Introduction
Boreal forests cover roughly 22 % of the earth's terrestrial landscape
(Chapin et al., 2000) and account for approximately 9% of the global
vegetation carbon (C) stock (Carvalhais et al., 2014). Most of the C in
boreal forests, however, is stored in the soil (Pan et al., 2011), where cold
and wet conditions have limited microbial decomposition; as a result, C
has accumulated over the past several millennia (Hobbie et al., 2000;
Trumbore and Harden, 1997). Recent estimates suggest that continuous and
discontinuous permafrost in the boreal region store around 137 Pg, or
40 % of near-surface permafrost (< 1 m) C (Loranty et al., 2016).
Despite the massive amount of C present in the boreal region, the quantity of
C stored here and the magnitude of the change in C stocks that will result
from climate change is one of the least understood carbon–climate feedbacks
(Schuur et al., 2015).
Over the past 50 years, air temperatures in the Arctic have risen nearly
twice the global average as a result of climate change (Christensen et al.,
2013), and this accelerated rate of warming means that the vast amount of C
stored in high-latitude systems is vulnerable to loss to the atmosphere
(Koven et al., 2015; Schuur et al., 2015). The amount of C released as a
result of thaw will be highly dependent on concurrent changes in topography
and hydrology (Liljedahl et al., 2016; Schneider Von Deimling et al., 2015),
vegetation (Guay et al., 2014; Sturm et al., 2005), fire regimes (Berner et
al., 2012; Kasischke and Turetsky, 2006; Rogers et al., 2015; Soja et al.,
2007), nutrient availability (Mack et al., 2004; Salmon et al., 2016), and
soil organic C lability (Harden et al., 2012; Schädel et al., 2014). Yet
despite the vulnerability of permafrost soils to increased thaw and C release
due to climate change, there is a lack of data quantifying the C stocks at
northern latitudes compared to other regions.
Permafrost C pool estimates tend to be dominated by sites located in Alaska
or western Russia, with very few data points from the Russian low Arctic or
Canadian high Arctic (Hugelius et al., 2014; Tarnocai et al., 2009). As a
result, many regions are under-represented in circumpolar permafrost C
estimates (Hugelius et al., 2014; Johnson et al., 2011; Mishra et al., 2013;
Tarnocai et al., 2009). Even in Alaska, which is one of the most densely
sampled Arctic sub-regions, Mishra and Riley (2012) found that the current
sample distribution is insufficient to characterize regional soil organic C
(SOC) stocks fully because of SOC variation across vegetation types,
topography, and parent material. Furthermore, permafrost regions are
characterized by high heterogeneity in soil C stocks due to variability in
soil-forming factors (Vitharana et al., 2017) and at small spatial scales due
to cryogenic processes (i.e., cryoturbation at the sub-meter scale). As a
result, sampling at higher spatial resolution may provide more accurate
estimates of soil C stocks (Johnson et al., 2011; Tarnocai et al., 2009).
Therefore, understanding variation in soil properties at the meter scale is
critical for reducing uncertainty in estimates of current and future
permafrost C pools (Beer, 2016).
Pleistocene-aged C and ice-rich permafrost (i.e., yedoma) deposits occur
across Siberia and Alaska (Strauss et al., 2013) and are particularly
important for regional soil C estimates. Yedoma deposits froze relatively
quickly in geologic history (Schirrmeister et al., 2011; Zimov et al., 2006);
as a consequence, these deep deposits (on average 25 m; Zimov et al.,
2006) are C rich compared to some other permafrost soils (Strauss et al.,
2013; Zimov et al., 2006). Approximately 30 % of high-latitude permafrost
C is found in these yedoma deposits, even though they comprise only 7 %
of the landscape (Walter Anthony et al., 2014). However, due to limited
sampling of deep (> 3 m) permafrost, establishing how much C is in these
deposits is difficult, leading to high uncertainty in estimates of soil C
pools in yedoma deposits (Strauss et al., 2013; Walter Anthony et al., 2014).
While vegetation stores a relatively small portion of the C pool in boreal
forests (approximately 20 %; Pan et al., 2011), it plays a crucial role
in local and global C cycling, and many future changes in C fluxes in this
biome will likely occur as a result of changes in vegetation (Elmendorf et
al., 2012; Euskirchen et al., 2009; Myers-Smith et al., 2015; Swann et al.,
2010). With increased temperatures, boreal forests are susceptible to insect
invasions (Berg et al., 2006; Kurz et al., 2008), moisture stress (Beck et
al., 2011; Trahan and Schubert, 2016; Walker et al., 2015), tree line advance
and retrogression (Lloyd, 2005; Pearson et al., 2013), and more frequent
forest fires (Kasischke and Turetsky, 2006; Rogers et al., 2015; Soja et al.,
2007), which all have the potential to alter C cycling significantly in the
region. Importantly, climate-change-driven alterations in forest cover,
composition, and structure will influence regional energy balance through
impacts on surface albedo, evapotranspiration, and ground insulation, which
will in turn affect ground thaw and soil C cycling (Chapin et al., 2005;
Euskirchen et al., 2009; Fisher et al., 2016; Jean and Payette, 2014; Loranty
et al., 2014).
However, the aboveground processes that regulate C dynamics are not
homogenous throughout the boreal biome (Goetz et al., 2007). For example, the
fire regimes of larch (Larix spp.) and pine (Pinus sylvestris) forests in Siberia are typically dominated by low- to
medium-intensity fires, whereas dark coniferous forests common in Alaska and Canada
are characterized by fires higher in intensity and severity (Rogers et al.,
2015; Soja et al., 2006, 2007; Tautenhahn et al., 2016). The dynamics of
larch forests are particularly important, as they store more than twice the
amount of SOC of all other boreal forest types in the continuous permafrost
zone combined (Loranty et al., 2016). Despite this, larch forests in Siberia
are notably understudied; indeed, the estimate of C stored in Russian
forests is the least well constrained of all forest systems globally (Shuman
et al., 2013).
In this study, we aim to reduce the uncertainty of regional C estimates by
providing a comprehensive assessment of vegetation, active-layer, and
permafrost C stocks in the Kolyma River watershed in northeast Siberia,
Russia. We present aboveground and belowground (to 1 m) C stocks from data
collected from 20 sites across the watershed along with deep permafrost C
pools to 15 m depth from a yedoma deposit and an alas (thermokarst
depression). We compare variation in soil C pools at meter to kilometer
scales in order to quantify the variability of permafrost C at small spatial
scales. Additionally, we examine the drivers of thaw depth and C density of
active-layer soils to understand environmental controls over these variables
across the watershed. Together, these analyses allow us to estimate C pools
and controls over changes in these pools that will likely occur with climate
change.
Methods
Site description
Our study area was a watershed (“Y4 watershed”, ∼ 3 km2;
Fig. 1) located within the Kolyma River basin, which is the largest river
basin (650 000 km2) completely underlain by continuous permafrost
(Holmes et al., 2012). The Y4 watershed is located near Chersky, Sakha
Republic, Russia, approximately 130 km south of the Arctic Ocean and is
underlain by yedoma, which is widespread across the region (Grosse et al.,
2013). The climate is continental with short, warm summers (July average:
12 ∘C) and long, cold winters (January average: -33 ∘C). Annual
precipitation is low (∼ 230 mm) and often occurs during summer
(Chersky Meteorological Station; S. Davydov, unpublished data). Mean summer
temperatures in this region increased by 1 ∘C from 1938 to 2009
(Berner et al., 2013).
Location of the Y4 watershed in relation to Russia (inset) and
location of the sampling sites within the Y4 catchment.
There are two main types of cryogenic deposits within the watershed. Upland
areas are Late Pleistocene syngenetic ice-rich deposits of yedoma. Drained
thaw lake depressions are underlain by alas consisting of lacustrine–wetland
sediments in the upper pedon and taberal (i.e., yedoma that thawed in a talik)
deposits in the lower part of the profile. Permafrost temperatures at 15 m
vary from -2.8 ∘C at the hilltops with relatively thin organic
layers to -4 ∘C in thermokarst depressions with thick (up to
20 cm) moss and peat layers (A. Kholodov, unpublished data).
Site characteristics. All sites were in forested areas except no. 17
(riparian); site no. 18 (alas) had few scattered trees located along one end
of the sampling transects.
Site
Latitude
Longitude
Slope
Aspect
Summer insolation
Stand age
number
(degrees north)
(degrees east)
(degrees)
(degrees)
(WH m-2)
(years)
1
68.74747
161.38988
5
160
4507
155
2
68.74529
161.38908
10
8
3950
167
3
68.74472
161.41486
14
249
4399
203
4
68.74164
161.41562
9
245
4409
23
5
68.74834
161.41350
10
357
3954
218
6
68.74939
161.41759
8
225
4509
205
7
68.74915
161.39000
5
57
4239
155
8
68.74932
161.38820
7
36
4132
208
9
68.75267
161.38544
8
340
4038
202
10
68.75352
161.39455
16
72
4008
211
11
68.74869
161.40834
10
222
4533
123
12
68.74837
161.40237
10
63
4121
71
13
68.74660
161.40433
17
61
3856
179
14
68.74513
161.40063
1
103
4361
40
15
68.75188
161.39095
3
237
4410
221
16
68.75519
161.40013
3
294
4307
200
17
68.74152
161.41411
8
225
4479
–
18
68.74632
161.38776
3
84
4314
–
19
68.74479
161.38410
6
61
4231
26
20
68.74333
161.40688
5
124
4429
–
Forests in the watershed are composed of a single larch species,
Larix cajanderi, with a well-developed understory of deciduous
shrubs (primarily Betula nana, Salix spp., and Vaccinium uliginosum), evergreen shrubs (e.g., Vaccinium vitis-idaea,
Empetrum nigrum, and Rhododendron subarcticum), forbs (e.g., Equisetum scirpoides, Pyrola spp., and Valeriana capitata), graminoids
(Calamagrostis spp.), moss (e.g., Aulacomnium palustre, Dicranum spp., and Polytrichum spp.), and lichen (e.g.,
Cladonia spp., Peltigera aphthosa, and Flavocetraria cucullata).
Site selection and sampling design
We selected 20 stands (i.e., “sites”) in the Y4 watershed that spanned a
range of aboveground tree biomass, as inferred from tree shadows mapped using
high-resolution (50 cm) WorldView-1 satellite imagery (Berner et al., 2012;
Fig. 1). All sites were located in forested stands except for one in a
Salix-dominated riparian zone (Site 17) and another in a
Sphagnum-dominated alas (Site 18; Table 1). Within each site, we
established three 20 m long by 2 m wide plots, each of which was separated
by 8 m and ran parallel to slope contours (Fig. S1 in the Supplement). In the absence of a
discernable slope, transects were aligned north–south. All sampling was
conducted in July 2012 and 2013 except stand age, which was sampled in 2016.
Stand age
To determine stand age, we collected a wood slab or core from the base
(∼ 30 cm above the organic layer) of 5–10 trees sampled randomly
within each stand. Wood samples were dried at 60 ∘C and then sanded
sequentially with finer grit sizes to obtain a smooth surface. Each sample
was then scanned, and the annual growth rings were counted using WinDENDRO
(Regent Instruments, Inc., Ontario, Canada).
Solar insolation and slope
Slope and aspect at each site were determined from a 4 m resolution digital
elevation model of the watershed created by the Polar Geospatial Center
(http://www.pgc.umn.edu/) using stereopairs of WorldView imagery.
Solar insolation was estimated using the Solar Radiation analyses toolset in
ArcGIS version 10 (ESRI, Redlands, CA, USA). The toolset used variability in the
orientation (slope and aspect) to calculate direct and diffuse radiation for
each pixel of the elevation model in the Y4 watershed using viewshed
algorithms (Fu and Rich, 2002; Rich et al., 1994). We report total insolation
on the summer solstice for each pixel.
Aboveground biomass
We measured diameter at breast height (DBH; 1.4 m height) or basal diameter
(BD; < 1.4 m height) of all trees and snags (i.e., dead trees standing
≥ 45∘ to the forest floor) within each 40 m2 plot (n=3 site-1). Live and dead aboveground tree biomass were determined based on
allometric equations developed from L. cajanderi trees harvested
near Chersky (Alexander et al., 2012). Tree biomass was converted to C mass
using a C concentration of 46 % for foliage (live trees only), 47 % for stemwood/bark and snag, and 48 % for branches (Alexander et al.,
2012).
We estimated understory percent cover in six 1 m2 subplots at each
site; subplots were placed at both ends of each of the three plots (at 0 and
20 m; Fig. S1). Understory vegetation was sorted into
functional types, which included shrub (evergreen and deciduous), herbs (forb
and graminoids), moss, lichen, and other (litter, woody debris, and bare
ground). At each site, understory vascular plant biomass was determined in
three 0.25 m2 quadrats, each of which was located within one of the
percent cover plots. We measured basal diameter of tall deciduous shrubs
(Alnus spp., B. nana, and Salix spp.) and used
published allometric relationships to derive biomass (Berner et al., 2015).
All remaining vascular plants were harvested and dried at 60 ∘C for
48 h for dry-mass determination. We converted live understory biomass values
to C pools by multiplying biomass by 48 % C content.
Following the line-intercept method for measuring woody debris (Brown, 1974),
we set a 20 m transect along the middle of each plot and counted the number
of times woody debris intercepted the transect for class I fine woody debris
(FWD; 0.0–0.49 cm diameter) and class II FWD (0.5–0.99 cm) along the
first 2 m; class III FWD (1.0–2.99 cm) along the first 10 m; and classes
IV FWD (3.0–4.99 cm) and V FWD (5.0–6.99 cm), and downed coarse woody debris
(CWD; > 7 cm diameter) along the entire 20 m length. We calculated the
mass of woody debris according to Alexander et al. (2012) using previously
published multipliers for softwood boreal trees from the Northwest
Territories of Canada for FWD (Nalder et al., 1997) and decay class and
density values for softwood boreal tree species within Ontario, Canada, for
CWD (Ter-Mikaelian et al., 2008). Mass values were converted to C pools based
on average C concentration of L. cajanderi boles (47 %). Total
aboveground biomass (AGB) is reported as the sum of the C pools in woody
debris, snags, trees, and understory biomass.
Canopy cover and leaf area index
We measured canopy cover under uniform, diffuse light conditions at the
center of each site in four cardinal directions using a convex spherical
densitometer, and leaf area index (LAI) using both hemispherical photography
and an LAI-2000 Plant Canopy Analyzer (Li-COR, Nebraska, NE, USA). The
LAI-2000 was placed ∼ 1 m above the ground at the center of each site,
and LAI estimates were divided by a factor of 0.68 (Chen et al., 2005) to
account for foliage clumping (Chen et al., 1997). Hemispherical photographs
were taken ∼ 1 m off the ground using a Sigma SD15 digital reflex
camera with Sigma 4.5 mm F2.8 EX DC circular fisheye lens. A N–S reflector
was used for N orientation, and photographs were taken using automatic
settings at the center of each of the three transects at each site. The
hemispherical photographs were analyzed using HemiView software.
Thaw depth/organic layer depth
We measured thaw depth using a metal thaw probe every meter along a 20 m
transect placed along the center of each plot (measured from 9 July through
3 August; does not represent maximum thaw). Organic layer depth (OLD) was
measured at 5 m intervals along each transect by cutting through the
active-layer soil with a serrated knife and visually identifying and measuring the
depth to the organic–mineral boundary.
Soil sampling and analysis
Active-layer soils were collected from all sites. Surface permafrost soils
(approximately the top 60 cm of frozen soil, which contained some frozen
active-layer soil) were sampled at seven sites (three cores per site), and deep
permafrost (15 m depth) was sampled at two sites (sites 18 and 19). We
collected six active-layer samples from each site, one at each end of the
20 m long plots. We used a serrated knife to collect an
8 cm × 8 cm sample from the organic layer and a 2 cm diameter
manual corer to collect the top 10 cm of mineral soil. When less than 5 cm
of mineral soil was thawed at the time of sampling, the mineral soil sample
was excluded from analysis (n=5). At the seven sites where surface
permafrost was sampled, we collected mineral soil to frozen ground (average
28 cm thawed mineral soil depth) using a manual corer and sampled
approximately 60 cm depth of frozen soil with a Soil Ice and Permafrost
Research Experiment (SIPRE) auger (7.62 cm diameter). We collected two deep
permafrost cores with a rotary drill rig (UKB-12/25, Drilling Technology
Factory); one deep core was collected from a site underlain by yedoma and the
other from an alas. Carbon pools presented for deep permafrost include C in
the active layer sampled at the drilling location. Carbon pools reported for
1 m depth were calculated using the seven surface permafrost samples as well
as the top 1 m of the deep core from the yedoma site. All permafrost samples
were kept frozen until analyzed as described below.
Surface permafrost cores were sectioned into 10 cm increments. Coarse roots
(> 2 mm) were removed from all active layer and surface permafrost soils,
and fine roots and organic soils were dried at 60 ∘C for 48 h, while
mineral soils were dried at 105 ∘C for at least 48 h. Gravimetric
water content (GWC) was determined as the ratio of soil water mass to soil
dry mass and was reported as a percentage (i.e., GWC × 100).
Organic matter (OM) content was measured as the percent mass lost from dried soil
after combusting for 4 h at 450 ∘C. Soil C content was analyzed on
a subset of soils (35 of 111 organic soils; 119 of 271 active layer and
surface permafrost mineral soil; and 30 of 149 deep permafrost samples) on a
Costech CHN analyzer at St. Olaf College or at the University of Georgia
Stable Isotope Ecology Lab. Carbon concentrations of the full set of soil
samples were then modeled using a linear relationship between organic matter
content and percent C (%C = 0.524 ⋅ %OM - 0.575;
R2=0.96 for active-layer and surface permafrost;
%C = 0.391 ⋅ %OM - 0.103; R2=0.86 for deep
permafrost samples). Carbon content of coarse roots was assumed to be
50 %. Sampled soils were reclassified as organic or mineral as needed
(< 1 % of samples) based on soil C content (C ≥ 20 % for
organic soils).
Bulk density (BD) was determined as the mass of dry soil per unit volume (g cm-3). Volume of active-layer soil samples was determined by measuring
the ground area and depth from where the soil sample was removed. Volume of
permafrost samples was quantified by water displacement. Ice volume was
determined based on soil water content and assuming an ice density of 0.9167 g cm-3.
Average carbon density of all sites in the Y4 watershed (a:
above- and belowground to 1 m; b: aboveground only). Bars indicate standard
error.
Soil C stocks at each depth increment were calculated as the product of
percent C, BD, and soil depth. For the deep permafrost samples, sub-samples used
for percent C, percent OM, and BD measurements were collected from adjacent depth
increments; therefore, for the percent C–percent OM regression and C pool
calculations, we used adjacent depth increments or interpolated values
between two adjacent depths.
Statistical analysis
To compare the variance in soil C among sites and studies, we used the
coefficient of variation (CV), which is the ratio of the standard deviation
to the mean. The CV is independent of the unit or magnitude and can be used
to compare intra-site variation (how variable the data are relative to the
mean value) among sites even if the mean of the sites is vastly different.
We also used percent variation, which was calculated by subtracting the
minimum value from the maximum value and dividing by the maximum value.
We used a linear model to determine the relationship between canopy cover,
LAI, and larch biomass, and the relationship between the different components
of AGB. To determine potential environmental drivers of thaw depth and soil
C, we fit a mixed-effects linear model using the nlme package in R (Pinherio
et al., 2013), using average plot-level data (three per site) as a replicate for each
site. The fixed effects were the environmental variables, and the random
effect was the nested study design (plots within sites). Both thaw depth and
soil C were log-transformed to meet the assumption of normality. After
collinear explanatory variables were removed from analysis using a variance
inflation factor of 3 (as suggested by Zuur et al., 2009), we considered
densitometry, organic layer depth, stand age, live shrub biomass, woody
debris, tree density, snag density, summer insolation, percent herbaceous
cover, percent moss cover, percent lichen cover, percent other cover, soil C,
BD, and root C, as explanatory variables for the thaw depth model. For the
soil C model the environmental variables considered were slope, summer
insolation, snag biomass, live tree biomass, live shrub biomass, woody
debris, tree density, percent herbaceous cover, percent moss cover, percent
lichen cover, percent other cover, thaw depth, organic layer depth, root
carbon, and moisture. The best model for each analysis was selected using
backwards stepwise reduction of variables to obtain the lowest Akaike
information criterion (AIC), and the residuals of all final models were
checked for normality and homogeneity of variance (Burnham and Anderson,
2002).
All reported errors are the standard error of the mean. All statistical
analyses were conducted using the statistical program R (R Core Development
Team, 2012).
Leaf area index (LAI), tree and snag density, and percent cover of the
20 plots in the Y4 watershed. Values in parentheses are standard error of the mean. “Other cover” includes woody debris and bare ground.
Site
LAI
LAI
Larch
Snag
Canopy
Understory
Herbaceous
Moss
Lichen
Other
number
(hemispherical
(LAI-2000)
density
density
cover
shrub cover
cover
cover
cover
cover
photography)
(trees m-2)
(snags m-2)
(%)
(%)
(%)
(%)
(%)
(%)
1
0.03 (0.00)
0.13
0.09 (0.05)
0.00
22.4 (3.2)
45.2 (2.7)
3.5 (1.7)
22.0 (3.4)
15.6 (4.9)
12.4 (3.4)
2
0.22 (0.02)
0.13
0.04 (0.00)
0.00
16.0 (4.0)
49.4 (5.4)
4.8 (2.4)
25.0 (4.4)
6.9 (2.9)
13.8 (6.0)
3
0.53 (0.03)
0.68
0.08 (0.03)
0.00
43.2 (7.4)
60.3 (9.0)
0.7 (0.3)
31.3 (9.4)
3.4 (2.6)
4.3 (0.6)
4
0.02 (0.01)
0.00
0.08 (0.07)
0.00
2.6 (2.6)
72.3 (7.9)
2.5 (1.6)
7.4 (2.4)
3.4 (2.1)
14.3 (5.7)
5
0.37 (0.05)
1.35
0.08 (0.02)
0.03 (0.01)
32.3 (7.6)
51.5 (4.9)
4.2 (1.4)
14.4 (2.9)
16.9 (4.1)
13.1 (2.4)
6
0.38 (0.03)
0.47
0.06 (0.01)
0.03 (0.01)
26.0 (4.6)
57.9 (7.2)
8.4 (5.9)
17.4 (5.2)
3.6 (1.3)
12.1 (3.8)
7
0.15 (0.08)
0.00
0.05 (0.02)
0.00
17.6 (8.4)
34.8 (3.5)
3.4 (0.8)
34.0 (7.1)
22.8 (6.4)
4.8 (1.9)
8
0.06 (0.04)
0.29
0.02 (0.00)
0.00
7.0 (2.1)
34.8 (4.5)
3.8 (1.8)
32.5 (7.9)
24.8 (9.5)
4.0 (2.3)
9
0.07 (0.02)
0.00
0.01 (0.00)
0.00
9.4 (1.6)
44.2 (5.5)
0.0
33.5 (5.0)
16.7 (7.6)
5.6 (1.6)
10
0.30 (0.09)
1.41
0.08 (0.04)
0.04 (0.02)
24.3 (6.2)
49.2 (10.6)
8.6 (2.9)
29.8 (8.8)
5.3 (1.4)
7.1 (2.5)
11
0.05 (0.03)
0.22
0.02 (0.01)
0.00
4.7 (1.5)
33.6 (6.9)
5.8 (3.0)
15.3 (4.5)
30.6 (8.0)
15.0 (5.9)
12
0.01 (0.00)
0.00
0.02 (0.01)
0.00
0.0 (0.0)
47.1 (7.4)
7.5 (4.0)
20.2 (3.7)
19.0 (5.3)
6.9 (3.2)
13
0.23 (0.07)
0.82
0.07 (0.01)
0.02 (0.01)
18.9 (3.0)
47.4 (8.1)
4.2 (2.6)
25.6 (8.2)
13.6 (6.2)
9.1 (0.8)
14
0.00 (0.00)
0.00
0.03 (0.02)
0.00
0.8 (0.8)
47.2 (12.0)
5.8 (3.7)
11.3 (3.8)
33.5 (13.9)
2.3 (1.1)
15
0.03 (0.01)
0.00
0.02 (0.01)
0.00
3.8 (1.0)
41.3 (3.9)
3.8 (1.7)
22.4 (4.5)
21.9 (4.6)
10.4 (5.5)
16
0.31 (0.13)
0.88
0.05 (0.01)
0.00
18.5 (7.7)
35.6 (7.6)
2.2 (0.6)
32.2 (11.6)
25.9 (9.0)
4.1 (1.5)
17
–
–
0.0
0.00
13.9 (13.9)
65.8 (15.1)
11.1 (4.4)
0.1 (0.1)
0.1 (0.1)
23.4 (11.5)
18
–
–
0.01 (0.01)
0.00
5.2
51.9 (6.5)
12.5 (4.1)
32.0 (5.0)
0.2 (0.2)
3.3 (1.9)
19
–
2.03
0.43 (0.28)
0.00
16.2 (2.2)
–
–
–
–
–
20
–
–
0.06 (0.03)
0.04 (0.02)
6.1 (1.3)
–
–
–
–
–
Results
Distribution of carbon pools
The majority of C in the watershed to 1 m depth was stored belowground
(92 %; 10.9 ± 0.8 kg C m-2 in top 1 m; Fig. 2), with
19 % in the top 10 cm of soil and 40 % in the top 30 cm. The top
10 cm of soil alone contained 58 % more C than the total aboveground C
stocks.
Aboveground biomass (g C m-2) at each site in the Y4 watershed.
Total aboveground biomass is the sum of the larch, understory vascular,
standing dead tree, and woody debris biomass. Understory vascular biomass
does not include lichen and moss. Values in parentheses are standard error of
the mean.
Site
Larch
Understory
Shrub
Standing
Woody
Total
Total
Total
number
vascular
dead tree
debris
live
dead
aboveground
1
392 (313)
112 (41)
52 (52)
0 (0)
322 (87)
504 (304)
322 (87)
826 (389)
2
603 (244)
140 (50)
75 (40)
0 (0)
76 (7)
744 (213)
76 (7)
820 (217)
3
743 (125)
320 (106)
209 (146)
0 (0)
86 (15)
1063 (230)
86 (15)
1149 (235)
4
67 (66)
611 (166)
529 (176)
0 (0)
59 (17)
679 (153)
59 (17)
737 (167)
5
1362 (516)
193 (27)
96 (32)
219 (96)
122 (28)
1555 (490)
341 (105)
1896 (579)
6
1340 (635)
257 (81)
146 (69)
386 (236)
131 (50)
1597 (560)
517 (218)
2114 (361)
7
263 (65)
271 (86)
209 (73)
0 (0)
24 (8)
533 (45)
24 (8)
557 (52)
8
471 (303)
170 (115)
124 (108)
27 (27)
10 (3)
641 (294)
37 (29)
678 (319)
9
122 (68)
176 (93)
64 (35)
0 (0)
37 (11)
298 (60)
37 (11)
335 (65)
10
697 (405)
183 (64)
51 (51)
262 (140)
106 (16)
880 (400)
368 (153)
1248 (501)
11
227 (201)
185 (87)
95 (95)
0 (0)
62 (17)
413 (285)
62 (17)
475 (278)
12
6 (6)
116 (39)
22 (13)
0 (0)
18 (4)
122 (45)
18 (4)
140 (45)
13
698 (124)
139 (25)
32 (18)
93 (69)
306 (189)
837 (126)
399 (146)
1236 (217)
14
5 (4)
253 (184)
169 (152)
0 (0)
16 (2)
259 (183)
16 (2)
275 (181)
15
142 (85)
180 (41)
82 (48)
0 (0)
71 (63)
322 (59)
71 (63)
393 (6)
16
984 (491)
470 (256)
417 (261)
0 (0)
56 (21)
1454 (628)
56 (21)
1510 (633)
17
0 (0)
2657 (2575)
2621 (2588)
0 (0)
118 (72)
2657 (2575)
118 (72)
2775 (2642)
18
2 (2)
263 (46)
245 (42)
0 (0)
16 (5)
265 (47)
16 (5)
281 (50)
19
35 (21)
465 (172)
382 (177)
0 (0)
116 (45)
500 (159)
116 (45)
615 (196)
20
585 (217)
321 (163)
156 (105)
47 (26)
158 (140)
906 (173)
205 (118)
1111 (244)
Stand density, stand age, and aboveground biomass
Stand density in the watershed ranged from 0.01 to 0.43 trees m-2 at
the forested sites (mean density was 0.07 ± 0.02 trees m-2;
Table 2). Mean stand age was 150 (±17) years (Table 1), but there was a
large range in tree ages among sites (23–221 years) and within sites
(average range: 78 years; maximum range: 238 years; minimum range: 7 years;
Table S1 in the Supplement).
Total C in AGB averaged 959 ± 150 g C m-2 across sites in the
watershed, with 53 % in larch biomass (460 ± 77 g C m-2),
30 % in understory biomass (254 ± 28 g C m-2), 11 % in
woody debris (94 ± 16.5 g C m-2), and 6 % in standing dead
tree mass (55 ± 19 g C m-2) (Fig. 2; Table 3). Among sites
across the watershed, aboveground C varied up to 95 %. Together, all C in
AGB contributed 8 % to the total amount of C stored above- and belowground
(to 1 m) across the watershed. Mean stand age was positively related to mean
stand AGB (R2=0.21, p < 0.001) and negatively related to mean stand
thaw depth (R2=0.58, p < 0.001).
Soil carbon in the Y4 watershed. Thawed soil cores were sampled from
six locations per site. Permafrost cores were sampled to 1 m at seven sites
(three per site). Root C and soil C values were normalized to 10 cm. The combined
soil C value is the amount of C in the top 10 cm of soil, regardless of soil
type (mineral/organic). Carbon pools from the permafrost cores include
active-layer soil (0–30 or 0–100 cm from top of ground surface). Values in
parentheses are standard error of the mean.
Site number
Thawed soil cores
Permafrost cores
Root C (g C m-2)
Soil C (kg C m-2)
C in top 30 cm
C in top 100 cm
(kg C m-3)
(kg C m-3)
Organic
Mineral
Organic
Mineral
Combined
1
137 (27)
0
2.60 (0.27)
2.03 (0.21)
2.34 (0.22)
4.69 (0.06)
9.36 (0.09)
2
97 (60)
0
1.35 (0.11)
1.46 (0.32)
1.32 (0.12)
3.67 (0.34)
10.16 (0.60)
3
108 (42)
0
1.86 (0.32)
1.43 (0.19)
1.83 (0.29)
4
169 (183)
0
2.06 (0.47)
2.06 (0.22)
2.49 (0.48)
5
453 (108)
0
4.47 (1.74)
1.57 (0.05)
3.42 (0.76)
6
230 (169)
0
3.86 (1.03)
2.22 (0.43)
3.71 (0.93)
7
44 (22)
0
1.13 (0.22)
2.31 (0.41)
1.14 (0.22)
4.29 (0.32)
10.48 (0.67)
8
69 (25)
0
1.25 (0.12)
2.79 (0.67)
1.38 (0.19)
9
177 (17)
45 (31)
2.51 (0.26)
1.54 (0.33)
2.41 (0.40)
4.85 (0.36)
8.63 (0.71)
10
278 (35)
0
2.12 (0.45)
1.36 (0.12)
2.10 (0.46)
4.82 (0.44)
9.39 (0.06)
11
520 (346)
6 (4)
1.63 (0.42)
2.02 (0.16)
1.66 (0.30)
12
271 (87)
0
1.39 (0.04)
3.26 (0.83)
1.51 (0.05)
13
267 (30)
0
1.65 (0.28)
1.96 (0.29)
1.66 (0.29)
14
252 (74)
6 (4)
3.12 (0.47)
1.31 (0.26)
2.74 (0.15)
15
103 (8)
0
2.04 (0.58)
2.15 (0.53)
1.84 (0.38)
16
189 (184)
20 (11)
1.70 (0.57)
2.08 (0.49)
1.66 (0.33)
5.32 (1.19)
11.90 (3.83)
17
0
97 (35)
–
2.37 (0.21)
2.76 (0.78)
18
95 (36)
0
2.19 (0.40)
2.66 (2.21)
1.49 (0.55)
19
205 (91)
203 (152)
3.51 (0.47)
2.74 (1.23)
2.85 (0.72)
20
0
0
2.44 (0.70)
1.41 (0.26)
1.85 (0.43)
5.70 (0.55)
11.91 (0.90)
Aboveground larch biomass was also highly variable across the watershed, with
some sites as low as 0 or 1.7 g C m-2 and others as high as 1340 and
1362 g C m-2. Of the three techniques used for estimating canopy
cover, LAI values from hemispherical photography (Table 2) showed the highest
correlation with larch biomass (R2=0.69, p < 0.001), but larch
biomass was also significantly associated with canopy density (R2=0.5,
p < 0.001). There was no relationship between larch biomass and
understory biomass (p = 0.4); however, the percent cover of tall shrubs
was negatively related to both moss (R2=0.2, p < 0.001) and lichen
cover (R2=0.2, p < 0.001).
Properties of thawed soil in the Y4 watershed. The mineral layer was
collected to approximately 10 cm below the organic layer (see Methods section). No
relationship existed between sample date and thaw depth or sample date and
water content. Values in parentheses are standard error.
Site
Thaw
Organic
Bulk density
Gravimetric water
Carbon content
number
depth
layer
(g cm-3)
content (%)
(%)
(cm)
depth (cm)
Organic
Mineral
Organic
Mineral
Organic
Mineral
1
23 (1)
13 (1)
0.078 (0.021)
0.52 (0.16)
198.9 (34.4)
64.7 (17.4)
37.6 (3.5)
6.9 (2.5)
2
22 (1)
11 (1)
0.040 (0.011)
0.64 (0.05)
203.8 (28.0)
33.9 (5.8)
38.3 (4.1)
2.4 (0.5)
3
24 (1)
14 (1)
0.062 (0.011)
0.70 (0.11)
103.3 (16.2)
29.1 (4.4)
30.4 (2.2)
2.3 (0.6)
4
41 (2)
10 (1)
0.148 (0.063)
0.54 (0.14)
107.3 (28.9)
61.0 (15.6)
26.6 (4.0)
8.7 (3.0)
5
23 (1)
8 (1)
0.120 (0.032)
1.02 (0.08)
220.2 (23.1)
25.6 (2.1)
39.2 (3.2)
1.6 (0.3)
6
21 (2)
9 (1)
0.113 (0.039)
0.63 (0.05)
182.0 (19.8)
34.2 (6.1)
39.0 (3.0)
3.8 (1.0)
7
21 (1)
12 (1)
0.026 (0.005)
0.76 (0.18)
348.5 (48.4)
43.6 (10.2)
44.4 (2.0)
3.9 (1.2)
8
16 (1)
11 (1)
0.027 (0.002)
0.68 (0.10)
304.9 (32.1)
46.4 (10.3)
46.7 (0.6)
4.4 (1.1)
9
26 (2)
13 (1)
0.082 (0.010)
0.64 (0.12)
171.3 (29.5)
46.5 (11.2)
30.9 (4.4)
5.5 (2.1)
10
23 (1)
11 (1)
0.048 (0.007)
0.89 (0.05)
272.6 (15.2)
26.5 (1.7)
43.6 (1.9)
1.6 (0.2)
11
35 (2)
10 (1)
0.060 (0.023)
0.84 (0.12)
142.8 (17.8)
39.4 (6.9)
30.5 (3.3)
3.6 (1.6)
12
29 (2)
10 (1)
0.053 (0.020)
0.67 (0.10)
247.7 (17.5)
58.3 (10.7)
43.5 (1.8)
5.0 (1.0)
13
29 (1)
12 (1)
0.042 (0.008)
0.71 (0.11)
194.1 (15.4)
48.6 (12.6)
40.0 (1.4)
4.0 (1.0)
14
42 (2)
8 (1)
0.103 (0.016)
0.82 (0.10)
165.8 (14.7)
31.0 (7.2)
32.4 (3.8)
3.0 (1.6)
15
28 (2)
12 (1)
0.150 (0.099)
0.92 (0.10)
419.1 (105.4)
39.9 (10.6)
38.3 (3.5)
2.6 (0.9)
16
24 (1)
12 (1)
0.042 (0.009)
0.76 (0.18)
256.3 (38.8)
49.5 (15.8)
40.2 (2.1)
5.9 (3.4)
17
45 (2)
9 (2)
–
0.46 (0.11)
–
50.9 (7.6)
–
8.7 (2.8)
18
26 (1)
18 (1)
0.059 (0.012)
0.39 (0.20)
346.8 (45.4)
123.2 (31.2)
39.9 (3.3)
8.7 (2.6)
19
36 (2)
14 (2)
0.078 (0.022)
1.40 (0.09)
204.9 (52.3)
22.8 (0.4)
33.5 (3.4)
1.0 (0.1)
20
29 (1)
9 (1)
0.118 (0.001)
0.65 (0.31)
252.9 (76.6)
76.1 (28.4)
29.9 (4.4)
8.6 (4.9)
Relationship between SOC in the top 10 cm of soil and moisture,
moss cover, and lichen cover. Each point represents the average SOC measured
at each transect (three transects per site) and its corresponding moisture
content or the average moss or lichen cover measured at that transect.
Surface soils
Average C content of the organic horizon was 37.6 (±0.8) % C, whereas
C content of the thawed mineral horizon (0–10 cm) was 4.6 (±0.48) % C. There was 2.24 (±1.22) kg C m-2 stored in the
organic layer (average organic layer depth = 11.2 ± 0.2 cm) and
1.96 (±0.07) kg C m-2 in the top 10 cm of the mineral layer
(Table 4).
There was large variation in BD, soil moisture (GWC), soil C content, and thaw
depth among sites (Table 5). Carbon content and GWC were more variable in
mineral soils than in organic (CVmineral=0.55 for percent C and
0.48 for GWC; CVorganic=0.15 for percent C and 0.36 for GWC), while
BD was more variable in organic soils (CVorganic=0.51;
CVmineral=0.3). While the CV of thaw depth was not particularly
high (0.28), the difference between the sites with the highest and lowest
thaw depth measured was still 65 %, underscoring the heterogeneity of
soil properties across the watershed. Variation in thaw depth was primarily
due to stand age (Fig. 3; Table S2).
Relationship between thaw depth and stand age. Each point represents
the average thaw depth measurement taken along a transect (three
transects per site) and the stand age of the entire site. Thaw depths were
measured in July/August of 2012 and 2013.
Soil C density in the top 10 cm of the ground surface (i.e., 0–10 cm soil
depth, which may have contained both organic and mineral soils) varied up to
93 % across the watershed (range: 0.51–7.14 kg C m-2; Tables 4
and S2), but the CV was larger within sites
(0.32) than it was between sites (0.26), indicating that soil C is more
variable at the meter scale than it is at the kilometer scale. The
distribution of soil C density in the top 10 cm was best explained by soil
moisture, percent moss, and percent lichen cover (Table S2); soil C density
was positively related to soil moisture and negatively related to percent
moss and lichen cover (Fig. 4).
Bulk density, carbon density, and ice content of the two deep
(15 m) permafrost soil cores.
Soil in the top 30 cm of the profile contained on average
4.8 ± 0.3 kg C m-2, but soil C density in the top 30 cm varied
by 56 % across the watershed as a whole. The average CV within a site was
0.16, whereas the CV among sites was 0.22, indicating C density at 30 cm is
similar or more variable across the watershed than at the meter scale. The
top 1 m of soil contained 10.9 ± 0.8 kg C m-2
(13.8 ± 3.0 kg C m-2 at alas site; Table S4). Soil C in the
top 1 m varied by 63 % across the watershed and by 44 % among sites.
The average CV within a site was 0.15, whereas among sites the CV was 0.20,
indicating soil C to 1 m is similarly variable at the meter and kilometer
scales. Ice content in the top 1 m was on average 68 ± 2 % by
volume, with a range of between 51 and 80 %.
Deep permafrost soils
Deep permafrost soils (including surface active layer to 15 m) contained
205 kg C m-2 (site 19; yedoma deposit, non-ice wedge) and
331 kg C m-2 (site 18; alas). Carbon density at each 1 m interval
ranged from 7.87 to 21.63 kg C m-3 in the yedoma deposit and
from 6.9 to 14.5 kg C m-3 in the deeper portion of the alas (Fig. 5;
Table S5). The top 2 m of the alas was characterized by particularly high C
density (∼ 30 kg m-3).
Highlighting the variability of C in deep permafrost, the total soil C
density in the two cores varied by 38 %. The alas site had higher GWC
than the yedoma site in the first 2 m (GWC: 385 ± 81 and
41 ± 8 %, respectively). Throughout the entire profile, GWC was
46 ± 2 % in the yedoma core and 100 ± 23 % in the alas
core. Overall, BD was similar between the two cores, and most of the
variation in BD occurred in the top 5 m (Fig. 5).
Discussion
Aboveground biomass
Aboveground C pools within the Y4 watershed represented only a small fraction
(8 %) of total C pools, likely due to low tree density at most sites
(< 0.09 trees m-2 at all but one site) and/or young stand ages at a
few sites. Low-density, mature (> 75 years old) stands with no recent fire
activity are common in this region (Berner et al., 2012); however, wildfires
can produce stands of considerably higher density (> 3 trees m-2),
which can substantially increase AGB and contribution to total C pools as
stands mature (Alexander et al., 2012). Aboveground C pools were similar to
those reported by Alexander et al. (2012) for 17 nearby stands of similar age
and density, but C in larch AGB was lower (∼ 23 %) than a
landscape-level estimate (∼ 600 g C m-2) across the Kolyma
River basin (Berner et al., 2012). Our estimate for C stored in larch AGB was
also 4 times lower than that of a mature (155-year-old), mid-density
(0.19 trees m-2) stand near Chersky and 2 times lower than a
mature, low-density (0.08 trees m-2) stand near Oymyakon, south of
Chersky (Kajimoto et al., 2006). In addition, our larch AGB estimates fell
within the low range of larch stands across other high-latitude
(> 64∘ N) regions and were generally 3–10 times lower than other
stands (Kajimoto et al., 2010). Our considerably lower estimates reflect both
the sparse, open-grown structure of our stands (Osawa and Kajimoto, 2010) and
the poor soil environment (e.g., shallow rooting zone, low soil temperature,
low N availability) found in stands near the latitudinal and altitudinal tree line
(Kajimoto et al., 2010). Despite the small contribution of AGB to total C
pools across our stands, aboveground vegetation composition and structure
were important factors related to soil C pools and permafrost thaw (see
below). In addition, characteristics of aboveground vegetation are major
determinants of land–atmosphere C fluxes (Bradshaw and Warkentin, 2015) and
thus remain essential components of C dynamics even when pools are relatively
low.
Variability of soil C pools
Soil C density is controlled by numerous biogeophysical factors such as
climate, local geomorphology, soil parent material, time since last
disturbance, and vegetation type, all of which lead to high variability in
soil C pools at the regional and local scale. Our soil C pool estimates for a
Siberian larch forest watershed fall within the range of published
assessments that characterize this area (Alexander et al., 2012; Broderick et
al., 2015) but are at the low end of other studies (Alexeyev and Birdsey,
1998; Hugelius et al., 2014; Matsuura et al., 2005; Palmtag et al., 2015;
Stolbovoi, 2006). For example, our mean estimate of
4.8 ± 1 kg C m-2 in the top 30 cm of soil is less than half of
a published assessment of C stored in soils across Russian larch forests
(10.2 kg C m-2; Stolbovoi, 2006) and less than one-third of the mean
estimate for Turbel soils across the permafrost region
(14.7 kg C m-2; Hugelius et al., 2014); however, variation in the
permafrost region Turbel soil C pool is high (CV = 0.85; Hugelius et al.,
2014), and our mean estimate falls within 1 standard deviation of this
regional mean.
Within larch forests, there is substantial variation in soil C pools at
regional scales, driven by variation in soil parent material and climate. For
example, larch forests in Northeastern Siberia store significantly more C
(16 kg C m-2) in the active layer and have more variable soil C pool
estimates than larch forests in central Siberia (6.3 kg C m-2)
(Matsuura and Hirobe, 2010). There is also considerable variation in soil C
pools within larch forests at smaller spatial scales. Indeed, the active
layer in larch forests located within 50 km from our study site contained
twice as much C as found in our study (4.8 ± 0.3 kg C m-2 to
30 cm); there was 8.3 kg C m-2 in the active layer (38 cm) of a
larch forest 44 km from the Y4 watershed (Matsuura et al., 2005) and
9.5 ± 2.9 (SD) kg C m-2 in the top 30 cm of soils from a forest
3 km away (Palmtag et al., 2015). This variation in soil C pools points to
the extreme variability in soil C throughout the landscape, even at the
kilometer scale. It also highlights the importance of sampling replication at
small scales; with 21 total soil cores at seven sites, our CV (0.13) was less
than half of other studies with lower site-level replication (Palmtag et al.,
2015).
As the climate warms, C in surface permafrost is becoming increasingly
vulnerable to thawing and subsequent decomposition and loss to the
atmosphere. As such, estimating variation in C pool size is critical for
understanding permafrost–climate feedbacks. The C stored in the top 1 m of
Y4 soils (10.9 ± 0.8 kg C m-2) was similar to the average 1 m
C pool reported for the Yakutia region, which comprises a range of ecosystem
types (8.1 kg C m-2; Alexeyev and Birdsey, 1998), but 37 % lower
than the 1 m soil C pool reported in a forest only 3 km away
(17.3 ± 5.7 kg C m-2; Palmtag et al., 2015). However, the
percent difference between our estimate and the nearby study (37 %) was
similar to the percent difference found between sites in the Y4 watershed
(44 %; Table 4), suggesting that these differences among studies are
likely due to natural variation in the landscape.
Carbon pool estimates from deep permafrost (> 3 m) are limited across the
Arctic (Hugelius et al., 2014; Schuur et al., 2015; Tarnocai et al., 2009),
yet these data are critical for assessing variation in and controls on C
density of yedoma, as these soils have particularly high C density at depth
(Strauss et al., 2013; Zimov et al., 2006). The average carbon density of
deep permafrost from yedoma deposits in the Y4 watershed
(13.5 kg C m-3) was similar to values reported for yedoma in
pan-Arctic summary studies (10 +7/-6 kg C m-3, Strauss et al.,
2013; 13.0 ± 0.75 kg C m-3 after correction for ice volume,
Walter Anthony et al., 2014) and at taiga sites within 100 km of Chersky
(12.3–15.4 kg C m-3 after correction for ice volume, Walter Anthony
et al., 2014, and references therein; 14.3 kg C m-3, Shmelev et al.,
2017). Carbon density was almost twice as high in the alas, which is
consistent with findings indicating that alas and thermokarst soils store
substantially more C (∼ 40–70 %; Walter Anthony et al., 2014;
Strauss et al., 2013; Siewert et al., 2015) than undisturbed yedoma, a
difference that is likely due to higher rates of recent (Holocene) C
accumulation at the alas site (Walter Anthony et al., 2014). Yedoma is
characterized by high landscape-level ice content due to the prevalence of
large ice wedges, which can comprise 31 to 63 % of ground volume (Ulrich
et al., 2014). Accounting for these deep ice deposits, which were not sampled
in this study, would reduce our landscape-level estimate of C content in the
top 15 m of yedoma from 205 to 76–141 kg C m-2,
which is still an order of magnitude more C than is stored in the active
layer and 2 orders of magnitude more C than is stored in biomass.
Micro-scale variation in soil carbon and thaw depth
In addition to the effects of parent material and climate on soil C storage,
soil carbon pools are determined by the balance between biological inputs and
losses due to microbial decomposition and lateral transport. These biological
processes are, in turn, also heavily influenced by climate on regional and
local scales. We found that soil samples with higher moisture content also
had higher C density, which is likely due to both the effects of soil
moisture on microbial activity and indirect effects of soil moisture on C
inputs to soils through effects on plant productivity. In wetter soils,
oxygen diffusion is limited, resulting in anaerobic conditions where
microbial decomposition is slower, and C can accumulate at a higher rate than
in more well-drained, well-aerated soils (Schädel et al., 2016). However,
this positive association between moisture and C density may also be a result
of increased C inputs and plant productivity associated with higher soil
moisture (Berner et al., 2013) or the lateral movement of dissolved organic C
into the wetter sites. It is likely that environmental controls on both C
inputs and losses are driving the patterns of C accumulation across the
watershed.
Plant species composition may also play an important role in soil C storage
in boreal forests (Hollingsworth et al., 2008) through the quality and
quantity of litter inputs and through vegetation effects on environmental
controls such as soil moisture and temperature. Lichens and mosses are
sometimes thought to encourage soil C storage through their promotion of low
soil temperatures, higher moisture, and a relatively acidic environment
(Bonan and Shugar, 1989). However, at our sites, increasing abundance of
lichen and moss was associated with lower soil C storage, which may have been
due to lower rates of C fixation (Turetsky et al., 2010), higher rates of
decomposition of vascular plant litter in moss and lichen patches (Wardle et
al., 2003), or impacts of vegetation functional types on soil moisture and
soil temperatures. Because the interactions between soil processes and
vegetation are bidirectional, the processes driving these observed patterns
are unclear, and further experimental work is needed to identify the
mechanisms.
Increasing thaw depth may result in increased C loss from boreal ecosystems;
as more soil is thawed, more organic matter is available for decomposition.
We found that thaw depth was negatively related to stand age; the deeper thaw
depth observed at the younger sites could be a result of more recent burning
events, which tend to increase thaw depth (O'Donnell et al., 2011; Yoshikawa
et al., 2002).