, e.g., harsh climate conditions.
Average daily flux (g C-CO2 m-2 day-1) from
interpolation for the period of 22 July–10 August (20 days) in both 2013
and 2014. Values in parentheses are cumulative flux (g C-CO2 m-2)
for the period of 22 July–10 August (20 days) in 2013 and 16 June–20 August
(66 days) in 2014. Results represent the fits of all data points
±SD from bootstrapping. NEE was calculated by
subtracting GPP from Reco: positive values are net CO2 emission to the
atmosphere, and negative values are net CO2 uptake by the terrestrial
ecosystem.
Year
Group
Reco
GPP
NEE
2013
D_Carex
2.03 ± 0.10 (41 ± 2)
3.42 ± 0.00 (68 ± 0)
-1.38 ± 0.09 (-28 ± 2)
D_CarexEriophorum
1.89 ± 0.16 (38 ± 3)
3.30 (66)*
-1.41 ± 0.24 (-28 ± 5)*
D_EriophorumShrub
1.88 ± 0.53 (38 ± 11)
4.81 (96)*
-2.93 ± 0.34 (-59 ± 7)*
C_CarexShrub
2.04 ± 0.08 (41 ± 2)
2.55 ± 0.02 (51 ± 0)
-0.51 ± 0.05 (-10 ± 1)
C_Eriophorum
1.76 ± 0.05 (35 ± 1)
3.41 ± 0.01 (68 ± 0)
-1.65 ± 0.04 (-33 ± 1)
C_EriophorumShrub
2.34 ± 0.17 (47 ± 3)
3.32 ± 0.00 (66 ± 0)
-0.98 ± 0.14 (-20 ± 3)
2014
D_wet (EriophorumShrub)
3.27 ± 0.16 (184 ± 9)
7.59 ± 0.11 (404 ± 6)
-4.31 ± 0.05 (-221 ± 3)
D_dry (CarexEriophorum)
3.51 ± 0.19 (200 ± 9)
5.14 ± 0.07 (274 ± 4)
-1.64 ± 0.11 (-74 ± 5)
C_wet (EriophorumShrub)
2.81 ± 0.24 (162 ± 14)
5.85 ± 0.03 (312 ± 1)
-3.05 ± 0.21 (-150 ± 13)
C_dry (CarexShrub)
3.98 ± 0.21 (222 ± 12)
6.20 ± 0.03 (331 ± 2)
-2.22 ± 0.18 (-109 ± 10)
* As no bootstrapping was conducted on data for GPP, error range in NEE
is only from Reco.
Increasing shrub abundance slightly compensated for lowered GPP rates in
drained areas following a reduction in E. angustifolium coverage.
EriophorumShrub of the control transect, which had 10 % shrub coverage,
had, on average, 4 % higher GPP rates than Eriophorum in 2013, although
this difference was not significant (Fig. 8b). This difference is expected to
be larger if the abundance of shrubs was not underestimated (see
Sect. 3.3.1). Also, this compensation may become larger with increasing
abundance and biomass of shrubs following continuing drainage. Increasing the
abundance of shrubs not only changes carbon exchange rates between the
atmosphere and the terrestrial ecosystem but also carbon storage patterns
within the terrestrial ecosystem (Shaver and Jonasson, 2001). In the drained
transect, living aboveground biomass – the sum of leaf and stem – was
larger than in the control transect, and in 2003, while the biomass of green
leaves decreased, that of stems increased, mostly due to the increased
abundance of shrub species (Fig. S2). When shrubs continue to expand, a large
portion of carbon will be stored in plants, especially in shrubs' stems, and
the proportion of litter added to the soil will decrease accordingly
(Fig. S2). The subsequent effects of these changes, such as litter quantity
and quality added to soils and its decomposability (Hobbie, 2008; Schädel
et al., 2014), need to be further investigated to better understand long-term
vegetation effects on CO2 fluxes.
Growing-season CO<inline-formula><mml:math display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> fluxes
Gap-filled growing-season CO<inline-formula><mml:math display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> fluxes
The modeled fluxes for both 2013 and 2014 had similar patterns to the
observed fluxes: Eriophorum-dominated plots (i.e., wet plots)
generally showed higher GPP rates than Carex-dominated plots (i.e.,
dry plots) in both transects (Table 4). In addition, the 10 % difference
in shrub cover between Eriophorum and EriophorumShrub from the control
transect did not significantly affect GPP rates (Table 4). Reco
rates were consistently greater in the dry plots, in part due to increased
Rh rates, and the cumulative Reco increased with
drainage by 5 % in 2013 and by 10 % in 2014 (Table 4). Combining the
effects of vegetation and Tsoil on GPP and Reco rates,
the net effects of drainage on CO2 fluxes (NEE) was -0.3 g
C-CO2 m-2 day-1 (i.e., 25 % more CO2 uptake) in 2013
and +0.98 g C-CO2 m-2 day-1 (i.e., 35 % less CO2
uptake) in 2014 when daily CO2 fluxes of 20 days, weighted by the number
of plots of each group, were compared (Table 4). This range was comparable to
other drainage studies presented in Table 1. However, the net CO2 flux
changes were in opposite directions in these two years due to control_dry
(CarexShrub) plots' sensitivity to dry and warm conditions during the 2013
observation periods (see Sect. 3.3.2), as well as pooling of the wet and the dry
plots to compare changes in flux rates for transect level. Despite the
variability between years for transect level, patterns of underlying
processes were consistent: after 10 years of drying manipulation, the
replacement of E. angustifolium by Carex sp., more aerobic
conditions, and increased Tsoil in shallow layers all weakened
CO2 uptake and increased CO2 emission (Table 4).
Model error from interpolation
Comparing observed against modeled flux rates for all individual measurements
in the database, the mean RMSE of Reco was 0.009 and 0.007 mg
C-CO2 m-2 s-1 for 2013 and 2014, respectively, and that of
GPP was 0.021 and 0.016 mg C-CO2 m-2 s-1 for 2013 and 2014,
respectively (Table S1 in the Supplement). Low uncertainty ranges imply that variations in
Reco and GPP can be mainly explained by Ta and PAR,
respectively. The uncertainty ranges of Reco were large compared to
those of GPP (Table S1), suggesting that Reco rates varied with
factors other than Ta, while GPP rates mostly varied with PAR.
Larger RMSE and MBE in the drained transect in 2013 compared to the control
transect can be attributed to the pooling of data points by vegetation group,
as well as to the limited number of data points; the large error in GPP for
the Carex group of the drained transect can be attributed to varying standing
biomass, and that of the EriophorumShrub in the drained transect stems from
the small number of data points (Table S1). In 2014, data points for each
group came from only a single plot, but varying WTD and thickening TD
over the growing season resulted in relatively large errors (Table S1).
Non-growing-season CO<inline-formula><mml:math display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> fluxes
Due to the low Ta and weak solar radiation, GPP in the
non-growing season was negligible. Although a limited amount of
photosynthetic activity could have theoretically taken place during this time
(Atanasiu, 1971), no significant differences were found between NEE and
Reco, implying that CO2 fluxes consisted mostly of CO2
released from the soil. The drained transect emitted an average of 4 times
more CO2 than the control transect: if the observed flux pattern is
representative of the entire month of November, the net CO2 emission of
this month would be 11 g C-CO2 m-2 in the drained transect and
3 g C-CO2 m-2 in the control transect.
Some plots in the drained transect showed sporadically high fluxes, the rates
of which were comparable to Reco rates from the growing season
(Fig. 9). These high fluxes in the drained transect could be linked to
vegetation groups, especially the abundance of E. angustifolium, as
well as to Pa and Tsoil (multiple linear regression,
adj. R2=0.46, P<0.001; Pa, P<0.001; Tsoil at 5 cm, P<0.001; Eriophorum,
P<0.001; Pa× Tsoil at 5 cm, P<0.001; Pa× Eriophorum, P<0.001). This may be a part of the physical processes outlined by Mastepanov
et al. (2008, 2013), through which
the freezing of soil pushes stored CO2 and CH4 gases in soil to the
atmosphere through cracks in soil or dead plant bodies. Although
Tsoil between 0 and 35 cm was consistently below zero,
Tsoil at 35 cm did not fall below -5 ∘C until the end
of November. Ongoing freezing at greater depths than 35 cm and low
Pa could have stimulated CO2 emission from the soil to the
atmosphere through dead E. angustifolium. The fact that the CO2
fluxes were influenced by Tsoil implies that high CO2
emissions were not exclusively triggered by the physical expression of
existing CO2 in soils, but also from ongoing respiration at relatively
mild Tsoil insulated by snow (Kelley et al., 1968; Webb et al.,
2016; Zimov et al., 1993, 1996). CO2 fluxes in the control transect were
also influenced by the abundance of E. angustifolium,
Pa, and Tsoil (multiple linear regression, adj. R2=0.21, P<0.001; Tsoil at 5 cm, P<0.01;
Tsoil at 15 cm, P<0.01; Tsoil at 25 cm, P<0.01; Eriophorum, P<0.05; Tsoil at
5 cm × Eriophorum, P<0.001; Tsoil
at 15 cm × Eriophorum, P<0.001;
Tsoil at 25 cm × Eriophorum, P<0.001), but the rates were relatively constant over time and without high
sporadic fluxes, unlike in the drained transect (Fig. 9).
Change in Pa and NEE from
November 2013 by vegetation type. Changes in Pa refer to changes
within 6 h before individual NEE was measured.
![]()
Although the CO2 fluxes in the non-growing season were partially
explained by vegetation group, Pa, and Tsoil, the
amount of variation (R2) together explained by these factors was low. We
also cannot firmly conclude that the observed sporadic high CO2 fluxes
in November were largely driven by these factors, because we did not observe
CO2 fluxes continuously along with Tsoil; what is more, these
high CO2 fluxes were only observed in the drained transect despite there
being similar conditions in the control transect. High uncertainties and
limitations in predicting both non-growing-season CO2 fluxes and
possible high CO2 fluxes during the thawing season (Friborg et al.,
1997) need to be addressed to determine the net effects of drainage on the
annual CO2 fluxes of this site. Nevertheless, the observed considerably
higher CO2 fluxes in the non-growing season for the drained transect
imply that drainage not only affects growing-season CO2 fluxes but also
has the potential to alter non-growing-season fluxes significantly.
Conclusion and final remarks
Drainage of a floodplain near Chersky resulted in an average WTD drop of
20 cm. This substantially altered both biogeophysical and biogeochemical
ecosystem properties over the span of a decade, with profound net impacts on
CO2 fluxes. The first change important for CO2 processes was that
vegetation community structure in the drained areas shifted significantly
toward increased Carex sp. and shrubs (B. exilis and
Salix sp.) and decreased E. angustifolium. The second
change was that WTD variation led to divergent Tsoil profiles by
depth, with the drained areas showing greater fluctuations in Tsoil
in shallow layers due to their low heat capacity, and with deeper soil
demonstrating colder Tsoil due to the low thermal conductivity of
the dry soil above it. Consequently, the drained areas had shallower TDs compared to the control areas.
These aboveground and belowground changes significantly affected CO2
fluxes. The drained areas showed higher Reco due to more aerobic
conditions, with a greater amount of organic carbon affected by warmer
Tsoil in shallow layers. Dominant plant species in the drained
areas took up less CO2 (i.e., Carex sp. engaged in less GPP)
than E. angustifolium, which is dominant in the control wet areas.
Increased abundance of shrubs slightly compensated for the decrease in GPP,
but, in our data sets, it could not fully balance out the losses. Overall,
drainage increased net CO2 uptake (NEE) by 25 % in 2013 but
decreased it by 35 % in 2014 during the 20 days of the growing season
when the two transects were compared. The opposite patterns of the two years
can be attributed to control_dry plots, which showed large variations with
climate. Despite the inter-annual variability, both years had consistent
trends toward the replacement of E. angustifolium with
Carex sp., more aerobic conditions, and increased Tsoil in
shallow layers, all of which weakened CO2 uptake and increased CO2
emission. In the non-growing season, CO2 emission was 4 times larger
in the drained than in the control areas, partially as a result of the
abundance of E. angustifolium, Pa, and Tsoil.
Ecosystem changes after 10 years of drainage on an Arctic floodplain
decreased CO2 uptake and increased CO2 emission in both the growing
and non-growing seasons. These findings highlight the importance of
considering the changes in ecosystem properties under persistent dry
conditions when investigating CO2 fluxes in response to global climate
changes. As ongoing global warming thaws ice-rich permafrost and makes some
regions drier, Arctic wetlands may accumulate less carbon in the terrestrial
ecosystem, respire more CO2 from shallow soil layers, and preserve
carbon in deep soil layers. Given that vegetation communities continue
changing after 10 years, with different areas then responding differently to
climates, further observations of this site, as well as of other ecosystems
in the Arctic, are needed over a longer term to better predict the fate of
the Arctic in the face of global climate changes.