MIMICS+ is based on the MIMICS framework where microbial groups, litter, and soil organic matter are represented as separate pools . In its current state, MIMICS+ is not coupled to a comprehensive land model and therefore needs prescribed C and N input and soil temperature and moisture, which it is set up to read from CLM history files. Mass balance equations, dP/dt=sources-sinks, determine the change at each time step for each pool, P. The model structure with pools and fluxes is illustrated in Fig. , and a detailed overview of mass balance and rate equations are provided in the Appendix; Tables  and  contain mass balance equations for C and N pools, respectively, while Tables  and  list C and N rate equations. Throughout the model description, fluxes referred to as CX or NX, where X is a number, can be found in the abovementioned tables and are illustrated as arrows in Figs.  and . A list of parameters is given in Table . By representing the same hydrologically and biogeochemically active layers as in CLM, MIMICS+ can represent the depth discretization of temperature- and moisture-dependent processes. For each layer the fluxes between the pools within the layer are calculated first, before the vertical transport is calculated. Unless otherwise stated, the equations below describe transport within one layer. The vertical transport is described in Sect. .

For comparison, a forest soil database collected in connection with the International Co-operative Programme on Assessment and Monitoring of Air Pollution Effects on Forests (ICP Forests) monitoring program level 1 sites was used . These data have been further analyzed by and provide a unique source of information about boreal soil conditions. A total of 1040 soil profiles were described, sampled, and analyzed between 1988 and 1992 . Soil profile descriptions were done according to standardized procedures and classified according to the Canadian System of Soil Classification (CSSC). Relevant information from the database includes C and N stocks, mean annual temperature (MAT), and mean annual precipitation (MAP). Specifically, the database contains C content down to 30, 50, and 100 cm, making it possible to compare vertically modeled C stocks to observations in these depth intervals. The dataset also contains separate measurements of C and N in the organic litter, fermented, and humic (LFH) layer and mineral soil. The organic layer consists of more or less decomposed litter, and although not directly comparable to modeled litter and SOM pools, the C:N ratio in organic vs. mineral soil is still a useful quantity for model evaluation purposes. A more detailed description of the database is given in . Because podzols are the most common soil category in Norwegian forests, we chose to focus on the podzolic sites in the dataset, giving a total of 578 sites. Due to computational resource limitations, we chose a subset of 50 representative sites (out of the 578) for the site simulations with CLM and MIMICS+. The remaining 528 sites were used for further comparison with the modeled carbon stocks. The 50 sites cover an area from 5 to 70° N latitude and from 5 to 29° W longitude. The MAT varies from -1.3 to 7 °C, while MAP ranges from 356 to 2510 mmyr-1.

For the subset of 50 sites, we performed single-site simulations using CLM5.1 in biogeochemistry (BGC) mode. Data from these simulations were used both to force MIMICS+ and to compare the C and N stocks as calculated by the standard decomposition model in CLM. The CLM variables that are used to force MIMICS+ are listed in Table . For the simulations we assume that all C allocated to active N uptake by plants in CLM is directed to mycorrhiza (in default CLM this C is assumed to directly respire).

The observations were performed during the years 1988–1992, so we ran the models up to and including 1992 and averaged model values over the 5 years. Unless otherwise stated, these averages are what is used for the comparisons. The three datasets each containing data from 50 sites are referred to as observations from the database (OBS), CLM simulations (CLM), and MIMICS+ simulations with CLM forcing (MIMICS+). An overview of the yearly mean input of carbon and nitrogen is shown in Fig. .

For the CLM simulations, a single-site configuration with 100 % natural vegetation was used together with atmospheric forcing from the Global Soil Wetness Project forcing dataset (GSWP3; https://hydro.iis.u-tokyo.ac.jp/GSWP3/, last access: 1 September 2023). This is the default atmospheric forcing for CLM and provides 3 h data with 0.5° resolution. Following CLM spin-up protocol , all sites were spun up for 500 years in “accelerated-decomposition” mode followed by 700 years of “regular spin-up” by recycling atmospheric forcing for 1901–1930. For the 1850–1900 period, the atmospheric forcing cycles the years 1901–1920, then historical forcing was used until the end of the simulation.

As with the CLM simulations, MIMICS+ needs to be spun up to equilibrium before running a historical period. The spin-up was performed from arbitrary initial concentrations by recycling monthly averages of soil temperature and moisture, N deposition, litter, and C input from the CLM history files for the years 1850–1869 (during which atmospheric forcing was used from 1901–1920) for 1000 years.

Observed and modeled soil carbon stocks are shown in Fig. . Both models capture the general trend of decreasing C concentration with increasing depth. The modeled mean C stocks of MIMICS+ across the 50 sites are closer to observations in the 0–30 cm depth interval, while the CLM simulations clearly underestimate C stocks (both models are significantly different from the subset of observations, p<0.05). The models both underestimate carbon at the 30–50 cm interval, while there is no significant difference between the modeled and observed C content in the deepest layer. Due to the heterogeneous nature of real soils and the impact of differences in litter production between the sites, a larger variability in the observations compared to the simulations is not unexpected. However, site-to-site comparisons with observations are poor for both models but marginally better for MIMICS+ (Fig. d–f). This is likely explained by subgrid variability in the observations that are not captured by the models and their forcing. As the model is intended to work on larger spatial scales within an ESM model, a good one-to-one match with specific sites is of less importance than being able to capture larger patterns in temperature and moisture. By looking at the collection of sites together, we remove some of the uncertainty related to the variability between the sites and focus on larger patterns in our analyses. There is no significant difference between the two observational subsets, meaning that the 50 sites chosen for the direct model comparison are representative of the broader region.

Looking at C:N ratios, the overall picture with a higher ratio in the forest floor (observations) and litter pools (models) than in the total soil is captured by both models, with MIMICS+ again being closer to the observed values (Fig. a and b). Both models have significantly lower C:N ratios in the total as well as in the mineral soil, but MIMICS+ has significantly higher values than CLM (p<0.05). For the litter pools, the pattern is the opposite, and the models have significantly higher C:N litter ratios than those observed in the LFH layer. The modeled litter pools are not directly comparable to the LFH layer, but we get an indication of how the modeled C:N ratio compares to the partly decomposed matter. Both models have higher mean values and greater variability than the observations (Fig. c). This is expected as the observed LFH layer is partly decomposed and would therefore have lost some C compared to the simulated litter pools which have not yet been affected by the decomposition processes. In addition, the modeled litter pools contain some low-quality (high C:N ratio) CWD, which is not included in the LFH samples.

The observed total C:N ratio ranges from 12–45 with a mean value of 28, while MIMICS+ and CLM have mean values of 23 and 11, respectively. The range of C:N values from the models is narrower than that of the observations, with MIMICS+ values ranging from 12–38 and CLM only between 11–12. The large variability among the observations indicates the influence of local conditions on a subgrid scale. The fact that MIMICS+ has a larger variability than CLM indicates that differences in soil quality are captured better with the improved modeling framework. Microbial competition for N and a higher fraction of directly plant-derived SOM are factors contributing to this difference between the modeled C:N ratios. Figure d shows the C:N ratios simulated with MIMICS+ at three different depth layers. As expected, the top layer with more litter has the highest ratio, while in the middle and lowest layers the ratios are significantly lower. For the CLM simulations the C:N ratio is constant around 11 for all three depth intervals. Since we do not have access to observed vertical N stocks, it was not possible to produce this plot for the observed sites.

In this section we look more in detail into model properties of MIMICS+. The sensitivity analysis showed that total soil C change using perturbed parameters was mostly within ±10 % of the default values (Fig. ). Modeled soil C was the most sensitive to the fraction of structural litter going directly to protected SOM, as well as mycorrhiza-related parameters (maximum CUE and mining decay rate KMO). The sensitivity of total C to parameter values related to inorganic N was small.

With the current model parameterization, the SOM pools contain about 78 % of the total soil C (all nine pools, ref. Fig. ), and 62 % of that are in the protected pools (SOMc and SOMp in Fig. ). The litter pools contain most of the remaining C, while 1.2 % are microbial biomass (Fig. ). The modeled percentage of microbes ranges from 0.3 %–2 % and is in agreement with the 1 %–3 % microbial biomass C typically reported for soils . The microbial respiration (HR) shows a clear seasonal pattern, with a stronger summer peak and winter limitation with MIMICS+ than with CLM (Fig. ). Figure b shows the relative magnitude of each pool within a pool category. Mainly due to the relatively high CWD contribution to the input, the structural pool is the largest litter C pool (18 % of total C, 85 % of total litter C), while metabolic litter consisting of leaf and fine-root litter accounts for ca. 3 % of the total C and 15 % of total litter. The saprotrophic microbial biomass C dominates over the mycorrhizal fungi biomass C, and the saprotrophic fungi dominate over saprotrophic bacteria (mean saprotrophic F : B biomass ratio of 2 and above 1 for all sites). This is largely a consequence of the parameter choices in the model and are further discussed in Sect. .

For the focus region of this study (boreal sites in Norway), total C (TOTC) is strongly correlated with both MAT and C input (+0.49 and +0.65, respectively), indicating that higher plant productivity at warmer sites is an important control on total soil C in the MIMICS+ simulations (Fig. ). The CUE presented in Fig.  is calculated as the ratio of the total microbial C uptake in biomass over the total C uptake (including respiration). CUE is positively correlated with available N, pointing to higher microbial efficiencies at sites with higher nutrient content. This is also illustrated by the positive relationship between the percentage of microbial biomass (pct_microbes) and available inorganic N (+0.41 for NNO3 and +0.62 for NNH4,sol). The negative correlation between CUE and MAT is likely explained by lower-quality litter input at warmer sites, as there is a positive relationship between the C:N ratio of the litter input and temperature (+0.46p<0.001, not shown). The lower litter quality causes reduced CUE and hence a negative relationship between temperature and CUE. The strong correlation between production (C_input) and HR (+0.81) indicates that most sites are close to equilibrium. Lower litter quality at high-production (and high-respiration) sites can explain the negative relationship between CUE and HR. There is a negative correlation (-0.64) between CUE and total C.

The fungal : bacterial saprotrophic biomass ratio (FB ratio) is negatively correlated to available inorganic N (-0.29 for NNO3 and -0.27 for NNH4,sol), reflecting the stricter stoichiometrical constrain on bacteria. There is a strong negative correlation between the percentage of microbes and the fungal : bacterial ratio (-0.78), reflecting that sites with more available N are more favorable for microbial growth in both pools but most beneficial for bacteria.

All three inorganic N pools are negatively correlated with MAP (-0.30 for NNO3, -0.29 for NNH4,sol, and -0.38 for NNH4,sorp) and NNH4,sorp also with soil water (-0.37). This indicates that the modeled microbes also respond to moisture conditions through the effects of moisture on inorganic N processes (leaching, runoff, and sorption of NH4), which contribute to making N unavailable, and not only through the modifications of the reverse Michaelis–Menten kinetics.

In Fig.  the 50 sites have been divided into two subsets of 25 sites based on the climate categories described in Sect. . (Figure S1 shows the result of the division of sites based only on the observed climate.) Figure a–c show lower carbon stocks for colder sites than for warmer sites for both models and observations for all three depth intervals, indicating that the models are broadly able to capture the temperature-dependent processes that govern the C storage in soils. As shown in Fig. , the modeled C input is positively correlated with MAT and total soil C, indicating that the difference is mainly caused by differences in litter input. The MIMICS+ simulations show a significant difference between the cold and warm mean C content (p<0.05) for all depth intervals, while the cold and warm means from the CLM simulations are not significantly different (0.14<p<0.29). This indicates that MIMICS+ temperature dependencies have a larger impact on soil C sequestration than the standard CLM formulation since the C inputs and soil temperatures are the same for the two models.

Figure d–f show that in the observations, the drier sites have a lower mean C stock than the wetter sites (but not significantly). This is the opposite of the modeled results; both models show higher mean C content for the drier sites than for the wetter sites. For MIMICS+ this discrepancy is only evident in the top layer, whereas for the lower layers, there are no significant differences between the drier and wetter sites. For the CLM simulations, the pattern is consistent and significant for all three depth intervals (p<0.05). The influence of moisture on decomposition is represented differently in the two models, which can explain some of the difference between the modeled values. This is further discussed in Sect. .

The responses to the N enrichment experiment are a result of how the extra reactive N (15 gNm-2 distributed evenly during 1 year) is distributed between the inorganic nitrogen pools after addition (Fig. a). All extra N is added to the NNH4,sol pool, which had the largest response ratio of the three inorganic N pools. Some of this N will gradually move to NNO3 via nitrification or to NNH4,sorp through sorption. While N is lost from NNO3 relatively fast via plant and microbial uptake and leaching, the extra sorbed N serves as a long-term supply of inorganic N, slowly releasing N back to the dissolved pool. This sustains the higher CUE of the microbes and leads to increased saprotrophic biomass for the duration of the 30-year simulation. Although NNH4,sol has the largest relative response to N addition, the change in mass of N is the largest in the NNH4,sorp pool.

Looking at each C pool response separately, we see the largest responses in the microbial pools (Fig. d and e). The extra inorganic N gives a relatively higher return on investment (ROI; Eq. ) for AM, resulting in more C allocated to AM and less to EcM. The initial large response declines gradually but remains positive throughout the simulation period. Although there is a shift to more AM, the EcM carbon pool is always larger than the AM pool.

Both saprotrophic C pools respond instantly and positively to the N enrichment, with a maximum increase of about 25 % for fungi and 30 % for bacteria at the end of the N addition year. The increase in saprotrophic biomass is a result of higher CUE made possible by more available N. After the N enrichment year, the response gradually decreases until it stabilizes at around 1 % after ca. 5 years. The long-term response is marginally higher for bacteria than for fungi.

The initial response in HR (Fig. f) is a result of a lower respired fraction, (1-CUE), leading to increased saprotrophic biomass and thus gradually increased rates of litter decomposition. After the initial negative response in HR in the N enrichment year, there is a positive response due to the higher decomposition rate. For bacterial HR, the response ratio stabilizes at a low positive value, while for fungi it stabilizes at a slightly negative value. Combined, the response ratio is close to 0 for HR after approximately 4 years.

The positive microbial biomass responses result in initial decreases in the substrate pools, LITm, LITs, and SOMa (Fig. c). Most microbial necromass ends up in either the physically protected pool (SOMp) or the available pool (SOMa), leading to a positive response for SOMp, while the increased decomposition of SOMa keeps the response ratio below 1. The chemically protected pool experiences a small negative response because increased microbial biomass increases the rate of the depolymerization process that moves chemically recalcitrant SOM to the available SOM pool (C11). The responses in SOM and litter pools are weak, and following 1 year of N enrichment the mean response of total C is a marginal decrease compared to the control. It is worth noting that some sites experience markedly larger responses in total C than others (shading in Fig. b).

The overall response of the model illustrates that shifts in N availability have consequences for microbial C and N dynamics, although not necessarily for the total C storage and respiration. It should be noted that in this experiment we did not increase plant productivity and thus carbon input to the soil, which is expected after N enrichment. This also means that possible changes in plant–microbe competition for N were not captured. The added value of this experiment is that we isolate the in-soil processes and quantify the effects of added nutrients available to microbes and how this affects the soil carbon pools.

Looking at the total distribution for the 50 sites, MIMICS+ is closer than CLM to the observations for the top layer (0–30 cm), and models are similar in the middle layer (30–50 cm), while none of the modeled means are significantly different from the observations in the bottom layer (50–100 cm). The site-to-site comparisons with observations were poor for both models, showing that there is a discrepancy between observed and modeled stocks at local scales. This challenge of local factors was illustrated by , who used the C-only version of MIMICS with optimized parameters based on local observations and showed reduced error in C stocks on smaller scales (catchments<50km2). Such methods would likely also reduce the errors in MIMICS+ at smaller scales. However, it is important to keep in mind that the intention with MIMICS+ is to develop a module that is simple and fast enough to be used in an ESM to simulate the soil carbon dynamics at a grid cell average scale. When forced with grid cell average input, it is not intended to and should not be expected to accurately describe local variation in soil carbon stocks. Upscaling of point observations of soil C stocks to a landscape level in our study area (Norwegian boreal forests) would be useful for comparison of ESM simulations with empirically based estimates of soil C stocks.

With the MIMICS-CN version, obtained soil C:N ratios that, although within observed ranges, had much lower maximums than the observed ratios. They suggested increasing the fraction of litter going directly to SOM, as forest soils (compared to agricultural and grassland soils) have been shown to contain a high fraction of C in plant residues . Our focus area is forested ecosystems, so we increased the fraction of litter going directly to protected SOM without going through microbial decomposition to 50 % for both structural and metabolic litter (these fractions also affect the total C; see Fig. ). This leads to a longer lifetime of soil C (stored in protected pools) before it becomes available for microbial decomposition and respiration. The higher directly plant-derived fraction in the SOM pools increases the soil C:N ratio, although it is still lower than observed for total and mineral soil (Fig. ). A recent study by indicates that the fraction of directly plant-derived SOM may be much higher than previously assumed, especially for forested sites and podzols. The high C:N ratios in our observational dataset point in the same direction, suggesting that the directly plant-derived fraction is an important factor to consider when modeling boreal soils. Our results demonstrate that we get closer to observed C:N ratios with MIMICS+ compared to the CLM formulation, a main reason being the high directly plant-derived fraction. In the CENTURY-based decomposition cascade in CLM, the C:N ratios of the SOM pools are fixed, which gives limited options to account for high C:N ratios and the implications that may have for soil C dynamics.

The division of C between the different pools in MIMICS+ shows that most soil C is in the SOM pools (78 %), whereof 62 % are protected. This again reflects the relatively high fraction of litter going directly to protected SOM but also the lifetime of C in the protected pools before it is either depolymerized or desorbed into the available SOM pool. Compared to MIMICS-CN we doubled the desorption coefficient (see Table ), but this is still 1 order of magnitude lower than the value used in the C-only version of MIMICS . In the abovementioned studies and the present study, this parameter has been adjusted to match the observed data. In the model formulation, the desorption coefficient is a function of soil clay content, and more observational studies constraining this parameter as a function of clay content or other observable variables would benefit further model development.

Saprotrophic fungi are the dominant microbial group in our simulations. Fungi are assumed to have a higher maximum CUE than bacteria in the model (0.7 vs. 0.4, respectively) and are more efficient at decomposing structural litter than the bacterial pool (higher Vmax for decomposition of LITs by SAPf than SAPb). This is based on the assumption that fungal decomposers are more specialized towards recalcitrant substrates, while bacteria thrive on labile, easily accessible metabolic litter . The fraction of CWD litter provided by CLM is relatively large at these forested sites, giving more substrate that is preferable for fungi. The Norwegian podzols we are looking at are nutrient poor, and fungal dominance is expected under N-limited conditions . Figure  indicates a negative relationship between available inorganic N and the F : B ratio, meaning a higher fraction of bacteria in more nutrient-rich conditions, in line with observations. Furthermore, the N enrichment experiment showed that bacteria had a larger positive response to the added N in the long term, which indicates that the model can capture shifts in microbial communities in response to N conditions.

The modeled saprotrophic fungal biomass C dominates over the mycorrhizal fungal biomass C. This is in contrast with an observational study on boreal forests that indicates that EcM can account for as much as 47 %–84 % of fungal biomass . Moreover, challenged the traditional view that C sequestration is mainly driven by the decomposition of aboveground litter by saprotrophs with their study that showed a dominance of root-associated fungi in deeper parts of the LFH in boreal forests. Few studies exist to inform models about fungal dominance in boreal systems, so parameters determining mycorrhizal growth and turnover are poorly constrained and not particularly adjusted for boreal conditions in this model iteration. The sensitivity analysis showed that the EcM mining rate (KMO) and maximum mycorrhizal CUE in particular impact total modeled C (Fig. ), highlighting the need for informing these parameters with representative observations. The C supply to mycorrhizal pools is prescribed directly from CLM output, and the growth of these pools is therefore governed by this input rate. Coupling MIMICS+ to the aboveground vegetation will allow the plant C supply to react to nutrient conditions in the soil and is a priority in future model development.

Regarding the correlations presented in this study (Fig. ), one should always keep in mind that correlation does not imply causation, especially in a coupled non-linear system like this model. The analysis should be regarded as a broad investigation into possible relationships within the soil dynamics. Recently, presented CUE as a strong predictor of SOC globally and argued for a positive correlation between CUE and soil carbon storage (SOC) based on a combination of global-scale datasets, a microbial-process explicit model, data assimilation, deep learning, and meta-analysis. In contrast, our analysis showed a negative correlation between microbial CUE and soil carbon storage, in addition to a strong correlation between total carbon and plant litter input. A relatively large fraction of the litter input in MIMICS+ (50 %) initially omits the microbial pathway (affected by CUE) as directly plant-derived organic matter into protected SOM pools, which weakens the relationship between microbial CUE and TOTC. A high fraction of microbial necromass ends up in SOMa (Eqs. C13–C18 in Table  and parameters in Table ). This leads to a relatively rapid recycling of the C that initially goes through the microbial pathway, which can also contribute to a weaker relationship between CUE and C storage than if larger fractions of the necromass ended up in the protected SOM pools. However, more microbially derived mass in the protected SOM pools will decrease the C:N ratio, taking modeled values further away from the observed C:N ratios in . used a process-guided deep learning and data-driven modeling (PRODA) approach to optimize parameters in a microbially explicit model using observations. Default model parameters prior to optimization gave a negative relationship between CUE and SOC, illustrating how model estimates rely on parameter choices. Using a similar approach to inform MIMICS+ can lead to more robust parameter values in future model iterations.

In MIMICS+ the availability of inorganic N is highly dependent on soil water processes because both N leaching from NNO3 and the Langmuir isotherm algorithm are dependent on soil moisture. This is evident from Fig. , where we see a negative correlation between inorganic N pools and moisture-related variables (MAP and SOILWATER). The available inorganic N pools are again positively correlated with the percentage of microbes, giving an indirect dependence of microbes on soil moisture. The total C is negatively correlated with the percentage of microbes and has a high correlation with the incoming C. With higher temperatures, the model gives a higher turnover rate and thereby more release of soil C to the atmosphere. However, increased temperatures also stimulate plant production, especially in boreal and arctic regions, which can exceed or offset the effect of higher decomposition rates . The correlation patterns from our simulation indicate that the effect of temperature on plant production dominates the effect of temperature on decomposition rates in the model. found that increased temperature sensitivity of the decomposition kinetics compared to the original MIMICS parameter values reduced error compared to their observational data, indicating that the temperature sensitivity in MIMICS and MIMICS+ may be too weak. However, the agreement between models and observations in Fig. a–c indicates that more plant production is the dominating effect of higher temperatures in Norwegian forests.

Although simple, dividing sites into different climatic categories serves as an idealized “space-for-time” investigation of climate change responses. Assuming that the climate in boreal forests in general and Norwegian forests specifically will get warmer and wetter in the future , the observations indicate higher soil C content at sites with higher MAP and MAT. The models indicate higher C content for warmer sites but lower C content for the wetter sites, especially in the 0–30 cm layer. There is a positive correlation between MAT and MAP, particularly for the observed climate forcing (Fig. ). When dividing the observed sites into the climate categories, a large fraction ends up as either cold and dry or warm and wet. We therefore did a simple “temperature-dependence removal” on the total podzol dataset (N=578) by dividing the sites into narrow temperature intervals of 0.5 °C (Fig. S2). This did not reveal a clear pattern between the wetter and drier sites, and it is therefore difficult to disentangle the effects of moisture from the effects of temperature in the observed data. Since the models use soil moisture and not MAP to define parameters, we also analyzed the results using a soil moisture variable from the CLM simulations (SOILWATER_10CM) instead of MAP to discriminate between drier and wetter sites to investigate any effects on the climatic pattern (Fig. S3). This showed the same pattern as in Fig. d–f (more C in drier soils for the models and less C in drier soils for observations) for all three distributions, except for the deepest layer, where the trend shifted for the observations, but not significantly. The CLM simulations show a negative correlation between MAP and total C (-0.63, p<0.001, Fig. ), while this is not evident for MIMICS+, indicating that it is different factors that determine the pattern from the two models. In MIMICS+, the moisture modifier on decomposition works on the fluxes from substrate to the microbial pools. The modeled microbes are the most abundant in the top 0–30 cm, which can explain why we observe a difference between drier and wetter sites only in this layer. In CLM, the moisture modification on decomposition rates works on every step in the decomposition cascade from litter to SOM pools. Since the SOM pools have more C in deeper layers, it can explain why we see the pattern in all three depth intervals for the CLM simulations. The moisture modifier used in MIMICS+ see rmoist, Table , and is a bell-shaped function of soil moisture, limiting decomposition in the case of both very wet and very dry soil conditions. If the optimal soil moisture conditions according to this function do not represent the optimal soil moisture value of the real soils, this could explain why MIMICS+ predicts the opposite pattern between the drier and wetter soils. Moreover, soil moisture can vary with subgrid features like slope and aspect, variations not expected to be captured by CLM. Therefore, discrepancy between actual and modeled soil moisture can also be a contributing factor.

Meta-analyses of observational N enrichment studies show that microbial biomass tends to decrease after enrichment, while the response in total soil C is relatively modest . The small modeled response of total soil C to N enrichment (Fig. b) is in line with these observations, but the modeled microbial biomass showed a marginal long-term increase after an initial high response (Fig. d and e). proposed several mechanisms for N effects on microbial growth (Fig. 1 in ), some leading to an increase and others leading to a decrease in microbial biomass. The sites studied in our model simulations are mainly N limited (N immobilization via mechanism 1 in Sect.  dominates), and we see an accumulation of microbial biomass as a direct consequence of the increased N availability, which is one of the mechanisms suggested by for an increase in microbial biomass. Mechanisms proposed to reduce microbial biomass in response to N enrichment are a decrease in soil pH, a decrease in ligninase activity, an increase in melanoidins, and a decrease in belowground NPP. In this offline iteration of MIMICS+ we are unable to capture potential decreases in belowground NPP allocation. Coupling to a vegetation model will enable this possibility and might affect the modeled response of N enrichment. When dividing results into separate biomes, Treseder's (2008) analyses indicate that for boreal forests the response for bacteria is positive (RR=1.061), while for fungi it is negative (RR=0.717) but with a confidence interval covering both positive and negative responses (0.0402–1.434). This points to uncertainties in observations of responses of N enrichment as well. To cover more of the possible mechanisms for microbial biomass decline in the model, one or more of the other mechanisms mentioned above could be included.

The strong N limitation in the model is partly a consequence of using low, constant C:N requirements for the saprotrophic pools (CNb=5 and CNf=8, ref. Table ). A less strict C:N requirement, or a dynamic C:N approach, as presented in the ORCHIMIC model , could lead to a weaker modeled N limitation and more microbial N mineralization, which can affect the response to N enrichment. This could also improve the modeled underestimation of the soil C:N ratio, as N in inorganic forms is subject to loss through direct plant uptake and leaching.

In the simulations, the largest loss of soil N is through the ectomycorrhizal pathway (N29). The parameter sensitivity analysis also shows a stronger sensitivity of total C to mycorrhizal parameters than to the plant uptake (Fig. ). The high microbial immobilization of N, together with the simplified representations of direct plant uptake (constant loss rate of available inorganic N), might cause an overestimation of the loss of organic N through mycorrhiza at the expense of direct plant uptake of inorganic N (N33). To model a more realistic scenario, with increased plant production and changes in plant N acquisition strategies as a response to the extra N, it is therefore necessary to couple MIMICS+ to a vegetation model. In such an experiment, both the increase in litter production and the shifts in plant C allocation will affect the soil dynamics. The enrichment experiment presented in this study showed that the model is able to capture microbial responses, and in a coupled system it can be used to further study plant–microbe competition for nutrients.

By expanding the MIMICS framework with extra microbial groups and an N cycle, we increase the possibilities to capture microbe–microbe interactions and, after coupling, plant–microbe interactions as well. However, we also introduce additional parameters and a more complex model structure that makes the model more prone to overfitting and equifinality issues. While acknowledging this possible drawback, we believe valuable insights can be gathered through a more detailed process representation, especially as new technologies allow for measurements that are suitable for constraining model parameters. Although the model produces results comparable to the observations from , there are still poorly constrained parameters in the model, especially related to mycorrhizal C and N transfer. Recent insights into the mycorrhizal role in soil C dynamics are valuable contributions to future model development . A more robust parameter optimization procedure like the PRODA approach or a Monte Carlo approach will contribute to constraining model parameters. The model should also be evaluated against observations from other ecosystems, which will increase confidence in the model structure and parameter choices. This offline version of MIMICS+ does not capture plant–microbe interactions and feedbacks, which is essential for capturing terrestrial responses to climate change. Therefore, coupling with a vegetation model is a priority in future model development.