Introduction
CO2 exchange between soil and atmosphere constitutes a major carbon (C) loss
from terrestrial ecosystems (Raich et al., 2002; Giardina et al., 2014). It
also plays an important role in the feedbacks between the global carbon cycle
and climate change (Rustad et al., 2000; Giardina et al., 2014; Karhu et
al., 2014). Arid and semiarid (dryland) ecosystems cover over 40 % of land
surface and contribute notably to interannual variations of terrestrial C
sink (Poulter et al., 2014). However, the contribution of soil CO2 flux
(FS) from those ecosystems to the global C budget is less studied
(Castillo-Monroy et al., 2011; Gao et al., 2012; Jia et al., 2014). The
temperature dependency of biological CO2 production (i.e. autotrophic
respiration and heterotrophic respiration) serves a conventional basis for
FS modelling in many terrestrial ecosystems (Raich and Tufekciogul,
2000; Ryan and Law, 2005; Song et al., 2015). Soil CO2 flux of dryland
ecosystems is also widely interpreted using temperature-response functions
modified by other environmental constraints, e.g. soil water content,
abundance of substrates and microbial activities (Curiel Yuste et al.,
2007; W. Wang et al., 2014; B. Wang et al., 2014, 2015).
Although many empirical studies have explained the dynamics of soil CO2
flux in specified space–time, their lack of mechanistic descriptions
represents a major difficulty in extrapolation under changing environmental
conditions (Fan et al., 2015). Soil CO2 flux is a “bulk” exchange
that comprises two main sets of processes, i.e. the CO2 production and
transport (Fang and Moncrieff, 1999; Fan et al., 2015). Hence, models
considering only autotrophic and heterotrophic respiration often fail to
account for the observed FS dynamics (Austin and Vivanco, 2006).
Gas-transport processes are important mechanisms regulating the magnitudes
and hysteretic features of soil CO2 fluxes (Ma et al., 2013). A
substantial fraction of respired CO2 may be transported to the atmosphere
via xylem and may not be measured by techniques like soil reparation
chambers (Bloemen et al., 2013, 2016). During wet periods, soil CO2
efflux could decrease significantly by water clogging of soil pores, which
restricts the diffusion of O2 and CO2 gases (Šimunek and
Suarez, 1993; Fang and Moncrieff, 1999). In dryland soils of high
salinity/alkalinity, CO2 transport and the water cycle are tightly coupled,
as large inorganic C fluxes can be driven solely by dissolution and
infiltration of CO2 and carbonates (Buysse et al., 2013; Ma et al.,
2013; Fa et al., 2014). Such inorganic processes may introduce
fluctuations not only to hourly or diurnal soil CO2 effluxes (e.g. Emmerich,
2003; Xie et al., 2009; Buysse et al., 2013) but also to terrestrial
CO2 sinks at much broader spatiotemporal scales (Schlesinger, 2009; Li
et al., 2015).
Key processes contributing to CO2 production in dryland soils also
extend beyond autotrophic and heterotrophic respiration. Although biocrust
organisms (lichens, mosses, bacteria, fungi and microfauna) mainly inhabit
in the top few centimetres of a soil profile, they constitute up to 70 %
of biomes in interspace areas (Belnap et al., 2003). These communities are able to
uptake C from atmosphere (Belnap et al., 2003; Castillo-Monroy et al., 2011;
Maestre et al., 2013), leading to greater concentrations of organic matter
in the soil crust than in the layer underneath (Ciais et al., 2013). Although crust
organisms could be inactive under stresses (e.g. drought; Green and Proctor,
2016), their photosynthetic potentials could be large (Zaady et al., 2000;
Lange, 2003), even comparable to temperate forests with closed canopies
(e.g. Zaady et al., 2000). The C uptakes of biocrusts are highly sensitive
to stresses like droughts, thermal extremes and excessive ultraviolet
radiation (Pointing and Belnap, 2012). Such variations can readily alter
crusted soils between considerable CO2 sinks and sources within a few
hours (e.g. Bowling et al., 2011; Feng et al., 2014). In addition, the
accumulation of debris from crust and canopy fuel photodegradation
represents an important abiotic C loss in arid conditions aside from biotic
decompositions (e.g. Austin and Vivanco, 2006; Throop and Archer, 2009).
Photodegradation is likely to dominate the mineralization during dry daytime
periods, when radiation is strong and microbial activities are prohibited by
low moisture content and high temperatures (e.g. Gliksman et al., 2016). On
an annual basis, photodegradation could consume more than 10 % of soil
organic matter (SOM) at the surface (e.g. Austin and Vivanco, 2006; Henry et
al., 2008; Brandt et al., 2010), even for the substrates (e.g. lignin) that
are difficult to degrade via biotic pathways (Henry et al., 2008).
The influences of the multiple C processes (i.e. autotrophic and
heterotrophic respiration, C exchanges by biocrust organisms, inorganic C
fluxes and photodegradation) on soil CO2 exchanges are highly
overlapped and tightly related to water-energy conditions. In dryland
ecosystems, patchy vegetation and large fractions of interspace area are
common features (Domingo et al., 2000), and the water-thermal conditions can
vary considerably from plant-covered areas to interspace within even a few
metres (Rodríguez-Iturbe et al., 2001; Caylor et al., 2008; Ma et al.,
2011). The water-energy dynamics at the different surfaces are linked by
multiple advection processes both above and below the ground (Gong et al.,
2016). Due to the complexity of water-energy processes, there may exist
high non-linearity of water-thermal responses to climatic
variabilities (e.g. Phillips et al., 2011; Barron-Gafford et al., 2013).
This will also complicate the C responses and consequently affect the
relationships between CO2 fluxes and environmental controls (e.g.
Jarvis et al., 2007; Song et al., 2015).
Global climate change is expected to increase annual mean air temperatures
considerably and alter precipitation regimes (Donat et al., 2016).
Understanding the responses of dryland ecosystems to such changes requires
mechanistic models that integrate the multiple biotic and abiotic processes
in soil C cycling. So far, only a few models have coupled the biotic
CO2 production with the transport of gases and heat (Šimunek and
Suarez, 1993; Fang and Moncrieff, 1999; Phillips et al., 2011; Ma et al.,
2013; Fan et al., 2015). Nevertheless, none of those models have described
the heterogeneous water-energy conditions in the soil–plant–atmosphere continuum
(SPAC) nor the unconventional C fluxes such as C uptake by biocrusts and
photodegradation, despite the importance of these processes to arid and
semiarid environments. Perhaps models by Porada et al. (2013) and Kinast et
al. (2016) represent the few existing works in this sense. However, both
models focus on the patterns at the regional scale with very simplified
ecosystem processes. In this study, we aim to (i) investigate the roles of
soil CO2 components in regulating soil CO2 effluxes in a dryland
ecosystem using a process-based modelling approach and (ii) estimate the
plant–interspace differences in the componential C processes.
Materials and methods
Model overview
The process-based model was build based on a semiarid shrubland ecosystem
located at the southern edge of Mu Us Desert (37∘42′1′′ N,
107∘13′7′′ E; 1560 m above sea level; Fig. 1a), Ningxia, China
(see Wang et al., 2014, 2015). The long-term (1954–2004) mean temperature
is 8.1 ∘C, and the mean annual precipitation is 287 mm, most of
which falls from July to September (Jia et al., 2014). The radiation and
evaporation demands are high in this area. The annual incoming shortwave
radiation is 1.4 × 105 J cm-2 and the annual potential
evaporation is 2024 mm. The vegetation is dominated by scattered crowns of
Artemisia ordosica (Fig. 1b). The soil is highly alkaline (pH of 8.2). Biocrust (mainly
lichens and algae) covers about 40 % of interspace soil (Fig. 1c–e).
The thickness of the crust layer was 0.5–2.5 cm (Gong et al., 2016).
Site position (a), overlook of measured ecosystem (b), appearance
of soil surface at collar C1 (c), C2 (d) and C3 (e), and layout of
representative land unit (RLU, adopted from Gong et al., 2016) (f).
![]()
In the modelling, the structure of ecosystem was considered as replications
of representative land units (RLUs; Fig. 1f; Gong et al., 2016), which
comprise the area covered by shrubs and the surrounding soil (interspace).
Vertically, the model simulates the C flows across soil profile and the
water-energy transport from the lower boundary of rooting zone to a
reference height in the boundary atmosphere. Horizontally, the SPAC
processes at the plant-covered and interspace areas are differentiated but
related via advection and diffusion flows, as driven by the gradients of
temperature, water potential and gas concentration. The mineralization,
uptake and transport of soil C and nitrogen (N) are also regulated by water-energy
conditions.
Figure 2 shows the framework of key processes and variables included in the
FS modelling. The model includes a set of submodels which describe
(i) CO2 dissolution, transport and efflux; (ii) autotrophic and
heterotrophic CO2 production in the soil profile; (iii) CO2 uptake
and emission by biocrust; (iv) surface energy balance and the soil temperature
profile; and (v) soil hydrology and water balance. These submodels are
linked by multiple feedbacks to represent the coupling of C, water, vapour
and energy transportation in the ecosystem. Submodels (iv)–(v) have been
developed and described in detail in our previous work (Gong et al., 2016),
which focused on (i) introducing the plant–interspace heterogeneity into
water-energy modelling and (ii) investigating the influences of such
heterogeneity on the ecosystem water-energy budgets for a dryland ecosystem.
Gong et al. (2016) also validated the model in regard to the diurnal to
seasonal dynamics of radiation balance, surface energy balance, soil
temperature and moisture content in the footprint area of a eddy-covariance
(EC) site (for details of measurement, see Jia et al., 2014). In this work, we
therefore focused on the development of submodels (i)–(iii) and the
model validation using FS data measured by multiple automatic
respiration chambers from crust-covered and non-crusted soils. Based on the
validated model, a series of sensitivity analyses was carried out
addressing (i) how the soil CO2 components
regulate FS in the studied ecosystem and (ii) how the plant–interspace
heterogeneities influence the componential C processes and FS.
Modelling approaches
Submodel (i): CO<inline-formula><mml:math id="M53" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> transport, dissolution and efflux
Conceptual
framework of process-based modelling. Solid arrows
represent flows of masses and dashed arrows represent flows of information.
![]()
For soil fraction x (see Fig. 1f for RLU settings), CO2 exchange
(FS, upward positive) was the sum of CO2 uptake by biocrust
(FCt), photodegradation (FP) and the emission from soil under the
biocrust layer (FT):
FSx=FCtx+FTx+FPx,
where FCt is the net balance between biocrust photosynthesis
(PCt) and respiration (RCt), and FCt=PCt-RCt (see
Sect. 2.2.3). FT was modelled based on the mass-balance functions
developed by Fang and Moncrieff (1999), which combined major transport
processes in both gaseous and liquid phases. We expanded the original
function from one dimension to two dimensions. For the soil layer (x, i) and
time step t, the CO2 concentration and C flows were calculated as
follows:
∂Cx,i∂t=∂∂ZFdgv+Fagv+Fdwv+Fawv+∂∂hFdgh+Fagh+Fdwh+Fawh+Sx,i,
where superscripts v and h denote the vertical and horizontal directions,
respectively (see also in Gong et al., 2016); C is the total CO2
content; Fdg and Fdw are the CO2 flows due to
diffusion/dispersion via the gaseous and liquid phases; Fag and
Faw are the flows in gaseous and liquid phases due to gas convection and
water movement; and S is the net CO2 sink of the layer. The calculation
schemes of Fdg, Fdw, Fag and Faw have been described in
detail by Fang and Moncrieff (1999). FT is the total exchange of gaseous
CO2 between the surface and topmost layer:
FTx=Fdgvx,1+Fagvx,1+Ex,1SCwx,1,
where Ex,1S is the soil evaporation at section x (see Eq. 17) in
Gong et al. (2016);
Cw is the equivalent CO2 concentrations in the soil
solution. For layer (x, i), Cw is linked to the gaseous CO2
concentration (Cg):
Cx,i=Cgx,iVx,i-θx,i+Cwx,iθx,i,
where V is the total porosity, and θ is soil water content.
Cg and Cw were further related via the dissolution–dissociation
balance of CO2 in the soil solution, following Fang and Moncrieff (1999)
and Ma et al. (2013):
CO2g+H2Ol=H2Ol+CO2aqKH=PC/CO2aqCO2aq+H2Ol=H2CO3K0=CO2aq/H2CO3H2CO3=H++HCO3-K1=H+HCO3-/H2CO3HCO3-=H++CO32-K2=H+CO32-/HCO3-,
where PC is the partial pressure of CO2 in pore air; KH is
Henry's law constant; K0, K1 and K2 are the equilibrium
coefficients of dissolution, the first- and the second-order dissociation
reactions of carbonic acid, respectively (for details, see Fang and
Moncrieff, 1999). The equilibrium [H+] was determined by soil pH and
the coefficients KH, K0, K1 and K2, which were functions of
temperature in each soil layer (Fang and Moncrieff, 1999). Cw was calculated
as the sum of CO2aq, H2CO3, HCO3- and
CO32-.
Submodel (ii): autotrophic and heterotrophic CO<inline-formula><mml:math id="M121" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> production along
the soil profile
For soil layer (x, i), Sx,i (Eq. 2) was calculated as the sum of
autotrophic and heterotrophic CO2 production, and the dissolved
CO2 was removed with the water taken up by roots:
Sx,i=Rsx,i+Rax,i-Ex,iCwx,i,
where E is the transpiration
uptake of water (Gong et al., 2016); Rs is the
CO2 production by heterotrophic SOM decomposition; Ra is the autotrophic
respiration of the rhizosphere, which comprises maintenance respiration
(Rm) and growth respiration (Rg):
Rax,i=Rmx,i+Rgx,i.
To simulate Rs, we simplified the pool-type model of Gong et al. (2013, 2014),
which originated from Smith et al. (2010) for simulating coupled C and N
cycling in organic soils. The SOM pool in each soil layer was divided into
debris (Mdeb, i.e. litter from roots and biocrust), microbes
(Mmic) and humus (Mhum), which are different in biochemical
recalcitrance and N content. During decay, mineralized masses transfer
from Mdeb and Mmic to a more resistant form (i.e. Mhum), leading to
a decrease in labiality (e.g. Li et al., 1992). The mineralization of
organic C followed first-order kinetics and was constrained by multiple
environmental multipliers, including temperature, water content and oxygen
content (Šimunek and Suarez, 1993; Fang and Moncrieff, 1999):
mx,ir=Mx,irkrfTsx,ifθx,ifOx,idt,
where superscript r denotes the type of SOM pool (r=1 for Mdeb, r=2 for
Mmic and r=3 for Mhum); m is mineralized SOM during time step
dt; k is the decomposition constant; dt is the time step; f(Tsx,i), f(θx,i) and f(Ox,i) are multiplier terms regarding the temperature,
water content and oxygen restrictions, respectively. f(Ox,i) was
calculated following Šimunek and Suarez (1993). f(Tsx,i) and
f(θx,i) were reparameterized with respect to the site-specific
conditions of plants and soil (see Sect. 2.4.3). The CO2 production
from mineralization was further regulated by the N starvation of microbes
following Smith et al. (2010):
Rsx,i=rEmx,ir,
where rE is the gas production rate (rEϵ[0,1]), and
(1-rE) is the proportion of organic matter passed to downstream SOM
pools. The evolution of each SOM pool was calculated as below:
Mx,ir=1-rEmx,ir-1-mx,ir+Ax,irdt,
where A is the SOM input rate (A=0 for Mmic and Mhum); superscript
r-1 denotes the source SOM pools.
Rmx,i was calculated in a way similar to Rsx,i (e.g. Chen et al.,
1999; Fang and Moncrieff, 1999). Rgx,i was calculated as a fraction of
photosynthetic assimilates, following Chen et al. (1999):
Rmx,i=Mx,iRkRfTsx,ifθx,ifOx,idtRgx,i=kgPgfrx,i,
where MR is the root biomass; kR is the specific respiration rate of
roots; kg is the fraction of photosynthetic assimilation consumed by
growth respiration; frx,i is the mass fraction of roots in the soil layer
(x, i). Pg is the photosynthesis rate of plants. Pg was estimated
using a modified Farquhar's leaf biochemical model (see Chen et al., 1999).
This model simulates the photosynthesis based on biochemical parameters
(i.e. the maximum carboxylation velocity, Vmax and maximum rate of
electron transport, Jmax), foliage temperature (Tc) and stomatal
conductance (gs). The values of Vmax and Jmax were obtained from in
situ measurements from the site (Jia et al., unpublished). Tc and gs were given
in the energy balance submodel, which was detailed in Gong et al. (2016).
N content bonded in SOM is mineralized and released to soil layers
simultaneously with decay. The abundance of mineral N (i.e.
NH4+ and NO3-) regulates the growth of microbial biomass
and rE following Smith et al. (2010) and Gong et al. (2014). Key
processes governing the dynamics of mineral N pools included
nitrification–denitrification (Smith et al., 2010), solvent transport with
water flows (Gong et al., 2014) and the N uptake by the root system. However, the
plant growth was not modelled in this work. Instead, Nupt was calculated
using the steady-state model of Yanai (1994), based on the transpiration
rate, surface area of fine roots and the diffusion of solvents from pore
space to root surface:
Nupt=2πroLαCodt,
where ro is the fine root diameter; L is the root length and 2πroL is the surface area of fine roots; α is the nutrient
absorbing power, which denotes the saturation degree of solute uptake system
(α∈0,1); Co is the concentration of solvents
at the root surface and is a function of bulk concentration of mineral N
(Nmin), inward radial velocity of water at root surface (vo=EE2πroL2πroL) and saturation
absorbing power α. Further details for calculating α and
Co can be found in the work of Yanai (1994).
Submodel (iii): CO<inline-formula><mml:math id="M209" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> exchange of biocrust and
photodegradation
Biocrusts are vertically layered systems that comprise the top crust (or
bio-rich layer) and underlying subcrust (or bio-poor layer), which are
different in microstructure, microbial communities and C functioning
(Garcia-Pichel et al., 2016; Raanan et al., 2016). Top crust is usually
a few millimetres thick, which allows the penetration of light and the
development of photosynthetic microbes (Garcia-Pichel et al., 2016). On the
other hand, the subcrust has little photosynthetic activity. Here, we focused
mainly on describing the C exchanges in the top crust but assumed the C
processes in the subcrust were similar to those in the underneath soil. We
developed the following functions to describe the C fixation and mass
balance in the top crust:
FCt=PCt-RCt,
where PCt is the bulk photosynthesis rate, and RCt is the bulk
respiration rate. PC and RC were further modelled as follows:
PCt=αCAPARPCmαCAPAR+PCmRCt=MCtkcrfRCTCtfRCθCt,
where αC is the apparent quantum yield; PCm is the maximal
rate of photosynthesis and was a function of moisture content (θCt) and temperature (TCt) in the top crust; APAR is the
photosynthetically active radiation (PAR); MCt is the total C in the SOM
of top crust; kcr is the respiration coefficient; f(θCt) and
f(TCt) are water and temperature multipliers. Here, we assumed zero
photosynthesis rate for subcrust. The heterotrophic respiration (RCs)
was calculated following Eq. (11), based on the C storages (Mx,1r),
temperature and moisture content of the crust layer (i.e. Tsx,1 and θx,1; see Eqs. 29 and 14 in Gong et al., 2016).
To consider different C losses and exchanges, and to calculate the C balance
in top crust and subcrust, respectively, we considered the following matters.
RCt includes the respiration from both autotrophic (MCA) and
heterotrophic (MCH) pools. When autotrophic organisms die, SOMs pass
from MCA to MCH and influence the turnover processes. A variety of
top crust organisms can reach into subcrust (e.g. through rhizines; Aguilar
et al., 2009) and export litter there. When the surface is gradually
covered by deposits, top crust organisms tend to move upward and recolonize
at the new surface (e.g. Garcia-Pichel and Pringault, 2001; Jia et al.,
2008), leaving old materials buried into the subcrust (Felde et al., 2014).
On the other hand, the debris left to soil surface is exposed to
photodegradation. Based on the above information, the C balance in top crust and subcrust was
calculated as follows, assuming the partitioning of respiration between
autotrophic and heterotrophic pools was proportional to their fractions:
MCt=MCA+MCHdMCAdt=PCt-RCtMCAMCt-kmMCA-kbMCAdMCHdt=kmMCA-RCtMCHMCt-kbMCH-FPdMCsdt=kbMCt-RCs,
where km is the rate of C transfer (e.g. mortality) from autotrophic pool
to heterotrophic pool; kb is the rate of C transfer (e.g. burying) from
top crust to subcrust; FP is the loss of SOM due to photodegradation.
Photodegradation tends to decrease surface litter masses in a near-linear
fashion with the time of exposure (Austin and Vivanco, 2006; Vanderbilt et
al., 2008). Considering the diurnal and seasonal variations of radiation,
FP was calculated as a function of surface SOM mass and solar
radiation:
FPx=MsurfkpRadx,
where Radx is the incident shortwave radiation at surface x (Gong et al.,
2016); Msurf is the surface litter mass; and kp is the
photodegradation coefficient.
Micrometeorological and soil CO<inline-formula><mml:math id="M244" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> efflux measurements
Meteorological variables were measured every 10 s and aggregated to
half-hourly resolution during 2013–2014. The factors measured included the
incoming and outgoing irradiance (PAR-LITE, Kipp and Zonen, the
Netherlands), PAR (PAR-LITE, Kipp and Zonen, the Netherlands), air
temperature and relative humidity (HMP155A, Vaisala, Finland). Rainfall was
measured with a tipping-bucket rain gauge (TE525WS, Campbell Scientific
Inc., USA) mounted at a nearby site (1 km away; see B. Wang et al., 2014). The
seasonal trends of the measured Ta and P can be found in Jia et al. (2016). No
surface runoffs were observed at the site, indicating the horizontal
redistribution of rainfall was mainly through subsurface flows.
Continuous measurements of FS were conducted using an automated soil
respiration system (model LI-8100A fitted with a LI-8150 multiplexer,
LI-COR, Nebraska, USA). The system was on a fixed sand dune of typical size
(B. Wang et al., 2014), which was located about 1.5 km south of the EC tower
described in Gong et al. (2016). Three collars (20.3 cm in diameter and 10 cm
in height, of which 7 cm were inserted into the soil) were installed, on
average, at 3 m spacing in March 2012. One collar (C1; see Fig. 1c) was set on a
bare-soil microsite with no presence of biocrust. Two other chambers (C2,
see Fig. 1d; C3, see Fig. 1e) were set on lichen-crusted soils. FS
was measured hourly from C1 and C2 by opaque chambers, whereas it was measured by
transparent chamber from C3 to include the photosynthesis and
photodegradation. Litter from the shrub canopies was cleared from the
collars during weekly maintenance. Hourly Ts and θ at 10 cm depth were
measured outside each chamber using the 8150–203 soil temperature sensor
and ECH2O soil moisture sensor (LI-COR, Nebraska, USA), respectively. Root
biomass was sampled near each collar (within 0.5 m) in July 2012, using a
soil corer (5 cm in diameter) to a depth of 25 cm. The samples were mixed
and sieved sequentially through 1, 0.5 and 0.25 mm meshes, and the living
roots were picked by hand. The comparison of the three microsites is shown
in Table 1. Methods used in data processing and quality control have been
described earlier in detail (see Wang et al., 2014, 2015). The quality
control led to gaps of 10–13 % in the FS dataset.
Configuration of soil collars used in this study.
Collar
C1
C2
C3
Surface type
Non-crusted
Lichen-crusted
Lichen-crusted
Chamber type
Opaque
Opaque
Transparent
Root biomass (g m-3)
420
106
92
Gap of data (%)
12.9
10.5
9.85
Annual C efflux (gC m-2)*
259
194
192
* The values were calculated from the measured hourly FS data excluding
data gaps.
Model setups
Parameterization of vegetation and soil texture
The parameterization schemes supporting the simulations of energy balance
and soil hydrology in submodels (i)–(v) have been described previously in
detail by Gong et al. (2016). As the water-energy budget is sensitive to
vegetation (i.e. canopy size, density and leaf area) and soil hydraulic
properties (see Gong et al., 2016), we hereby re-estimated these parameters
for the FS site. Measurements based on four 5 m × 5 m plots
showed that the crown diameter D (86 ± 40 cm) and height H (47 ± 20 cm)
at this site were similar to those measured from the eddy-covariance
(EC) footprint by Gong et al. (2016). However, the shrub density was 50 %
greater, leading to higher shrub coverage (42 %), shorter spacing distance
L (40.2 cm) and greater foliage area. On the other hand, the subsoil at the
FS site is sandy and much coarser than that at the EC footprint.
Therefore, we collected 12 soil cores from 10 cm depth and measured
saturated water content (θsat), bulk density and residual water
content (θr) from each sample. Then, the samples were
saturated, and covered and drained by gravity. We measured the water content
after 2 and 24 h draining, which roughly represented the matrix
capillary water content (10 kPa) and field capacity (33 kPa) (Armer, 2011).
The shape parameters n and αh (see Eq. 26 in Gong et al., 2016)
for the water-retention function were estimated from these values (Table 2).
Parameters for soil water retention and C turnover.
Parameter
Equation
Unit
Value
αh
–a
–
0.0355b
n
–a
–
1.5215b
k1
(11)
g g-1 day-1
0.01c
k2
(11)
g g-1 day-1
0.08d
k3
(11)
g g-1 day-1
0.001d
kg
(15)
g g-1
0.15e
kcr
(19)
g g-1 s-1
0.0014f
kp
(24)
g g-1 yr-1
0.23g
kR0
(25)
g g-1 day-1
0.002e
a
(26)
–
0.1h
b
(26)
–
24h
c
(26)
–
0.89h
QCt
(32)
–
1.585f
aRC
(32)
–
-0.0525f
bRC
(32)
–
2.602f
cRC
(32)
–
-1.653f
aPt
(33)
–
0.9837f
bPt
(33)
–
-0.1385f
cPt
(33)
–
0.0095f
dPt
(33)
–
-1.6318E-4f
aPw
(33)
–
-0.3501f
bPw
(33)
–
5.5884f
cPw
(33)
–
-7.1783f
dPw
(33)
–
2.6837f
a See Eq. (26) in Gong et al. (2016). Sources of parameter values:
b This study; see Sect. 2.3.2. c Lai et al. (2016). d Gong
et al. (2014). e Chen et al. (1999). f This study; see Sect. 2.4.4 and Fig. 3.
g B. Wang et al. (2014).
Parameterization of soil C and N pools
The sizes and quality of soil C pools were parameterized based on a set of
previous studies. The total soil organic carbon (SOC)
in the root-zone soil (i.e. 60 cm depth, bulk
density of 1.6 g cm-3) was set to 1200 g m-2, based on the values
reported from previous studies in Yanchi area (e.g. Qi et al., 2002; Chen
and Duan, 2009; Zhang and Hou, 2012; Liu et al., 2015; Lai et al., 2016).
The mass fraction of the resistant SOM pool (Mhum) was set to 40–50 %
of the total SOM, following work of Lai et al. (2016). The vertical distribution
of the SOM pools was described following Shi et al. (2013). At the ecosystem
level, the aboveground biomass was linearly related to the crown projection
area (MS=0.2917×π(0.5D)2; see Zhang et al., 2008). The
total root biomass was then calculated as proportional to the aboveground
biomass using a root–shoot ratio of 0.47 (MR=0.47MS; Xiao et
al., 2005). The vertical profile of root biomass was parameterized as
decreasing exponentially with depth, using the depth profile reported by Lai
et al. (2016). In the horizontal direction, root biomass was set to decrease
linearly with the distance from the centre of a shrub crown (Zhang et al.,
2008). The N content was parameterized following the measurement of Wang et
al. (2015).
Based on the above settings, the specific decomposition rate of debris was
estimated from the litterbag experiment done by Lai et al. (2016), which
showed a 16 % decrease in the mass of fine-root litter during a 7-month
period in 2013 at the Yanchi site. The photodegradation coefficient
(kp) was set to 0.23 yr-1, which was the mass-loss rate reported by
Austin and Vivanco (2006). Msurf was set to 33 % of MCH in top crust,
assuming the depth of light penetration was about 2 mm and C concentration
was homogeneous in top crust. The surface litter from canopy was not
considered in this modelling, as the plant litter was cleaned from the
collars during weekly maintenance. The specific respiration rate of roots
(kR), however, could be much greater during vegetative growing stage
than other periods, e.g. at the defoliation stage (Fu et al., 2002; Wang et
al., 2015). Here, we linked kR to the development of foliage in modelling
using the approach of Curiel Yuste et al. (2004):
kR=kR01+nRLlnRLlLmaxLmax,
where kR0 is the “base” respiration rate (Table 2); Ll is the
green leaf area, which is a function of the Julian day (Gong et al., 2016);
Lmax is the maximum Ll; nR is the maximum percentage of
variability and is set to 100 %.
Parameterization of soil CO<inline-formula><mml:math id="M363" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> production
Based on the empirical study of B. Wang et al. (2014), the steady-state
sensitivity of CO2 production to soil temperature and water content
(i.e. fTsfθ; Eq. 11) can be
described as a logistic-power function:
fTsfθ=fTs,θ=1+expab-Ts-1θθθsatθsatc,
where a, b and c are empirical parameters. This function represents the
long-term water-thermal sensitivity of CO2 production over the growing
seasonal, yielding an apparent temperature sensitivity Q10 of 1.5 for
the emitted CO2 (B. Wang et al., 2014). However, this could underestimate
the short-term sensitivities of CO2 production. The apparent Q10
could be much greater at the diurnal level than at the seasonal level (B. Wang
et al., 2014). In this work, we firstly calculated the base sensitivity
using the long-term scheme (Eq. 26) with a 1-day moving average of
water-thermal conditions. Then, the deviation of hourly sensitivity from
the base condition was adjusted by the short-term Q10:
fTsfθ=fTsshort,θshort+fTs,θ-fTsshort,θshortQ10Ts-TsshortTs-Tsshort1010Q10=maxQ10Tsshort,Q10θshortQ10Tsshort=-0.42Tsshort+12.4Q10θshort=18010θshort3.721+1.604,
where Tsshort and θshort are the 1-day moving averages of
Ts and θ, respectively; Q10 (Ts) and Q10 (θ) are the
adjustment functions for short-term apparent Q10, regarding the
short-term Ts and θ.
Further non-linearity of soil respiration responses refers to the rain-pulse
effect (or the “Birch effect”; Jarvis et al., 2007) that respiration
pulses triggered by rewetting can be orders of magnitude greater than the
value before the rain event (Xu et al., 2004; Sponseller, 2007). Such a response
could be very rapid (e.g. within 1 h to 1 day, Rey et al., 2005) and
sensitive to even minor rainfall. It also seems that the size and duration
of a respiration pulse not only depend on the precipitation size but also
on the moisture conditions prior to the rainfall (Xu et al., 2004; Rey et
al., 2005; Evans and Wallenstein, 2011). As numerical descriptions of such
an effect remain unavailable at the moment, we simply multiplied Eq. (26) to
a rain-pulse coefficient (fpulse):
fpulse=max1,θθθ72 hθ72 hnp,
where θ72 h is the
3-day moving average of soil moisture content; np is a shape
parameter and set to 2 in this study. θ72 h is the 72 h
moving average of θ. For tests on model sensitivities to different
parameterizations of fpulse, see Sect. 2.5.3.
Parameterization of biocrust photosynthesis and respiration
In submodel (iii), Eqs. (17)–(19) were parameterized based on the
experiment of Feng et al. (2014). In the experiment, 50 lichen (top crust)
samples of 0.5–0.7 cm thickness (100 % coverage; average C content of 1048 µmol C cm-3)
were collected from a 20 m × 20 m area. The
samples were wetted and incubated under controlled TCt (i.e. 35, 27, 20, 15 and
10 ∘C). These samples were divided into two groups to measure the
net primary productivity (NPP) and dark respiration (Rd) separately. Gas
exchanges and light response curve for each crust sample were measured using
an LI-6400 infrared gas analyser equipped with an LI-6400-17 chamber and an
LI-6400-18 light source (LI-COR, Nebraska, USA). Measurements were taken
at ambient CO2 values of 385 ± 35 ppm. Saturated top crust samples
were placed in a round tray and moved to the chamber. CO2 exchange was
measured during the drying of samples until the CO2 flux diminished.
During drying, θCt was measured every 20 min. For more
details, see Feng et al. (2014).
Measured and fitted bulk respiration (a) and photosynthesis (b) of
the lichen top crust as functions of temperature and water content.
![]()
Fitting measured Rd to TCt and θCt (see Fig. 3a) was based
on the Matlab® (2012a) curve-fitting tool. The
obtained multipliers in Eq. (19) are as follows:
fRCTCtfRCθCt=QCtTCt-2010aRC+bRCθCt+cRCθCt2,
where QCt, aRC, bRC and cRC are the fitted shape parameters
(Table 2).
The parameterized Eq. (19) was then used to simulate the Rd for the NPP
samples, based on the correspondent TCt and θCt from each
measurement. PCm was determined by subtracting the simulated respiration
rate from the NPP measured under light-saturated conditions. Then, PCm
was fitted to TCt and θCt in the Matlab® (2012a) curve-fitting tool using the following equations (Fig. 3b):
PCm=fPtTCtfPwθCt=aPt+bPtTCt+cPtTCt2+dPtTCt3×-aPw+bPwθCt-cPwθCt2+dPwTCt3,
where aPt, bPt, cPt, dPt, aPw, bPw,
cPw and dPw are fitted shape parameters (Table 2).
It should be addressed that TCt and θCt could change more
rapidly than the mean conditions of the crust (i.e. Tsx,1 and θx,1). In this work, TCt was calculated from the surface
temperature (Tx; see Eq. 13 in Gong et al., 2016) and Tsx,1 by
linear interpolation. The calculation of θCt, on the other
hand, depended on the drying–rewetting cycle. During drying phases,
θCt was interpolated linearly from θx,1 and
surface moisture content (θx), whereas during wetting phases
the mass balance of water input P and evaporation loss (Ex,1s; see
Eq. 17 in Gong et al., 2016) was considered:
TCt=TxZCt+Tsx,1Zsx,1ZCt+Zsx,1θCt=max()θxZCt+θx,1Zsx,1ZCt+Zsx,1,θCt+P-Ex,1sZCt,
where Zsx,1 is the thickness of the biocrust, and ZCt is the
thickness of the top crust. θx was calculated from the surface
humidity and the water retention of the crust layer using Eqs. (25)–(26)
by Gong et al. (2016).
Calculation of litter input to soil and SOC transport in
biocrust
The litterfall added to each soil layer (Ax,i1; Eq. 13) was linked
to the mortality of roots, which was calculated following Asaeda and
Karunaratne (2000).
Ax,i1=kmoQmoTsx,i-20Mx,iR,
where kmo is the optimal mortality rate at 20 ∘C;
Qmo is the temperature sensitivity parameter (Asaeda and Karunaratne,
2000). Similarly, we attributed the C transport rate (ACm) from
MCA to MCH mainly to the mortality of autotrophic organisms. We
assumed that most mortality of crust organisms occurred during abrupt
changes in wetness, as microbial communities may adapt slow moisture changes
or remain inactive during drought (e.g. Reed
et al., 2012; Garcia-Pichel et al., 2013). Here, we introduced a water-content multiplier, fm(θCt), to describe the impact of abrupt θCt changes on
km:
ACm=kmcQmoTCt-20fmθCtMCAfmθCt=max0.01,1-minθCt,θCt7θCt,θCt7maxmaxθCt,θCt7,
where kmc is the optimal mortality rate at 20 ∘C;
Qmo is the temperature sensitivity parameter (Asaeda and Karunaratne,
2000); θCt7 is the forward 7-day moving average of θCt.
C transport from top crust to subcrust was calculated as driven mainly by the
sand deposition and burying of top crust SOM. Assuming the C content in
top crust was homogeneous and the thickness ZCt was near constant, the
transport rate (kb) was then proportional to the sand deposition rate:
kb=ksandρbulk1ZCt,
where ρbulk is the bulk density of soil; ksand is the sand
deposition rate in Yanchi area, which is a function of wind velocity (Li and
Shirato, 2003).
Model validation and sensitivity analyses
Simulation setups
In the model simulations, soil depth was set to 67.5 cm to cover the rooting
zone (Gong et al., 2016), including the crust layer (2.5 cm) and sandy
subsoil (65 cm, stratified into 5 cm layers). Water contents measured at 70 cm depth was used as the lower boundary conditions for hydrological
simulations (Jia et al., 2014). The calculation of soil temperature extended
to 170 cm depth with the no-flow boundary, in regard to the probably strong heat
exchange at the lower boundary of rooting zone (Gong et al., 2016).
The zero-flow condition was set for the lower boundary of CO2 and O2
gases, whereas dissolved CO2 was able to leach with seepage water.
Based on presumed similarity of RLU structures, we assumed no-flux
conditions for transport of water, heat, solvents and gases at the outer
boundary. In the simulation, we assumed instant gas transport via top crust,
whereas considered the CO2 released by subcrust (RCs) was subject
to the dissolving-transport processes. In this work, we aggregated the C
processes in subcrust with those in the soil profile. The initial ratio of MCA : MCH was set to 2 : 3. The C concentration of organic matter was set to
50 %.
The model was run with half-hourly meteorological variables including the
incoming shortwave radiation, incoming longwave radiation, PAR, Ta, relative
humidity, wind speed and precipitation. Initial temperatures and soil
moisture contents for each soil layer were initialized following the work of
Gong et al. (2016). Surface CO2 concentration was set to 400 ppm. The
initial gaseous CO2 concentration was set to increase linearly with
depth (5 ppm cm-1). The initial CO2 concentration in liquid form
was then calculated based on Eqs. (4)–(8). The initial content of
mineral N content was set to 40 mg g-1, which was within the range of the
field observations. The two-dimensional transpiration of water, energy and
gases along the soil profile was solved numerically using the
predict–evaluate–correct–evaluate (PECE) method (Butcher, 2003). In order
to avoid undesired numerical oscillations, the transport of water, energy
and gases was calculated at 5 min substeps.
Model validation
First, we validated the modelling of soil temperature and moisture content
for the FS site (Test 0). The simulated hourly soil temperature and
moisture content at 10 cm depth were compared to the measured values for
each collar. The validation was based on the same meteorological data as
used by Gong et al. (2016), who validated the model in regard to the diurnal
to seasonal dynamics of radiation balance, surface energy balance, soil
temperature and moisture content at the EC site.
The validity of the modelled FS was then examined in three separate
tests. In Test 1, modelled FS was validated for non-crusted soils. In
this case, FT in Eq. (1) was the only term affecting FS
(FB=0 and FP=0), and the crust influences on C–water
exchanges were excluded. The biocrust-related processes were considered in
Test 2 and Test 3. Test 2 considered the dark respiration of biocrust
(RCt) and set FB=RCt and FP=0. Test 3 considered
all the flux components (FT, FP and FB). In these tests,
different values of root biomass were assigned to the model, regarding the
different collar conditions (Table 1). In Tests 1–3, half-hourly
FS was simulated and averaged to hourly, and compared to those measured
from the collars C1–C3, respectively. Linear regressions were used to
compare the modelled and measured values. The biases (ζ) of the
simulated values were calculated by subtracting the measured values from the
modelled ones. Gap values in the measurements were omitted in the validation
and the bias analyses.
Simulating componential CO<inline-formula><mml:math id="M497" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> fluxes and their parameter
sensitivities
Simulated component CO2 fluxes (gC m-2 yr-1) for
areas with plant cover and without (interspace).
Surface type
FS
FT
Rs+Raa
Ra
PCt
FCt
FP
FCt-FPb
Interspace
244
249
295
113
54.6
31.1
26.1
5.0
Plant covered
214
218
263
108
36.3
18.2
14.6
3.6
a Rs + Ra represents the total CO2 production from soil respiration.
Ra
is the total autotrophic respiration (Ra = ∑iRai; see
Eq. 10)
and Rs is the total heterotrophic respiration (Rs =∑i Rs
i; see Eq. 12). b FCt-FP represents the net CO2
exchanges of top crust; see Eqs. (17) and (24) for correspondent
algorithms of the variables. For definitions of other fluxes, see Eq. (1)
for FS, Eq. (3) for FT, Eq. (17) for FCt, Eq. (18) for
PCt and Eq. (24) for FP.
Using the validated model, we simulated the temporal trends of C flux
components (i.e. PCt, RCt, FP, FT, Ra and Rs) in Test 4, in
order to find out how the different flux components may have contributed to
the total efflux (Table 3). The simulation used the same model setups and
climatic variables as Test 3. It should be noted that although the model
was built as an abstract for ecosystem-level processes, the simulation
setups and validation were performed at a point level corresponding to
respiration chambers. Therefore, understanding the uncertainty sourced from
parameterization could be helpful for future development and applications.
In Gong et al. (2016), we have studied the sensitivities of modelled soil
temperature and moisture content to the variations in soil texture, water
retention properties, vegetation parameters and plant–interspace
heterogeneities. In this study, we also tested the sensitivity of FS and
componential fluxes to the changes in a number of site-specific parameters
(Table 4). These parameters included pH, nitrogen content, water-thermal
conditions, root biomass, production rates and decomposition rates of
litter, which are often key factors for regulating the soil C processes but
likely to vary within and among ecosystems (see, e.g. Ma et al., 2011; Gong
et al., 2016; Wang et al., 2015). Furthermore, we tested the model
sensitivities to several newly defined parameters (i.e. nR, np and
fm) in order to understand their effects on model uncertainties. FS and
componential fluxes at interspace were simulated by varying the single parameter
value by 10 or 20 %. The sensitivity of each tested flux was described
by the difference (dF) in the annual flux rate simulated using manipulated
parameters, as compared to the rate simulated under no-change conditions.
Sensitivity of simulated FS and its componential fluxes to
manipulations of parameter values.
Change of parameter
FSa
FT
Ra +RS
Ra
PCt
FCt
FP
FCt-FP
nR+20 %
+3.3b
+3.2
+2.7
+7.9
/c
/
/
/
nR-20 %
-2.9
-2.8
-3.4
-8.8
/
/
/
/
nPd+20 %
+1.6
+1.6
+1.0
/
/
/
/
/
nP-20 %
/
/
-1.4
/
/
/
/
/
fm+20 %
/
/
/
/
+2.9
+3.8
+3.4
+6.0
fm-20 %
/
/
/
/
+1.2
/
-5.7
+30
Ts+2 ∘C
+9.5
+9.6
+7.1
+11
+4.9
+3.9
+1.5
+16
Ts-2 ∘C
-9.0
-9.2
-8.1
-11
-1.3
-2.9
/
-20
θ+10 %
+3.6
+5.6
+7.5
+14
+41
+28
+14
+102
θ-10 %
-5.0
-5.6
-8.1
-14
-16
-13
-8.4
-34
Mtot+10 %
+2.9
+2.8
+2.0
/
/
/
/
/
Mtot-10 %
-2.5
-2.4
-3.1
/
/
/
/
/
MR+10 %
+7.0
+6.8
+6.8
+8.8
/
/
/
/
MR-10 %
-7.0
-6.8
-7.1
-8.9
/
/
/
/
Ntot+10 %
/
/
/
/
/
/
/
/
Ntot-10 %
/
/
/
/
/
/
/
/
k1+10 %
+2.9
+2.8
+2.4
/
/
/
/
/
k1-10 %
-2.5
-2.4
-3.1
/
/
/
/
/
kmo+10 %
+4.1
+4.0
+3.4
/
/
/
/
/
kmo-10 %
-3.3
-3.2
-3.7
/
/
/
/
/
kmc+10 %
/
/
/
/
/
/
+1.5
-8.0
kmc-10 %
/
/
/
/
/
/
-2.3
+8.0
MCt+10 %
/
/
/
/
/
/
/
/
MCt-10 %
/
/
/
/
/
/
/
/
MCA : MCH+10 %
/
/
/
/
/
/
/
/
MCA : MCH-10 %
/
/
/
/
/
/
/
/
pH + 5 %
-8.6
-8.4
/
/
/
/
/
/
pH -5 %
+7.0
+6.8
/
/
/
/
/
/
a Definitions of fluxes; see Table 3 and Sect. 2.5.3. b Value
represents the percentage (%) of change (dF) in correspondent C flux with
manipulated parameter value, as compared to the no-change condition. A
positive value represents the percentage of increase in the simulated flux,
whereas a negative value represents the percentage of decrease. c The
change in simulated C flux was smaller than 1 %.
Plant–interspace differences in the sensitivities of C fluxes to
changes in soil temperature (Ts), water content (θ) and root biomass
(MR).
Change of parameter
FSa
FT
Ra +RS
Ra
PCt
FCt
FP
Ts +2 ∘C
+0.38b(+4.0)
+0.45(+4.7)
+0.87(+12)
+0.49(+4.6)
+1.7(+34)
+2.7(+70)
+1.2(+78)
Ts-2 ∘C
-1.6 (+18)
-1.3 (+14)
-0.61 (+7.4)
/c
-2.8 (+222)
-0.40 (+14)
-3.1 (+615)
θ+10 %
-1.8 (-49)
-1.5 (-27)
/
-0.27 (-1.9)
+8.7(+21)
+13(+46)
+2.9(+20)
θ-10 %
+0.84(-17)
+0.58(-10)
/
+1.5(-10)
-3.5 (+22)
-4.5 (+36)
+0.9(-11)
MR+10 %
-0.44 (-6.3)
-0.41 (-6.0)
-0.31 (-4.7)
-0.52 (-5.8)
/
/
MR-10 %
+0.44(-6.3)
+0.41(-6.0)
+0.65(-9.2)
+5.2(-5.8)
/
/
a Definitions of fluxes; see Table 3 and Sect. 2.5.3. b Values
shows the plant–interspace difference in parameter sensitivities by value
(outside bracket, dFp-dFi) and by percentage (inside bracket, 100 × (|dFp|-|dFi|)/|dFi|),
where dFp and dFi are parameter sensitivities (dF; for
definitions, see Table 4 and Sect. 2.5.3 ) for plant-covered and interspace areas,
respectively. A positive percentage (inside bracket) indicates a greater
sensitivity (|dF|) of the flux at the plant cover than at interspace,
whereas a negative value indicates a lower sensitivity. For definitions of
fluxes and sensitivities, see Table 3 and Sect. 2.5.3. c The
difference in sensitivity is smaller than 0.1 % by value.
Comparing model sensitivities between plant cover and
interspace
In order to study the effects of plant–interspace heterogeneity on soil
CO2 efflux, Test 5 simulated annual FS and componential fluxes at
the plant cover and compared the values to interspace. The simulation setups
were almost the same as those employed in Tests 1–3; the only exception was that the same
initial values of SOC storages (650 gC m-2) and root biomass (200 g m-2)
were used for under-canopy and interspace areas for comparison purposes.
Based on Test 4, we further compared the plant–interspace differences in the
carbon flux (C-flux) sensitivities to most important site-specific parameters, i.e. soil
temperature (Ts), water content (θ) and root biomass (MR) (see
Sect. 3.2). The differences in parameter sensitivities were calculated by
comparing the absolute values of sensitivities (|dF|; see
Sect. 2.5.3 and Table 4) from the area with plant cover to that without
(interspace).
Results
Model validity
Measured and modelled soil temperature (a) and soil moisture
content (b) at 10 cm depth for the FS site and as compared to the EC site
in the year 2013 by Gong et al. (2016).
![]()
Compared to the EC site in previous work (Gong et al., 2016), the soil in
this study was much coarser and the measured θ at 10 cm depth was
constantly lower (Fig. 4), indicating the necessity of reparameterization
and validation of the water-energy algorithms. Figure 4a shows the modelled
hourly Ts and θ at 10 cm depth with the mean values measured from the
FS site during 2013. Based on the site-specific vegetation and
soil texture parameters, our model explained 97 % of the variations in the
measured hourly Ts. The model underestimated the temperature mainly in summertime
(i.e. days 150–250; Fig. 4a). The underestimation was more pronounced
around the noontime in the diurnal cycle. As the water-content sensors may
not accurately capture the moisture dynamics during the freezing period, only
the simulations during the ice-free period were compared to measured data
(Fig. 4b). During the ice-free period, the model explained 83 % of the
variations in the measured mean water contents at 10 cm depth. The biases in
the modelled temperature and moisture content were less than the spatial
variations observed in this area (e.g. Wang et al., 2015). Therefore, our
model could be able to reproduce the time series for the measured
water-energy fluxes at the site.
Our model explained 87 and 83 % of the variations in the hourly
FS measured on the non-crusted surface in the years 2013 and 2014
(Fig. 5a). The root mean square errors (RMSEs) were 0.43 and
0.29 µmol m-2 s-1, respectively. The model mainly
underestimated the daytime FS during the freezing periods. During
the ice-free periods, the model mainly overestimated the efflux in early
spring. The biases in modelling largely showed a diurnal pattern (Fig. 5b):
FS was mainly underestimated in the afternoon hours (i.e. from
10:00 to 15:00 LT) but slightly overestimated in the afternoon and evening.
At the daily level, our model explained 94 % of the variations in measured
daily efflux during the 2-year period (Fig. 5c).
Measured and modelled hourly FS for non-crusted soil (a),
the temporal pattern of the bias of simulated hourly FS (b) and the
comparison of measured and modelled daily FS (c) during 2013–2014.
![]()
Compared to the non-crusted soil (C1), the simulated FS for crusted
surfaces (C2 and C3) showed greater deviations from measured data. At the
hourly scale, our model explained 75 % (year 2013) and 68 % (year
2014) of variations in measured FS from C2 (Fig. 6a), and 68 % (year
2013) and 61 % (year 2014) of variations in the FS measured from C3
(Fig. 6b). For the 2-year period, the RMSEs of the modelled hourly FS were
0.25 and 0.35 µmol m-2 s-1 for C2
and C3, respectively. The magnitudes of biases (|ζ|) were generally
greater during the rainfall period (i.e. from the start of rainfall to 24 h
after the end of rainfall) than the inter-rainfall period (Fig. 7). The
simulated FS for C2 showed a similar diurnal pattern of biases as compared
to C1, suggesting ineligible contributions from the biases in simulated
subsoil emissions. Introducing photosynthesis and photodegradation of
biocrust to the system (C3) led to greater overestimations in FS, and
these were more obvious in the afternoon hours (i.e. from 12:00 to 18:00 LT)
and during the wetting period. Nevertheless, at the daily scale, the model
explained 91 % (C2; Fig. 5c) and 86 % (C3; Fig. 5d) of the variations in
the measured FS during the 2-year period. There were no significant
systematic deviations between the measured and the modelled daily values, as
indicated by the regression slopes close to 1 and the intercepts close to 0
(Figs. 4 and 5).
Measured and modelled FS of lichen-crusted soils for opaque
(a, c) and transparent chambers (b, d) at hourly (a, b) and daily (c, d)
scales during 2013–2014.
![]()
The results above showed that the model was able to describe the seasonal
variations of FS for both non-crusted and lichen-crusted soils.
Moreover, the model captured the strong variability of hourly/daily
FS in wetting–drying cycles. Compared to earlier statistical modelling by
B. Wang et al. (2014) and W. Wang et al. (2014b), this model showed equal or improved accuracy. In
this sense, we assume that our model has included the main mechanisms
controlling the FS dynamics in the soil system and could be used for
further analysis of componential C processes and their parameter
sensitivities.
Modelled C flux components of <inline-formula><mml:math id="M815" display="inline"><mml:mrow><mml:msub><mml:mi>F</mml:mi><mml:mi mathvariant="normal">S</mml:mi></mml:msub></mml:mrow></mml:math></inline-formula>
Test 4 showed that Rs was the main contributor to the root-zone CO2
production, which accounted for a major source of effluxes (FS). Our
measurements showed large diurnal and seasonal variations in FS
regardless the existence of crust covers (Figs. 5 and 6). Particularly,
the FS dynamics depended strongly on rain events. Even at non-crusted
soil (i.e. C1), FS dropped significantly from the pre-rainfall level
even to near zero but rebounded rapidly and peaked after the rain stopped (Fig. 5a). This could relate to the mismatched trends of CO2 production (Rs +
Ra) and emission (FT) from the rooting zone with respect to the
wetting–drying cycles (Fig. 8a). Compared to CO2 production, the
responses of FT to rainfall were generally lagged and smoothed (see
examples in Fig. 8b–d), irrespective of the size of rain events. In the
simulation, soil rewetting increased CO2 production rapidly but
depressed FT, which increased after rain ceased. In all the examples
(Fig. 8b–d), FT exceeded RP within 48 h after the end of
rain events. At the annual level, the total RP was larger during wetting
period (i.e. rainfall days plus 1 day after rainfall) than during the rest of the days of
the year (i.e. the drying period), whereas the total FT was greater during
the drying period (Fig. 8e).
Diurnal patterns of biases (ζ) in the simulated hourly
FS for lichen-crusted soils using opaque (a) and transparent chambers
(b), and the cumulative probability of the biases during wetting and drying
periods (c) during 2013–2014. The wetting period included the rainy days
and a 1-day period after each rainfall. The drying period included the rest
of the years other than the wetting period.
![]()
On the annual basis, CO2 production (Rs+Ra) and emission (FT) from
root-zone soil were mismatched (Table 3), and the former was more than 15 % greater than the latter. Such a gap was mainly due to the root uptake and
transport of dissolved CO2 (i.e. 36 gC m-2 yr-1), whereas
the loss of dissolved CO2 via seepages or pore-mediated horizontal
flows was limited (i.e. 7.4 gC m-2 yr-1). The photosynthesis
rate of top crust was 31.1 gC m-2 yr-1 at interspace. After
rainfall, the C uptake by the top crust increased significantly and even turned the
soil from net C source to sink during a few hours to 1 day (Figs. 6, 8).
However, at the annual scale, the C losses via respiration and
photodegradation accounted for 90 % of the photosynthetic products,
leading to a near-zero contribution of top crust to FS during the 2-year period (i.e. < 5 gC m-2 yr-1).
Analysis of parameter sensitivity showed that the modelled FS and the
component fluxes were more sensitive to ±2 ∘C in
Ts or ±10 % in θ, as compared to the effects of ±10 or ±20 % in the other parameters (Table 4). Varying θ
by 10 % produced greater impacts on the simulated RP and
crust-related fluxes (i.e. PCt, RCt and FP), as compared to
changing Ts by ±2 ∘C. Increasing θ by 10 %
enhanced the simulated PCt and FCt by 41 and 28 %, and
doubled the net C sequestration (FCt–FP) by the top crust. However,
the contribution of such changes to annual FS was minor and accounted
for only 2.0 % of the total efflux. Besides Ts and θ, the simulated
efflux was also sensitive to changes in root biomass (MR). Manipulating
root biomass by ±10 % changed the annual FT and FS by
about 7 %, and such effects were 100 % greater than ±10 %
changes in Mtot in the soil. Adjustment of other parameters, e.g. np
(Eq. 31) and fm (Eq. 38), had little influence on the modelled
FS and the componential fluxes (Table 4). In addition, the model was
robust to the adjustment of several crust-related parameters, i.e.
kmc, MCt and MCA : MCH.
Hence, algorithms corresponding to those parameters could be simplified in future developments.
Modelled plant–interspace differences in C flux components
At either plant-covered or interspace area, Rs was a major contributor to
root-zone CO2 produced, and FT dominated the total effluxes
(Table 3). The C loss at interspace was 14 % faster than under canopy on an
annual basis if root biomass and SOC were homogeneous at plant cover and
interspace. The lower FS rate at plant cover is mainly attributed to the
lower CO2 production (RS+Ra) from subsoil. The C loss via
seepage and root transport, which is the gap between subsoil CO2
production and emission (FT), was slightly higher under canopy (17 %)
than at interspace (15 %). Compared to interspace, the photosynthesis of
biocrust (PCt) was 34 % lower under canopy. This reduced the
under-canopy FCt by 42 % in comparison with interspace. However, such
a difference was largely offset by the reduced photodegradation rate under
canopy, leading to limited plant–interspace differences in net sequestration
by the top crust (i.e. by 1.4 gC m-2 yr-1).
Simulated component CO2 exchanges by biocrust and root-zone
soil (a), the simulated CO2 fluxes before and after example
rain events of 2.3 mm (b), 7.6 mm (c) and 12.8 mm
(d) sizes, and the comparison of FT and
RP during wetting and drying periods during 2013–2014 (e). The wetting
period included the rainy days and a 1-day period after each rainfall. The
drying period included the rest of the years other than the wetting period.
![]()
We further compared the flux sensitivities at plant cover and interspace to
the changes in three most effective parameters (i.e. Ts, θ and
MR; see Table 5). For subsoil-mediated fluxes (i.e. FS, FT,
Ra, Rs), the sensitivity values differed by less than 2 % from
plant cover to interspace. On the other hand, the sensitivities of
crust-related fluxes (i.e. PCt, FCt, FP) showed greater
differences between plant cover and interspace. Compared to interspace,
FS and FT in plant-covered areas were more sensitive to Ts changes but
less sensitive to manipulations in θ. On the other hand, the plant
cover reduced the sensitivity of CO2 effluxes to changes in root
biomass. PCt, FCt and FP were generally more sensitive to
warming and θ manipulations at plant cover than at interspace, except
that plant cover decreased the sensitivity of FP to -10 % changes in
θ. Nevertheless, their contribution to the sensitivity of FS was
marginal due to the low flux rates of crusts.
Discussions
The roles of componential C processes in regulating soil CO<inline-formula><mml:math id="M928" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>
efflux
Our process-based model provided a tool to separate the multiple soil C
processes and investigate their roles in regulating FS dynamics in
dryland ecosystems. So far, efforts to quantify the soil C loss in
terrestrial ecosystems have considered soil C efflux as a synonym of
respired CO2. However, based on this work, precautions must be taken when
extrapolating the FS responses from the chamber to ecosystem scale and
from short-term to long-term periods. Processes other than autotrophic and
heterotrophic respiration could significantly modify the FS responses to
climatic variability. Our simulation highlighted decoupled CO2
production and emission during the wetting–drying cycle, as regulated by the
CO2 transport in the soil profile. The simulated CO2 production in
the soil profile were much greater than effluxes during rain pulses (e.g. Fig. 7).
This indicated that the low FS during rewetting was mainly due to
the increase in CO2 dissolution, instead of the reduced respiration
rates by low O2 supply (e.g. Fang and Moncrieff, 1999). This finding is
further supported by the measurement of Maier et al. (2011), which showed
that 40 % of the respired CO2 could be stored temporally in the soil
pore space after rainfall. The dissolved CO2 was then released gradually
with the evaporation of pore water, leading to lagged responses of efflux as
compared to respiration. Given that a major fraction of CO2 was
produced during the wetting periods (Fig. 5e), such a lagging effect should
be carefully examined when analysing the climatic sensitivity of FS. Our
simulations showed that a considerable fraction of CO2 produced could
be removed by root uptake and leave the volume measured by the respiration
chamber. Bloemen et al. (2016) showed that the CO2 concentration in
root xylems could be higher than in soil solutions. This implies that such a
“missing source” might be even greater than the model estimation, although
knowledge is still limited about the efficiency of the removal and the
diffusion/release of CO2 during the transport (Bloemen et al., 2016).
The contributions of biocrusts as C sinks or sources have remained largely
unknown (Castillo-Monroy et al., 2011). This is mainly due to the difficulty
to separate the CO2 exchanges of crust organisms from the background
respiration (Castillo-Monroy et al., 2011; Sancho et al., 2016). As
demonstrated in our work (Fig. 5b–d), the photosynthesis of top crust
could be masked by background emission quickly (e.g. within 1 day) after
rain events. The simulated FCt was 31 gC m-2 yr-1 at
interspace. Considering a 30 % coverage of lichens over the sampling area
(Feng et al., 2014), the interspace-level NPP was 9.3 gC m-2 yr-1. This value was largely greater than the lab-based estimation
for the site (Feng et al., 2014). However, it was in range of the values
reported from several other dryland ecosystems (i.e. 5.3–29 gC m-2 yr-1; Sancho et al., 2016). Our simulations also suggested that
photodegradation might offset about 48 % of the CO2 photosynthesized
by biocrust. It could explain the much higher FS measured from the
transparent chamber (C3) than from the opaque chamber (C2) during dry daytime
periods (e.g. Fig. 9). It should be also noted that the litter from shrub
canopy was not included in the measurement nor the modelling. Also, the
interactions between photodegradation and biotic decay were not
considered either. Hence, the contribution of photodegradation to soil C
balance could be greater than our estimation at the ecosystem level (see,
e.g. Gliksman et al., 2016). Although the contribution of surface exchanges
was only marginal as compared to the annual CO2 efflux, removing the
biocrust processes would substantially reduce the model validity. For
example, the goodness of fit (i.e. R2) in Test 3 dropped from 0.65
to 0.45 for the 2-year period, if FCt and FP were neglected.
Therefore, delineating the gas exchange of biocrust could be helpful in
order to upscale the modelling of C balance from chamber to ecosystem level,
where the distribution of the crust cover may vary from one site to another.
Comparison of the measured FS from lichen-crusted surfaces
using opaque and transparent chambers during a dry period (days 83–103) in
spring 2013.
![]()
Plant–interspace differences in soil C fluxes
Clumped distributions of foliage and biomass are critical features for the
adaptation and functioning of vegetation in arid and semiarid environments.
Previous studies have mainly emphasized the shrub effects on ecohydrology
(e.g. Gong et al., 2016) and enrichment of sediments and
nutrients, known as the “resource island” effect (Reynolds et al., 1999;
Rietkerk et al., 2004). Our simulations showed that the presence of shrub
canopy also influenced soil C exchanges. The presence of shrub cover
affected the C functioning of biocrust mainly through shading, which reduced
photosynthesis more than respiration and photodegradation. Compared to
interspace, the simulated annual FS was 13 % lower under canopy (Test 5). As we ruled out the differences in SOC and root biomass and limited the
C-flux differences contributed by biocrusts between plant-covered and
interspace areas, such a decrease in plant-cover FS was probably due to
the cooling effect of canopy (Gong et al., 2016). This effect was close to
the modelled responses of FS to ±2 ∘C in soil
temperature or ±10 % in soil water content. As the root density and
litter production rate are commonly larger under canopy than interspace
(e.g. Zhang et al., 2008), the lower respiration rate under canopy tends to
facilitate the accumulation of biomass and organic matter and feedback to the
functioning of resource islands during prolonged periods.
Our simulation further indicated considerable differences in the C-flux
sensitivities between areas with plant cover and those without. As the C processes
and initial conditions were set to be homogeneous, those differences could
mainly result from the different water-thermal conditions at plant cover and
interspace. For example, the higher temperature sensitivities of FT,
FS and FCt may relate to the cooling effect of canopy (see Gong et
al., 2016), which may lead to a greater Q10 value for respiration
estimations (i.e. Eq. 27). Moreover, the slower decomposition in
under-canopy soil could also lead to the lower sensitivities of FT and
FS to changes in root biomass and SOM contents. On the other hand, water
advection from interspace to plant cover, which may support over 30 % of
water loss from under-canopy soils (Gong et al., 2016), could help to lower
the FS sensitivity to water content changes at plant cover. The
increased water-thermal sensitivities of C exchanges of biocrust could be
explained by the less-stressful environment for curst organisms, e.g. higher
moisture content but lower radiation and temperature, although the
photosynthesis of lichens (PCt) could be reduced by shading (Table 3).
Such heterogeneity of C-flux sensitivities thus should be considered in
future studies on the ecosystem-level responses to climate change and
extreme climatic events.
Modelling uncertainties and future research needs
Our model showed its ability to describe the dynamics of soil temperature,
moisture content and C effluxes measured for the studied semiarid ecosystem.
Major uncertainty of the modelling, however, may refer to the concept of
equifinality (Beven, 1993, 2006). The question of equifinality arises from
the fact that the structures and mechanisms being modelled are based on
insufficient information. Consequently, alternative models using different
functions and parameter sets may fit equally well to observations; thus, the
mechanisms quantified in modelling are difficult to be justified or
falsified. Regarding this work, the modelling equifinality and uncertainty
could relate to several aspects.
Firstly, the RLU was a statistical simplification to the
target ecosystem at footprint scale (Gong et al., 2016) and may not fully
capture the spatially explicit schemes of soil environment and biogeochemistry at
ecosystem scale. For example, the model assumed Poisson probability of
mutual shading (Bégué et al., 1994), and the probability of shading
increased continuously with solar zenith (Gong et al., 2016). However, for
explicit space–time, shading is binary. This possibly leads to the biases in
the estimations of net radiation (Gong et al., 2016) and collar temperature
around midday, which sequentially affected the simulated diurnal pattern of
FS (see Fig. 3b). Moreover, field observations showed considerable
spatial variations of soil temperature, water content and biogeochemistry
(e.g. pH, litter quality and root biomass) within a distance of 3–5 m.
Such variations could well exceed a magnitude of 10 % and even over 100 % (e.g. Zhang et al., 2008; Feng et al., 2013; Wang et al., 2015).
Therefore, the variation of FS driven by the spatiality of soil factors
could be greater than the responses to ±2 ∘C in soil
temperature or ±10 % in soil water content. Therefore, future
modelling may need to consider spatially explicit settings, in order to
further minimize the gaps between model settings and reality.
Secondly, the high sensitivity of simulated FS to soil pH indicated that
unconsidered processes of inorganic C could strongly affect the accuracy of
modelling. Our modelling for CO2 transport considered gaseous and
liquid phases. However, the solid phase was not included, despite the high
lime content (2300–5400 kg ha-1) in the soil (Feng et al., 2013; Wang
et al., 2015). Based on soil samples of similar lime content (2700 kg ha-1), Buysee et al. (2013) showed that neglecting the inorganic C
exchanges by solids may lead to underestimation of FS during the heating
phase of the day but overestimation of FS during the cooling phase. This
is very similar to the diurnal pattern of biases in simulated FS (Fig. 3d). Therefore, further improvement on the modelling may need to consider
the solid phase as well.
Thirdly, the current model still lacked descriptions on growths of plant and soil
microbes. Compared to many other ecosystems, drylands often feature a high
root–shoot ratio (Jackson et al., 1996) but low SOC storage. Changes in
plant physiology and growth can readily influence root metabolisms, and
labile SOC pools hence modify FS dynamics (Wang et al., 2015). On the
other hand, a large fluctuation of diurnal and seasonal temperature may drive
the microbial communities to shift between those that are warm adapted and those that are cold adapted (Van
Gestel et al., 2013), which could largely change soil respiration and its
sensitivity to freeze–thaw cycles (Van Gestel et al., 2013; Liu et al.,
2016). Both the biotic controls are mixed with the legacy effects of
climatic variability over annual and interannual courses (Sala et al.,
2012; Jia et al., 2016; Shen et al., 2016), and could affect the C–water
simulations cumulatively through the feedbacks between biomass accumulation
and soil biogeochemistry (Bradford et al., 2016). This may explain the
decreasing trend of model validity from 2013 to 2014 (Figs. 3, 4).
Therefore, the dynamics of plant and microbial communities are required in
future modelling in order to improve the FS simulations regarding
interannual and long-term periods.
In addition, proper field data are still needed to support the future modelling work.
The dataset used in our model validation mainly separated the influences of
biocrusts from subsoil respiration. However, some processes like
photodegradation and lateral CO2 transport by root or water flows still
require more support from observations. Also, respiration data from
shrub-covered soil remain unavailable, as the settlement of soil collars
and respiration chambers under canopy could easily interrupt the
biophysiology of shrubs. The C functionality of crust organisms is especially
sensitive to water content (Table 4). However, tracking the water content in
the very thin layer of top crust can be very challenging using hour-based
meteorological data. Nocturnal water inputs (e.g. dewfalls) are important for
the metabolisms of crust organisms (e.g. Liu et al., 2006), but they are
hard to be quantified precisely by EC measurement or models derived from EC
data. Moreover, we presumed structural homogeneity for the crust layer and
employed a constant regime for crust processes. In reality, there may not be
clear boundaries between top crust and subcrust, and even top crust itself may
contain significant variations in microstructure and communities even within
1 cm (Williams et al., 2012; Raanan et al., 2016). The C
sequestration of biocrust can be strongly modified by microbial communities
directly (Belnap et al., 2003; Pointing and Belnap, 2012; Feng et al., 2014;
Maestre et al., 2015) and through other factors, e.g. surface albedo (Chamizo
et al., 2012), dew falls (Liu et al., 2006) and soil pore formation (Williams et
al., 2012; Felde et al., 2014). So far, many questions remain unanswered
about the mechanisms that control the colonization, adaption and succession
of microbial communities and the structure function of biocrust (Pointing
and Belnap, 2012). Further knowledge on these mechanisms will be helpful to
validate or falsify the modelled C functionality in response to climate change
and extreme climatic events.
Conclusions
This work represents a first attempt to integrate the CO2 production,
transport and surface exchanges (e.g. biocrust photosynthesis, respiration
and photodegradation) in FS modelling for dryland ecosystems with high
plant–interspace heterogeneities. Our model reproduced the FS dynamics
measured from non-crusted and lichen-crusted soil collars during
2013–2014, although introducing the gas exchanges of lichen crust decreased
the model performance at the hourly scale. However, further model
development may still be required on several aspects, e.g. by including (i) the spatially explicit schemes for surface conditions and soil
biogeochemistry; (ii) influences of lime and solids on CO2 transport;
(iii) growth dynamics of plants; (iv) high-resolution dynamics of surface
water-thermal conditions; and (v) the dynamics of microstructure and microbial
communities of biocrusts.
Our model simulations highlighted that the transport processes of inorganic
C and the metabolisms of biocrusts could strongly modify the CO2
efflux, and these influences are closely linked to soil hydrology. Soil
rewetting could enhance CO2 dissolution and delay the emission of
CO2 produced from the root zone. In addition, an ineligible fraction of
respired CO2 could be removed via lateral flows and root uptake, and
become “missing” from volumes under respiration chambers. The
lichen-crusted soil could temporally shift from net CO2 source to sink
during rewetting, as driven by the photosynthesis of lichens and the
restrained CO2 emissions from subsoil, whereas after rain events the
CO2 exchanges of lichens could be easily masked by background emissions
from subsoil. Based on our modelling, the annual NPP was 9.3 gC m-2 by
the top crust at interspace. However, the net C sequestration by the top crust could
be marginal if the photodegradation was accounted. Our modelling further
showed different componential C fluxes and sensitivities between
plant-covered and interspace areas. The presence of plant cover tended to
decrease the root-zone CO2 production and biocrust C sequestration but
increase the temperature sensitivities of these fluxes. On the other hand,
the sensitivities of root-zone emissions to water content were lower under
canopy. This may be due to the advection water flows from the interspace to
plant cover. To conclude, the complexity and plant–interspace
heterogeneities of soil C processes should be carefully considered when
extrapolating findings from chamber to ecosystem scales, in order to predict
the ecosystem responses to climate change and extreme climatic events. Our
model can serve as a useful tool to simulate the soil CO2 efflux
dynamics in dryland ecosystems.