The Fernow Experimental Forest (Fernow Forest) is a broadleaf deciduous forest located in the central Appalachian Mountains near Parsons, West Virginia (39.03∘ N, 79.67∘ W). Elevations at the Fernow Forest range from 530–1115 m with steep slopes between 20 %–50 % grade. The predominant soils at the Fernow Forest are shallow (<1 m) Calvin channery silt loam (Typic Dystrochrept) underlain with fractured sandstone and shale parent material. Mean monthly temperatures range from about -1.8 ∘C in January to about 25 ∘C in July, and annual precipitation is about 146 cm with an even distribution across seasons (Kochenderfer, 2006).

The Fernow Forest is the site of a long-term whole-watershed N addition experiment. N additions to the experimental watershed catchment area (watershed 3; 34 ha) were applied annually by aerial applications of 35.4 kgNha-1yr-1 as ammonium sulfate from 1989–2019 (30 years). The experimental N addition rate was about double the ambient N deposition measured in throughfall concentrations at the start of the experiment, and about 4 times the rate of N deposition by the end of the experiment (https://nadp.slh.wisc.edu/, last access: November 2022; https://www.epa.gov/CASTNET, last access: November 2022). Aerial application of (NH4)2SO4 was distributed in three applications per year to simulate the seasonal, ambient N deposition rates. An adjacent watershed (watershed 7; 24 ha) of similar topography and forest age is used as a reference, receiving only ambient N deposition.

The vegetation at the Fernow Forest is classified as mixed mesophytic forest. The fertilized watershed was harvested using selection harvesting and patch clear-cutting from 1958–1968 before being clear-cut in 1970 and allowed to regrow naturally for 19 years before fertilization treatment began. The adjacent reference watershed was clear-cut in two sections, the upper half in 1963 and lower half in 1966. Following cutting, both sections of the reference watershed were kept barren with herbicide treatment until 1969 when the vegetation was allowed to regrow. No legacy effects of the herbicide treatment were observed 10 years into regrowth (Kochenderfer and Wendel, 1983). The Fernow Forest has relatively diverse vegetation, and tree species are similar in both watersheds, dominated by Prunus serotina, Acer rubrum, Liriodendron tulipifera, and Betula lenta, although the fertilized watershed has a greater percentage basal area of Prunus serotina and less Liriodendron tulipifera than the reference watershed.

The observational data from the Fernow Forest used in this study were collected over various timescales and locations in the fertilized and reference watersheds, with most of these data described and summarized by Eastman et al. (2021). In brief, tree aboveground net primary productivity (ANPP) measurements were estimated from 25 permanent growth plots per watershed. The aboveground biomass of all trees >2.54 cm diameter at breast height (DBH) was estimated 6 times during the 30-year experiment using measurements of DBH and allometric equations (in the years 1990, 1996, 1999, 2003, 2009, and 2018). Autumnal fine litterfall was also measured at these plots annually from the start of the experiment (1989) through 2015 and in 20 additional plots per watershed from 2015–2017. Fine-root biomass was measured several times throughout the experiment in various sets of plots using soil cores ranging in depth from 0–10 to 0–45 cm (in the years 1991, 2012, 2013, 2015, and 2016). Fine-root production (0–10 cm) was estimated in 2016–2017 using in-growth cores. Soil organic horizon C and N stocks were measured in 2012 and 2013 (which included all organic horizons), and mineral soil C and N stocks were measured from soil pits (0–45 cm depth) in 2016. The top 0–45 cm of soil typically included the A horizon and most of the B horizon. At the study site, the A, B, and C horizons are typically found between depths of 0–12, 12–56, and 56–75 cm (expert opinion, Mary Beth Adams, unpublished data). Soil respiration was measured at 80 locations per watershed approximately weekly during the growing season and monthly during the dormant season for 2 years (2016–2017) using an infrared gas analyzer. Stream inorganic N export has been monitored at the Fernow Forest from continuous streamflow measurements and weekly or biweekly stream water chemistry samples since 1983 by the US Forest Service. Additionally, we used measurements of the partitioning of mineral SOM into different soil density fractions in the fertilized and reference Fernow Forest watersheds to compare observed versus modeled SOM distributions and stoichiometry (Eastman et al., 2022). These mineral soil samples were collected in 2018 at 20 plots per watershed, in 4 subplots per plot, to a depth of 15 cm.

The soil biogeochemical model test bed, developed by Wieder et al. (2018, 2019a), provides a framework to compare the performance of two structurally different soil C and N biogeochemical models by coupling them to a common vegetation model. The soil model test bed was originally developed to facilitate the comparison among three structurally distinct soil C models in their abilities to predict global soil C stocks and their responses to environmental change. Two of these models in the test bed include the N cycle and its interactions with the C cycle: one first-order soil C and N model, the Carnegie–Ames–Stanford Approach (CASA-CN; Potter et al., 1993; Randerson et al., 1996; Wang et al., 2010), and one microbially explicit soil C and N model, MIcrobial-MIneral Carbon Stabilization (MIMICS-CN) (Wieder et al., 2014, 2015c; Kyker-Snowman et al., 2020). While both models were developed and parameterized to run at the global scale, the test bed allows these models to be run at single-point scale for comparisons against site-level, empirical data.

The soil biogeochemical model test bed provided a computational framework for comparing the response to elevated N inputs of a first-order decay model to a microbially explicit representation of soil biogeochemical cycles. After calibrating these models to our study site (Sect. 2.3), we ran three 30-year N addition experiments that simulated the long-term N addition study at the Fernow Forest. The first experiment was performed using the default models calibrated to the study site. In the second experiment, we addressed the assumptions in the common vegetation model about fixed plant allocation. And in the third experiment, we tested the mechanism that N additions can directly inhibit enzyme activity and the decomposition of chemically recalcitrant POM.

Based on preliminary results, we modified several parameters in the vegetation and soil model components so that historic simulations (through 1988) better matched observed ecosystem C and N stocks and fluxes at the Fernow Forest (Sect. 2.1; based on Eastman et al., 2021). All vegetation and soil parameter modifications for site-specific configuration are detailed in Tables A1–A3, and these modifications are supported by observational data from the long-term experimental data (Eastman et al., 2021). Briefly, changes in the CASA-CNP vegetation parameters were made to decrease vegetation C stocks and increase the baseline N limitation in the model, which was defined by a positive NPP response to N additions (Table A1).

Modifications to CASA soil component parameters reduced the total litter plus soil C:N ratio and total litter plus soil C stocks, again better capturing observed values (Table A2; Eastman et al., 2021, 2022). In contrast, modifications to the MIMICS-CN soil parameters were needed to increase total litter plus soil C:N ratios and total litter plus soil C stocks to better reflect observed values and reduce model-to-model differences (Table A3; Eastman et al., 2021, 2022). After both CASA-CN and MIMICS-CN soil model parameters were calibrated to the Fernow Forest site for the end of the historic period, the models with these calibrated parameters became the “default” models that were used in experimental simulations (1989–2019) that are the focus of this study.

We performed three soil model test bed experiments that simulated the experimental N additions at the Fernow Forest (1989–2019). Similar to historic simulations, experimental simulations used GSWP3 climate and atmospheric CO2 data that were extended with an anomaly forcing for the years 2015–2019. Each experiment consisted of a control simulation with ambient N deposition rates used in CLM 5.0 and a “+N” simulation that received an additional 3.5 gNm-2yr-1 distributed evenly across every day of the year (Table 1). This annual rate of additional N deposition matched the annual rate of experimental N additions at the Fernow Forest whole-watershed fertilization experiment (Adams et al., 2006). In the first experiment, “default +N”, the N perturbations were the only modifications made to the site-calibrated models (Table 1).

The default +N simulation did not capture observed responses to N fertilization that included increases in wood biomass, increases in the total litter plus soil C:N ratio, and a reduction in soil heterotrophic respiration (Fig. 2). In the second experiment, “allocation shift +N”, we modified the CASA-CNP vegetation model to address assumptions about plant C allocation. It is well established that more nutrient availability leads to less belowground C flux and thus increases in aboveground NPP (Vicca et al., 2012; Litton et al., 2007; Fernández-Martínez et al., 2017), but this dynamic allocation pattern in response to nutrient enrichment is one that many models do not capture, including CASA-CNP (Wieder et al., 2019a; Thomas et al., 2015). To improve model representation of observed ecosystem responses at the Fernow Forest and to test our second hypothesis that reduced soil heterotrophic respiration resulted from shifts in plant allocation away from belowground C inputs, we adjusted carbon allocation of vegetation in CASA-CNP (Table 1). We adjusted the parameters of the fixed allocation scheme in the CASA-CNP vegetation model to shift 10 % of GPP C away from roots and towards wood production under conditions of +N. This 10 % shift is a conservative estimate of the observed response, where total belowground carbon flux (estimated using a mass balance approach) was ∼13 % lower in the fertilized watershed (Eastman et al., 2021). Results from the adjusted allocation scheme experiment are presented here and referred to as “allocation shift +N” models and simulations hereafter.

In the third experiment, “enzyme inhibition +N”, we built on the allocation shift +N parameterization to test the additional effect of direct enzyme inhibition: the hypothesis that reduced microbial enzyme activity from elevated soil N led to an accumulation of POM and a subsequent increase in the mineral soil C:N ratio. In the MIMICS-CN model, this could be approached in multiple ways (see Wieder et al., 2015a), but here we focus on the direct effects that N additions may have by suppressing ligninolytic enzyme activity, which is supported by observations at the Fernow Forest and other sites (Carreiro et al., 2000; Xia et al., 2017; Carrara et al., 2018; Tan et al., 2020). MIMICS-CN includes a transition of chemically protected SOM (SOMc, which we equate with POM) to microbially available SOM (SOMa). This transition from SOMc to SOMa in MIMICS-CN follows reverse Michaelis–Menten kinetics but is not parameterized as a function of soil N availability. Therefore, to represent potential nitrogen inhibition on POM decomposition, we increased the half-saturation constant for the oxidation of the chemically protected SOM pool during experimental N additions, essentially reducing the rates of decomposition of this pool (Table 1). In CASA-CN, we adjusted the turnover time of the slow pool, increasing it by 30 %. These increases in the turnover time of the SOMc (slow pool) were intended to reflect observed declines in decomposition and increases in POM (Eastman et al., 2022). These declines in decomposition were, in part, a result of a 25 %–57 % reduction in ligninolytic enzyme activity in the fertilized watershed, the primary agent of decomposition of POM (Table 1; Carrara et al., 2018). Results from this experiment are presented here and referred to as enzyme inhibition +N models and simulations hereafter.

To compare the sensitivity of observed and modeled responses to N enrichment, we calculated response ratios for different C and N pools and fluxes following 30 years of N additions. Response ratios were calculated for key observations and model outputs, using the most recently observed values and the annual mean value from the last 10 years of the experimental simulations. Response ratios were estimated by dividing the +N watershed observed or modeled value by the ambient (control) observed or modeled value. Thus, a response ratio of 1 indicated that there was no effect of N additions on the pool/flux, whereas a response ratio greater than or less than 1 indicated an increase or a decrease in that flux/pool with N additions.

Modeled total soil C and N stocks were estimated as the sum of C in both litter pools, the three SOM pools, and the microbial biomass pools (in MIMICS-CN only), unless otherwise noted. Observed total litter plus soil C and N stocks included the litter layer, organic horizon, and 0–45 cm of the mineral soil horizon, unless otherwise noted. Because of limitations in measuring total NPP, we compare ANPP between models and observations. ANPP is calculated as the sum of wood C production and leaf C production (measured as litter C flux in observational data).

Baseline models calibrated to the Fernow Forest had overall good agreement of key carbon and nitrogen pools and fluxes in comparison to observations. Table 2 summarizes baseline-calibrated model output from the last 10 years of the historic transient simulations (1979–1988) with comparisons to observations. We compare baseline models to the recent measurements from reference watershed 7 because this was the watershed and time period with the most complete observational data at the site (see Eastman et al., 2021). Because of higher than observed vegetation nitrogen concentrations (especially in wood) represented in the CASA-CNP vegetation model, models had greater aboveground NPP (ANPP) and plant N uptake fluxes than observations but slightly lower wood C pools. The CASA-CNP vegetation model also simulates much larger fine-root C pools than observed (Table 2). The discrepancy in fine-root C pools is in part due to the depth difference in modeled (45 cm) versus observed (15 cm) values, but CASA-CNP still likely overestimates this total pool (over 3 times observed; Table 2) that is typically concentrated in the first 20 cm of soil (Jobbágy and Jackson, 2000). As intended, calibrated soil pools, mineral soil C:N ratios, and simulated soil respiration by baseline models were very similar to observations (Table 2). The CASA-CN soil model attributes more of the total soil C to the litter layers than MIMICS-CN and observations, and MIMICS-CN predicts slightly lower soil respiration fluxes (Table 2).

Both default versions of the models exhibited a positive response in aboveground plant productivity but were not as sensitive to nitrogen additions as observations, as shown by relatively small increases in ANPP (Fig. 2a and Table A4). Because the default version of CASA-CNP uses fixed plant allocation, changes in leaf, wood, and root C pools were all positive, reflecting increases in NPP that were associated with N fertilization. Overall, the vegetation response to N addition was stronger with the MIMICS-CN soil model than with the CASA-CN soil model. Belowground, both soil models predicted little to no change in soil C stocks, a slight increase in heterotrophic soil respiration, and very slight decreases in soil C:N ratios (Fig. 2b). These modeled, positive soil responses were the opposite of mean negative observed responses (Fig. 2b). The modeled responses were due to the overall positive response of root and leaf production and, thus, plant matter inputs to the soil without a reduction in decomposition. However, even with no change in the total soil C stock, this response fell within the wide range variability of the observed response. Observed increases in soil C stocks were found in the surface mineral soil (0–10 cm) but not at greater depths (Eastman et al., 2021).

To elicit a vegetation response in CASA-CNP that reflected the observed shift in plant C allocation, we modified allocation parameters for fertilized experiments in the CASA-CNP vegetation model. As intended, this “plant allocation shift” modification to the CASA-CNP vegetation model parameterization improved model–observation agreement through a more positive ANPP response, enhanced woody biomass C stocks, and reduced fine-root production with N additions in both coupled vegetation–soil models (Fig. 2a and Table A4). This change in the vegetation response influenced soil biogeochemical responses by both models as well.

Notably, the significant (∼20 %) reduction in root C inputs to the soil with N additions leads to a small reduction in soil respiration (∼6 %) in both soil models (Fig. 2b). However, the combination of large reductions in soil C inputs and small reductions in soil C outputs (respiration) resulted in an overall 3 %–4 % reduction – rather than the observed stimulation – in the total soil C pool with both models (Fig. 2b). Soil models diverged in their soil stoichiometric response to N additions, with a slight decrease in total soil C:N simulated in the CASA-CN plant allocation shift experiment – similar to the default experiment – but a very slight increase in total soil C:N resulting from the MIMICS-CN plant allocation shift +N experiment, which was more similar to the mean observed response (Fig. 2b). These increases in the soil C:N ratio resulting from the plant allocation shift and N additions in MIMICS-CN coincided with a subtle accumulation of POM (SOMc; Fig. 3).

Based on observed increases in light particulate organic matter (POM) and soil C:N ratios with N additions in the surface soil at the Fernow Forest (Eastman et al., 2022), we examined whether the distinct soil models could capture this pattern with an additional parameter modification that reflected a reduction in soil enzyme activity with the elevated N perturbation (Fig. 1 and Table 1). These enzyme inhibition +N experiments generated similar plant productivity responses as in the plant allocation shift +N simulations (Fig. 2a). By increasing the turnover time of the CASA-CN slow pools and reducing the oxidation rate of SOMc in MIMICS-CN, both models simulated increases in the total soil C stocks, consistent with observations (5 % and 8 %, respectively) and, particularly, in the POM pools (slow and SOMc; Figs. 2b and 3). While both models captured total soil C:N responses within the range of observations, only MIMICS-CN captured an increase in the total soil C:N ratio that closely approximated the observed mean value (Fig. 2b and Table A4). Similar to observations, the positive response of the bulk soil C:N ratio (not including litter pools) that occurred with N additions was concurrent with an increase in the relative abundance of the POM pools in enzyme inhibition +N simulations (Fig. 3). However, the relationship between the fraction of soil C in POM and bulk soil C:N ratios captured by the models was weak compared to the actual relationship found in surface mineral soil (0–15 cm) samples collected from both watersheds at the Fernow Forest (Fig. 3). The weak relationship between POM abundance and bulk soil C:N ratios was due to the low C:N ratios of the POM pools in the CASA-CN and MIMICS-CN models.

Our model efforts suggest that capturing the shifts in plant C allocation in response to N additions is the most impactful way to improve the modeled N fertilization response. The default parameterizations with static plant C allocation were not sensitive to N additions, suggesting that the models underestimate N limitation by plants (Fig. 2). We modified the fixed allocation parameterization in the CASA-CNP vegetation model under elevated N inputs so that both models captured the often-observed increase in wood production and reduction in belowground carbon flux (root inputs) with N additions (de Vries et al., 2014; Fernández-Martínez et al., 2014; Frey et al., 2014; Zak et al., 2008). We note this model experiment was a post hoc modification to the CASA-CNP allocation parameters for fertilized simulations, but it allowed us to test the soil model responses to shifts in plant C allocation and underscores the importance of future work to develop more robust model processes that moderate plant C allocation as a function of ecosystem fertility status (e.g., Parton et al., 2010; Shi et al., 2016). Such developments in models that do not already account for dynamic allocation shifts are critical for making more accurate projections of plant NPP responses to global change drivers and the role of terrestrial ecosystems in sequestering atmospheric CO2 (Shi et al., 2019). Notably, with improved vegetation responses to fertilization in our plant allocation shift +N simulations, reduced soil heterotrophic respiration rates followed from reductions to belowground C inputs in both the model and the experimental results (Fig. 2b). The ability of the plant allocation shift +N model experiment to capture the observed responses at the Fernow Forest mirrors other recent model–experiment integration efforts. For example, a recent model–data synthesis of forest responses to elevated CO2 showed that the models that performed the best had dynamic representations of C allocation that were responsive to water and nutrient availability (De Kauwe et al., 2014). As such, there remains a clear need to prioritize models that employ dynamic allocation approaches based on data syntheses.

High soil N availability encourages shifts in plant nutrient acquisition strategies by reducing belowground C flux to mycorrhizae that is typically required for nutrient acquisition (Gill and Finzi, 2016; Eastman et al., 2021). At our study site, shifts in nutrient acquisition strategy and C allocation led to reduced mycorrhizal colonization, reduced rates of SOM decomposition, and an accumulation of POM. When we shifted the overall C allocation of plants (through a parameter change in the second allocation shift +N experiment) to increase wood production and reduce root production, this parameter change did reduce soil respiration relative to the control run but does not account for all mechanisms that reduce soil respiration. Rather than an overall reduction in decomposition and accumulation of POM (as observed), this allocation shift reduced litter inputs from the roots to the soil and thus caused a relative decrease in total soil C (Fig. 2). One shortcoming of our model efforts was the inability to represent a meaningful rhizosphere priming response in the control simulation that would lead to reduced priming in the elevated N simulations. Neither the CASA-CN nor the MIMICS-CN represents mycorrhizae and the plant–soil interactions that occur in the rhizosphere and with priming of soil microbes. Although MIMICS-CN does represent a K-type microbial pool, it does not distinguish between mycorrhizae that closely interact with plants versus free-living saprotrophs.

Future modeling attempts could implement a root exudate flux in a way that reflects C allocated to a microbial community that targets POM and may mobilize plant-available N (e.g., K-type microbes or a new pool of microbes that more closely resemble mycorrhizae). Incorporating a N component of the exudate flux that stimulates microbial growth and activity may also be necessary to avoid microbial N limitation and increase plant-available N in the models. Some ecosystem models do consider plant exudate inputs to the soil that prime the rhizosphere community for N acquisition (e.g., FUN-CORPSE; (Sulman et al., 2017). Because N acquisition comes with a C cost, such a transactional representation of N acquisition and uptake may better predict the plant C allocation response to elevated N inputs (Thomas et al., 2015). As we were not able to address this shortcoming in this study, we directly targeted a parameter controlling decomposition rates and simulated the direct inhibition to decomposition by N additions instead.

We tested the enzyme inhibition hypothesis by modifying parameters that control the decomposition rates of POM pools with N addition. Augmented N likely increases the turnover time of the POM pool through reduced oxidative enzyme activity and less microbial priming (Von Lützow et al., 2008; Eastman et al., 2022; Craine et al., 2007; Chen et al., 2018). While our modeling efforts did successfully increase the litter plus soil C stocks in both CASA-CN and MIMICS-CN, it only led to an increase in the total litter plus soil C:N ratio response in the MIMICS-CN model (Fig. 2b). The ∼8 % increase in soil C with N addition predicted by the MIMICS-CN enzyme inhibition +N model was similar to the mean enhancement in surface mineral soil (0–15 cm) at the Fernow Forest (∼11 %; Eastman et al., 2021), as well as increases in surface soil C stocks in other long-term N addition experiments (Frey et al., 2014; Zak et al., 2008). The CASA-CN model predicted a more moderate enhancement in soil C stocks (5 %) with N additions and these parameter modifications, which still fell within the range of observation.

Beyond accurately predicting changes in the total soil C stocks and fluxes, the distribution of SOM between POM and MAOM pools is of high importance to the future land C sink (Lavallee et al., 2020; Whalen et al., 2022). Changes in the distribution of these SOM pools may impact overall soil stoichiometry (Mikutta et al., 2019; Eastman et al., 2022), which drives important soil C and nutrient cycling processes, such as net N mineralization rates (Aber et al., 2003; Venterea et al., 2004). An increase in the relative proportion of POM constituting SOM stocks in the fertilized watershed at the Fernow Forest raises compelling questions about the future of C and N accumulations due to chronic N additions in a changing world. For example, how will N-induced increases in the relative importance of POM impact forest recovery from N deposition and progressive N limitation under elevated CO2 conditions (Craine et al., 2018; Groffman et al., 2018; Norby et al., 2010)? Indeed, a recent global analysis by Hartley et al. (2021) found evidence of greater vulnerability of POM decomposition under conditions of soil warming compared to MAOM.

MIMICS-CN offers a potential advantage over CASA-CN because of the diagnostic soil stoichiometry and more mechanistic decomposition dynamics, which allowed for greater shifts in soil organic matter composition (i.e., POM accumulation) and soil C:N ratios compared to CASA-CN. Capturing shifts in bulk soil C:N (not including the litter layer) requires representation of multiple pathways of SOM formation, which MIMICS-CN includes, such as microbial biomass turnover and the direct physical transfer of litter-derived organic matter that has bypassed microbial decomposition (Cotrufo et al., 2019, 2015). Observed C:N ratios of POM at the Fernow study site were ∼25, but the C:N ratios in CASA-CN and MIMICS-CN were between ∼14–20. In CASA-CN, the prescribed C:N ratio of POM is lower than observed, leading to very small changes in total bulk soil C:N even as the fraction of SOM in the POM pool increases. Thus, even if the POM turnover time in CASA-CN was further increased, it would still not capture the mean observed increase in soil C:N with POM accumulation (Fig. 3 and Table A4). In MIMICS-CN, the increase in total soil C:N was mainly driven by the relative increase in the fraction of POM and decrease in the low C:N MAOM fraction. Still, the increase in bulk soil C:N (not including the litter layer) with POM accumulation in MIMICS-CN was not as strong as observations suggest (Fig. 3). This was likely due to the mechanism targeted in the enzyme inhibition +N experiment: reducing the oxidation of SOMc to SOMa. In MIMICS-CN, most litter inputs pass through a microbial pool, and SOM pools are mostly made up of microbial necromass – with a lower C:N ratio – though a small number of litter inputs bypass microbial pools (Fig. 1). This underscores challenges in assessing plant vs. microbial contributions to SOM formation and persistence (Whalen et al., 2022). Under conditions of elevated N inputs, it is thought that more litter inputs bypass microbial decomposition. Therefore, the direct transfer of litter inputs to soil pools is a key pathway that may better achieve observed responses of the SOM stocks and composition to N amendments in MIMICS-CN.

While the microbial explicit foundation of MIMICS-CN holds promise, there still appears to be uncertainties in how plant–soil interactions and their responses to environmental change should be presented and parameterized in the models. For example, our post hoc adjustment of plant C allocation and microbial decomposition with N enrichment were intended to represent ecosystem responses that are commonly observed in nitrogen enrichment studies (Janssens et al., 2010). This experiment allowed us to identify certain mechanisms and processes in the models that exert strong control over the formation and stabilization of SOM and that are influenced by N deposition. However, out results cannot necessarily be generalized to other ecosystems or environmental changes. Future model developments, therefore, should focus on constructing more process-based representations of these mechanisms (i.e., dynamic plant carbon allocation and reduced ligninolytic enzyme activity with N addition) to better predict the often-observed reductions in decomposition and soil respiration under N addition that are currently hard to capture with most soil biogeochemical models.

Given the widespread empirical evidence of a reduction in lignin-degrading enzyme activity with elevated N inputs (Treseder, 2004; Pregitzer et al., 2008; Frey et al., 2014; Carrara et al., 2018) and the resulting impacts on soil stoichiometry (Chen et al., 2018), additional efforts to improve mechanistic representations of decomposition parameterizations with available data should be a focus area for future model improvement. Such responses, however, are nuanced across ecosystems. As such, Rocci et al. (2022) found no consistent changes in soil stoichiometry with nutrient additions in grassland ecosystems, despite an increase in the relative fraction of POM compared to MAOM. Additionally, the microbial community composition, as approximated by the relative abundance of MICr : MICK simulated by MIMICS-CN, was not sensitive to N additions or shifts in plant allocation and inputs (not shown). By contrast, N addition experiments in forest ecosystems have found reductions in fungal decomposer biomass, reduced ligninolytic enzyme activity, (Frey et al., 2014; Argiroff et al., 2019), and a shift in community function with reduced ability to decompose recalcitrant SOM (Ramirez et al., 2012) – including at our study site (Carrara et al., 2018; Moore et al., 2021). In these studies, this shift in microbial community and function results in accumulation of SOM. Microbially explicit models like MIMICS-CN need further development to accurately represent these changes in community composition as resource availability and stoichiometry shift to simulate the downstream effects on soil biogeochemistry. Currently, microbial communities in the model may have too much access to SOM and litter inputs, resulting in more rapid decomposition rates and lower C:N ratios of SOM pools than are often observed (see also Kyker-Snowman et al., 2020). Specifically, constraining carbon use efficiencies (CUEs), nitrogen use efficiencies (NUEs), and C:N ratios for microbial communities against data and observations is warranted to capture their responses to environmental changes.