2.1 Model description

JSM is a soil biogeochemical model built on the backbone of the vertically explicit C-only SOC model COMISSION (Ahrens et al., 2015). The COMISSION model was further developed from the conventional one by introducing a scalable maximum sorption capacity based on soil texture for dissolved organic C and microbial residues (Sect. S1.4 in the Supplement) as well as temperature and moisture rate modifiers for microbe-mediated processes and sorption (Sect. S1). We will investigate in a separate study how the maximum sorption capacity for mineral-associated organic C contributes to the observed patterns of SOC content and radiocarbon age. A schematic overview of JSM is presented in Fig. 1, and the mathematical description of the processes is provided in the Supplement. The model is integrated into the QUINCY (Thum et al., 2019) TBM modelling framework and can either be applied as a stand-alone soil model or coupled to the representation of the vegetation and surface processes. In this study, we applied JSM as a stand-alone model. JSM neither describes the energy and water processes at the atmosphere–soil interface or in the soil profile nor simulates the production of litter. Model inputs (soil temperature, moisture and water fluxes as well as plant litter data) were derived from the QUINCY model.

JSM describes the formation and turnover of SOM along a vertical soil profile, which is explicitly represented as exponentially increasing layer thickness with increasing soil depth (Fig. 1). The biogeochemical processes and pools of C, N and P are represented in each layer. Vertical transport of biogeochemical pools between the adjacent layers due to percolation and bioturbation is also modelled. To reflect the development of an organic layer, the model also includes an extra advective transport term which accounts for the upwards/downwards shift of the soil surface when the surface SOM accumulates/diminishes.

Figure 2Simulated dynamics of (a) SOC and (b) non-occluded inorganic P contents in topsoil (30 cm) and subsoil (30–100 cm) for 10 000 years. The three vertical dashed lines represent 200, 1000 and 5000 years.

Download

SOM is represented as pools of soluble, polymeric or woody litter as well as of dissolved organic matter (DOM), mineral-associated DOM, living microbial biomass, microbial residue (necromass) and mineral-associated microbial residue, each of which contains organic forms of C, N and P. The flows of organic N and P follow the pathways of C, with additional nutrient-specific processes, such as mineralisation and plant uptake, to link organic matter turnover with inorganic nutrient cycles. Microbial biomass is assumed to maintain a fixed stoichiometry in the model. It assimilates organic forms of C, N and P from DOM with fixed element use efficiencies and inorganic forms of N and P from soluble mineral pools. Microbes are assumed to aim to maximise their growth by maintaining high C use efficiency; however, when growth is limited by nutrients, microbes reduce their C use efficiency and increase nutrient mineralisation accordingly (See Sect. S1.5). The stoichiometry of all other SOM pools depends on the $\mathrm{C}:\mathrm{N}:\mathrm{P}$ ratios of influx and efflux, and these fluxes retain the stoichiometry of their source pools unless the formation processes involve respiration. In addition, when microbes decay, nutrients are preferentially recycled to the DOM pool due to the low C-to-nutrient ratio in the cyctoplasma, as proposed by Schimel and Weintraub (2003). The inorganic pools of N and P include soluble inorganic ammonium (referred to as NH₄), nitrate (referred to as NO₃), soluble inorganic phosphate (referred to as PO₄), as well as adsorbed PO₄, absorbed PO₄, occluded PO₄ and primary PO₄. The inorganic P cycle follows the QUINCY model (Thum et al., 2019) and accounts for microbial interactions. JSM explicitly traces ¹³C, ¹⁴C and ¹⁵N following Ahrens et al. (2015) and Thum et al. (2019).

Enzymes are not explicitly modelled in JSM, although these are described implicitly to regulate processes such as depolymerisation and nutrient acquisition. For enzyme allocation within depolymerisation processes, we extended the microbial adaptation approach of the SEAM model (Wutzler et al., 2017) by including P dependence of enzyme allocation and the assumption of a steady state of enzyme production, leading to the prediction that the total enzyme level is always proportional to the microbial biomass. The fractions of enzymes allocated to different depolymerisation sources (litter and microbial residue) are dynamically modelled to maximise the release of the most limiting microbial elements. JSM tracks three potential fractions of enzyme allocation, which represent cases in which microbes only maximise the depolymerisation release of C, N or P, respectively, and then updates the microbial enzyme allocation fraction by acclimating gradually to the potential fraction of the most limiting element (See Sect. S1.5.2). For nutrient competition between plant, microbes and soil adsorption sites (only for phosphate), the potential rates are calculated on the basis of the respective richness and half-saturation level of enzymes and the impacts of other competitors, following the ECA approach (See Sect. S2.2).

The impacts of soil conditions on biogeochemical processes are also represented in JSM. The temperature response of different processes (e.g. microbial growth, decay, and nutrient uptake in Sect. S1.4) are represented by the Arrhenius equation with different activation energies. Moisture responses are described by two rate modifiers – one representing the effects of oxygen limitation (e.g. litter turnover in Sect. S1.2) and the other representing the effects of diffusion limitation (e.g. depolymerisation in Sect. S1.3). JSM also considers the effects of SOM content to correct bulk density (Sect. S3), which in turn affects other processes such as organic matter (Eq. S7) and phosphate (Eq. S25) sorption.

2.2 Site description and data for model analysis

The Vessertal (VES) site is a mature beech (Fagus sylvatica) forest stand with an average tree age of >120 years, located in the central uplands of Germany (Thuringian Forest mountain range). The intermediate elevation is 810 m a.s.l., with a high annual precipitation of 1200 mm and a mean annual temperature of 5.5 ^∘C (Lang et al., 2017). It is one of the Level II intensive monitoring plots in the Pan-European International Co-operative Program for the assessment and monitoring of air pollution effects on forests (ICP Forests). Since 2013, the VES site has also been one of the main study sites in the German Research Foundation (DFG) priority programme 1685 “Ecosystem Nutrition: Forest Strategies for Limited Phosphorus Resources”.

Soil at the VES site is classified as Hyperdystric Skeletic Chromic Cambisol (WRB, 2015), with loamy topsoil and sandy loamy subsoil, overlain by a moder organic layer. The current soil developed on trachyandesite, and the development started at the end of the last ice age, 10–12 000 years ago (Lang et al., 2017). The soil was sampled up to 1 m, with layer depths of 5–10 cm, for the measurements of total C, N, and organic and inorganic P and basic physical properties such as bulk density and soil texture. Soil from the A horizon alone was extracted for the estimation of microbial C, N and P pools. Detailed sampling and measurement approaches are described in Lang et al. (2017).

The soil contains 19 kg m⁻² C, 1.1 kg m⁻² N and 464 g m⁻² P up to 1 m soil depth, including the forest floor (Lang et al., 2017). The soil C content decreases from 510 g C kg⁻¹ soil in the forest floor to 126 g C kg⁻¹ soil in the A horizon to 5.9 g C kg⁻¹ soil at 1 m depth. The C:N ratio of SOM slightly decreases from 19.5 in the forest floor to 14.75 at 1 m depth, whereas the C:P ratio decreases more steeply from 348.7 in the forest floor to 46.6 at 1 m depth. The organic P fraction of total P also decreases from 67 % in the organic layer to 13 % at 1 m depth. The microbial C content decreases from >2000 µg C g⁻¹ soil C in the forest floor (Zederer et al., 2017) to 764 µg C g⁻¹ soil C in the top mineral soil (Bergkemper et al., 2016). The microbial biomass shows a C:N ratio of 13 and a very low C:P ratio of 10.3 (Lang et al., 2017).

2.3 Model protocol, model experiments and sensitivity analysis

2.3.2 Model experiments

To further test the effects of different model features, we implemented several model experiments, including a SEAM-off scenario in which the enzyme allocation to polymeric litter and microbial residue are both fixed to 50 % (Eq. 1b), and an ECA-off scenario in which the ECA-based plant and microbial uptakes of inorganic N and P and soil adsorption of phosphate were switched off and replaced by a demand-based microbial uptake of inorganic N and P that ignored P adsorption flux (Eq. 1c). All model experiments used the same parameterisation from the calibrated model with full model features, which is denoted as the Base Scenario in this study.

The differences between Base Scenario and SEAM-off and ECA-off are listed below.

Base Scenario:

$$\begin{array}{}\text{(1a)}& \begin{array}{rl}& {\mathrm{Enz}}_{\mathrm{frac}}^{\mathrm{poly}}\mathit{\&}{\mathrm{Enz}}_{\mathrm{frac}}^{\mathrm{res}}\mathrm{calculated}\mathrm{as}\mathrm{Eq}.\left(\mathrm{S}\mathrm{15}\right);\\ & U{*}_{X,y}\mathrm{for}\mathrm{microbes},\mathrm{plant}\mathrm{and}\mathrm{adsorption}\\ & \mathrm{calculated}\mathrm{as}\mathrm{Eq}.\left(\mathrm{S}\mathrm{23}\right).\end{array}\end{array}$$

SEAM-off Scenario:

$$\begin{array}{}\text{(1b)}& {\mathrm{Enz}}_{\mathrm{frac}}^{\mathrm{poly}}={\mathrm{Enz}}_{\mathrm{frac}}^{\mathrm{res}}=\mathrm{0.5}.\end{array}$$

ECA-off Scenario:

$$\begin{array}{}\text{(1c)}& \begin{array}{rl}& U{*}_{X,\mathrm{plant}}=f(T_{\mathrm{soil}},\mathrm{\Theta })v_{\max ,\mathrm{plant}}^X{C}_{\mathrm{fine}\mathrm{\_}\mathrm{root}}\left[X\right]\\ & (K_{m\mathrm{1}}^{\mathrm{upt}}+{\displaystyle \frac{\mathrm{1}}{K_{m\mathrm{2}}^{\mathrm{upt}}+\left[X\right]}}),\\ & U{*}_{X,\mathrm{mic}}=F_{\mathrm{mic},X}^{\mathrm{demand}},\\ & U{*}_{P,\mathrm{adsorp}}=\mathrm{0}.\end{array}\end{array}$$

The plant uptake of inorganic N or P ($U{*}_{X,\mathrm{plant}}$) in the ECA-off scenario (Eq. 1c) uses the equations and parameters from the QUINCY model (Thum et al., 2019).

Table 1The annual soil C, N and P fluxes of model experiments at the study site. Positive values infer accumulation in the soil, and negative values infer loss from the soil. The values are the accumulated sum of the whole soil profile, calculated based on data from the last 10 years of model experiments. SOP: soil organic phosphorus; SON: soil organic nitrogen.

Download Print Version | Download XLSX

2.3.3 Calibration and model sensitivity

We calibrated the Base Scenario in two main steps. In the first step, we matched the model results with the measured SOC profile, mainly by calibrating the depolymerisation, organic matter sorption and litter turnover processes; in the second step, we matched the model results with the measured soil organic P profiles by calibrating the microbial growth and decay, nutrient acquisition, and soil inorganic P cycling. The two steps were not performed iteratively; however, during the second step, we revised the parameters from the first step as necessary. Other observed soil profiles, such as the soil organic N and the bulk density, were used as additional criteria to select parameterisation, although not specifically used to calibrate the model. During the calibration processes, parameter values were gradually changed and the goodness of model fit was visually evaluated on the basis of observations.

To test the effects of different microbial stoichiometry, we ran a “Glob Mic Stoi” scenario in which the global average microbial stoichiometry ($\mathrm{42}:\mathrm{6}:\mathrm{1}$, Xu et al., 2013) was used to parameterise the model instead of the observed microbial $\mathrm{C}:\mathrm{N}:\mathrm{P}$ ratio ($\mathrm{10.3}:\mathrm{0.8}:\mathrm{1}$, Lang et al., 2017). To further test the model responses to different initial conditions, we ran the model with different initial SOM contents (50 %, 75 %, 150 % and 200 % of the default initial content) for 1000 years to ensure that the soil reached a more stable state.

We also tested the sensitivity of JSM to parameterisation using a hierarchical Latin hypercube design (LHS; Saltelli et al., 2000; Zaehle et al., 2005). We selected 28 parameters from calibration (Table S1) and varied each parameter between 80 % and 120 % of the Base Scenario values given in Table S2, which were obtained through LHS sampling from a uniform distribution to form a set of 1000 LHS samples and used in model sensitivity and uncertainty analyses presented in this paper. We evaluated the model output from these simulations in terms of main biogeochemical fluxes (e.g. respiration, net N and P mineralisation, microbial uptake of inorganic N and P, N and P losses, and P biomineralisation) and the main SOM pools (up to 1 m depth) (e.g. total C, N and P in SOM; total soil inorganic P and microbial C, N and P). We measured parameter importance as the rank-transformed partial correlation coefficients (RPCCs) to account for potential non-linearities in the association between model parameters and output (Saltelli et al., 2000; Zaehle et al., 2005).

Figure 3Simulated and observed (a) SOC content, (b) C:N ratio in SOM, (c) C:P ratio in SOM, (d) organic-P-to-inorganic-P ratio in soil, (e) microbial C content, (f) microbial N content, (g) microbial P content and (h) soil bulk density at the study site up to 1 m soil depth. Black lines and dots: observations; coloured lines and shades: simulated mean values and ranges of standard deviation by different model experiments in Sect. 2.3. Microbial C, N and P values were only measured in the top 30 cm of soil. Simulated means and standard deviations were calculated using data from the last 10 years of model experiments.

Download

3.1 Model stability and quasi-equilibrium state

We tested JSM with different initial SOM contents and different microbial stoichiometry. The simulated SOM profiles, including SOC; C:N and C:P ratios of SOM; microbial C, N and P contents, and bulk density, did not respond strongly to changes in initial SOM contents (Fig. S2) but were notably affected by the assumed microbial stoichiometry (Fig. S1). We further examined the effects of simulation time on soil profile development (Figs. 2 and S1). SOC in the topsoil (30 cm) reached a stable state (ca. 70 kg C m⁻³) after approximately 150 years and the subsoil (30–100 cm) reached a stable state (ca. 30 kg C m⁻³) after approximately 1000 years. The accumulation rate of SOM decreased with time, but the complete soil profile had not yet reached a steady state (Table 1) because C continues to accumulate slowly, particularly in deeper soil layers (>1 m). Although the organic P dynamics follow the soil C dynamics, the inorganic P pools inevitably deplete in the long-term simulation (Fig. 2) due to high uncertainties in initialisation and P cycling processes. Therefore in this study, we focussed on the stable state of the topsoil (30 cm) at the end of the 200-year simulations and referred to it as a “quasi-equilibrium state” since slow changes are still occurring, particularly in soil inorganic P pools and SOM in deeper soil layers (Figs. S3, S4 and S5).

3.2 Simulated SOM stocks and fluxes at the study site

We first compared the simulated profiles with the in situ observed ones (Fig. 3). The modelled results agreed well with observed stock sizes and vertical patterns, indicating that the stocks (here we define the term “stock” as the total amount of all (model) pools within a larger set) of C, N and P pools in SOM show smaller temporal variations than the microbial pools at the quasi-equilibrium state (Fig. 3a to c) due to strong seasonal variations in microbial biomass. We also found a greater variation in the simulated organic-P-to-inorganic-P (Po-to-Pi) ratio (Fig. 3d) than for the individual organic and inorganic P stocks (data not shown), inferring that the seasonal dynamics of microbes also impose a seasonal pattern of P immobilisation (from Pi to Po) and mineralisation (from Po to Pi).

The distribution and radiocarbon profile of total organic matter in the simulations varied across soil depths (Fig. 4). The first layer (0 cm, O–A horizon) is dominated by the plant litter and microbial component (living or dead microbes), and while the microbial component decreases strongly from ca. 40 % at 0 cm to almost zero at 50 cm soil depth, the litter component still constitutes ca. 10 % of the total SOC at 1 m soil depth. The mineral-associated C (MOC) component switches from a minor component in the O–A horizon (ca. 20 %) to the dominant component (ca. 90 % at 1 m) in deeper soil layers.

Figure 4Simulated SOC fractions (upper panels) and their respective radiocarbon profiles (bottom panels) at 1 m soil depth. Column (a): mineral-associated C (MOC), including adsorbed DOM and adsorbed microbial residue; column (b): litter, including woody, polymeric and soluble litter; column (c): live and dead microbes. Data points are derived using data from the last 10 years of the model experiments. All model experiments used 200-year simulations and were not validated against the measured Δ¹⁴C.

Download

The simulated radiocarbon (Δ¹⁴C) profile qualitatively agreed with the observed one (Fig. S1e); the Δ¹⁴C content increased within the O horizon and started decreasing with increasing soil depth from mineral soil, i.e. the A horizon. This pattern indicates that the “bomb pulse” Δ¹⁴C signal significantly affects the apparent ¹⁴C age in the organic layer and its impact decreases with soil depth due to the slow turnover in deeper soil. Our simulations further indicated that such a vertical pattern is caused by MOC and microbial components, while the litter component stays modern throughout the profile (Fig. 4). Increases in litter ¹⁴C with depth suggest that more bomb-derived SOC is still found in subsoils due to slower litter turnover, while it is already replaced by more recent, ¹⁴C-poorer SOC in the topsoil. Although the Base Scenario did not reproduce the measured radiocarbon profile, despite producing its vertical pattern, a much better fit with the measured radiocarbon profile and an increase in soil ¹⁴C age, driven by MOC, were indeed observed as simulation time increased (Figs. S1e and S4).

The simulated biogeochemical fluxes show strong seasonal and vertical patterns (Figs. 5 and 6), in which the flux rates in summer and in the topsoil are generally higher than those in winter and in the subsoil, respectively. Meanwhile, microbial inorganic N uptake shows a different seasonal pattern, with the lowest rates observed in August and September (Fig. 6c). In fact, this pattern is supported by the seasonal pattern of net N mineralisation flux, in which the peak is observed in August and September (Fig. 5b). This result indicates that organic N in DOM is the most abundant for microbial growth during August and September, leading to a large reduction in the microbial inorganic N uptake and an increase in net N mineralisation. In contrast, organic P content in DOM is the lowest during August and September, leading to net P immobilisation and microbial inorganic P uptake elevation (Figs. 5d and 6a). While the vertical pattern of plant N uptake parallels root distribution (Jackson et al., 1996), plant P uptake is lower in the organic layer than in the topsoil due to strong competition from microbes in the organic layer (Figs. 6 and 8).

Figure 5Simulated seasonal and vertical distribution of (a) respiration, (b) net N mineralisation, (c) biochemical P mineralisation and (d) net P mineralisation at the study site at 1 m soil depth. Points represent the mean values and error bars represent the standard deviations, both calculated using data from the last 10 years of model experiments.

Download

Figure 6Simulated seasonal and vertical distribution of (a) microbial inorganic P uptake, (b) plant P uptake, (c) microbial inorganic N uptake and (d) plant N uptake at the study site at 1 m soil depth. Points represent the mean values and error bars represent the standard deviations, both calculated using data from the last 10 years of model experiments.

Download

The sources and sinks of soluble inorganic N and P also show very different patterns (Fig. 7). The main source and sink for inorganic N in solution are gross mineralisation and plant uptake of NH₄, respectively, whereas for P, microbial uptake is the main sink and biomineralisation is a larger source than gross mineralisation in each scenario.

Figure 7Simulated yearly budget of (a) N and (b) P in soil solutions. In (a), sourcing fluxes of N are presented in the order of gross mineralisation of NH₄ and NO₃ and N deposition (in the bar plot: from right to the zero point; in the legend: from the top to the separation line); sinking fluxes of N are presented in the order of plant and microbial uptakes of NH₄, plant and microbial uptakes of NO₃, N leaching (both inorganic and organic), and changes in soluble N content (delta_sol_N) (in the bar plot: from left to the zero point; in the legend: from the separation line to the bottom). In (b), sourcing fluxes of P include weathering, gross mineralisation of PO₄, biochemical mineralisation of PO₄ and P deposition; sinking fluxes of P includes adsorption (Exchange_fast), microbial and plant uptakes, P leaching (both inorganic and organic), and changes in soluble P content (delta_sol_P). (The order of presented processes follows the same rule as N.) The budgets are calculated using data from the full simulation (200 years) of the model experiments.

Download

3.3 Model features: nutrient acquisition competition and enzyme allocation

In general, the SEAM-off scenario did not differ much from the Base Scenario in terms of the main soil stocks and biogeochemical fluxes (Figs. 3 and 5); however, the ECA-off scenario produced a lower microbial biomass and Po-to-Pi ratio in the organic layer and topsoil. Total SOC may not be influenced in either scenario, although its composition and radiocarbon profile were both altered (Fig. 4).

We presented the uptake of inorganic PO₄ and competition between phosphate adsorption, uptake of inorganic P at three different depths (Fig. 8), and seasonal and vertical uptakes of inorganic N and P for both microbes and plants (Fig. 6). The simulations showed that microbes outcompeted roots for inorganic P uptake in JSM at all depths. However, the relative competitiveness of roots for phosphate uptake increased with increasing soil depth because the plant P uptake rate decreases less strongly than the microbial P uptake with increasing soil depth. In contrast, the phosphate adsorption rate increased strongly with increasing soil depth and outcompeted biological processes (plant and microbial uptake) in deeper soil layers. The relative competitiveness of phosphate adsorption against microbial and plant uptake also strongly decreased in summer in the topsoil due to high biological activity in warm months (Fig. 8b). With respect to competition for inorganic N, plants outcompeted microbes along the entire soil profile and throughout the year, particularly in summer when microbes assimilate N mainly from DOM (Fig. 6c and d).

Figure 8Simulated (a) microbial and plant P uptake rates and (b) relative competitiveness (in fractions) of P adsorption, microbial P uptake and plant P uptake at depths of 0 (O–A horizon, upper panels), 15 (A–B horizon, middle panels) and 80 cm (B–C horizon, bottom panels). In (a), monthly mean values at different depths are presented throughout the whole year; in (b), relative competitiveness is calculated as the fraction of the individual rate to the sum of all three rates (P adsorption rate, microbial P uptake and plant P uptake). All data points are derived from data from the last 10 years of model experiments.

Download

Turning off the model's feature for nutrient acquisition competition, i.e. ECA-off scenario, led to a notably lower microbial biomass and Po-to-Pi ratio in the topsoil (Fig. 3). This is caused by concurrent changes in microbes and plant inorganic P uptake, particularly in the topsoil layer where plants take up more inorganic P than in the Base Scenario (Fig. 6) due to reduced competition with microbes. Moreover, there were differences in spatial and temporal variations in uptake and mineralisation fluxes between the ECA-off scenario and the other two scenarios. For instance, the seasonal variation in fluxes was notably lower in the ECA-off scenario. Decrease in P flux rate with soil depth may be weaker in the ECA-off scenario, but the decrease in net N mineralisation with soil depth is marginally stronger (Fig. 5). This difference is because geophysical processes, such as adsorption and absorption, play more crucial roles in the soil P cycle than in the N cycle and because these show rather different seasonal and vertical patterns from the biochemical processes, such as mineralisation.

Table 2The five most important parameters (Par) and their respective RPCCs for each output variable and the overall model importance (OVI). RPCCs were calculated for each output variable, and the overall importance of parameters was measured by calculating the mean of the absolute RPCCs across all output variables, weighted by the uncertainty contribution of these model outputs. The parameters are listed in Table S2 and explained in Table S1. SOP: soil organic phosphorus; SIP: soil inorganic phosphorus; SON: soil organic nitrogen.

Download Print Version | Download XLSX

The modelled enzyme allocation for depolymerisation is presented in Fig. 9. We compared the enzyme allocation curve of polymeric litter (${\mathrm{Enz}}_{\mathrm{frac}}^{\mathrm{poly}}$ in Eq. S17) with three potential allocation curves (${\mathit{\alpha }}_{\mathrm{poly}}^X$ where X stands for C, N and P in Eq. S15), which represent cases in which microbes only maximise C, N or P release from depolymerisation. All modelled fractions of enzyme allocation to polymeric litter were well below 50 % for the whole soil profile, indicating that polymeric litter is less preferred than microbial residues for depolymerisation in the soil, particularly in very deep soil layers where no roots are present and microbes would thus only produce enzyme to depolymerise microbial residues because the content of residue is much higher than that of polymeric litter. The simulated curve of allocation overlaps with the curve of potential allocation to maximise P release, indicating that depolymerisation is solely driven by P demand. This explains why microbial residues are preferred over polymeric litter since the C:P ratio of microbial residues is much lower than that of polymeric litter (data not shown). Despite rather different enzyme allocation fractions shown in Fig. 9, the majority of the modelled stocks and fluxes were not significantly influenced when enzyme allocation was turned off (Figs. 3 and 5). More profound differences were observed in the composition and radiocarbon profile of SOC; there was less litter and more SOC in the SEAM-off scenario than in the Base Scenario, resulting a systematic difference in the radiocarbon profiles between the two scenarios (Fig. 4).

Figure 9Simulated enzyme allocation to polymeric litter compared with potential enzyme allocations to polymeric litter given that the element C, N or P is the most limiting. Black lines: fraction of enzyme allocated to polymeric litter; orange, blue or green lines: potential enzyme allocation to polymeric litter to maximise C, N or P, respectively. (a) Base Scenario; (b) SEAM-off scenario; (c) ECA-off scenario. All data points are derived from data from the last 10 years of model experiments.

Download

3.4 Model sensitivity and uncertainties

The interquartile range of outputs (Fig. 10) from model sensitivity analysis revealed that all outputs were well centred around the results of the parameterisation of the Base Scenario (Table S2), except microbial inorganic N uptake and N losses. In general, the soil stocks were more stable than the microbial pools and biogeochemical fluxes. N mineralisation was surprisingly insensitive while microbial inorganic N uptake was very sensitive to parameterisation. N mineralisation in JSM was mainly driven by the C:N ratio of DOM, which remains rather stable due to the similar C:N ratios of plant litter, microbes and microbial residues. The very sensitive response of microbial inorganic N uptake was attributed to the high-affinity (low K_m,mic value) N uptake transporters of microbes (Kuzyakov and Xu, 2013) and their sensitivity to changes in NH₄ concentration. The RPCC of parameters with outputs (Table 2) also demonstrated that the C and N contents of SOM as well as the C and N fluxes were more sensitive to changes in C processes, such as depolymerisation, organic matter sorption and litter partitioning, while the microbial dynamics and the P fluxes were more sensitive to changes in microbial and nutrient processes, such as maximum biomineralisation rate (v_max,biomin) and P recycling during microbial decay (${\mathit{\eta }}_{\mathrm{res}\to \mathrm{dom}}^{\mathrm{P}}$). Overall, most of the selected outputs were strongly influenced by microbial stoichiometry. The five most influencing parameters in JSM were microbial C:N ratio (${\mathit{\chi }}_{\mathrm{mic}}^{\mathrm{C}:\mathrm{N}}$), microbial N:P ratio (${\mathit{\chi }}_{\mathrm{mic}}^{\mathrm{N}:\mathrm{P}}$), microbial mortality rate (τ_mic), soluble litter C fraction transformed into DOM (${\mathit{\eta }}_{\mathrm{C},\mathrm{sol}\to \mathrm{dom}}$), and P fraction recycled from res to dom during microbial decay (${\mathit{\eta }}_{\mathrm{res}\to \mathrm{dom}}^{\mathrm{P}}$).

Figure 10Normalised output variations in the LHS sensitivity analysis. The selected output variables include respiration; total soil organic C, N and P; microbial C, N and P; net N mineralisation; microbial N uptake; net P mineralisation; biomineralisation of P; microbial P uptake and N and P losses. All the calculations are performed for the topmost 1 m of soil based on data from the last 10 years of 1000 LHS simulations.

Download

4.1 Features of nutrient cycling

4.2 Key features and model limitations

We applied the ECA approach described by Tang and Riley (2013) to simulate inorganic nutrient competition. In general, our model simulations indicated that microbes take up more inorganic P than plants, which supports the findings of ³³P tracer experiments at two other beech forests in Germany (Spohn et al., 2018). However, our study showed that plants take up more inorganic N than microbes (Figs. 7a and S1). This pattern seems to disagree with the findings of field studies of ¹⁵N addition (e.g. Bloor et al., 2009; Dannenmann et al., 2016) and a modelling study using the same approach to simulate competition (Zhu et al., 2017). The reason for this disagreement is that in JSM, we assumed high microbial N use efficiency from DOM, and the majority of microbial N assimilation was actually fulfilled by DOM uptake. Therefore, plants take up more inorganic N than microbes. However, in ¹⁵N tracer experiments and a model study by Zhu et al. (2017), there was no distinction between assimilation from organic and inorganic sources; thus, microbes outcompete plants in the sense that the total N assimilated by microbes exceeds the total N taken up by plant roots, which was also true in our study. Another uncertainty related to the plant–microbe competition for inorganic N is the microbial stoichiometry we used in parameterisation. As discussed in the previous section, a change in microbial stoichiometry from the observed field value to the global average value resulted in a switch from microbes outcompeting plants for inorganic N to the opposite trend. Additionally, the choice of microbial nutrient use efficiencies not only affected the microbial demand for inorganic nutrients and the concentrations of inorganic N and P, thereby influencing the potential uptake rates of microbes and roots.

We extended the enzyme allocation approach of the SEAM model (Wutzler et al., 2017) by including P dependence and vertical explicitness and by assuming a steady state of enzyme production. Due to the very small microbial C:P ratio used in model parameterisation, our results indicated that depolymerisation is solely driven by P demand; thus, microbial residues are the preferred substrate because they have a much lower C:P ratio than polymeric litter. This is also supported by the massive P biomineralisation flux (Fig. 7) independent of depolymerisation and gross mineralisation, indicating that microbial growth is strongly P limited. Even in the scenario using the global microbial stoichiometry, depolymerisation was still solely P driven, and P biomineralisation fulfilled over half of the microbial P demand (Fig. S5). This result is partly supported by the global enzymatic activity data, in which global ratios of specific C, N and P acquisition activities converged on $\mathrm{1}:\mathrm{1}:\mathrm{1}$ (Sinsabaugh et al., 2008), while the global microbial stoichiometry was much higher, indicating that relatively more resources are allocated to acquire P than to acquire N and C. This result actually reveals a caveat in the current implementation of enzyme allocation in JSM: the main process via which organic P is hydrolysed, biomineralisation and the mobilisation of sorbed inorganic P due to root exudation are not included in the enzyme allocation calculation. It also explains the very small difference between the Base Scenario and the SEAM-off scenario.

JSM demonstrated a capacity to reproduce the vertical patterns of soil stocks (Fig. 3) and to satisfactorily produce both vertical and seasonal patterns of biogeochemical fluxes (Figs. 5 and 6). While the seasonal patterns are primarily driven by the temperature response of the represented processes, the vertical patterns are shaped by the combined effects of biochemical and geophysical factors represented in the model. As seen in Figs. 3 and 4, although total SOC decreased with soil depth, the microbial, litter and MOC components showed very different patterns. Following the COMISSION model (Ahrens et al., 2020), we constrained the capacity of organo-mineral association with silt and clay contents and soil bulk density in JSM. In the organic layer and topsoil, the continuous litter input sustains a large microbial biomass and microbial residue pool; however, due to the very low bulk density and relatively low silt and clay contents, sorption is weak and MOC content is very low. As soil depth increases, bulk density and silt and clay contents increase such that microbial residues and DOM stabilise to a greater extent. This hinders microbial DOM assimilation and nutrient immobilisation, leading to a strong decline in microbial biomass and an increase in MOC. As a consequence of the decreasing microbial biomass and litter inputs, much less microbial residue and DOM are available for sorption to the mineral soil, which explains the observed decrease in total SOC in deep soil layers.

Nonetheless, certain caveats of this study and JSM should be discussed. A main challenge is the different simulation times for different purposes. Our results indicated that in the upmost 30 cm of soil, SOM content stabilises after 150 years, while in the upmost 1 m SOM stabilises after 1000 years of simulation (Fig. 2), regardless of the initial SOM content (Fig. S2). However, with respect to the radiocarbon profile, as indicated by Ahrens et al. (2015), a very long simulation time (13 500 years) was required to match both the measured Δ¹⁴C and SOC profiles at a nearby Norway spruce forest site. In our study, a 10 000-year simulation time was still not sufficient to match the measured Δ¹⁴C profile, indicating that an even longer simulation time is required. Although JSM is very stable in the long term in terms of SOM development and storage, long-term simulation of soil P balance as a result of continuous weathering and occlusion remains a significant challenge (Fig. 2, Table 1). Such a long simulation time is unrealistic for the P cycle due to the unknown conditions of the initial soil P pools and the un-equilibrated soil inorganic P cycling processes (Yang et al., 2014b). Although we used a much shorter simulation length in this study, noticeable uncertainties remain due to inorganic P cycling parameters (Table 2). Additionally, the long simulation time required to match the radiocarbon profiles is also problematic for future coupling to TBMs because these models typically examine centennial timescales. A possible solution is to spin up radiocarbon (>10 000 years) independent of the plant–soil spin-up (1000 years), although this approach needs to be properly tested in the future.

Another caveat involves the model's representation of microbial adaptation schemes. In JSM, we describe enzyme allocation, which is one of the schemes of microbial adaptation proposed by Mooshammer et al. (2014); however, as discussed above, enzyme allocation to phosphatases might be essential and might thus need to be included. Additionally, we found out that another adaptation scheme, the microbial community shift between fungi and bacteria, is crucial for reproducing the vertical pattern of soil stoichiometry. Although we mimicked such a shift in this study by calibration and parameterisation, a more mechanistic representation is necessary in the future for representing the acclimation of microbial functional properties to climate and environmental changes.

Concerning the model's description of N dynamics, in the current version, N processes such as nitrification–denitrification and abiotic ammonium adsorption are not yet implemented. Although the simplified N dynamics will probably not alter the main findings of this study, it is important to investigate these in the future since plants often have a preference for ammonium uptake (Masclaux-Daubresse et al., 2010). Finally, given the good quality of the input data, JSM could adequately reproduce the soil stocks and flux rates at the selected study site; however, its capacity to extrapolate to other climate and soil conditions needs to be further investigated in the future.

JSM is highly non-linear and sensitive to the parameters controlling microbial growth and decay (Table 2). The C and N stocks in SOM as well as respiration and net N mineralisation are highly sensitive to the parameter changes of depolymerisation and organo-mineral association, whereas the organic or inorganic P stocks and P mineralisation are highly sensitive to the microbial processes. These trends support, and also explain, the finding of Yang and Post (2011) and Tipping et al. (2016) that the P cycle is decoupled from the C and N cycles in the soil. A more in-depth explanation of this difference, based on our results, is that the gross mineralisation associated with microbial DOM uptake can supply microbes and plants with sufficient N but not P; thus, a large amount of P needs to be mobilised, particularly from SOM as well as from mineral pools, to sustain microbial growth. Therefore, the microbial pools and soil P stocks or fluxes are highly sensitive to microbial processes.