We apply our model, QUINCY-JSM, to the Eucalyptus Free Air CO2 Enrichment (EucFACE) experiment. Our goal is to quantify competing soil mechanisms on C storage for increased root exudation under eCO2 in a P-limited forest. Therefore, we implemented root exudation, based on plant labile pool dynamics, in the model. We parameterize and apply our model to the EucFACE experiment (Crous et al., 2015; Ellsworth et al., 2017). We first evaluate our model against field measurements for ambient (aCO2) conditions, then evaluate our results against measured CO2 responses of plant C allocation to see if our model can simulate the suggested increase in root exudation under eCO2. We further quantify the simulated increase in belowground C allocation and its effect on gross microbial growth. To assess feedback via microbial nutrient acquisition strategies we identify sources for gross microbial growth in C, N, and P and potential shifts in sources with eCO2. We conclude by comparing simulated influx and outflux of microbial necromass C, N, and P pools and their turnover rates, for aCO2 and eCO2 conditions. QUINCY-JSM has two components, a terrestrial biosphere model (QUINCY – QUantifying Interactions between terrestrial Nutrient CYcles and the climate system, Thum et al., 2019) coupled to the integrated JSM (Jena Soil Model, Yu et al., 2020) and simulates coupled ecosystem N, P, C, and water dynamics (Fig. 2 for an overview). In the following sections, we first summarize key processes represented in QUINCY and JSM, before introducing the new root exudation module.

QUINCY simulates the function of vegetation, which is described by five structural pools (leaves, fine roots, coarse roots, sapwood, and heartwood) and three non-structural pools (labile, storage, and fruits) that all contain C, N, and P. Photosynthesis is calculated after Friend and Kiang (2005) along a vertical canopy gradient (Kull and Kruijt, 1998), with photosynthetic activity assumed proportional to leaf N content, but attenuated in the presence of sink limitation on growth (Thum et al., 2019). Acquired C is added to the labile pool, from which it is used to cover the cost of respiration. Maintenance respiration is calculated for each tissue based on tissue N content and temperature (Lloyd and Taylor, 1994). In addition, QUINCY accounts for growth respiration (Atkin et al., 2014) and nutrient uptake respiration (Zerihun et al., 1998). The remaining C in the labile pool is then used for root exudation, growth, storage (Fig. 2, upper panel). Nitrogen and P taken up by plant roots or recycled from senescing tissues are added to the labile pool. Excess labile C, N, and P are stored in a reserve pool, from which these elements can be retrieved to buffer low levels of the labile pool, e.g., to ensure the regrowth of foliage and roots, root exudation (Sect. 2.4) and structural components at the start of the growing season. Upon senescence of structural pools, a fixed fraction of nutrients is recycled and returned into the labile pool. The remaining biomass (C, N, P) enters the soil via litterfall in the form of three litter pools (soluble, polymeric, wood). Vegetation and soil processes are calculated at a 30 min time-step. All physical and biogeochemical soil processes are modelled for 15 discrete soil layers, with exponentially increasing depth and a minimum layer depth of 6.5 cm. The model accounts for differences in vegetation by twelve plant functional types (PFTs), of which for this study we use only the dry broadleaved evergreen type to represent vegetation at the site.

Soil biogeochemistry is modelled by a microbially explicit and vertically resolved SOM model (JSM, Yu et al., 2020; Fig. 2, lower panel; Table 1). All biogeochemical soil pools and processes are modelled for 15 discrete soil layers, along a vertical soil gradient. Soil layer thickness increases exponentially with depth. JSM considers five different SOM pools: dissolved organic matter (DOM), microbial biomass, microbial necromass, and organo-mineral-associated DOM and microbial necromass (Table 1). Depolymerization, DOM uptake and associated biological nutrient mineralization, and inorganic nutrient uptake are affected by the abundance of microbial biomass, either modelled by using forward Michaelis-Menten or reverse Michaelis-Menten kinetics (Schimel and Weintraub, 2003). Microbes take up C from DOM (Fig. 2, process 1). Nutrients are acquired from DOM or by immobilization of soluble nutrients (Fig. 2, process 2). DOM taken by microbes either enters microbial biomass or becomes mineralized, based on element-specific use-efficiencies. The carbon-use efficiency of the microbial biomass varies as a function of nutrient availability between 30 % and 50 %, mimicking a change in microbial decomposer community. Mineralized C is emitted from soil as heterotrophic respiration. Soil organic matter and inorganic forms of nutrients are transported between soil layers via water transport (for nutrients and dissolved organic matter) and bioturbation. Microbial decay is modelled as first-order decay with a constant turnover time of ∼150 d. Dead microbial biomass is partitioned into DOM and microbial necromass (Fig. 2, process 3). The fraction of N and P entering DOM is higher than for C, to represent the lower C-to-nutrient ratio in cytoplasm (Schimel and Weintraub, 2003). Additionally, a constant fraction of N is transferred into inorganic soluble pools to mimic grazing on living microbial biomass (Wutzler et al., 2022) (Eqs. A1–A5, Table A2). Decomposition of soluble and woody litter follows first-order decay, whereas the depolymerization of polymeric litter (Fig. 2, process 4a) and microbial necromass (Fig. 2, process 4b) into DOM is a function of microbial biomass pool and in addition, varying enzyme allocation based on microbial nutrient status (Wutzler et al., 2017; Yu et al., 2020). Organo-mineral-associated DOM and necromass are protected from depolymerisation, but are exchanged with the “free” DOM/necromass pool via first-order sorption kinetics (Fig. 2, process 5). Soil solution NH4 and PO4 are in adsorption/desorption interaction with soil surfaces, simulated by Langmuir equilibrium (Yu et al., 2023). Adsorption, plant uptake and microbial uptake of soluble nutrients are in direct competition with each other (Tang and Riley, 2013). In addition to biological nutrient mineralization that results from microbial DOM uptake, PO4 is further mobilized through biochemical mineralization via acid phosphatase (Fig. 2, process 6). Biochemical mineralization is regulated by the production of phosphatase, which is assumed to be affected by root and microbial biomass, as well as the C allocation for production of other enzymes. This process allows for an additional pathway of access to P from microbial necromass and both organo-mineral-associated pools that cannot be directly depolymerised. As biochemical mineralization does not need complete decomposition of organic substances, it can affect the stoichiometry of target pools. Biochemical mineralization is assumed to be controlled by temperature, moisture, and the C:P stoichiometry of the target pools.

New N enters the system via deposition and asymbiotic N fixation (Fig. 2, process 7) and is lost by gaseous emission and lateral loss. New P enters the system via atmospheric deposition and weathering from parent material, which is controlled by abiotic and biotic factors. Phosphorus is made irreversibly unavailable by occlusion or is lost from the system by lateral loss.

To improve the representation of soil-microbe interactions in QUINCY-JSM, we implemented a root exudation module (Fig. 3). Root exudation is modelled as direct, respiration-free flux from the plant labile pool to the soil DOM pool, with a C:N stoichiometry corresponding to the current C:N ratio of the plant labile pool (Fig. 2). As we assume that root exudation mainly consists of carbohydrates, amino acids and carboxylates (Jones et al., 2004), we do not model exudation of P. Implementation of root exudation thus reduces the C and N available for tissue growth and storage and thereby respiration despite no change in the respiration calculations. From the DOM pool, the possible pathways for root exudates are: direct consumption by microbes, adsorption to the mineral surface or leaching.

The rate of root exudation [gCm-2s-1] is modelled as a function of fine root respiration [gCm-2s-1] and plant nutrient status (Eq. 1). We assume that physiologically active roots have increased exudation rate, and chose root respiration as a proxy for tissue activity. Root respiration is simulated as a function of root N mass and root temperature. Observations support the positive relationship between root exudation and root respiration or root N content (Akatsuki and Makita, 2020; Li et al., 2021; Sun et al., 2017, 2021). Distribution of root respiration and exudation across the vertical soil profile follows the simulated root distribution. In addition, the slope of the linear relationship between root exudation and root respiration is assumed to be a function of nutrient status of the plant (Ataka et al., 2020; Prescott et al., 2020): C-to-nutrient ratios in the plant labile pool that are higher than the average C-to-nutrient ratios required for tissue production will result in higher exudation (Eq. 2). 1CEX=s⋅βmavg⋅Rfr Where CEX is C root exudation in [gCm-2s-1], Rfr is fine root respiration in [gCm-2s-1], s [unitless] is a unitless parameter that determines maximum slope and βmavg [unitless] is the 7 d moving average of a scalar β reflecting nutrient shortage and surplus C in the labile pool (range 0–1) (Thum et al., 2019), calculated by: 2β=1-e-(λ⋅Φ)kex Where λ [unitless] and kex [unitless] are parameters (Table 2 for values used in this study) and the modifier Φ represents C excess, or N and P deficiency, respectively: 3Φ=ΦC⋅Max(ΦN,ΦP,1) Where ΦC is calculated as the ratio of the actual labile C pool size (Clab [molCm-2]), compared to the target pool size (Clab∗ [gCm-2]). ΦN and ΦP are calculated as the ratio of labile pool stoichiometry ((C/N)lab, (C/P)lab [unitless]) and labile pool target stoichiometry ((C/N)lab∗, (C/P)lab∗ [unitless]). The target labile C pool size (Clab∗) is calculated as the maximum of either the weekly sum of GPP or maintenance respiration. The N and P target labile pool sizes are calculated based on the last 7 d stoichiometric growth requirements in N and P to convert Clab∗ into plant biomass. The modifiers are given by: 4ΦC=MAX1,ClabClab∗5ΦN=(C/N)lab(C/N)lab∗6ΦN=(C/N)lab(C/N)lab∗ As a result, Φ increases under conditions in which labile C accumulates in the labile pool, and if the stoichiometry of the labile pool is higher than the target stoichiometry. The smallest value that Φ can take is 1, leading to values of β close to zero. We constrained β to values larger than 0.0125 (βmin), to ensure a minimal root exudation flux equal to 5 % of fine root respiration, even at nutrient-saturated conditions.

The EucFACE site is located in Western Sydney, Australia (33°37′ S, 150°44′ E, 30 m in elevation) in a mature eucalypt woodland. For a detailed description of the site see Drake et al. (2016) and Ellsworth et al. (2017). The site has a subtropical transitional climate with an annual average temperature of 17 °C and mean precipitation of 800 mmyr-1. Overstorey vegetation is dominated by Eucalyptus tereticornis, while understorey vegetation consists of a diverse mixture of shrubs, graminoids, and forbs dominated by Microlaena stipoides (Piñeiro et al., 2020).

The soil is characterized as Aeric Podosol with a loamy sand texture up to 50 cm and a sandy clay loam texture in deeper horizons. The soil has a slightly acidic pH and low fertility (Crous et al., 2015; Ross et al., 2020). Previous experiments showed that plant growth is P-limited (Crous et al., 2015). Available soil P and plant P to soil P ratio are similar to tropical and subtropical ecosystems (Jiang et al., 2024b).

The EucFACE experiment consists of 6 FACE rings, with a 25 m diameter. Three rings received aCO2 levels and three rings were exposed to eCO2 levels (ambient +150 ppm). In 2012, CO2 was stepwise increased over a period of six months with an increase of 30 ppm every 5 weeks. In February 2013, the CO2 concentration in the experiment rings reached their target of +150 ppm (Drake et al., 2016). Since then, the ecosystem has been continuously exposed to eCO2. Measurements used in this study were taken between 2013–2019. Major C pools, biomass production (BP) and net ecosystem production (NEP) under ambient conditions were measured between 2013–2016. A detailed description of measurement methods is provided in Jiang et al. (2020) and Jiang et al. (2024b). Soil respiration is taken from Drake et al. (2018) and supplemented by additional information from Jiang et al. (2020).

We parameterized the model following in situ-based observations for E. tereticornis, taken under ambient conditions from 2013–2019, provided by Jiang et al. (2024a) (Table A1). As no direct measurements of root exudation exist for this site, we calibrated s and βmin to reproduce GPP, biomass and soil respiration for ambient conditions from 2013–2016 (Jiang et al., 2020) (Table B1). This limits the size of the root exudation flux. We ensured that the modelled exudation flux still responded to variations in the size and stoichiometry of the labile plant pool, meaning that the flux was not controlled by empirical bounds of maximum exudation, implying that it is sensitive to changes in plant carbon and nutrient status induced by eCO2.

Simulations were done for 500 years of spin-up to reach quasi-equilibrium, with a subsequent simulation length of 325 years afterward from 1700–2024, following the modelling protocol from Jiang et al. (2024a): Meteorological forcing for 1700–2012 and 2019–2024 was gained from repeated and randomized observations from 2013–2018, while CO2, N, and P deposition followed historic values. Forcing for spin-up repeated the first 30 years of the forcing data under 280 ppmCO2. As a result of forcing from randomized observations and meteorological extremes in observation years, and contrary to observations, our model showed a downward trend in total vegetation C from 5900 to 5700 gCm-2 between the years 2012 and 2019 (Fig. B1). The trend did not cause major pulses in nutrient input and therefore unlikely affects eCO2 effects on nutrient-dependent root exudation and microbial C cycling.

The mean yearly change in pools and mean yearly fluxes for simulations were calculated for 2013–2019. We calculated the effect of eCO2 by subtracting results under aCO2 conditions from results under eCO2. We compared the results to observed effects of eCO2 on the C budget described in Jiang et al. (2020).

We trace the fate of the additionally assimilated C under eCO2, through the ecosystem. We standardize fluxes as the percentage of additional overstorey GPP for both simulations and observations. The model currently only simulates forests without understorey vegetation. To maintain comparability, we use observations from overstorey aboveground biomass production, overstorey autotrophic respiration, and overstorey changes in vegetation C pools. However, for belowground fluxes, i.e., heterotrophic respiration, root respiration, root biomass production and changes in soil C pools, we cannot differentiate between overstorey and understorey origin. We still standardize them as the percentage of additional overstorey GPP but are therefore likely to overestimate the observed change per additional overstorey GPP under eCO2 in soil fluxes.

For model analysis, microbial growth was calculated by the flux from the DOM to the microbes (FDOM→MICX [gXm-2s-1]), adjusted by the current carbon-use or nutrient-use efficiency (XUE [unitless]) for each soil layer separately. Most measurements for soil processes at EucFACE were taken in topsoil layers, which are also typically more active. We therefore separate model outputs between topsoil (upper 50 cm) and deep soil (below 50 cm). We estimate the contribution of fluxes from root exudation, microbial recycling upon death, and decomposition of soluble litter, polymeric litter, and microbial necromass to microbial growth Fi→growthX [gXm-2s-1] for topsoil based on their relative production of DOM: 7Fi→growthX=FDOM→MICX⋅XUE⋅Fi→DOMX∑iinIFi→DOMX With I being the set of all contributing pools and Fi→DOMX [gXm-2s-1] being the contributing fluxes from these pools to DOM. Similarly, we calculated heterotrophic respiration derived from the depolymerization of necromass C, Rhpriming [gCm-2s-1] as follows: 8Rhpriming=FDOM→MICC⋅(1-CUE)⋅Fnecromass→DOMX∑iinIFi→DOMC

Our simulations indicate that increased root exudation is a major pathway for belowground C allocation under eCO2, but the magnitude is difficult to validate due to the absence of direct measurements at EucFACE. Our model estimated a root exudation-to-GPP ratio of 21 % under ambient conditions, which is substantially higher than global estimates (Chari et al., 2024). Chari et al. (2024) suggest a global mean of 5.6 % based on upscaling measurements of free root exudation, primarily taken with the cuvette method (Phillips et al., 2008). Due to the lack of such measurements at EucFACE, we estimate root exudation from the ambient C budget. Our estimate therefore also includes other belowground allocations, e.g., mycorrhiza and losses through mucilage, and root exudation in deeper soil layers, which likely explains the differences between our results and global estimations (Brunn et al., 2025; Johansson et al., 2009; Jones et al., 2004).

Simulated C allocation to root exudation increased by 30 % under a 22 % increase in GPP under eCO2, consistent with the observation-based hypothesis of increased belowground C allocation at EucFACE. Elevated CO2 also enhanced mean annual root exudation rate per unit fine root mass by 18 % (Fig. B2). Our estimates are lower than increases reported at other FACE experiments. In Duke FACE allocation to exudates and fungi, increased by 40 % in upscaled annual flux per unit land area (Drake et al., 2011). At BIFoR FACE, specific C exudation (µgg-1finerootd-1) increased by 41 %–135 % (P=0.037) (Reay et al., 2025), and upscaled annual flux per unit land area by 43 %–63 % (P=0.042) (Norby et al., 2024).

Uncertainties in exudation rates at EucFACE remain considerable. We parameterized our model using ambient GPP and plant pool observations and calculated root exudation flux based on fine root N concentration. The deviation from simulated BP and fine root C:N from observations lead to parameter uncertainty. The compiled C budget suggests increased C allocation belowground under eCO2 (Jiang et al., 2020), but a direct evaluation of root exudation is not possible, also due to large uncertainties associated with other components of the C budget. Additionally, our model only simulates the response in overstorey vegetation, whereas the understorey vegetation at EucFACE showed a strong BP increase that likely affected soil processes (Jiang et al., 2020; Piñeiro et al., 2023). Overall, more empirical data on belowground C allocation would aid the development and evaluation of TBMs.

The model also simulated a 33 % increase in overstorey BP under eCO2, which was not observed in EucFACE. In the simulation, C accumulated primarily in woody biomass and the labile plant pool (Fig. B10), but this had only a minor influence on soil processes because plant litter contributed little to microbial growth in C, N, and P in the treatment period (Fig. 7). Instead, root exudation dominated the additional C input and the supply to microbial growth under eCO2. Observational studies suggest that plant P limitation suppressed plant growth under eCO2 (Jiang et al., 2020), suggesting that the model likely underestimates P constraints or overestimates plant adaptability to P limitation.

Despite these limitations, our model closes the gap between increased GPP and observed increased heterotrophic respiration under eCO2, underlining the importance of root exudation as a key pathway for plant C allocation that deserves greater empirical and model attention.

Our model identified root exudation as the dominant driver of higher heterotrophic respiration under eCO2, mainly as rapid microbial respiration of newly supplied labile C. Simulations showed that more than half of the additional root exudation-derived C was directly respired as overflow respiration. Although there is no partitioning in heterotrophic respiration by substrate origin in EucFACE, this result is consistent with other studies showing that under eCO2, most soil respiration originates from recently fixed C (van Groenigen et al., 2017; Taneva et al., 2006). Microbial CUE was a key modulator of soil C sequestration and respiration (Spohn et al., 2016; Tao et al., 2023, 2025). Simulated annual microbial CUE ranged between 40 % and 50 % and did not change under eCO2 (Fig. B11). A decline in microbial CUE under greater C supply or nutrient limitation would reduce microbial growth and increase the fraction of root exudates respired (Schimel and Weintraub, 2003). Our results are conservative in that respect but still support the hypothesis that the majority of additional heterotrophic respiration originates from new plant-derived C.

In the model, root exudation disproportionately stimulated microbial C growth, and consequently gross necromass production, under eCO2. This suggests that microbes might change their diet under eCO2 to more easily available and more degradable substrates for C gain (Ai et al., 2023; Dijkstra et al., 2013; Zhou et al., 2021). However, higher gross microbial growth in C also increased microbial nutrient requirements, influencing decomposition and nutrient acquisition processes.

Our simulations suggest that, to meet increased nutrient demands under eCO2, microbes enhanced necromass decomposition and biochemical mineralization; however, observations of enzyme activity only partially support this pattern. For N, demand was met by increased uptake of available inorganic N and decomposition of necromass. Priming of microbial necromass increased, but without a substantial shift in its relative contribution to growth. Here our model partially disagreed with field observations based on soil enzymes: Pihlblad et al. (2023) found no increase in N-degrading enzymes or microbial nutrient limitations under eCO2. However, it is questionable if this effect would be statistically detectable in a FACE experiment, with inherently low sample size and difficulties in detecting changes in the soil (Filion et al., 2000).

For P, our model suggests that additional microbial P demand was primarily satisfied by increased biochemical mineralization of organic sources, including necromass. As a consequence, eCO2 increased available PO4 by 20 % (Fig. B12). This is partially consistent with observations which report increased PO4 concentrations under eCO2 in rhizosphere or deeper soil layers (Hasegawa et al., 2016; Ochoa-Hueso et al., 2017; Pihlblad et al., 2023). To our knowledge it is not yet resolved which processes caused the increased PO4 concentration in EucFACE. In accordance with the model, observations from other ecosystems report increased phosphatase activity under P limitation (Marklein and Houlton, 2012) or under high soil organic P content (Margalef et al., 2017). In EucFACE, eCO2 caused no significant increase in phosphatase at the start of the experiment (Hasegawa et al., 2016), after 1 1/2 years (Ochoa-Hueso et al., 2017) or after 4 1/2 years (Pihlblad et al., 2023).

Model uncertainties derive from assumptions on nutrient acquisition strategies and root exudation stoichiometry. NH4 availability and recycling of microbial nutrients upon microbial turnover strongly contributed to microbial N acquisition under eCO2, but remain poorly constrained and require further field observations.

Additionally, we do not account for P acquisition by increased carboxylate exudation, which could have increased PO4 concentrations via ligand exchange with Fe-Al-oxides at soil surfaces (Hinsinger, 2001; Wang and Lambers, 2020) and has been reported for common EucFACE understorey species (Hasegawa et al., 2023). We currently lack field observations to implement such mechanisms in the model, and therefore the model resolves enhanced microbial P demand with increased biochemical mineralization.

Finally, simulated microbial nutrient demand is substantially controlled by root exudation stoichiometry. In the model root exudation C:N ratio was 176 (no P exudation), whereas microbial C:N ratio was 4.3 (N:P=4.1). Simulated C:N ratio is higher than root exudation measurements (Li et al., 2021; Su et al., 2022; Zhang et al., 2016) but ensures that modelled root exudation does not cause an extensive N loss in plants. As a consequence, increased root exudation enhanced microbial nutrient demand, and plant-microbial competition for N and P (Thurner et al., 2024). Under a root exudation stoichiometry closer to microbial stoichiometry, increased root exudation may further promote microbial growth. Microbial N supply may be less dependent on decomposition of microbial necromass, resulting in stronger decomposition of litter for C acquisition. Instead, a substantial amount of N would be stored in SOM, unavailable to plants. Root exudation C:N ratio remains a key variable in plant-microbe nutrient dynamics, but further research regarding the fate of exudate N is needed to improve model assumptions in plant C-for-N trade mechanisms (Drake et al., 2013; Rumeau et al., 2025).

The model shows that under increased root exudation, microbial growth can simultaneously increase gross necromass production and necromass decomposition, resulting in no substantial net effect in soil C storage (Drake et al., 2011; van Groenigen et al., 2014; Kuzyakov et al., 2019). In the model, priming offset gross necromass production and contributed to increased heterotrophic respiration under eCO2. There are no continuous measurements from EucFACE analyzing changes in soil organic matter age composition. However, measurements using a combination of litter bag and isotopic signature measuring (Castañeda-Gómez et al., 2020) showed that eCO2 increased losses of old SOC and input of new SOC at EucFACE in summer months, agreeing with our simulations.

Uncertainties in C storage arise from model representation of C cycling in deep soil layers and mineral sorption of necromass. The model predicts increased soil C in deep soil layers under eCO2, caused by increased root exudation in these soil layers. As the model assumes root exudation following root biomass distribution, eCO2 enhances root exudation in deeper soil layers with low microbial activity (Fig. 4). As a consequence, additional labile C is adsorbed to the mineral surface and thereby sequestered. While specific exudation rate may differ with depth and function of roots, we are currently missing data to further calibrate this.

Additionally, necromass is partially protected from decomposition by association with soil mineral surfaces. Our model accounts for this process but we are lacking measurements for site-specific sorption capacity and affinity. Less mineral protection would translate into stronger microbial necromass cycling, as the simulated depolymerization flux depends on availability of decomposable material. Further, the model simulates increased desorption of mineral-associated necromass under eCO2. This is partly caused by additional root exudation competing with necromass for sorption sites, but also by model assumptions about the amount of available mineral soil sorption sites with increasing organic matter and litter. This model assumption increased microbial available N-rich material and may have enhanced the eCO2 effect on necromass decomposition in the model.

We further simulated small shifts in P pools from slow cycling microbial necromass to fast cycling microbial biomass (<2 % change in necromass). A recent analysis of the P budget for EucFACE showed that eCO2 treatment led to no significant changes of soil organic P pools (Jiang et al., 2024b), but changes of this magnitude are likely below the detection limit of field observations.

Implementing root exudation in QUINCY-JSM improved the predicted CO2 response of C allocation, but results still indicate that further development is necessary. A recent model inter-comparison study demonstrated that most C-N-P cycle models, including QUINCY-JSM, overestimate the ecosystem C sequestration of the EucFACE ecosystem under eCO2 (Jiang et al., 2024a). In contrast to observations, most models simulated a positive effect on net ecosystem productivity (NEP) under eCO2. Additionally, all except one model allocated most additional C fixed under eCO2 to either autotrophic respiration, storage or structural biomass, and underestimated the increase in soil heterotrophic respiration. The exception was GDAYP, which implemented root exudation but no microbial dynamics.

Implementing root exudation improved the allocation of additional GPP: Compared to other models in Jiang et al. (2024a) and to QUINCY-JSM without root exudation, and consistent with observations, QUINCY-JSM with root exudation allocates a substantial part of additional GPP to heterotrophic respiration instead of autotrophic respiration. Additionally, less C remains in the system in the form of increased vegetation pools (Fig. B8). With consideration of root exudation, the model explains weak increases in topsoil C under eCO2 by respiration of root exudation and priming, rather than by low litter C input (Fig. B9).

Implementing root exudation improved the simulated CO2 response in heterotrophic respiration, but not in NEP, due to the effect on BP and deep soil C accumulation. Regardless of root exudation, QUINCY-JSM simulates increased net P mineralization under eCO2. Therefore, it remains difficult to evaluate the actual effect of the implementation on PO4 mineralization. Further model adjustment is needed to fully understand the feedback of plant-microbe interactions on plant nutrient uptake under eCO2.

In the following, we provide methodological directions for upcoming FACE experiments and terrestrial biosphere models to improve understanding of root exudation and microbial representation under eCO2.