Soils on the Qinghai–Tibetan Plateau (QTP) have distinct physical properties
from agricultural soils due to weak weathering and strong erosion. These
properties might affect permafrost dynamics. However, few studies have
investigated both quantitatively. In this study, we selected a permafrost
site on the central region of the QTP and excavated soil samples down to
200 cm. We measured soil porosity, thermal conductivity, saturated hydraulic
conductivity, and matric potential in the laboratory. Finally, we ran a
simulation model replacing default sand or loam parameters with different
combinations of these measured parameters. Our results showed that the mass
of coarse fragments in the soil samples (diameter
Permafrost underlies 25 % of Earth's surface. Degradation of permafrost has been reported extensively in Alaska, Siberia and the Qinghai–Tibetan Plateau (QTP; Boike et al., 2013; Jorgenson et al., 2006; Wu and Zhang, 2010). Permafrost thaw has global impacts by releasing large quantities of soil carbon previously preserved in a frozen state and enhancing concentrations of atmospheric greenhouse gases, which will promote further atmospheric warming and degradation of permafrost (Anisimov, 2007; McGuire et al., 2009). Permafrost dynamics also have local to regional impacts on ecosystems by altering soil thermal and hydrological regimes (Salmon et al., 2015; Wang et al., 2008; Wright et al., 2009; Ye et al., 2009; Yi et al., 2014a). In addition, degradation of permafrost affects infrastructure, such as QTP railways and roads (Wu et al., 2004) or the Trans-Alaska Pipeline System in Alaska (Nelson et al., 2001). Therefore, it is critical to develop mitigation and adaptation strategies in permafrost regions for ongoing climate change. Accurate projection of the degree of permafrost degradation is a prerequisite for developing these strategies.
Significant effort has been made to improve modeling accuracy and efficiency
of permafrost dynamics along two primary lines of inquiry. One is to create
suitable freezing and thawing algorithms for different applications,
including land surface models (Chen et al., 2015; Oleson et al., 2010; Wang
et al., 2017), permafrost models (Goodrich, 1978; Langer et al., 2013; Qin
et al., 2017), and other related models (Fox, 1992; Woo et al., 2004). The
other line of inquiry is focused on schemes of soil physical properties
(Chen et al., 2012; Zhang and Ward, 2011), which play a critical role in
permafrost dynamics. For example, porosity determines the maximum amount of
water that can be contained in a soil layer, thermal properties determine
the heat conduction within soil layers, and hydraulic properties determine
the exchange of soil water between soil layers. The soil water content also
determines the large amount of latent heat lost or gained by freezing or
thawing. On the QTP, soil is coarse due to weak weathering and
strong erosion (Arocena et al., 2012). Soils with gravel content (particle
diameter
In this case study we investigated the characteristics of soil physical properties at a site on the central QTP and their effects on permafrost dynamics. We first measured soil physical properties of excavated soil samples in a laboratory. We then conducted a sensitivity analysis with an ecosystem model by substituting the default soil physical properties with those that we measured. We aimed to emphasize the effects of coarse-fragment content on soil physical properties and on permafrost dynamics, rather than develop general schemes of soil physical properties for use in modeling studies on the QTP.
Locations of
The site (34
The site is on top of upland plain landforms, which are formed from fluvial
and deluvial sediments. The surficial sediments are dominated by fine to
gravelly sands and stones (Fig. 2; Yin et al., 2017). Soils at this site
are Inceptisols (Wangping Li, Lanzhou University of Technology, personal
communication, 18 April 2018) that are commonly underlain by mudstone. The plant community
type is mainly alpine meadow, which is dominated by monocotyledonous species,
primarily Poaceae and Cyperaceae. The dominant species are
Images of site conditions:
A weather station was set up in 2002 (Fig. 2a) to measure air temperature
and relative humidity (2.2 m, HMP45C-L11/L36, Campbell Scientific Inc., USA),
solar radiation (MS-102, EKO, Japan), and precipitation (QMR102, Vaisala
Company, Finland). Soil temperatures were measured at depths of 5, 10, 20,
40, 80, and 160 cm using a PT-100 (EKO, Japan); soil moisture were measured
at depths of 20, 40, 80, and 160 cm using a CS616-L50 (EKO, Japan). A CR3000
data logger (Campbell Scientific Inc., USA) was used to store these data at
30 min intervals. These readings were averaged or summed (e.g.,
precipitation) into monthly values to drive and validate the model. Based on
measurements, multi-year mean annual air temperature, precipitation,
downward solar radiation, and relative humidity were
A borehole was drilled in 2002, and thermistors made by the State Key
Laboratory of Frozen Soil Engineering, Chinese Academy of Sciences were
installed at 0.5 m intervals from 0.5 to 10 m, at 2 m intervals from 12 to
30 m, at 4 m intervals from 34 to 50 m,, and at 55 and 60 m. Temperature
accuracy of this type of thermistor is
Time series of data measured at the Beiluhe weather station,
Qinghai–Tibetan Plateau, 2003 to 2011:
Permafrost dynamics are affected by atmosphere, vegetation, and soil
textures; therefore, we excavated soil close to the weather station and
borehole (Fig. 2a) down to 2 m (Fig. 2b) in August 2014. We used cut
rings (10 cm diameter, 6.37 cm height and 500 cm
We used the KD2 Pro (Decagon, US) to measure thermal conductivity of soil
samples. The steps we took to determine soil properties for each sample were
as follows: (1) the soil sample was dried in an oven and weighed (0.001 g
precision) to calculate bulk density; then (2) the soil sample was exposed to
a constant temperature (20
We used pressure membrane instruments (1500F1, Soilmoisture Equipment Corp, US) to measure the matric potential of soil samples (Azam et al., 2014; Wang et al., 2007), using both 15 bar and 5 bar pressure chambers. Pressure values were set to 0, 10, 20, 40, 60, 80, 100, 150, 200, 300, and 400 kpa. It usually took 3–4 days to finish one measurement at one pressure level. We used a soil permeability meter (TST-70, Nanjing T-Bota Scietech Instruments & Equipment Co., Ltd. China) to measure saturated hydraulic conductivity of soil samples (Gwenzi et al., 2011). Finally, soil samples were sieved through a 2.0 mm mesh, and soil particle size distribution was determined with a laser diffraction analyzer (Malvern-2000, Worcestershire, UK).
To simulate soil temperatures, soil liquid water content, temperature in rock layers, active layer depth (ALD) and permafrost low boundary depth (PLB) dynamics we used a dynamic organic soil version of the Terrestrial Ecosystem Model (DOS-TEM). Models from the TEM family simulate the carbon and nitrogen pools of vegetation and soil, and their fluxes among atmosphere, vegetation, and soil (McGuire et al., 1992). They have been widely used in studies of cold region ecosystems (e.g., McGuire et al., 2000; Yuan et al., 2012; Zhuang et al., 2004, 2010). The DOS-TEM consists of four modules: environmental, ecological, fire disturbance, and dynamic organic soil (Yi et al., 2010). The environmental module operates on a daily time interval using mean daily air temperature, surface solar radiation, precipitation, and vapor pressure, which are downscaled from monthly input data (Yi et al., 2009a). The module takes into account radiation and water fluxes among the atmosphere, canopy, snowpack, and soil.
Earlier versions of TEM did not simulate soil temperature (McGuire et al.,
1992). Zhuang et al. (2001) incorporated Goodrich (1978) permafrost model
into TEM. Yi et al. (2009b) incorporated a two-directional Stefan algorithm
to simulate soil freezing and thawing for complex soils with changes in the soil
organic and moisture content. Temperatures of all soil layers in the DOS-TEM
are updated daily. Phase change is calculated first before heat conduction.
A two-directional Stefan algorithm is used to predict the depths of freezing
or thawing fronts within the soil (Woo et al., 2004). It first simulates the
depth of the front in the soil column from the top downward, using soil
surface temperature as the driving temperature. It then simulates the front
from the bottom upward using the soil temperature at a specified depth
beneath a front as the driving temperature (bottom-up forcing). The latent
heat used for phase change is recorded for each soil layer. If a layer
contains
The version of the DOS-TEM in this study uses the Côté and
Konrad (2005) scheme to calculate thermal conductivity (Yi et al., 2013; Pan et
al., 2017), which is also been used by other studies on the QTP (e.g., Chen
et al., 2012; Luo et al., 2009), and is as follows:
Surface runoff, infiltration, and water redistribution among soil layers are
simulated in a similar way to the Community Land Model 4 (Oleson et al., 2010).
Soil matric potential (
The DOS-TEM has been verified against the Neumann equation for water, mineral and organic soil under an idealized condition (Yi et al., 2014b), and validated against field measurements for various locations in Alaska, the Arctic, and the QTP (Yi et al., 2009b, 2013, 2014a).
We used the monthly averaged air temperature, downward radiation,
precipitation, and humidity as input to drive the DOS-TEM. Leaf area index
(LAI), leaf area per unit ground surface area, was specified to be
0.6 m
Soil temperature and moisture were initialized at
The soil textures on the QTP mainly consist of loam, sand, and coarse
fragment soils (Wu and Nan, 2016). We used a uniform sand or loam soil
profile to represent coarse and fine soil textures, respectively. Sands are
the coarsest texture considered in most modeling studies (e.g., Oleson et
al., 2010). Therefore, we used our measured parameters to substitute the
parameters of sand and loam to investigate the effects of coarse-fragment
soil parameters on permafrost dynamics. We first ran DOS-TEM using the
default porosity, soil thermal conductivity (Eq. 5), hydraulic
conductivity (Eq. 9), and matric potential schemes of these two default
soil textures (Eq. 8). The default parameters
We did not measure the heat capacity. The maximum and minimum heat
capacities of mineral soil types considered in land surface model are 2.355
and 2.136 MJ m
The mean (standard deviation in brackets) of measured soil bulk
density (
The particle size diameter fractions (for
Results from laboratory analysis of the soil samples are shown in Tables 1
and 2. The mean mass ratio of the coarse-soil fraction (particle size
diameter
The mean (standard deviation in brackets) of the measured frozen
and unfrozen dry and saturated soil thermal conductivity (W m
The results of the thermal conductivity determinations are shown in Table 3.
The unfrozen
The relationship between soil saturation (solid and dotted lines
represent frozen and unfrozen cases) and soil thermal conductivity (
Results from determining thermal conductivities using the Côté and
Konrad (2005) scheme are shown in Fig. 4b. The default frozen and unfrozen
The relations between
The mean (standard deviation) of measured saturated hydraulic
conductivity (
The mean
The correlation coefficients between calculated and fitted
Comparisons of soil temperatures (
The mean root mean square errors (RMSEs) between monthly measured soil
temperatures and model runs with measured parameters using different
combination of soil thicknesses (3.25, 4.25, and 5.25 m) and slopes (0, 5,
and 10
Comparisons of soil volumetric liquid water content (
The standard deviations of soil temperatures among different slopes and soil
thicknesses using measured parameters were larger than those using the
default parameters (Fig. 6), and they increased from 0.40
The mean RMSEs between monthly measured
The standard deviations of
The mean RMSEs between measured ALDs (derived from linear interpolation of
soil temperatures) and modeled ALDs (simulated explicitly) were about 1.06,
1.72, and 0.28 m for model runs with sand, loam, and measured parameters
(Fig. 8a). The mean standard deviations were about 0.088, 0.026, and
0.28 m. All simulations using sand and loam parameters underestimated ALDs. When
The mean RMSEs between measured PLBs (derived from linear interpolation of
temperatures) and modeled PLBs (derived from linear interpolation of
simulated bed rock temperatures) were about 10.25, 10.23, and 6.71 m for
model runs with sand, loam, and measured parameters (Fig. 8b). The mean
standard deviations were about 1.89, 1.51, and 6.62 m. All simulations using
sand and loam parameters overestimated PLBs. When
Deep soil layers used in models are usually specified as being thick. For
example, a 1 m thick soil layer was used in our simulations starting around
3 m soil depth. Soil temperatures at this depth are usually close to 0
Contour plots showing
Root mean square errors between measurements and model
simulations (with different combinations of measured porosity (I), thermal
conductivity (II), hydraulic conductivity (III), and matric potential (IV)
substituted for default sand parameters) for 20 and 100 cm soil temperatures
(
Root mean square errors between measurements and model
simulations (with different combinations of measured porosity (I), thermal
conductivity (II), hydraulic conductivity (III), and matric potential (IV)
substituted for default loam parameters) for 20 and 100 cm soil temperatures
(
Replacing default sand or loam porosity with
Root mean square errors between measurements and model
simulations (with different combinations of measured porosity (I), thermal
conductivity (II), hydraulic conductivity (III), and matric potential (IV)
substituted for default sand parameters) for 10, 40, 80, and 160 cm
soil volumetric liquid water content. O and All represent model runs without
substitution of default parameters and with all four parameters substituted. Mean and standard deviation of model simulations with
three different soil thicknesses at each slope (0, 5, and 10
Root mean square errors between measurements and model
simulations (with different combinations of measured porosity (I), thermal
conductivity (II), hydraulic conductivity (III), and matric potential (IV)
substituted for default loam parameters) for 10, 40, 80, and 160 cm
soil volumetric liquid water content. O and All represent model runs without
substitution of default parameters and with all four parameters substituted. Mean and standard deviation of model simulations with
three different soil thicknesses at each slope (0, 5, and 10
Replacing default sand or loam thermal conductivity with measured parameters
reduced mean RMSEs of soil temperatures from 1.18 or 1.84
Replacing default sand or loam hydraulic conductivity with measured
parameters had very small effects on mean RMSEs of soil temperatures and
ALDs (Figs. 9 and 10). The same was true for matric potential. When
hydraulic conductivity of default sand or loam was substituted, mean RMSEs
of PLB decreased or increased, respectively. However, when matric potential
was substituted, mean RMSEs of PLBs increased or decreased.
When hydraulic conductivity or matric potential parameters were substituted
in default sand or loam parameters, mean RMSEs of
Model performance when default sand parameters are substituted with combinations of measured porosity (I), thermal conductivity (II), hydraulic conductivity (III), and matric potential (IV).
Best column shows the model simulations (individual parameter
substitution) with the smallest root mean square error (RMSE) for 100 cm
soil temperature (ST,
Model performance when default loam parameters are substituted with combinations of measured porosity (I), thermal conductivity (II), hydraulic conductivity (III), and matric potential (IV).
Best column shows the model simulations (individual parameter
substitution) with the smallest root mean square error (RMSE) for 100 cm
soil temperature (ST,
We compared model simulations with different combinations of measured parameters (porosity, thermal conductivity, hydraulic conductivity, and matric potential) to those with one substituted measured parameter. We ranked those model runs with less RMSEs than the best of the model runs with one parameter substituted with a measurement-derived value (Tables 5 and 6). We did not consider the 10 cm soil temperature, which was similar across all model runs.
For sand, model simulations with porosity and thermal conductivity and/or hydraulic conductivity substituted had four outcomes with lower RMSEs (Table 5 and Figs. 9 and 11). Only two out of seven outcomes had lower RMSEs with all four parameters substituted. Among all the 18 cases with RMSEs less than the individual “best” RMSE, porosity was included 18 times, and thermal conductivity and hydraulic conductivity were included 10 times.
For loam, model simulations with porosity and thermal conductivity substituted had five outcomes with lower RMSEs (Table 6 and Figs. 10 and 12). Among all the 27 cases with RMSEs less than the individual best RMSE, porosity was included 27 times, thermal conductivity was included 16 times, and matric potential 14 times.
Changes in slope alone had small effects on simulated soil temperatures and
ALDs (Figs. 9 and 10). An increase in slope generally reduced RMSEs of
Soil thickness had small effects on 20 and 100 cm soil temperatures and
10–160 cm
Although the effects of coarse-fragment soils on permafrost dynamics have been considered in a few modeling studies, the thermal and hydraulic properties of coarse-fragment soils were calculated without validation or calibration (Pan et al., 2017; Wu et al., 2018). To our knowledge, this is the first study measuring physical properties of coarse-fragment soil samples from the permafrost region of the QTP.
The weight fraction of coarse fragment (diameter
The measured thermal conductivities of saturated soil samples were relatively close to those estimated by the Côté and Konrad (2005) scheme. However, they were much less than those estimated by the Farouki scheme (Fig. 4). Several other studies also found that Farouki scheme overestimated soil thermal conductivity (Chen et al., 2012; Luo et al., 2009).
One important finding of this study is the relatively small value of
porosity. The
Soil water not only affects soil thermal properties (e.g., thermal
conductivity and heat capacity) but also affects the amount of latent heat
lost or gained for freezing or thawing, respectively (Goodrich, 1978;
Farouki, 1986). Soil water is determined by infiltration,
evapotranspiration, water movement among soil layers, subsurface runoff, and
exchange with a water reservoir. Therefore, processes or parameters that
affect soil water dynamics will also affect permafrost dynamics. This study
quantitatively assessed the effects of soil water on permafrost dynamics.
For example, when default loam parameters with high porosity and low
saturated hydraulic conductivity were used, soil layers were almost
saturated (Fig. 7). The simulated ALDs were about 1.58 m, which was less
than half of measured ALDs (Fig. 8a). When the slope was 0
Land surface models generally represent soil water dynamics (e.g., Chen et al., 2015; Oleson et al., 2010; Wang et al., 2017). However, the thermal processes in permafrost models usually use specified thermal properties, which were static during model simulations (Li et al., 2009; Nan et al., 2005; Qin et al., 2017; Zou et al., 2017). As shown in this study, variation in soil water content in coarse-fragment soils strongly affects the thermal and hydrological properties; thus it is critical to simulate soil water dynamics to properly project permafrost dynamics in the future.
We used cut rings with 10 cm diameter to sample soil and weathered mudstones. However, it is very likely that there could have been much bigger coarse-fragment soils. Therefore, larger containers should be used to take samples for further laboratory analysis in the future.
During our laboratory work, we found two phenomena. First, we used the QL-30 thermophysical instrument (Anter Corporation, US) to measure thermal conductivity. It worked properly in an unfrozen condition. However, when frozen, the surface of the soil sample was usually uneven due to frost heave, which reduces the contact between the QL-30 plate and the soil sample surface. The measured frozen thermal conductivities were smaller than unfrozen thermal conductivity even for the case of saturation, which was definitely wrong. Thus we used the KD2 Pro to determine thermal conductivities. The second phenomenon was that there seems to be a threshold of soil saturation, below which unfrozen soil thermal conductivity is greater than frozen soil thermal conductivity (Fig. 4a). This pattern was somewhat exhibited in estimates of the Côté and Konrad (2005) scheme (Fig. 4b), but not in the estimates of the Farouki scheme (Fig. 4c). More measurements using instruments with higher accuracy should be made in the future.
The measured porosities are generally smaller than those calculated from
bulk density. We made additional model simulations using porosities
calculated from bulk density in combination with other measured parameters.
Our results showed that the RMSEs of ALD and PLB were 0.55 and 4.78 m
(figures not shown), whereas those calculated using
Although the DOS-TEM using measured parameters provided satisfactory
results, there are some aspects requiring further improvement in the future.
For example, the measured soil moisture at 40 cm depth was less than
0.1 m
The TEM family models use monthly atmospheric data to drive both site and regional applications. In this study, 30 min and daily driving data are available. Although it is possible to lose fidelity after daily interpolations, we still decided to use monthly driving data for the following reasons: (1) Zhuang et al. (2001) performed a test with daily and monthly driving data sets, and the results showed that the RMSEs of ALD were about 3 cm; and (2) we intend to apply the model over large regions where reliable daily data sets might not be available.
The coarse-fragment content of soil affects its physical properties. For example, soil porosity and saturated hydraulic conductivity are determined by the fraction of gravel, diameter, and degree of mixture (Zhang et al., 2011). Thus soil texture plays an important role in permafrost dynamics (Fig. 8). The dominant soil textures on the QTP from Wu and Nan (2016) are loam, sand, and gravel. The specification of loam in simulations results in estimates of ALD that are much smaller than the measurements (Yi et al., 2014a). To properly simulate the distribution and dynamics of permafrost on the QTP under climate change scenarios, it is important to develop proper schemes of soil physical properties in relation to coarse-fragment content (including gravel) and to develop regional data sets of soil texture for input.
Organic soil carbon content in mineral soil on the QTP affects soil porosity and thermal conductivity (Chen et al., 2012). However, in the site considered in this study, the amount of organic soil carbon in the soil was small (Fig. 2), and we did not explicitly consider the effects of organic soil carbon on soil properties. Alpine swamp meadow, alpine meadow, alpine steppe, and alpine desert are the major vegetation types on the QTP (Wang et al., 2016; see also Fig. 1b). Alpine swamp meadow and alpine meadow usually contain fine soil particles and high organic carbon density, while the other two types usually contain coarse-soil particle and low organic carbon density (Qin et al., 2015). More laboratory work is needed to develop proper schemes for representing mixed soil with fine mineral, coarse fragment (including gravel), and organic carbon in permafrost models. It is the first priority to develop schemes that make use of porosity data sets due to its importance and simplicity of measurement.
The development of a spatially explicit data set of soil texture is also required for regional projections of permafrost changes on the QTP. Currently, a preliminary data set for gravel exists (Wu and Nan, 2016), though gravel soil has only been mentioned in a few papers on the QTP (Chen et al., 2015; Wang et al., 2011; Yang et al., 2009). One way to improve the regional data set is to collect relevant data through extensive field campaigns (e.g., Li et al., 2015). Ground-penetrating radar is a feasible tool that retrieves soil thickness above the coarse-fragment soil layer (Han et al., 2016), and coarse-fragment soils can be identified in photos taken with unmanned aerial vehicles (Chen et al., 2017; Yi, 2017). In combination with ancillary data sets (e.g., geomorphology, topography, vegetation), it is possible to improve the accuracy of spatial data sets of soil texture on the QTP (Li et al., 2015; Wu et al., 2016). Another way is to retrieve soil physical properties using data assimilation technology, as done by Yang et al. (2016), who assimilated porosity using a land surface model and microwave data.
In this study, we excavated soil samples from a permafrost site on the central QTP and measured soil physical properties in the laboratory. Coarse fragments were common in the soil profile (up to 65 % of soil mass) and porosity was much smaller than the typical soil types used in land surface models. We then performed a sensitivity analysis of these parameters on soil thermal and hydrological processes within a terrestrial ecosystem model. When default sand or loam parameters were substituted with measured soil properties, the model errors of active layer depth were reduced by 74 % or 84 %, whereas those of the low permafrost boundary were reduced by 35 % or 34 %. Our sensitivity analyses showed that porosity played a more important role in reducing model errors than the other soil properties examined. Though it is unclear how representative this soil is in the QTP, it is clear that soil physical properties specific to the QTP should be used to properly project permafrost dynamics into the future.
The laboratory data set is included in Tables 1–4. Any other specific data can be provided by the authors on request.
YD and QW designed the study, JC, YQ and SY collected the soil samples, and YH and XG performed laboratory measurements. SY did modeling work. All authors contributed to the discussion of the results. SY led the writing of the paper and all co-authors contributed to it.
The authors declare that they have no conflict of interest.
We would like to thank Dave McGuire of University of Alaska Fairbanks for his careful editing; Yi Sun for vegetation classification; Xia Cui of Lanzhou University, Guangyue Liu for determining depth of zero annual amplitude, Yan Qin for measurements of soil particle size distribution; Chien-Lu Ping of University of Alaska and Wangping Li of Lanzhou University of Technology for helping on soil taxonomy; and the editor and two anonymous reviewers for valuable comments. This study was jointly supported through grants provided as part of the National Natural Science Foundation Commission (41422102, 41690142, and 41730751), and the independent grants from the State Key Laboratory of Cryosphere Sciences (SKLCS-ZZ-2018). Edited by: Peter Morse Reviewed by: two anonymous referees