Introduction
Along with climate warming, permafrost warming has been widely observed in
the Arctic and sub-Arctic as well as in midlatitude alpine regions like the
Alps and the Tibetan Plateau (Romanovsky et al., 2013; Harris et al., 2009;
Cheng and Wu, 2007). Permafrost thaw
has caused considerable changes for surface and subsurface hydrologic routing
(Kurylyk et al., 2014), geotechnical failures (Harris et al., 2009) and
carbon dioxide and methane release (Schuur et al., 2015). Therefore,
understanding permafrost warming is essential to cope with such consequences.
The mean annual ground temperature (MAGT) at a depth of zero annual amplitude
is often used to indicate permafrost warming (e.g. Wu et al., 2012). The
warming rate is controlled by a variety of factors such as weather regimes,
geography/geology and ecosystems. Generally, cold permafrost has a higher
warming rate than warm permafrost (Wu et al., 2012). However, permafrost
temperature can differ greatly in the same region due to local factors like
topography, soil properties and vegetation. Responses of permafrost
controlled by these local factors to climate change is thus also expected to
differ. For instance, the permafrost along the Qinghai–Xizang (Tibet) railway
experienced a mean warming rate of 0.02 ∘C yr-1 at a depth of
6.0 m over the period from 2006 to 2010, and the highest warming rate even
reached 0.08 ∘C yr-1 in the Fenghuo Mts. area (Wu et al.,
2012). Given the same change in climate variables, these local factors still
cause the underlying permafrost to develop differently. For instance, one
recent study in the central QTP shows that permafrost at 10 sites experienced
highly differing warming rates over the period of 2002–2014 (Wu et al.,
2015). Compared to the rate of 0.03 ∘C yr-1 for the past 5
decades over the QTP (Piao et al., 2010), there was no extraordinary increase
in air temperature (0.02 ∘C yr-1) at the investigated area.
Thus, an average increasing rate of 0.01 ∘C yr-1 at 10 m
depth in permafrost temperature sounds rather high. Wu et al. (2015)
suggested that this was due to the increasing rainfall and the asymmetrical
seasonal changes in subsurface soil temperatures. Under the warming of the
atmosphere, the high warming rate of the permafrost is even similar to that
of air temperature rise, 0.02 ∘C yr-1. Thereby, we expect a
dominating role of subsurface processes in the active layer that amplify the
climate warming input.
As a buffer layer, the active layer regulates the energy transfer between
the atmosphere and permafrost in addition to vegetation and snow cover. In
this study, we focus on the high-altitude permafrost on the QTP, which is
characterized by a thick unsaturated active layer and sparsely vegetated
surface. In contrast to the commonly thin active layers in the Arctic, the
active layers on the QTP are usually over 1.5 m thick, and soil
architectures consist of a fine-grained layer without (or only thin) surface
organic horizons overlaying coarse immature soils, which is characterized as
low content of fine-grained materials like silt and clay (Huang et al.,
2006). These stratified active layers are commonly found on the QTP along
with a dry interlayer between top soil and bottom soil. Furthermore, with
precipitation concentrated in the summer season as rainfall, the subsurface
soil hydraulic properties play an important role in ground heat transfer due
to seasonal soil moisture change (e.g. Hayashi et al., 2007). In contrast
to the well-studied permafrost in the Arctic, the applicability of the
analytic models of the climate–permafrost relationship based on the
seasonally variable thermal conductivity might be challenging. For instance,
thermal offset was firstly identified as higher mean annual temperature in
the upper portion of the active layer than in the underlying permafrost by
Burn and Smith (1988) at several sites in Yukon, Canada and it is
attributed to the difference between frozen and thawed thermal
conductivities of the soil. Then a model of thermal offset effect is
proposed to link ground surface temperature to mean annual temperature at
the top of the permafrost table (TTOP) (Romanovsky and Osterkamp, 1995).
Based on this concept, the TTOP is estimated using the TTOP model developed
by Smith and Riseborough (1996, 2002), which also uses surface offset that
links air temperature to ground surface temperature. Normally, a wet active
layer has a larger thermal offset than a dry active layer, which has small
or even vanishing thermal offset (e.g. Romanovsky and Osterkamp, 1995;
Hasler et al., 2011). However, a reversed thermal offset, namely TTOP > MAGST,
has been reported by Lin et al. (2015) on the QTP. To
evaluate the applicability of this concept, further exploration of the
hydraulic and thermal mechanisms in the active layers on the QTP is
required. This might enable us to understand the impact of the active layer
on permafrost warming.
In this study, we use observations over a 9-year period and numerical
simulations to investigate recent permafrost warming at a site in a warm
permafrost region on the QTP and to demonstrate the role of a typical
stratified active layer in permafrost warming. Our goals are (1) to
characterize the rapid permafrost warming at the study site, (2) to reveal
the unique phenomenon of the reversed seasonally variable thermal
conductivity in the active layer that challenges the application of the
analytic models by comparing with observations and physically based
modelling and (3) to emphasize the importance of incorporating structural
soil hydraulic properties in permafrost projections given a rain-dominated
weather pattern on the QTP.
Material and methods
Site descriptions
The Chumaer site is located on a high plain of the Chumaer River catchment
in the north-eastern QTP with an average altitude of over 4450 m (Fig. 1). In
the catchment area, the land surface is covered by bare soil or sparse
vegetation. Measurements at the study site comprise a monitoring station and
several boreholes, with discontinuous ground temperature measurements since
2006 (Pan et al., 2014). The monitoring station has been complemented by
soil–weather measurements since 2006. The weather data from 2006 to 2014
show an average air temperature of about -16.0 ∘C in January and
7.2 ∘C in July, and precipitation is dominated by summer monsoon
from June to September, which brings about 350 mm precipitation annually,
falling mostly as rainfall. Irregular thin snow cover occurs in late spring
or early winter, lasting usually just a few weeks. The stratigraphy includes
a fine top soil (30 cm) and a middle layer of alluvial sandy and gravelly
sediment up to 3 m that lies over deeply weathered mudstone. The borehole
data indicate that the permafrost has a thickness about 25 m, and the
temperature at a depth of 10 m is less than -1.0 ∘C. The
active layer thickness is around 2.5 m.
Study site location and permafrost distribution on the
Qinghai–Tibet Plateau. The background map shows the permafrost classes from Li
and Cheng (1996).
![]()
Surface–subsurface monitoring scheme
The surface–subsurface interaction has been investigated since 2006. Regular
meteorological variables including air temperature, precipitation, relative
humidity, wind speed and direction and net radiation were monitored at an
automatic weather station. Subsurface hydraulic and thermal dynamics within
the active layer were monitored by measuring soil temperature and soil water
content at a variety of depths (soil temperature: 0.05, 0.10, 0.15, 0.20,
0.30, 0.50, 0.70, 0.90, 1.10, 1.30, 1.50, 1.70, 1.92, 2.08, 2.18, 2.30, 2.50,
2.70, 3.00, 3.30, 3.60 m; soil water content: 0.10, 0.20, 0.40, 0.65, 0.89,
1.19, 1.54, 1.92, 2.10 m). They were recorded with a time interval of
60 min. Soil temperature was measured with thermistors, which provides an
accuracy of 0.05 ∘C. Liquid soil water content was measured with
CS616 sensors (Campbell Scientific Ltd.), and the total water content in
frozen soils was deduced from the value measured just before freezing. Here
we assume negligible soil water redistribution during freezing due to the
coarse soils. Thus, its accuracy is around ±5 % (Pan, 2011). More
detailed technical description of the instrumentation can be found in
Pan (2011). In addition, permafrost temperature was investigated with two
boreholes about 30 m away in a flat area. The shallow borehole is 15 m in
depth and the deep one is 60 m, which penetrates through the permafrost.
GEOtop model
GEOtop (version: 1.45), used in this study, is a process-based energy and
mass-balance model (Rigon et al., 2006; Endrizzi, 2009), which has a number
of advantages for a wide range of permafrost applications. It simulates
hydrological fluxes from the energy balance in complex terrain with
snow-covered and snow-free regimes (e.g. Simoni et al., 2008; Endrizzi and
Marsh, 2010, Endrizzi et al., 2014). Soil temperature and moisture dynamics
are simulated using a robust and energy-conserving model of freezing in
variably saturated soil (Dall'Amico et al., 2011). It invokes a relation
between the soil-freezing characteristic and the soil water characteristic
and assumes a rigid soil scheme without change in volume for water phase
transition (Kurylyk and Watanabe, 2013). The model's versatility enables the
user to investigate the responses of permafrost degradation on the QTP, where
permafrost is characterized as a thick and stratified active layer with
pronounced hydraulic dynamics due to the rainfall-evaporation-dominated land
surface fluxes. In this study, we apply a one-dimensional (1-D) model at a
single site, where the spatial factors, e.g. topography and snow, are not
important. A proper representation of subsurface processes is essential for
our research questions. Details are introduced in the following subsections.
Model set-up
Considering the features of land surface energy exchange and subsurface
hydraulic and thermal dynamics at the study site, a 1-D conceptualized model
was set up with GEOtop. The soil profile domain was generated with element
size of 10 cm for the shallow soils (0–3.0 m), reducing to 0.5 m and
1.0 m for the underlying soils. There are 63 elements in total. In order to
solely diagnose the effect of the stratified active layer on permafrost
degradation, model simulations were driven by the same atmospheric forcing.
In addition, some assumptions for this model include (1) no lateral flows
exist like surface run-off and subsurface groundwater flow, (2) surface
features like vegetation, soils and associated parameterization are
constant in the long-term simulations. This might lead to certain
deviations in the simulated results.
Input parameters
The required climate forcing for the GEOtop models includes precipitation
(snow and rain), air temperature, wind speed, relative humidity and
incoming short and longwave radiation. The bottom boundary conditions for
energy and water balance are set as follows. Considering the weak impact of
the bottom mudstone on surface water flux, the bottom drainage rate through
the mudstone was simply set to zero. While the bottom thermal condition is
essential to the permafrost warming rate, as well as surface energy fluxes.
The geothermal flux at the depth of 30 m was determined from the measured
ground temperature gradient (0.07 ∘C m-1) and estimated
thermal conductivity of the mudstone. Given a thermal conductivity of soil
particles 2.0 W m-1 K-1 and a porosity of 0.2 for the mudstone,
the geothermal flux at the bottom boundary was set to 0.14 W m-2. This
high geothermal flux is consistent with other observed values in the same
Kunlun Mountains area (Wu et al., 2010).
Taking the analysis of the sensitivities and uncertainties of the GEOtop in
mountain environments by Gubler et al. (2013) into account, we set the
following surface and subsurface parameters based on field observations and
relevant literature without doing any site-specific sensitivity analysis.
The vegetation coverage for the sparsely vegetated ground surface was set to
0.3 and the surface roughness length, used for calculating turbulent fluxes,
to 120 mm. The latter was chosen in agreement with studies on the QTP
(Ishikawa et al., 1999; Yang et al., 2008; Ma et al., 2008). They were set
constant in the simulations. For subsurface, the hydraulic and thermal
parameters of shallow soils (0–3.0 m) and underlying soils (0.3–30.0 m) are
listed in Table 1. More details are given as follows.
Soil properties of shallow soils (A: 0–3.0 m) and
underlying soils (B: 3.0-30 m) for three soil profiles (A1/B, A2/B and
A3/B). Ks is saturated hydraulic conductivity, α and n are van
Genuchten parameters, θr and θs are residual and
saturated soil water content, respectively, λsp is thermal
conductivity of soil particles and C is thermal capacity.
Soil architecture
A1
A2
A3
B
0–3.0 m
0–3.0 m
0–0.3 m
0.3–3.0 m
3.0–30 m
Soil texture %
sand
66.3
92.2
66.3
92.2
–
silt
12.0
3.8
12.0
3.8
–
clay
21.7
4.0
21.7
4.0
–
Hydraulic properties
Ks (m d-1)
0.19
4.68
0.19
4.68
2.2 × 10-3
α (cm-1)
0.03
0.03
0.03
0.03
0.01
n (–)
1.33
2.85
1.33
2.85
1.5
θr (m3 m-3)
0.06
0.05
0.06
0.05
0.10
θs (m3 m-3)
0.38
0.38
0.38
0.38
0.2
Thermal properties
λsp (W m-1 K-1)
5.0
2.0
C (J m-3 K-1)
2 × 106
Three soil architectures of the shallow soils were used in the simulations.
For the realistic case, the soil architecture (A3) consists of two layers:
sandy loam (0–0.3 m) and coarse sand (0.3–3.0 m). For comparison,
another two forged single-layered architectures, A1 and A2, were employed
and corresponding soil properties are the same as sandy loam and coarse sand, respectively.
The required van Genuchten parameters were derived from actual soil texture
information using the neural network routine (Schaap and Bouten, 1996). Soil
textures for sandy loam and coarse sand are available from König (2008), while no data are
available for the bottom layer of gradually weathered bedrock. Thus,
hydraulic properties for the third layer are assumed to be the same as the
typical clay (Domenico and Schwartz, 1990).
The soil thermal parameters of soil thermal conductivity and heat capacity
were determined from the four volumetric components: water, ice, air and
soil particles (Dall'Amico et al., 2011). The bulk thermal conductivity
(λb) was estimated with the following equation proposed by
Cosenza at al. (2003):
λb=(1-ϕ)λsp+θwλw+θiλi+θaλa2,
where
λsp, λw, λi and λa are thermal conductivities of soil particles, water, ice and air,
respectively; ϕ is soil porosity; θw, θi, and
θa are the volume fractions of water, ice and air,
respectively. Thermal properties of the soil particles in Table 1 were set to
common values for different soil types (Farouki, 1986).
Simulation protocol
The role of soil architecture in regulating hydraulic and thermal dynamics
in the stratified active layer, as well as long-term permafrost change, was
investigated with six numerical simulations in Table 2. The first three
simulations used the parameters listed in Table 1, and another three used
the same parameters except for the thermal conductivity of soil particles for
the shallow soils, which was reduced to 2.5 W m-1 K-1. Thus, six
simulations with corresponding model settings were projected with
atmospheric forcing for a long period of 1980–2100. Besides, similar
simulations were also conducted for comparison of the hydraulic and thermal
pattern in 2008 by replacing the meteorological forcing with the observed
air temperature and precipitation.
Six simulations with different combinations of soil
architecture and thermal conductivity of soil particles (λsp)
for the shallow soils (0–3.0 m). A1, A2 and A3 stand for three types of soil
architecture in Table 1.
λsp (W m-1 K-1)
Soil architecture
A1
A2
A3
5.0
1
2
3
2.5
4
5
6
The atmospheric forcing used in this study was produced from the fifth
Coupled Model Intercomparison Project database of GCM output (CMIP5) (Taylor
et al., 2012). The projected climate scenario of the Representative
Concentration Pathway 8.5 (RCP8.5) was dynamically downscaled using the
CanESM2/CGCM4 model (Verseghy, 1991), which corresponds to a usual warming
scenario with 8.5 W m-2 forcing by 2100. Figure 2 provides the
projected changes in mean annual air temperature (MAAT) and annual total
precipitation from 1900 to 2100. Generally, a pronounced increase in air
temperature started in the 1980s and there is also a noticeable change in
precipitation. These features are generally consistent with the regional
trend of air temperature and precipitation obtained from local observations
(Guo and Wang, 2013; Hu et al., 2014). Considering the rapid warming in
permafrost during the past few decades (Cheng and Wu, 2007), a reasonable
hypothesis is to presume the climate was steady for the 80 years before the
1980s. Accordingly, we assume that the thin permafrost around the 1980s was
in pseudo equilibrium. Therefore, the model was spun up with the atmospheric
forcing by using a repeated 10-year period from 1970 to 1979 that kept the
mean annual soil temperature change less than 0.01 ∘C in all soil
layers. The initial condition for the spin-up was a constant ground
temperature of -0.5 ∘C and water pressure in static equilibrium
with a water table at 1.0 m below ground.
Time series of projected mean annual air temperature (MAAT)
(a) and annual total precipitation (b) from 1900 to 2100. The black section
represents historical data and the red section represents projected data. The
period in shadow (1970–1979) was repeatedly used for spin-up.
![]()
Results and discussion
Relationship between air temperature and near-surface soil
temperature
Figure 3 shows the relationship between daily mean air temperature and
near-surface soil temperature (5 cm below ground surface) over the period of
2006–2014. A few sporadic outliers are related to abrupt cold weather, e.g.
summer/autumn freezing. The linear fit indicates that the freeze–thaw
process does not exert significant impact on heat transfer in the very
shallow subsurface, which means a small change in seasonal thermal
properties. This might be attributed to the seasonal total water change. In
addition, the average temperature difference was about 5.0 ∘C,
while the mean annual air temperature was even higher than -5.0 ∘C.
Thus, the mean annual near-surface temperature should fluctuate around
0 ∘C.
Relationship between the daily mean air temperature
(Ta) and near-surface soil temperature (T5 cm, measured at 5 cm
below the ground surface) over the period from 2006 to 2014.
![]()
Surface and thermal offsets
Figure 4 shows the mean annual thermal profiles measured over the period of
2007–2013 from 1.5 m above ground surface to 2.18 m in subsurface. The
interaction between the lower atmosphere and permafrost can be characterized
with surface and thermal offsets (Smith and Riseborough, 2002). Limited by
available measurements, surface offsets were approximately estimated from
the difference between mean annual near-surface (10 cm) temperature (MAGST)
and MAAT, and the thermal offsets were calculated from the difference
between mean annual temperature (MAT) close to the permafrost table (2.18 m)
and MAGST. Here we should mention that these thermal offsets were
overestimated slightly due to the used bottom MAT, which is not exactly at
the permafrost table. Calculations show that the surface offsets in 2008,
2009 and 2013 were 4.40, 3.78 and 4.30 ∘C, respectively. These values indicate a weak coupling between the lower
atmosphere and ground surface. While the thermal offsets changed from
positive values, i.e. 0.47 ∘C (2007) and 0.33 ∘C (2008)
to negative ones, -0.18 ∘C (2009) and -0.15 ∘C (2013).
Surprisingly, the positive thermal offsets occurred in colder weather
conditions. A similar phenomenon has been found in a nearby region by Lin et al. (2015).
This result seems to conflict with the fact that permafrost
commonly exhibits a negative thermal offset (Smith and Riseborough, 2002).
This might be related to the unique hydraulic and thermal dynamics in the
active layer, which can cause reversed seasonally variable thermal
conductivity.
All available thermal profiles from 2007 to 2013. The
temperatures at 1.5 m are the mean annual air temperatures (MAAT). Note:
missing years were caused by data gaps.
![]()
Pattern of hydraulic and thermal regimes in the stratified active
layer
Figure 5 shows a typical annual evolution of the active layer to the weather
condition (MAAT is -4.40 ∘C and total rain 316 mm) in 2008. Figure 5a
reflects a typical climate regime with dominant rainfall in the rainy
season from June to September on the north-eastern QTP. Note that the
precipitation measurements mainly include rainfall, and light snowfall was
detected by an acoustic sensor to measure distance change. Figure 5b shows
the active layer during an annual freeze-thaw cycle. The distribution of
liquid soil water indicates that a large amount of suprapermafrost
groundwater existed during the period from late May to the end of January
with a maximum thickness of this saturated layer in excess of 1 m. The
groundwater table roughly fluctuated around 1.0 m below the ground surface
and was mainly recharged by rainfall infiltration during the thawing period
(late April to late October).
Typical dynamics of the active layer during an annual
cycle (data from 2008). (a) Daily mean air temperature and daily rainfall.
(b) Liquid water content (colours) and 0 ∘C-isotherm (black line).
![]()
A noteworthy pattern of the water content distribution is the dry
interlayer around 0.7 m. This layer was occasionally wetted by rain
infiltration during the rainy season, but otherwise was rather dry, as was
also the case during the freezing period. This situation results from a
fine-textured and less permeable layer overlying a coarse-textured one. We
anticipate that the seasonal contrast in total soil water (reduction from
the thawing period to the freezing period) in this dry layer will modify the
seasonal thermal properties of the active layer.
Effect of seasonal total water content reduction on the ratio of thermal
conductivity
For active layers with mineral soils, the ratio of thermal conductivity in
the thawed and frozen states (λt/λf) is assumed
to be less than or equal to one (Riseborough and Smith, 1998). However, the
factor of seasonal total water content reduction is not negligible at our
study site. Figure 6a compares the change in total water content between
summer and winter in 2008. The maximum seasonal water content reduction
occurred around 0.7 m in the active layer. Assuming a thermal conductivity
of 5.0 W m-1 K-1 for the sand with high content of quartz, the
thermal conductivities at different depths were calculated with Eq. (1)
in Fig. 6b. There are two locations with λt > λf
at 0.65 and 0.89 m depth. Significant reduction of the
total water content also occurred at other depths, all accompanied by
corresponding reductions of λf. The interannual variation of
the maximum λt/λf in the profile is listed in
Table 3. In most years, it is very close to or over 1, e.g. 2008 and 2009. The
smaller value 0.90 in 2013 is attributed to the wet year with extraordinary
rainfall.
Seasonal change in soil water content (a) and thermal
conductivity (b) calculated with Eq. (1) in the thick active layer in
2008. θt and θf are the mean total water content
during the summer period and winter period, respectively, and λt and λf are the corresponding thermal conductivities.
![]()
Interannual variation of the maximum seasonal thermal
conductivity ratio (λt/λf) in the monitoring
profile. * Hydrological year occurs from 1 May to 30 April.
Year*
2007
2008
2009
2010
2011
2012
2013
λt/λf
0.99
1.01
1.01
0.97
–
1.00
0.90
In order to exceed the ratio of 1, the seasonal total water content has to
fall below a certain threshold, which depends on soil thermal conductivity
and water content in a thawed state. For instance, the soils with high thermal
conductivity of soil particles will need larger total water content
reduction than the soils with small thermal conductivity of soil
particles. However, given the same amount of total water content reduction,
the soils with a low soil water content in thawed state will be prone to
reach a ratio over 1. Generally, the soil water content condition in a thawed
state depends on soil type and soil structure. Considering the unique
precipitation characteristics on the QTP, seasonal total water content
reduction is common in this kind of permafrost region. Unfortunately, the
role of soil architecture in thermal conductivity parameterization has been
rarely addressed to date.
Comparison of observed and simulated permafrost warming rates
In this section, the model was validated by comparing with the observations.
Figure 7 compares the observed and CMIP5-projected air temperature over the
period of 2006–2014. Generally, the patterns are very similar in Fig. 7a,
but there is a daily averaged upshift of 0.96 ∘C of the projected
values according to the linear regression of all available values (Fig. 7b).
Since there were several data gaps in the observed air temperature, it is
difficult to derive a trend of the measured MAAT change in Fig. 7c. However,
a linear-fitted warming rate of 0.07 ∘C yr-1 of the projected
MAAT from 2006 to 2100 can be derived from Fig. 2a.
Air temperature change over the period of 2006–2014. (a) Comparison
of observed and projected daily averaged air temperature (Ta
and Ta′). (b) Relationship between Ta
and Ta′ with a linear regression. (c) Comparison of the observed and projected
annual mean air temperature (MAAT) after correction with (b). A
linear-fitted (dashed line) warming rate of 0.07 ∘C yr-1 is
derived from the projected one.
![]()
Figure 8 compares the observed and simulated permafrost temperature changes
over the period from 2006 to 2014. The measured temperature Tobs,06 was
taken from a shallow borehole on 30 August 2006 and Tobs,14 was from a
nearby deep borehole on 22 February 2014. Here we assume that permafrost
ground temperature distribution was roughly uniform within a small area (30 m × 30 m)
due to similar surface and subsurface properties. Thus,
permafrost warming can be deduced from ground temperature change from the
boreholes. Considering the small annual fluctuation of ground temperature
change at the depth of 10 m, two corresponding measurements irregularly
conducted once a year in 2006 and 2014 can roughly provide a permafrost
warming rate of 0.05 ± 0.1 ∘C yr-1,Ḃy
assuming a constant warming rate in permafrost temperature at the depth of
about 10 m, a value of 0.02 ∘C yr-1 was calculated from
Tsim,06 and Tsim,14. This is just about half of the observed one.
Compared to the projected warming rate of air temperature at the same period
(0.07 ∘C yr-1), the simulated warming rate is underestimated.
Comparison of observed and simulated permafrost
temperature changes at the Chumaer site over the period from 2006 to 2014.
The measured temperatures, Tobs,06 and Tobs,14, were
taken from a shallow borehole and a nearby deep borehole on 30 August 2006
and 22 February 2014, respectively, while the simulated ones
Tsim,06 and Tsim,14 show all values from the corresponding
years.
![]()
The evident discrepancy between observed and simulated permafrost warming
rate is mainly attributed to the following three factors. First, snow
process is not represented reasonably in the model due to the limitation of
the meteorological forcing. Permafrost is extremely sensitive to snow cover,
which has a much higher albedo (> 0.9) than regular ground
surface (0.1–0.4). Field observations show that snow cover only lasts a few
weeks in pre-/post-winter, and the missing snow cover is mainly caused by
evaporation and sublimation during the diurnal thawing. However, the
projected meteorological data are in daily resolution and the precipitation
is also not generally accurate. Second, the atmospheric forcing was
downscaled from large-scale climate modelling, and it differs from site
observations. In particular, the observed increasing rainfall and lower
amount of snowfall are not well predicted in the projected meteorological forcing.
Third, the model, using the van Genuchten parameters derived from neural
network routines, had difficulty reproducing the site-specific water dynamics
in the active layer and it resulted in smaller ratios of seasonal thermal
conductivity compared to observed ones. Nevertheless, the model can reasonably
mimic hydraulic and thermal regimes and current permafrost thermal status,
and can help us to investigate the effect of the stratified active layer
on permafrost warming.
Role of the stratified thick active layer in permafrost warming
In this section, the effect of the stratified active layer on permafrost
warming was investigated with modelling, and subsurface controlling factors
in the active layer including soil architecture and thermal conductivity of
soil particles were examined with thermal offset and permafrost temperature.
Sect. 3.6.1 demonstrates the simulated hydraulic and thermal pattern in
the active layers, and Sects. 3.6.2 and 3.6.3 present the evolution of
the unique thermal offset and permafrost temperature, respectively.
Simulated hydraulic and thermal pattern in the active layers
Given a typical observed meteorological forcing in 2008, the simulated
hydraulic and thermal patterns of the active layers with different soil
architectures (A1, A2 and A3) in 2008 are shown in Fig. 9. Generally, the
thermal conductivity of the soil particles dominates the active layer
thickness, regardless of the soil architecture. However, the soil hydraulic
patterns are controlled by hydraulic properties (mainly soil architectures
here) and influence the active layer thickness. Notice that the order
of the maximum thawing depth is A3 > A1 > A2, which is
similar in both panels. In addition, the one with realistic soil
architecture, A3 in Fig. 9c, presents a similar hydraulic and thermal pattern
to the one observed (Fig. 9c'), and this shows also that the model can
reasonably capture the hydraulic and thermal dynamics in the active layer.
However, the simulated downward thawing in early summer is slower than the
observed one. This is mainly related to the different soil water content
distributions between observed and simulated. The overestimated soil
water storage in the shallow layer in the simulation lagged its thawing
rate.
Comparison of the simulated unfrozen soil water content
(colour) and the 0 ∘C-isotherm (black line) in the active layer
with different soil architectures and thermal conductivities (λsp) in 2008. The rows correspond to different architectures of the
soil: (a, d) single fine-grained layer (A1), (b, e) single coarse-grained
layer (A2), (c, f) two layer structure (A3). The panels correspond to
λsp= 5.0 W m-1 K-1 (left) and λsp = 2.5 W m-1 K-1
(right). (c') Observed case (same as Fig. 5b).
![]()
Evolution of the unique thermal offset
Based on the above different hydraulic and thermal patterns we investigated
the impact of soil architecture on the warming of the underlying permafrost
by using the thermal offset. Figure 10 shows the evolution of the thermal
offset with the change of MAAT; the latter the result of climate warming.
The thermal offset is calculated as the annual temperature difference
(Ttop-T0.1 m) between the top of the permafrost table and the
near-surface (0.1 m) and disappears when talik presents, disconnecting
the permafrost from the seasonal frost layer. Generally, all the thermal
offsets decrease with increasing MAAT. However, the thermal offsets of A3
in Fig. 10a and b are both positive at the beginning of the simulation,
when the permafrost is close to thermal equilibrium, whereas the thermal
offset of A2 is always negative and the thermal offset of A1 is in between
A2 and A3.
Thermal offset as a function of mean annual air
temperature (MAAT) obtained from the simulations for the period 1980–2100 of
the three soil architectures, A1, A2 and A3. The lines connect the means of
1.0 ∘C-bins, the bars indicate the corresponding standard
deviations. The upper frame is for λsp= 5.0 W m-1 K-1,
the lower one for λsp= 2.5 W m-1 K-1.
![]()
Thermal offset (TTOP-MAGST) originates from different heat transfer
efficiencies of an active layer between summer and winter in a permafrost
region at equilibrium state. These mainly result from thermal conductivity
of ice being 4 times that of water. With this, the annual mean
temperature profile is shifted towards the summer profile and the thermal
offset is typically negative. Its value is modified by several aspects, for
instance it is decreased through snow cover and increased by a soil water
content that is higher in summer than in winter. For instance, the positive
thermal offset at equilibrium state in Fig. 10 was mainly led by the high
ratio of λt/λf around 1 via seasonal water
content reduction. This phenomenon is prone to appear in the thick
stratified active layers on the QTP with a summer-monsoon-dominated
precipitation pattern. Generally, the schematic mean annual ground
temperature profile is closer to the one described by Brown (1970)
than the one suggested by Smith and Riseborough (2002) as shown in Fig. 11.
In addition, the concept of thermal offset will be invalid at disequilibrium
conditions. For instance, the thermal offsets in Fig. 10 decrease
dramatically with climate warming. It is clear that the decreasing
negative thermal offsets were not caused by λf > λt
but by the lag between surface and subsurface warming. This
is corroborated by the observed thermal offsets (Fig. 4) for which positive
values occurred in 2007 and 2008 and decreased to negative values in 2009 and
2013. Therefore, the concept of the normal offset is not suitable for the
studied case, and the plausible “normal” thermal offset might not
necessarily be attributed to λf >λt,
but to permafrost disequilibrium.
Schematic mean annual ground temperature for two types
of permafrost. Black curve shows the studied permafrost with positive thermal
offset and red curve shows common permafrost with negative thermal offset.
![]()
Subsurface factors controlling the permafrost warming rate
Apart from the change rate of climatic forcing, the evolution of thermal
regime in active layers is also not negligible for permafrost warming, and
it is related to subsurface factors like thermal conductivity of soil
particles and soil hydraulic properties. In particular, the thermal regime
in the studied thick active layer with unique soil architecture strongly
relies on the hydraulic regime. Given three different soil architectures and
two different thermal conductivities of soil particles (λsp = 5.0 W m-1 K-1
and λsp= 2.5 W m-1 K-1), the differences of the permafrost warming are shown in Fig. 12.
Generally, the permafrost with a higher thermal conductivity of the matrix
(left panel of Fig. 12) degrades faster than that in the right panel with a
lower thermal conductivity, and the influence of soil architecture in the
left panel is negligible. In contrast, the role of soil architecture emerges
in the right panel, and the stratified active layer (A3) leads to a faster
permafrost warming rate. This is mainly attributed to the effect of seasonal
total water content change on seasonal variation of thermal conductivity.
For the active layer with a high λsp, the ratio of seasonal
thermal conductivity is always close to 1.0, and the impact of seasonal
total water content reduction is rather weak. However, for the active layer
with a low λsp, the ratio of seasonal thermal conductivity
depends strongly on the seasonal total water content reduction. Besides,
soil architecture with low soil water content will have a higher ratio of
seasonal thermal conductivity, given the same amount of seasonal total water
content reduction.
Comparison of the influence of soil architecture and
thermal conductivity of soil particles (λsp) on permafrost
degradation with simulations over the period from 1980 to 2100.
(a–e) Annual mean ground temperature (MAT) of A1, A2 and A3 with a high λsp= 5.0 W m-1 K-1 at selected years; (f–j) the same
as (a–e) but with a low λsp= 2.5 W m-1 K-1.
![]()
The effects of soil architecture on permafrost warming evolve in time. First
of all, the Asian summer monsoon, caused by the elevated heat source driven
by the QTP (e.g. Yeh et al., 1957; Yanai et al., 1992) may not disappear in
the near future, but the monsoon intensity might shift due to climate change
(Duan et al., 2011). Secondly, thickening active layers can weaken the role
of soil architecture in permafrost warming via exerting impact on the
suprapermafrost water level as well as soil water content distribution.
Given a monsoon-dominated precipitation pattern in the projected climate
model data, Fig. 13 shows the evolution of seasonal λt/λf of the shallow soils (0–1.5 m) over the simulating
period of 1980–2100. Decadal mean λt/λf were
calculated for all the simulations. Compared to the same soil architectures
but different thermal conductivities of soil particles, the maximum decadal
mean λt/λf in the left panel with λsp= 5.0 W m-1 K-1 are higher than that in the right
panel with λsp= 2.5 W m-1 K-1. However, the
differences of λt/λf among the architectures in
the right panel are bigger than in the left panel. Besides, the maximum
values of λt/λf in A3 are bigger than the other
two from the 1980s to 2040s, then they gradually become smaller. Generally, the
shrinking differences in decadal mean λt/λf
among the three architectures indicate that the effects of the stratified
active layers on permafrost warming are significant in the early state of
permafrost degradation and will decrease afterwards. These results are
all consistent with the thermal regimes in Fig. 12.
Comparison of the influence of soil architecture and
thermal conductivity of soil particles (λsp) on the seasonal
thermal conductivity ratio λt/λf in the shallow
soils over the period from 1980 to 2100. (a–e) Decadal mean λt/λf of A1, A2 and A3 with a high λsp= 5.0 W m-1 K-1
at selected decades; (f–j) the same as (a–e)
but with a low λsp= 2.5 W m-1 K-1.
![]()
Conclusions
In summary, this study presented an interesting case which showed the effects of
stratified active layers with a high ratio of seasonal thermal conductivity,
λt/λf≥ 1.0, on permafrost warming on the
QTP. Key findings are listed in the following.
An observed extraordinary permafrost warming rate (> 0.5 ∘C
per decade) was found at the study site with sparse
vegetation and annual precipitation 300–400 mm. Apart from the climate
drivers and the unusual high geothermal flux from the bottom, a high ratio
of seasonal thermal conductivity in the stratified active layer is
instrumental in regulating the interaction between climate and permafrost.
Observation and simulation suggest that the concept of the thermal
offset proposed by Smith and Riseborough (2002) is not suitable for the
studied permafrost on the QTP. In contrast to the normal negative thermal
offset caused by the low ratio of seasonal thermal conductivity, a reversed
thermal offset at equilibrium state is formed due to the remarkably high
ratio of seasonal thermal conductivity (≥ 1).
Furthermore, the specific soil architecture plays a non-negligible role
in forming a dry interlayer while raising the ratio λt/λf and resulting in a higher permafrost warming rate
than the active layers with uniform soils.
Considering the importance of rainfall in the mechanism of the hydraulic and
thermal dynamics in the active layers, there is no doubt that permafrost
warming would be influenced by the increasing precipitation in recent years
and in future on the QTP. Consequently, soil hydraulic properties,
particularly soil architecture, become more and more important for the
thermal conductivity parameterization in land surface and permafrost
modelling. In particular, the empirical permafrost models using a ratio of
seasonal thermal conductivity smaller than 1.0 might underestimate the effect
of climate warming. However, this study is mainly based on a specific site.
More field investigations are required to reveal the regional difference in
permafrost degradation over the QTP.