Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
Root and dissolved organic carbon controls on subsurface soil carbon dynamics: A model approach
Research Group for Environmental Science, Nuclear Science and Engineering Directorate, Japan Atomic Energy Agency, Ibaraki, Japan
Corresponding author: M. Ota, Research Group for Environmental Science, Nuclear Science and Engineering Directorate, Japan Atomic Energy Agency, Tokai, Ibaraki, 319–1195, Japan. (email@example.com)
 Although subsurface horizons store more than half the total soil organic carbon (SOC) globally, the sources and dynamics of subsurface SOC are still unknown. Root litter input and dissolved organic carbon (DOC) transport are possible sources. Using a vertically extended soil C model, we explore the role of root C input and DOC transport in controlling subsurface SOC dynamics. The model involves organic matter decomposition and DOC leaching in the aboveground litter layer, the belowground input of C from roots, and SOC turnover and DOC transport along water flows throughout the soil profile for three SOC pools (active, slow, and passive — characterized by a turnover time of years, decades, and millennia, respectively). Model simulations with a range of rooting profiles demonstrate that a large proportion (36% – 78% — greater in deeper rooting profiles) of SOC is apportioned to the subsurface horizons (below the top 30 cm). A significant part (39% – 73%) of subsurface SOC was found to be associated with C pools that turn over on time scales of decades or less. DOC transport appeared to be dominant in distributing the added C to the deeper soil layers, making the SOC content profile deeper than that of the root litter (C) input. The results suggest that current soil C studies focusing on the surface alone, significantly underestimate the stock of decadally cycling C, especially for deeply rooted ecosystems. Studies ignoring subsurface C dynamics therefore underpredict the responses of soil C to changes in climate, land use, and vegetation.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Of all the terrestrial ecosystems, soil holds the largest reservoir of carbon (C), storing more C (~1600 Pg-C) as soil organic carbon (SOC) than both terrestrial vegetation and the atmosphere combined [Schimel, 1995]. Thus, the response of the soil C cycle to future climate and land use change is attracting significant scientific attention. In general, soil C studies and global change models focus on the top 20–30 cm of soil [Parton et al., 1987, 1993; Jenkinson et al., 1992], because this region of surface soil is considered to be the most dynamic in the cycling of C, and therefore the most susceptible to changes in environmental conditions [Trumbore et al., 1995; Van Dam et al., 1997; Gill et al., 1999; Gill and Burke, 2002].
 However, recent studies have shown the importance of subsurface soils (below the 20–30 cm surface layer) in ecosystem C cycling. More than half of the total soil C is stored in subsurface soils [Jobbágy and Jackson, 2000], and this region plays an important role in the global C cycle as it contains a significant amount of SOC that cycles on an annual to decadal time scale [Nepstad et al., 1994; Trumbore et al., 1995; Richter et al., 1999; Koarashi et al., 2012]. For example, in an Amazonian forest, where roots penetrate the soil down to a depth of several tens of meters, a substantial fraction (~15%) of deep SOC turns over annually to decadally [Trumbore et al., 1989, 1995, 2006; Nepstad et al., 1994]. Koarashi et al.  estimated that a large proportion (up to 73%) of mineral-associated SOC in California grassland and forest subsurface soils turns over decadally, with a significant time lag (>20 years) between C fixation and incorporation into the subsurface SOC pool. The resultant C flux from the subsurface SOC amounts to 3% – 16% of the total soil respiration [Koarashi et al., 2012].
 Recent studies also indicate that C stored in subsurface soils is able to respond more significantly and dynamically to changes in environmental conditions than previously thought. For example, SOC that has been stabilized in deep soil layers for a long time (centuries to millennia) can readily be decomposed by soil microbial decomposers when the soil is disturbed [Baisden et al., 2002a; Xiang et al., 2008; Salomé et al., 2010] or when fresh SOC is supplied [Fontaine et al., 2007; Kuzyakov, 2010]. This implies that the subsurface to deep soil layers can release a significant amount of stabilized C when experiencing environmental changes. Other studies have shown that chemically recalcitrant SOC is more sensitive to temperature than labile SOC [Davidson and Janssens, 2006], suggesting that SOC at different decomposition stages at different depths may contribute differently to soil C loss driven by global warming. To predict the response of global soil C to changes in environmental conditions therefore a comprehensive understanding of SOC cycling through the entire (surface to subsurface to deeper) soil profile is necessary [Rumpel and Kögel-Knabner, 2011; Schmidt et al., 2011].
 The possible sources and mechanisms that regulate the subsurface C cycling that have been considered to date are: — the input of C through above- (leaf litter and plant debris) and belowground (root litter and exudates) litter, and the subsequent transport of soil C as dissolved organic carbon (DOC) along the soil profile [Neff and Asner, 2001; Baisden and Parfitt, 2007; Fröberg et al., 2007; Kramer et al., 2010; Rumpel and Kögel-Knabner, 2011]. However, it has recently been discovered that the input of C by aboveground litter plays a minor role in regulating the subsurface soil C dynamics [Fröberg et al., 2007, 2009]. This has been made clear from evidence showing that the annual C input from aboveground litter to the underlying mineral soil is small in comparison with the actual soil C stocks [Fröberg et al., 2007, 2009; Kramer et al., 2010; Tate et al., 2011], and that the soil C delivered from aboveground litter is mostly retained in the top (~10 cm) soil layers [Fröberg et al., 2007, 2009]. In contrast, the input of belowground C by plant roots, and the subsequent transport of DOC through the soil profile, have been highlighted as large sources of both the stabilized and dynamic aspects of deep SOC [Trumbore et al., 1989, 2006; Nepstad et al., 1994; Neff and Asner, 2001; Rasse et al., 2005; Kramer et al., 2010]. Jobbágy and Jackson  showed that on a global-average scale, plant- (or ecosystem-) specific root-biomass distribution partly explains the belowground SOC profile; ecosystems with a deeply rooted profile tend to have a deeply distributed SOC profile [Jobbágy and Jackson, 2000]. It has also been shown that on a global-average scale, SOC is distributed somewhat deeper in the soil profile than the plant root biomass [Jobbágy and Jackson, 2000], and this pattern has also been observed on a local ecosystem scale [Weaver et al., 1935; Gill et al., 1999]. However, the mechanism of this underground pattern, which involves the living roots (as a source of C) and the SOC, is not well understood.
 Many studies have suggested that the downward percolation of DOC into the soil profile is the potential driver of the redistribution of C that has been added by the root litter, into the deeper soil layers [Jobbágy and Jackson, 2000; Neff and Asner, 2001; Baisden and Parfitt, 2007; Rumpel and Kögel-Knabner, 2011], thereby creating a profile of SOC deeper than that of living roots. Several observations have shown the presence of “atomic bomb” 14C, (produced by atmospheric nuclear weapons' testing), in the deep soil layers where plant roots are absent [Richter et al., 1999; Baisden et al., 2002b; Baisden and Parfitt, 2007], which supports the theory of the downward transport of soil C by DOC percolation from the overlying layers. However, despite the critical role of root C input and the DOC transport in soil C dynamics, their role in the development and cycling of the subsurface soil C still remains poorly understood [Jobbágy and Jackson, 2000; Rasse et al., 2005; Baisden and Parfitt, 2007; Rumpel and Kögel-Knabner, 2011]. This lack of understanding is mainly due to difficulties in the ability to make direct observations of the deep SOC cycle over extensively long time scales.
 To clarify the role of DOC transport in controlling the belowground C dynamics, soil C models involving this process have been developed [Elzein and Balesdent, 1995; Currie and Aber, 1997; Neff and Asner, 2001; Baisden et al., 2002b; Michalzik et al., 2003; Braakhekke et al., 2011]. For example, the DocMod model by Currie and Aber  links the generation of DOC to biological decomposition in the aboveground litter, and simulates the leaching of DOC to the underlying mineral soil. The DyDOC model by Michalzik et al.  considers the leaching of DOC from the surface bulk soil to the underlying horizon. The SOMPROF model by Braakhekke et al.  is a vertically extended soil C model which considers the turnover of C and the transport of DOC in the soil profile. However, models that address the complete processes which regulate the subsurface C dynamics (i.e., C input by root litter, subsequent C turnover and DOC (and thus water) transport from the surface to deep horizons) have not been developed, and the role of root C input and DOC transport in controlling subsurface C dynamics is still poorly explored [Neff and Asner, 2001; Fröberg et al., 2009].
 The aim of the present study is therefore to clarify the role of root litter input and DOC transport in controlling the dynamics and distribution of SOC from the subsurface to deeper soils. Here we hypothesize that the downward DOC flow conveys the root-derived C to deeper parts of the soil profile, thereby providing a decadal cycling of C in the subsurface and deeper soil horizons. To test this hypothesis, the belowground transport and turnover of C were modeled and linked to a well-established land-surface hydrological model (SOLVEG-II, developed by Yamazawa  and Nagai [2002, 2003, 2005]). After testing the model's performance, numerical experiments simulating chronosequence SOC development in hypothetical ecosystems with a range of rooting patterns were carried out.
2 Model Descriptions
2.1 General Description of SOLVEG-II
 The hydrological model SOLVEG-II consists of multilayered soil, vegetation, and atmospheric submodels. The soil water submodel calculates the soil water content and flux throughout the soil profile [Yamazawa, 2001], and the vegetation submodel calculates the physiological variables, such as leaf transpiration [Nagai, 2002, 2003, 2005]. The use of SOLVEG-II for the calculation of heat, water, and CO2 transport has been validated for a range of land-surface ecosystems (wheat field to grassland) [Nagai, 2002, 2003, 2005].
2.2 Aboveground C Submodel
 The concept of the C turnover model developed in the present study is shown in Figure 1. For the C budget (kg-C m−2 s−1) in the aboveground organic layer (hereafter referred to as the O-horizon), the C input was considered to be determined by litter fall (“aboveground C input” in Figure 1), and the C losses determined by DOC leaching and microbial decomposition:
where Ao (kg m−2) is the dry mass of the organic material (including the organic C) in the O-horizon per unit ground area, χos (kg-C kg−1) is the C content of the O-horizon material, t (s) is the time, Fabove (kg-C m−2 s−1) is the aboveground C input, P (m3 m−2 s−1) is the vertical water flux (= precipitation intensity at the ground surface) percolating through the O-horizon, χow (kg-C m−3) is the DOC concentration in the percolating water, and ko (s−1) is the decomposition constant for the O-horizon material.
 For the leaching of DOC (Pχow), it was assumed that the concentration of DOC (χow) in the water that percolates from the O-horizon to the underlying mineral soil layer is directly proportional to the C content (χos) in the O-horizon:
where Kdo (m3 kg−1) is the proportionality factor. This direct relationship is supported by a laboratory DOC extraction experiment, showing that organic C contained in material within the O-horizon dissolves in water according to a partition rule [Christ and David, 1996]. Using four-year-averaged observations of the C content (χos) in the O-horizon and the DOC concentration (χow) in the percolating water in a subtropical cypress forest [Schmidt et al., 2010], the proportionality factor was calculated to be Kdo = 12 m3 kg−1 (Table 1).
Table 1. List of Parameters
aSilt + clay content (q) is parameterized by the soil texture (section 2.3.1).
Dry mass of organic material in the O-horizon (Ao)
Model performance test: 0.0351) m Numerical experiment: 0.112), 0.412), and 0.993) m
1)Baisden et al. [2002b] 2)Jackson et al.  3)Trumbore et al. 
e-folding depth of root-water uptake (zr)
Model performance test: 0.362) m Numerical experiment: 0.112), 0.412), and 0.993) m
 The turnover time to (year) for the microbial decomposition of the O-horizon material is assumed to be 3 years (i.e., ko ≡ to−1 = (3 years)−1), based on measurements of the decomposition rates of a range of plant litter materials [Berg and McClaugherty, 2003]. The value of to = 3 years is similar to the turnover time for litter or plant debris assumed in well-accepted conceptual models (e.g., CENTURY by Parton et al. [1987, 1993] and Roth-C by Jenkinson et al. ).
 In this model, DOC that is leached from the O-horizon is sent to an active C pool (see section 2.3.1 for the definition) in the topmost mineral soil layer immediately below the O-horizon (Figure 1). A fraction (ao1 in Figure 1) of the decomposing organic C in the O-horizon is also added into the active C pool in the topmost mineral soil layer. The downward flow of the decomposing organic C simulates the sedimentation of physically degraded organic materials contained in the O-horizon [Trumbore et al., 1995; Elzein and Balesdent, 1995; Rubino et al., 2007]. The fraction ao1 is set to 0.45, a value adopted for the transfer of the decomposing structural or metabolic C into the active SOC pool in the CENTURY model [Parton et al., 1987]. The modeled transfer of the decomposing and leaching organic C into the most labile active C pool (not into the slow or passive C pools) is based on observations; fresh litter material does not immediately contribute to stable SOC pools [Trumbore, 2000; Tate et al., 2011], and DOC leached from undecomposed organic materials is labile [Kalbitz et al., 2005]. It has recently been shown that enhanced decomposition of the aboveground litter by photodegradation [Austin and Vivanco, 2006] and subsequent rain events [Ma et al., 2012] can affect the transfer of C from the O-horizon to mineral soil, but this process was not considered in the present model.
2.3 Soil C Submodel
2.3.1 SOC-DOC Distribution
 In the model used in this study, the soil organic C in the mineral soil layers was modeled in a similar way to that of the CENTURY model [Parton et al., 1987]: as active, slow, and passive SOC pools with turnover times of 1.5, 25, and 1000 years, respectively (see Figure 1 and Table 1). Conceptually, the active SOC represents the soil microbes, microbial products, and the rapidly degradable plant materials; the slow pool represents the resistant plant materials (e.g., plant lipids and lignin) and SOC weakly stabilized by physical and chemical processes; and the passive pool represents SOC strongly stabilized by interactions with mineral surfaces and metal ions. For these three SOC pools, (active, slow, and passive), it was assumed that at every time step and every grid within the soil, an immediate equilibrium can be achieved for sorption and desorption of soil C between the solid (SOC contained in the soil constituents) and dissolved (DOC contained in the soil water) phases:
where χss,i (kg-C kg−1) is the SOC content per dry soil mass in the ith C pool (subscript i = 1, 2, and 3 for the active, slow, and passive C, respectively), Kds (m3 kg−1) is the distribution coefficient, and χsw,i (kg-C m−3) is the DOC concentration in the soil water in the ith C pool. The coefficient Kds is the distribution between the solid and dissolved organic carbon, not a depth distribution of the organic carbon in the soil profile. Equation (3) assumes that soil C added to each C pool (active, slow, and passive) is immediately distributed to the solid and dissolved phases of the corresponding C pool, and is transported, decomposed, or stabilized to the other C pools (see sections 2.3.2 and 2.3.3). The partition between the SOC and DOC in equation (3) also assumes that Kds is independent of the quality (active, slow, and passive pools) of the SOC. This assumption may be partly supported by Ussiri and Johnson , who reported a relatively small difference, (only by, at most, a factor of two), between the distribution coefficients measured for hydrophilic C and hydrophobic C, which can be easily associated with soil minerals, and hence, is thought to be a part of the slow to passive SOC pool [Kalbitz et al., 2000].
 It was assumed that Kds depends on the silt + clay content q (kg kg−1) of the soil. The empirical formulation for Kds (Figure 2) was then derived by fitting a logarithm function of q to Kds as measured by Moore et al. :
 Twenty-four measurements out of the total 45 measurements (53%) (see Figure 2) agreed with the modeled Kds within a factor of 1.5, and 37 measurements out of the total 45 measurements (82%) agreed with the modeled Kds within a factor of 2. The silt + clay content q is parameterized using the U.S. Department of Agriculture classification of soil texture [Dunne and Willmott, 1996]: q = 0.08 for sand, 0.18 for loamy sand, 0.34 for sandy loam, 0.82 for silt loam and silt, 0.59 for loam, 0.42 for sandy clay loam, 0.90 for silty clay loam, 0.68 for clay loam, 0.49 for sandy clay, 0.94 for silty clay, and 0.82 for clay.
2.3.2 SOC Turnover
 When considering the C input and decomposition for each C pool (active, slow, and passive), the change in the SOC content for the ith C pool at a certain depth in the soil profile is modeled by:
where ρb (kg m−3) is the dry bulk density of the soil, Sss,i (kg-C m−3 s−1) is the input rate of C (as SOC) into the ith SOC pool, and kss,i (year−1) is the decomposition rate of the SOC in the ith C pool. The variable kss,i is characterized as a reciprocal of the turnover times (section 2.3.1): kss,1 = (1.5 years)−1, kss,2 = (25 years)−1 and kss,3 = (1000 years)−1 for the active, slow, and passive SOC pools, respectively.
 The root litter input (including root exudates and root debris) to the active C pool is considered as the primary, direct source of SOC in the soil profile (“belowground C input” in Figure 1). Generally, root biomass and root litter input decrease with increasing soil depth [Trumbore et al., 1995, 2006; Gill and Burke, 2002]. Thus, an exponential distribution is assumed for the belowground C input Sbelow (kg-C m−3 s−1) of root origin [Elzein and Balesdent, 1995; Baisden et al., 2002b]:
where Fbelow (kg-C m−2 s−1) is the input of belowground C measured per unit ground area, zb (m) is the e-folding depth of the belowground C input, and z (m) is the soil depth (downward positive) measured from the O- and A-horizon interface.
 On the basis of the concept of the CENTURY model [Parton et al., 1987], at a certain depth in the soil a small fraction (a13 = 0.004 in Figure 1 and Table 1) of the C originating from the decomposition of active SOC is assumed to be stabilized and settled in the passive SOC pool, while a large fraction, F(q) in Figure 1, is lost as CO2, where the fraction F(q) is parameterized by the silt + clay content q [Parton et al., 1987]:
 For example, F(q) is calculated to be 0.62 for sandy loam soil (q = 0.34; see section 2.3.1). The remaining fraction, a12 = 1 – a13 – F(q), of the decomposed active SOC is incorporated into the slow SOC pool (Figure 1). Similarly, for the decomposition of the slow SOC, a small fraction a23 = 0.03 is assigned to the passive SOC pool, while a large fraction a21 = 0.42 is again allocated to the active SOC pool [Parton et al., 1987]. All of the decomposed C from the passive SOC pool is assumed to be lost as CO2. The input rates of C Sss,i in equation (5) to the three SOC pools are summarized as:
2.3.3 DOC turnover-transport
 In our model, it is assumed that at a certain soil depth a fraction of organic C in each C pool (active, slow, and passive) is as mobile as DOC (defined by equation (3)). We also assume that the DOC in each C pool is independently transported within the soil profile according to the vertical water flow, and that the DOC removed from one layer is added into the next (underlying, or overlying) layer (Figure 1). The transport of DOC for the ith C pool in the soil is then modeled by considering the diffusion and advection of the DOC, as well as the loss of DOC via root-water uptake and microbial decomposition:
where ηw (m3 m−3) is the soil water content, Dw (m2 s−1) is the effective diffusivity of the DOC, Ew (m3 m−2 s−1) is the vertical soil water flux, er (m3 m−3 s−1) is the root-water uptake, and ksw,i (s−1) is the decomposition rate for DOC defined according to each DOC pool. In equation (9), ηw and Ew are calculated using the soil water submodel of SOLVEG-II (section 2.1).
 The effective diffusivity Dw is calculated by considering the molecular diffusion and dispersion of the DOC as:
where the factor ηw/1.5 is the tortuosity in the bulk soil [Penman, 1940], Dm = 1.23 × 10−10 m2 s−1 is the molecular diffusivity of the DOC in water [Burdige et al., 1999], and Adis = 0.05 m is the dispersion coefficient of the soil water [Nakano, 1991].
 For the root uptake er, the live root density, (and hence the root-water uptake), generally decreases with soil depth [Feddes et al., 2001]. Therefore, the site- (or plant-) specific e-folding depth zr (m) of the root-water uptake is defined, and er in equation (9) is modeled as:
where Ftra (m3 m−2 s−1), which is the leaf transpiration flux measured per unit ground area, is calculated using the vegetation submodel of SOLVEG-II (section 2.1).
 In relation to the microbial decomposition of DOC (ksw,i in equation (9)), there have been no experimental studies linking the biodegradability of DOC with that of the SOC contained in soil [Kalbitz et al., 2000]. Therefore, it is assumed that the decomposition rate of the DOC in the ith C pool is identical to that of the corresponding SOC pool; i.e., ksw,i = kss,i (see Figure 1 and Table 1). This assumption may be partly supported by measurements [Zsolnay and Steindl, 1991; Nakanishi et al., 2012], which show that the water-extracted C from mineral soil is a mixture of labile (decomposed over months) and less-degradable C fractions. The entire amount of C produced by DOC decomposition is assumed to be lost as CO2 (Figure 1).
 The SOC turnover and DOC transport shown in equations (5) and (9), respectively, are linked by the exchange (immediate equilibrium represented by equation (3)) of C between the SOC and DOC pools, and solved under the following boundary conditions: SOC sedimentation (represented by flux ao1koAoχos, see section 2.2) and DOC leaching (represented by flux Pχow in equation (1)) from the O-horizon into the active SOC pool in the uppermost soil layer, and a zero vertical gradient in the C concentration at the bottom of the mineral soil.
3 Model Inputs and Parameters
 The performance of the developed model was tested by a simulation of SOC development in a chronosequence grassland ecosystem [Baisden et al., 2002a, 2002b]. The site (Post Modesto; 37.51°N, 120.46°W, elevation ~300 m above sea level) is located in eastern California where the mean annual temperature is 16°C and the mean annual precipitation is 300 mm. The temperate Mediterranean climate (hot, dry summers and cool, wet winters) at the site supports a flora dominated by invasive annual grasses (Boromus mollis, Bromus rigidis, etc.) and forbs, with a secondary component of deep-rooted oaks accessing deep water sources all the year-round [Jones and R.G. Woodmansee, 1979; Baisden et al., 2002a, 2002b]. The age of the underlying sandy loam soil is <3 ky. In 1997, soil (including the O-horizon) was sampled down to a depth of 1.5 m [Baisden et al., 2002b], and the sampled soil was separated into the O-horizon and mineral soil material (below the O-horizon). For the mineral soil samples, the contained root litter was removed and the SOC content (kg-C kg−1) was measured [Baisden et al., 2002b].
 In the calculation, a 2.5 m high atmospheric layer and a 5.5 m thick sandy loam soil layer were established in a 1 m2 cell. The atmospheric layer and soil layer are subdivided into eight and 27 layers, respectively. The boundary heights of the subdivided atmospheric layer were: 0.1, 0.3, 0.5, 0.7, 1, 1.5, 2, and 2.5 m. The bottom six (z = 0 – 1.5 m) boundary heights of the atmospheric layer contained the C4-grass vegetation canopy. The boundary depths of the subdivided soil layer were: 0.005, 0.015, 0.03, 0.05, 0.07, 0.10, 0.14, 0.19, 0.26, 0.35, 0.45, 0.57, 0.70, 0.84, 1, 1.18, 1.38, 1.58, 1.78, 2, 2.5, 3, 3.5, 4, 4.5, 5, and 5.5 m.
 In the soil model, an e-folding depth, zr, of 0.36 m was employed for the belowground root-water uptake (er in equation (11)), which is the globally averaged live-root distribution for temperate deciduous and coniferous forests [Jackson et al., 1996]. This zr remained constant throughout the entire sequence of the calculation to simulate the year-round water uptake by the deep-rooted oak at the site (see above). The bottom soil layer at z = 5.0 – 5.5 m was assumed to be saturated with soil water, simulating the deep water source at the site. On the basis of the measurements by Baisden et al. [2002b], the dry bulk density ρb of the soil was set as a function of the soil depth (Figure 3).
 Half-hourly meteorological data observed at 2.5 m above the ground at a grassland ecosystem in 2001 (Ameri-flux observation site in Vaira Ranch, CA., 38.41°N and 120.95°W, elevation 129 m above sea level and 113 km apart from Post Modesto; data are available from the Ameri-flux webpage at http://public.ornl.gov/ameriflux/) were used as the input meteorological data at the model-assumed top (z = 2.0 – 2.5 m) atmospheric layer. The mean annual temperature and annual precipitation at the Vaira Ranch site in 2001 were 16°C and 415 mm, respectively. To precisely simulate the water regime at the objective Post Modesto site, the annual precipitation of the Vaira Ranch data set was reduced to 300 mm (= mean annual precipitation at Post Modesto, see above), calculated by multiplying a factor 300 mm/415 mm to the whole half-hourly precipitation data for the Vaira Ranch data set in 2001.
 For the organic C calculation, above- and belowground C inputs measured at the Post Modesto site [Baisden et al., 2002b] were used: Fabove = 0.27 kg-C m−2 yr−1 for equation (1) and Fbelow = 0.45 kg-C m−2 yr−1 for equation (6) (see Table 1). The e-folding depth (zb in equation (6)) of the belowground C input was set as zb = 3.5 cm, a value given by Baisden et al. [2002b]. The dry mass of the O-horizon (Ao in equation (1)) at the site was not available, and thus was assumed to be constant (Ao = 2.275 kg-C m−2) throughout the entire sequence of the calculation. The value 2.275 kg-C m−2 was determined as follows. First, it was assumed that the O-horizon is made up of C6H12O6, which has a C content (χos in equation (1)) of C6/C6H12O6 = 0.4 kg-C kg−1. The model calculations were then carried out iteratively, by tuning Ao until the model-calculated C content (χos) in the O-horizon reached a steady-state condition (i.e., ~10 years after initiating the C input, see section 4.1) at 0.4 kg-C kg−1.
 Assuming that the meteorological conditions did not change interannually, and that all of the yearly litter input occurred at the end of the year (31 December), a consecutive calculation was carried out over a period of 3000 years, by repeatedly using the meteorological data set from 2001 (precipitation-modified). The initial SOC content at the beginning of the first year were all set to be zero for both the O-horizon and the entire soil profile, and the calculated SOC profile at the end of one year (31 December) was used as the initial SOC profile at the beginning of the next year (1 January). The time step of the calculation was 15 min.
 In the numerical experiments, the development of SOC at the Post Modesto site was calculated by assuming a range of e-folding depths zb of the root litter input, and zr of the root-water uptake; zb = zr = zroot = 0.11, 0.41, and 0.99 m (zroot is hereafter referred to as the “rooting depth”). The employed rooting depths were chosen on the basis of the rooting patterns found in temperate climate ecosystems. The rooting profile of zroot = 0.11 m corresponded to shallow-rooted profiles observed in a grassland in Central Missouri Prairie [Dahlman and Kucera, 1965]. The profile of zroot = 0.41 m was based on the rooting profiles in a temperate coniferous forest in Scotland [Wright, 1955; Jackson et al., 1996]. The deepest profile of zroot = 0.99 m was determined from observations for deeply accessing roots in a coniferous forest in Florida [Heyward, 1933; Canadell et al., 1996]. To investigate the effect of zroot on the belowground C dynamics, all the settings (input meteorological data, annual litter input, etc.) other than zroot remained unchanged from those used in the model validation. SOC development was calculated over 1 ky for each zroot case.
 Some additional calculations were made in the model validation and the numerical experiments to check the sensitivity of the calculation results to certain parameters that had a range of uncertainties: (1) distribution coefficient (Kds) in soil; (2) decomposition rates (ksw,i) of DOC; (3) depth profile of decomposition rates (kss,i) of SOC, and (4) water budget at the ground surface. The distribution coefficient Kds assumed in the model has an uncertainty of a factor of 2, at most (section 2.3.1). Thus, sensitivity tests were carried out by doubling Kds; i.e., by replacing “Kds” in equation (3) with “2Kds.” There is no experimental evidence related to the decomposition rate of DOC to support the assumption used in the model, that the decomposition rate for the DOC is fully identical to that of the corresponding SOC pool (i.e., ksw,i = kss,i, see section 2.3.3). In fact, Zsolnay and Steindl  observed that a large fraction (up to 50%) of water-extracted DOC from soil samples is rapidly decomposed within several months. Thus, calculations were performed with accelerated decomposition rates for the slow and passive DOC pools: ksw,2 = ksw,3 = ksw,1 = 1.5 years−1 over the entire 5 m of soil. For the depth profile of SOC decomposition, the present model uses a fixed SOC decomposition rate (kss,i) throughout the soil profile (section 2.3.2). However, reduced SOC decomposition rates in deeper horizons have been observed in numerous studies (see section 4.1). Then additional calculations were performed assuming a depth-dependent decomposition rate for the SOC:
where the reference decomposition rates kssREF,i at the ground surface are defined as kssREF,i = 1.5, 25, and 1000 years for the active, slow, and passive SOC pools, respectively, and the e-folding depth of zdec = 1.7 m was estimated from measurements by Gill and Burke . For the water budget at the ground surface, the present model does not consider the rainwater interception by the O-horizon. At a forest floor, it is possible that surface litter layer (O-horizon in this study) intercepts a fraction of rainwater by, at most, c.a. 1 mm [Bristow et al., 1986; Findeling et al., 2006], which may affect water (and DOC) transport into the underlying mineral soil. To check this effect of the rainwater interception by the O-horizon, we performed an additional calculation by reducing half-hourly precipitation input (i.e., precipitation of the input meteorological data) by 1 mm.
 In numerical experiments, the vertical water flow and root-water uptake in soil were found to be able to control the belowground C dynamics (see sections 4.2.1 and 4.2.4). To further clarify the impact of these processes on the soil C dynamics, calculations were performed by tuning the vertical water flow and root-water uptake in the soil. For the soil water flow, calculations using double the amount of soil water flux were carried out by replacing “Ew” in equation (9) with “2Ew.” With respect to the root uptake, calculations hypothetically assuming a zero root-water uptake in the soil were performed by setting er = 0 m3 m−3 s−1 in equation (9).
4 Results and Discussion
4.1 Model Validation
 Owing to the annual litter fall, the C content (χos) in the O-horizon increased each year (data not shown) and reached an equilibrium value of 0.4 kg-C kg−1 by ~10 years. In this equilibrium state, the transfer of C from the O-horizon to the underlying mineral soil amounted to 0.118 kg-C m−2 yr−1 for SOC sedimentation, and 0.005 kg-C m−2 yr−1 for the DOC leaching. 3 ky after initiating the C input, the calculated C transport and turnover in the top 1.5 m of soil entered a quasi-steady state, with negligible year-to-year increases in the SOC content for all three (active, slow, and passive) pools. The magnitude and the vertical profile of the total SOC content (sum of χss,i of the active, slow, and passive pools) obtained using the model calculation agreed well with the observed data (Figure 4), except for an underestimation of the calculated SOC content in the surface (z = 0–0.06 m) and deeper (z = 0.95–1.35 m) soil layers.
 Three possible reasons are considered for the underestimation of the surface SOC content in the simulation. First, the reference SOC measurement can have uncertainties. At the study site, abundant amounts of very fine live roots have been observed in the top several centimeters of soil [Baisden et al., 2002a, 2002b]. In such a soil, a clear definition of the SOC content would therefore be difficult, because C-rich fine roots are often incorporated into the soil matrix and may not be completely removed before the C measurements, thereby resulting in an overvaluation of the SOC content.
 Second, underestimation of the distribution coefficient (Kds) in the model simulation can be considered as a possible reason for the underestimation in the surface SOC content by the simulation. An increase (or decrease) in Kds suppresses (or enhances) the downward transport of the added C through DOC percolation (see section 4.2.1 for a detailed discussion of the belowground C transport), thereby leading to a larger (or smaller) SOC content beneath the ground surface. In fact, the calculation employed in this study, which used double the value of Kds (see section 3), produced an increased total SOC content of 0.032 kg-C kg−1 in the uppermost (z = 0 – 5 mm) soil layer; compare with the total SOC content of 0.025 kg-C kg−1 in the uppermost soil layer (Figure 4), calculated with the original settings.
 The last possible reason is insufficient modeling and parameterization for DOC leaching from the O-horizon to the mineral soil. We acknowledge that the simulated DOC input (0.005 kg-C m−2 yr−1) from the O-horizon to the mineral soil can be highly variable, depending on the factor Kdo used in equation (2). While this model used a fixed value of Kdo = 12 m3 kg−1 estimated from a field observation (section 2.2), Kdo decreases to 0.02–0.04 m3 kg−1 when Kdo is measured by in vitro DOC extraction experiments [Christ and David, 1996]. This implies that, if precipitating water in the O-horizon is in good contact with organic materials in the O-horizon (e.g., a situation which would be expected to be found in a field where there is dense organic material in the O-horizon), then there would be an increase in the concentration of DOC in the water filtering to the underlying soil layers. Indeed, research has shown that the addition of litter material to the forest floor increases the DOC concentration in forest floor leaches [Park and Mazner, 2003; Yano et al., 2003]. With a biological decomposition model (DocMod) using the aboveground litter, Currie and Aber  also demonstrated that the forest-floor DOC flux proportionally increases with an increase in the amount of the litter aboveground.
 The modeled rate of DOC leaching from the O horizon to the soil can also be affected by insufficient modeling of water flux at the ground surface. The model in the present study (SOLVEG-II) does not consider the water budget in the O-horizon. Therefore, the model may overvalue the amount of water (and thus DOC) percolating from the O-horizon to the top mineral soil layer, because a fraction (at most 1 mm, see section 3) of rainwater can be intercepted by the litter materials and does not enter the underlying mineral soil. However, the result calculated with the reduced-precipitation input (section 3) showed that the effect of the rainwater interception by the O-horizon is minor in the present simulation. The SOC content in the top (z = 0–5 mm) soil layer calculated with the reduced-precipitation decreased only by ~1%, compared with the calculation result for the regular precipitation input (Figure 4).
 For the underestimation in the simulated SOC content (mainly composed of the slow and passive SOC, Figure 4) in the deep soil layers (z = 0.95 – 1.35 m), two possible causes are considered. First, an insufficient parameterization of the allocation of root-originated C can be considered. In the model, it was assumed that the root-derived C was allocated totally to the active SOC pool in the soil (section 2.3.2). However, certain plant-derived molecules, such as plant lipids, lignin, and other structural tissues, often persist longer (on decadal time scales) in soils [Nierop, 1998; Rasse et al., 2005]. Thus, a fraction of the root-derived C would be directly incorporated into the slow SOC pool (25 year residence time in the present model), thereby bypassing the active pool [Parton et al., 1987, 1993]. Therefore, the current model probably underestimates the C input to the slow SOC pool at all depths, because a large fraction, F(q) = 0.62 (equation (7)), of the C added to the active SOC pool from the direct source is mineralized without further stabilization into the slow pool (Figure 1). Another mechanism for the decadal-time scale stabilization may be the occlusion of root-derived C by aggregation, particularly in deeper soils [von Lützow et al., 2006].
 An overestimation of the decomposition of SOC in the deeper horizons by the simulation is another possible cause for the underestimation in the deep SOC content. In the current model, a constant decomposition rate (kss,i) for each SOC pool (active, slow, and passive) was used throughout the entire soil profile (section 2.3.2). However, numerous studies have shown a reduced decomposition rate of SOC in subsurface horizons [Van Dam et al., 1997; Gill et al., 1999; Trumbore, 2000; Gill and Burke, 2002; Koarashi et al., 2009]. The primary mechanism considered to date is the reduced accessibility of SOC to decomposer organisms, which occurs because of the sparse distribution of the decomposer organisms and the enhanced protection of the SOC by organo-mineral association in deeper soil layers [Dungait et al., 2012]. The model results with depth-dependent decomposition rates (equation (12)) showed that the effects of the reduced decomposition rate in the deeper soil layers on the deep SOC stock are small in the present calculation (e.g., the total SOC content in Figure 4 increased only by ~20% for the z = 0.9 – 1.1 m soil layers), because the SOC is distributed superficially within the top soil layers (several tens of centimeters) at the objective site (Figure 4).
 The rainwater interception by the O-horizon may affect the deep SOC content. The simulation results with the reduced-precipitation input (section 3) showed that the rainwater interception by the O-horizon certainly affects the deep SOC content; e.g., the total SOC content calculated with the reduced-precipitation decreased by ~30% for the z = 1.0 – 1.18 m soil layers. The decrease in the deep SOC content was delivered by the decreased water (and thus DOC) flow in the soil (see section 4.2.1 for the detailed role of soil water transport in the subsurface soil C dynamics).
4.2 Soil C Dynamics for Different Rooting Patterns—Results from Numerical Experiments
4.2.1 Factors Determining SOC Profiles
 The development of SOC belowground, calculated with the rooting depth zroot = 0.11 m, significantly differed between the three (active, slow, and passive) SOC pools (Figure 5). In the active pool, the SOC was mostly apportioned to the upper layer (several tens of centimeters), and the development of this pool was nearly completed by ~10 years after initiating the C input (Figure 5(a)). In contrast, the distribution of the slow and passive SOC pools was deeper in the soil profile than that of the active pool, and the formation of the profiles of the slow and passive SOC pools occurred over decades, to centuries, or even to millennia (Figures 5b–5c). The deeper penetration of the more recalcitrant C in the soil profile is consistent with the results from multi-C-pool models that consider vertical transport of DOC in soil [Elzein and Balesdent, 1995; Baisden et al., 2002b]. Although not shown in the figure, the time scales for the development of SOC calculated with zroot = 0.41 and 0.99 m were approximately identical to those calculated with zroot = 0.11 m.
 Under a quasi-equilibrium state (1 ky after initiating the C input) with negligible year-to-year increases in the SOC content for all three (active, slow, and passive) pools, the depth distribution of the total (active + slow + passive) SOC content differed considerably, depending on zroot (Figures 6a–6c), with a larger fraction of SOC being apportioned to the deeper parts of the soil profile with a larger zroot. However, in all the zroot cases, the depth distribution of the total SOC content extended deeper than that of the belowground C input (Sbelow, indicated in Figure 6(d)), which is characterized by zroot. The results are consistent with simulations of Neff and Asner  showing that SOC penetrates deeper in the soil profile than the root C input.
 Clearly, these simulations demonstrate that the downward water flux efficiently transports C (as DOC) into deeper parts of the soil profile (Figure 5) and enables SOC to be distributed to lower depths than the belowground root litter (C) input reaches (Figure 6). This result supports the hypothesis we made: downward DOC flow conveys root-derived C to deeper pars of soil (section 1). The result also provides an answer to the question that has arisen from several other studies [Weaver et al., 1935; Gill et al., 1999; Jobbágy and Jackson, 2000]: why is the SOC in soils generally distributed at depths deeper than the root biomass? A number of studies have discussed this subject [Van Dam et al., 1997; Gill et al., 1999; Jobbágy and Jackson, 2000; Neff and Asner, 2001; Gill and Burke, 2002; Rasse et al., 2005; Baisden and Parfitt, 2007; Rumpel and Kögel-Knabner, 2011; Koarashi et al., 2012], and have suggested four potential explanations for this inconsistency: (1) a downward transport of added SOC through DOC percolation; (2) an increase in live-root turnover with depth, causing a higher input of C in the deeper soil layers; (3) a mixing of SOC by soil faunal organisms (bioturbation); and (4) a decrease in SOC turnover with soil depth.
 The belowground transport of C simulated in the present study supports the first explanation as being the main cause for the inconsistency. Because of the small effective-distance (Ladv,1 ~ 0.1 m, see Appendix A) of active soil C transport by soil water flow, the depth profile of the steady-state active SOC content becomes similar to that of the root litter input represented by zroot (Figure 6). In the present simulations, however, the surface soil (several tens of centimeters) stores SOC that is more active (Figure 6), mainly because of the C input from the overlying O-horizon. In contrast, in relation to the slow soil C (decadally cycling), the effective distance (Ladv,2) is calculated to be ~0.6 – 1.6 m (Appendix A). Therefore, this SOC can be allocated to deeper soil layers for decades after initiating the SOC input (Figure 5b), and the steady-state slow SOC profile becomes deeper than its source profile (i.e., the active SOC) (Figure 6). Similarly, the passive soil C can be accumulated at much greater depths than the sources (the active and slow SOC), because of its long residence times (tres,3) that increase with depth on a time scale up to millennia (see Figure A1 in Appendix A). Overall, the slower-cycling SOC, with a longer residence time, forms a more deeply distributed profile in the soil through an extended downward transport of the DOC. This mechanism eventually makes the total SOC content deeper than the root litter input. The increased fractions of subsurface SOC in all three zroot cases as a result of the doubled soil water flux (Table 2; the fractions of subsurface SOC with a doubled soil water flux increased to 50% – 84%, in comparison with the original fractions of 36% – 78%) further support the role of the DOC transport in distributing the added C to deeper parts of the soil. This result is consistent with observations showing that heavier rainfall drives larger DOC fluxes within the soil profile [Schmidt et al., 2010].
Table 2. SOC Stock (kg-C m–2) at 1 ky After Initiating SOC Development
 The present model assumes that the root litter (C) input (Sbelow by equation (6)), or the turnover of live roots, decreases with soil depth. Trumbore et al.  observed that fine root production steeply decreases with depth by two to three orders of magnitude over the top 6 m of soil in Amazonian tropical forests, while the turnover time of the fine roots remained nearly unchanged throughout the soil profile. Similar results have also been observed in other ecosystems [Gill and Burke, 2002; Joslin et al., 2006], suggesting that higher root litter inputs generally occur in the upper soil layers. Although course roots (>2 mm in diameter) may be an important source of deep C, they are likely to be only small contributors to the overall C inputs, because of their longer lifespans [Gill and Jackson, 2000; Joslin et al., 2006]. These observations do not support a faster turnover of live roots (and thus a higher C input) in deeper horizons as being the main cause of the inconsistency.
 With reference to the mixing of SOC through bioturbation, Trumbore et al.  showed that by comparing the profiles of Cs and 14C observed in a forest soil, bioturbation is a less important process in the belowground allocation of C than the process of advective DOC transport. In this study, the effective distance (Lbio,i) due to bioturbation was calculated (Appendix B). It was found that Lbio,1 = 0.05 m as a result of bioturbation for active soil C is smaller than the effective distance Ladv,1 ~ 0.1 m for active soil C (see above), based on the advective DOC transport estimated for an extensively dry climate with the annual precipitation of only 300 mm (section 3). Therefore, the mixing of SOC through the process of bioturbation can be ruled out as the main cause of the inconsistency.
 The last possible explanation, the reduced turnover of the SOC with depth, has been reported in several studies [Van Dam et al., 1997; Gill et al., 1999; Trumbore, 2000; Gill and Burke, 2002; Koarashi et al., 2009]. Although the additional calculations with reduced rates of SOC decomposition with depth (which were performed in this study) clearly demonstrate a larger accumulation of C in the subsurface soil layers (Table 2; the fractions of the subsurface SOC with depth-dependent decomposition rate of SOC increased to 40% – 86% in comparison with the original fractions of 36% – 78%), the responses of soil C to changes in environmental conditions can only be reasonably predicted if there is a mechanistic understanding of the processes that inhibit C turnover in the soil. With depth, the association of SOC with mineral surfaces [Torn et al., 1997; Six et al., 2002] and soil particles (occlusion within soil particles) [Salomé et al., 2010] increases, and density of the decomposer organisms decreases [Fiere et al., 2003], which can make the SOC less degradable. Recent studies [Fontaine et al., 2007; Kuzyakov, 2010] suggest that the stability of SOC in the deeper soil layers is maintained in the absence of an energy source for the soil microbes. Koarashi et al.  suggest that the dynamics of decadally cycling C in the subsurface soil layers may be largely controlled by interactions with soil minerals, not by climatic conditions, (as generally observed in surface soil layers). Increasing evidence of how the regulatory processes differ between the surface and the subsurface soil horizons suggests that a more quantitative understanding of the relative importance of the processes along the soil depth is required [Rumpel and Kögel-Knabner, 2011; Schmidt et al., 2011], and that the improved understanding should be incorporated into soil C models that explicitly consider the subsurface C budget.
4.2.2 Subsurface Soil as a Large Pool of Decadally Cycling C
 We found that the subsurface horizons store a large fraction of the total SOC, and a significant proportion of the subsurface SOC consists of decadally cycling C pools (Table 2). Even in soil with the shallowest rooting depth (zroot = 0.11 m), a significant fraction of the total SOC (36% — data from original settings in Table 2) was found in the subsurface (z = 0.3 – 5 m) layers. The fraction of the SOC in the subsurface increased up to 78% with increasing rooting depth (Table 2). The three-pool-model results further demonstrate that a large proportion of the subsurface SOC (39% – 73%, depending on zroot) consists of the active and slow C (data from original settings in Table 2), which turns over on a time scale of decades or less. The results are consistent with previous observations, suggesting that subsurface C cycling can be significant on decadal time scales in a range of ecosystems [Richter et al., 1999; Trumbore et al., 1995; Baisden et al., 2002b; Koarashi et al., 2012].
 Owing to the large stock of the fast-cycling (active, and slow) C pools in the subsurface to deep horizons, a surprisingly large proportion of the mineralized C in the 5 m of soil comes from the subsurface SOC (Table 3). In the shallowest zroot = 0.11 m case, 17% of the total C (as CO2) flux from the entire 5 m soil profile was delivered from the subsurface layers (Table 3). The contribution of the subsurface increased significantly to 70% with increasing rooting depth. Clearly, the results indicate the importance of subsurface horizons as a large source of mineralized C.
Table 3. Mineralized C Fluxa (g-C m−2 yr−1) at 1 ky After Initiating SOC Development
aFlux includes the SOC and DOC decomposition within the sectioned soils.
bSum of the C fluxes from the active, slow, and passive C.
cRatio for the total (active + slow + passive) C flux.
zroot = 0.11 m
zroot = 0.41 m
zroot = 0.99 m
 With global environmental changes, a further release of C from the subsurface soil layers can be expected, due to several processes. In areas where land use change brings a shift of vegetation [Jackson et al., 2002], the SOC which is physically protected by the interaction with soil constituents (corresponding to the slow to passive SOC in this model) could potentially be disrupted through the expansion of plant roots and fungi hyphae [Baisden et al., 2002a], and the addition of fresh C (root litter and root exudates) can also stimulate the decomposition of SOC anciently stored in soil (priming effect) [Fontaine et al., 2007; Kuzyakov, 2010]. Furthermore, several studies indicate that the more recalcitrant SOC is sensitive to temperature, and hence, the decomposition of recalcitrant C can be accelerated by warming [Davidson and Janssens, 2006]. Together, these studies imply that a large amount of SOC stored in the subsurface horizons can be destabilized with changes in climate, land use, and vegetation, and thus could have significant impacts on global C cycling.
4.2.3 Implications from the Vertically Extended Soil C Study
 Our results relating to the development of SOC have implications for a better prediction of the responses of soil C to global changes. The rapid development of the active SOC profile (see section 4.2.1 and Figure 5) indicates that this pool would be expected to follow any changes in environmental conditions within a couple of years. In contrast, the slow and passive SOC would be expected to respond to any environmental changes with a significant time lag of decades to centuries (see the development of these pools in Figure 5). In addition, the regulatory processes for maintaining stability of SOC would differ between the surface and the subsurface horizons (section 4.2.1). Together, the findings indicate the need to extend current soil C studies and models to the subsurface to deeper horizons, to enable a precise prediction of the timing and impact of soil C responses to potential environmental changes.
 The demonstrated ability of soil water transport to control the belowground C dynamics indicates the important role of regional water cycles in determining the ecosystem balance of C. Given the findings that a large fraction of soil C consists of slow and passive C pools (Table 2), and that the dynamics of the slow and passive SOC strongly depend on soil water transport (section 4.2.1), changes in regional water cycles (e.g., caused by changes in plant distribution [Kelly and Goulden, 2008] and precipitation patterns [Walther et al., 2002; Dore, 2005]) may significantly affect the ecosystem C balance over the next century. In our simulations, the doubled soil water flux led to the increased amount of subsurface soil C (e.g., fraction of the subsurface soil C increased from 36% to 50% at zroot = 0.11 m; see Table 2), and this was largely due to the increased amount of C in the slow and passive SOC pools (Table 2). Clearly, there is a need to improve our ability to predict the soil C response to global change by using coupled water and C cycle models, and the C cycle models should explicitly consider different dynamics, (including the sources, size, turnover, transport, and their controlling factors), of different SOC pools in vertically extended soil profiles.
4.2.4 Root-uptake Control of Belowground C Cycle
 Our results demonstrate that the uptake of soil water (and thus DOC) by roots may control belowground C cycling. These results were not expected from our hypothesis that the decadal cycling of C in the subsurface and deeper soil horizons are delivered by the downward DOC flow (section 1). Under a quasi-equilibrium state, the total SOC stock in the entire 5 m of soil differed by zroot (5.18, 6.25, and 7.13 kg-C m−2 for zroot = 0.11 m, 0.41 m, and 0.99 m, respectively — data from the original settings in Table 2). This difference is attributed to the higher root uptake of soil C in the shallower zroot case (the amount of soil C taken up by the root in the entire 5 m of soil was 114, 70, and 47 g-C m−2 yr−1 for zroot = 0.11, 0.41, and 0.99 m, respectively), which is caused by a higher root uptake of the soil water (er) coupled with a higher content of the soil C in the shallower soil layers (Figure 6). The results indicate that the turnover of organic C in the soil is not only controlled by microbial decomposition of soil C, but also by the uptake of the soil water (DOC) by the roots. Indeed, the total C stock in the entire soil profile increased by 71% – 108% (Table 2), when a zero root uptake of the soil water was assumed (section 3).
 Several experimental studies show that during short-term (several to ten days) laboratory incubations, soil C losses caused by root uptake are much smaller than those caused by the microbial decomposition of the SOC [Boddy et al., 2007; Rusmussen et al., 2010]. The results of the present study are consistent with these findings; the amount of C taken up by the roots (114 g-C m−2 yr−1 - 47 g-C m−2 yr−1, see above) is substantially smaller than that mineralized in the soil (422 g-C m−2 yr−1 - 505 g-C m−2 yr−1, Table 3), and the latter is derived mostly from the decomposition of the active C. However, a root uptake effect on the C dynamics appears to be significant for the slow and passive C, as can be seen by the largely increased SOC stocks for these two pools in soil with a zero root uptake (Table 2).
 Given that the root uptake control of the soil C dynamics may be surprisingly large, plant roots should be highlighted in global biogeochemical models as being, not only as a belowground C supplier, but also as a belowground C consumer. However, additional simulations with accelerated decomposition rates (ksw,i) for the slow and passive DOC pools (section 3) demonstrated weaker controls of the root uptake on the soil C dynamics. For example, for the passive C pool in the soil with zroot = 0.99 m, the mineralization of the C from the entire 5 m of soil increased from ~2 g-C m−2 yr−1 (with original settings, see Table 3) to 5 g-C m−2 yr−1 (with accelerated decomposition rates of DOC), while the root uptake of this pool decreased from 7 g-C m−2 yr−1 (with original settings) to 3 g-C m−2 yr−1 (with accelerated decomposition rates of DOC). The result indicates that the root-uptake control would be less important in ecosystem C cycling if the DOC was decomposed within years. Therefore, further experimental studies linking the degradability of the DOC, (not the turnover time), with that of the SOC — as a source of DOC (i.e., linking the parameters ksw,i and kss,i in the current model) — are needed to generalize the root uptake effect on the soil C dynamics.
 To investigate the role of root litter input and DOC transport in controlling subsurface soil C dynamics, a vertically extended soil C model predicting C turnover and transport throughout the soil profile was developed. The numerical study using a range of rooting profiles demonstrated that the downward percolation of DOC in the soil profile efficiently distributes soil C added by the root litter to the deeper parts of the soil, thereby making the profile of the SOC deeper than that of the root litter input. Model simulations also demonstrated that a large proportion of SOC contained in the entire 5 m of soil (36% – 78%, more for deeper rooting profiles) is apportioned to the subsurface horizons (below the first 30 cm), and the resultant C (as CO2) flux from the subsurface soil can be significant; amounting to 17% – 70% of the total C flux from the entire soil. Furthermore, a large proportion of the subsurface soil C (39% – 73%) is found to be associated with fast-cycling C pools (time scales of decades or less), suggesting that the subsurface soil C actively participates in global C cycling. The results indicate a need to extend the current soil C studies and models that focus on the surface alone, to the subsurface or even deeper horizons, in order to precisely predict the response of soil C to changes in climate, land use, and vegetation.
Appendix A: Estimation of the Effective Distance for Soil C Transport via DOC Percolation
 By the definition in equations (5) and (9), the SOC turnover in soil is regulated by the microbial decomposition of SOC and DOC, and root-water (DOC) uptake. The turnover times of soil C at a certain soil depth are thus expressed by:
for the microbial decomposition of C in the ith pool, and for the root-water uptake in all three pools:
 Because the microbial decomposition and the root uptake act in parallel to determine the turnover of soil C, the overall residence time (tres,i) for the soil C at the ith C pool is determined by:
 In the present simulation, tdec,i from equation (A1) becomes tdec,i = 1/kss,i = tss,i = 1.5, 25, and 1000 years for the active, slow, and passive C, respectively, because it was assumed that ksw,i = kss,i. For the root uptake process, troot was calculated with the annual averages of ηw and er in each zroot case, and it was found that troot exists for several decades in the surface (z ~ centimeters) soil layers and exists for an excess of 1000 years in the deep (z ~ meters) soil layers. With the obtained tdec,i and troot, the overall residence time tres,i was calculated for each zroot using equation (A3) (Figures A1a–A1c). For the active C, throughout the soil profile, the overall residence time (tres,1) is nearly identical to the turnover time (tdec,1 = 1.5 years) of the microbial decomposition, because the turnover time troot for the root-water uptake is significantly larger than the tdec,1 of 1.5 years. In contrast, for the slow and passive C, the overall residence times tres,i are reduced to tens to hundreds of years in the upper soil layers (smaller than their turnover time tdec,i of microbial decomposition of 25 and 1000 years, respectively), because of a small troot there.
 The efficiency of soil C transport by the downward percolation of the DOC is characterized by the effective distance (Ladv,i) where the SOC of the ith C pool is transported by the soil water flow before it is decomposed by microbes or taken up by roots:
 With a soil water flux of Ew = 6 × 101 mm yr−1 (annual mean flux over the top meter of soil), the effective distance Ladv,1 for the active C with tres,1 ~ 1.5 years (Figure A1) is calculated to be ~0.1 m at nearly all depths. For the slow C with tres,2 ~ 10 – 25 years in the top meter of soil (Figure A1), Ladv,2 is calculated to be ~0.6 – 1.6 m. A similar analysis is also valid for the passive C, with long overall residence times (tres,3) that increase with depth up to millennia (Figure A1).
Appendix B: Estimation of the Effective Distance of Soil C Transport via Bioturbation
 Translocation of soil material within the soil profile by bioturbation acts like diffusion because of the random movement of soil faunal organisms [Elzein and Balesdent, 1995; Braakhekke et al., 2011]. Thus, the effective distance Lbio,i (m) over which the SOC of the ith C pool is transported via bioturbation during its overall residence time (tres,i, see Appendix A and Figure A1) can be expressed by:
where Dbio (m2 s−1) is the effective diffusivity for bioturbation. Given the maximum effective diffusivity of 17 cm2 yr−1 (Dbio = 0.4 – 17 cm2 yr−1 obtained in a range of ecosystems by Elzein and Balesdent  and Van Dam et al. ), the effective distance is calculated to be Lbio,1 = 0.05 m for the active SOC with an overall residence time tres,1 of 1.5 years (Figure A1).
 We thank Drs. M. Atarashi-Andoh and T. Nakanishi for their helpful discussions relating to the simulation results, and Drs. D. Baldocchi, T. Hehn, and S. Ma for providing the meteorological data set used in the model validation.