Dynamics of Fungal and Bacterial Biomass Carbon in Natural Ecosystems: Site‐Level Applications of the CLM‐Microbe Model

Abstract

Explicitly representing microbial processes has been recognized as a key improvement to Earth system models for the realistic projections of soil carbon (C) and climate dynamics. The CLM‐Microbe model builds upon the CLM4.5 and explicitly represents two major soil microbial groups, fungi and bacteria. Based on the compiled time‐series data of fungal (FBC) and bacterial (BBC) biomass C from nine biomes, we parameterized and validated the CLM‐Microbe model, and further conducted sensitivity and uncertainty analysis for simulating C cycling. The model performance was evaluated with mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) for relative change in FBC and BBC. The CLM‐Microbe model is able to reasonably capture the seasonal dynamics of FBC and BBC across biomes, particularly for tropical/subtropical forest, temperate broadleaf forest, and grassland, with MAE <0.49 for FBC and <0.36 for BBC and RMSE <0.52 FBC and <0.39 for BBC, while R² values are relatively smaller in some biomes (e.g., shrub) due to small sample sizes. We found good consistencies between simulated and observed FBC (R² = 0.70, P < 0.001) and BBC (R² = 0.26, P < 0.05) on average across biomes, but the model is not able to fully capture the large variation in observed FBC and BBC. Sensitivity analysis shows that the most critical parameters are turnover rate and carbon‐to‐nitrogen ratio of fungi and bacteria and microbial assimilation efficiency. This study confirms that the explicit representation of soil microbial mechanisms enhances model performance in simulating C variables such as heterotrophic respiration and soil organic C density. The further application of the CLM‐Microbe model would deepen our understanding of microbial contributions to global C cycle.

Introduction

Global climate change is primarily caused by human‐induced increases in atmospheric greenhouse gases, modulated by terrestrial ecosystems through a multitude of climate‐ecosystem feedbacks (Gruber & Galloway, 2008; Keohane, 2015; Peters et al., 2012; Zaehle et al., 2010). Heterotrophic respiration, the second largest carbon (C) flux in terrestrial ecosystems, is primarily driven by soil microbes (Gougoulias et al., 2014; Sulman et al., 2014; Van Der Heijden et al., 2018). Although progress has been made in understanding how microbes affect C cycling, more research is needed on accurately projecting microbial feedbacks to climate change. Large uncertainties in global C projections challenge the current model framework (Luo et al., 2015; Taylor et al., 2011), and the implicit representation of soil microbes may partially reduce those uncertainties (Fang et al., 2005; Wieder et al., 2013, 2015; Xu et al., 2014).

The importance of soil microbes in governing the terrestrial C cycle has received growing attention, and soil microbial processes have been implicitly represented in ecosystem models (Schimel & Weintraub, 2003; Treseder et al., 2012). The development of soil microbial models, such as the SCAMPS model (Sistla et al., 2014), DAMM model (Davidson et al., 2012), microbial‐enzyme model (Allison et al., 2010), and MEND model (Wang et al., 2013), proved to be valuable in simulating microbial feedbacks to soil C processes. Soil microbial traits such as enzyme production, temperature sensitivity, carbon use efficiency (CUE), microbial and abiotic interaction, and priming effects were incorporated into soil microbial models (Allison, 2012; Allison et al., 2010; Tang & Riley, 2015). Recently, soil microbes and microbial traits have increasingly been incorporated into Earth system models (ESMs) (Sulman et al., 2014). However, these models assumed that biological responses of soil microbial community are functionally equivalent and exert invariant effects on soil processes (Bradford & Fierer, 2012). Given the large temporal variations in soil microbial community (Cleveland et al., 2007; Díaz‐Raviña et al., 1995; Lipson et al., 2002; Lipson & Schmidt, 2004), the assumption of a static soil microbial community is increasingly questioned (Wang et al., 2015; Wieder et al., 2014, 2015).

Developing models that explicitly represent soil microbial processes poses an important challenge for ESMs. Recently, the classification of soil microbes in the TRIPLEX‐Microbe model (active and dormant components) and the MIMICS model (K‐ and r‐strategists) advanced the representation of the soil microbial community and its functions in ESMs (Wang et al., 2017; Wieder et al., 2014). However, the distinct roles of broad soil microbial groups (e.g., fungi and bacteria) have not yet been considered in models. Bacteria and fungi have different physiological traits; for example, bacteria prefer to decompose litter low in carbon‐to‐nitrogen (C:N) ratio, while fungi tend to decompose litter with higher C:N ratio (Paul, 2016). These differences may cause considerable distinct trajectories of C responses to changing environments such as atmospheric nitrogen (N) deposition, elevated carbon dioxide (CO₂), and precipitation change (Bell et al., 2014; Hopkins et al., 2006; Rousk & Bååth, 2011; Strickland & Rousk, 2010). The classification of the soil microbial community into K‐ and r‐strategists based on functional traits improves the representation of distinct roles of soil microbial groups in biogeochemical processes; however, this characterization is largely theoretical and may therefore limit the effort of directly applying observational data to constrain the microbial parameters.

Fungi and bacteria, two major measurable soil microbial groups playing distinct roles on soil processes, comprise over 90% of the total soil microbial biomass and are the major agents responsible for soil organic matter mineralization (Beare, 1997). Although fungi are widely believed to decompose low quality compounds such as lignin, bacterial ligninases are also commonly found, both fungi and bacteria decompose plant residues and soil organic matter (Burns et al., 2013). For example, as litter quality decreases, fungi are expected to play more important roles (Van Der Heijden et al., 2018). Soil fungal and bacterial biomass are important components of soil microbial community (He et al., 2020), representing the microbial ability to conduct biochemical transformation of C and nutrients (Xu et al., 2013). Therefore, variations in fungal:bacterial (F:B) biomass ratio can imply changes in the population of decomposers as well as changes in soil microbial community composition and function (Six et al., 2006).

To fill the research gap of explicitly representing soil microbial community functions in ESMs, we developed the CLM‐Microbe model based on the framework in Xu et al. (2014). Fungi‐ and bacteria‐regulated processes such as the decomposition of plant and microbial residues were added into the CLM4.5 to mechanistically represent major soil microbial processes (Figure 1). To distinguish the physiological traits of fungi and bacteria in soil processes, different parameters for fungal‐ and bacterial‐regulated processes were developed and tested. In this study, we reported the model parameterization, sensitivity analysis, and uncertainty analysis at the site level for nine natural biomes, and further analyzed the simulated dynamics of microbial biomass C and their controls. The key objectives were to: (1) parameterize the CLM‐Microbe model using observed time‐series data of fungal (FBC) and bacterial (BBC) biomass C in diverse biomes, (2) evaluate the performance of the CLM‐Microbe model in simulating FBC and BBC dynamics, and (3) identify the key parameters and processes controlling variations in FBC and BBC.

1 Figure. Conceptual diagram showing the key processes and the roles of fungi and bacteria in the CLM‐Microbe model. CWD, coarse woody debris; SOM, soil organic matter; B, bacteria; F, fungi; DOM, dissolved organic matter. In the CLM‐Microbe model, number in the box means turnover time of each pool. Black solid lines indicate transitions in the CLM‐Microbe model, which generally represents processes such as: (1) decomposition of coarse woody debris, (2) litter 1 decomposition, (3) litter 2 decomposition, (4) litter 3 decomposition, (5) soil organic matter 1 decomposition, (6) soil organic matter 2 decomposition, (7) soil organic matter 3 decomposition, (8) soil organic matter 4 decomposition, (9) fungal and bacterial lysis, (10) dissolved organic matter adsorption, (11) dissolved organic matter uptake by fungi and bacteria, and (12) fungal and bacterial respiration. Red dash lines represent the regulatory role of fungi and bacteria on processes, including fungal and bacterial regulation on (13) litter 1, (14) litter 2, (15) litter 3, (16) soil organic matter 1, (17) soil organic matter 2, (18) soil organic matter 3, and (19) soil organic matter 4 decomposition.

Methodology

Data Sources

Due to the large variation in soil microbial community among biomes, we parameterized the model by biome (Xu et al., 2014). We selected time‐series observed data of FBC and BBC from nine natural biomes (i.e., tropical/subtropical forest, temperate coniferous forest, temperate broadleaf forest, boreal forest, shrub, grassland, desert, tundra, and wetland), with at least two sites in each biome. Then, we randomly selected one site for model calibration and the others for model validation. Finally, nine sites were used for model calibration and 12 sites for model validation for nine natural biomes. Site information, including geographic location, biome type, site name, site ID, sampling years, and the measurement methods, was presented in Table 1.

1 TableSite Information of the Observational Data

Phase	Biome	Site ID	Location	Country	Measurement	Depth (cm)	Sampling year	Reference
Calibration	Tropical/subtropical forest	TSF‐HS	22.57°N, 112.83°E	China	PLFA	0–10	2012–2014	Zhao et al. (2017)
Temperate coniferous forest	TCF‐NJ	39.92°N, 74.6°W	US	PLFA	0–2.3	2006–2008	Landesman and Dighton (2010)
Temperate broadleaf forest	TBF‐VA	48.21°N, 16.36°E	Austria	PLFA	0–5	2006–2008	Kaiser et al. (2010)
Boreal forest	BRF‐WC	65.32°N, 142.31°W	US	DMO	5–20	1973	Flanagan and Van Cleve (1977)
Shrub	SHB‐OB	40.18°N, 112.46°W	US	PLFA & FAME	0–10	2009–2010	Docherty et al. (2015)
Grassland	GRS‐IA	42.18°N, 93.5°W	US	DMO	0–10	2001–2002	Dornbush et al. (2008)
Desert	DST‐CH	29.08°N, 103.17°W	US	CF & FAME	0–15	2004–2006	Bell et al. (2014)
Tundra	TUN‐MH	68.36°N, 18.50°E	Sweden	PLFA	0–10	2006	Björk et al. (2008)
Wetland	WET‐EM	51.91°N, 11.98°E	Germany	PLFA	0–20	1999–2000	Moche et al. (2015)
Validation	Tundra	TUN‐ES	64.83°N, 111.63°W	Canada	DMF	0–5	2007	Buckeridge et al. (2013)
Temperate coniferous forest	TCF‐NT	47.58°N, 11.64°E	Australia	PLFA	0–5	2008–2009	Schindlbacher et al. (2011)
Desert	DST‐GB	44.28°N, 87.93°E	China	PLFA	0–5	2011–2013	Huang et al. (2015)
Desert	DST‐JN	32.59°N, 106.84°W	US	PLFA	0–30	2017	National Ecological Observatory Network (2020)
Boreal forest	BRF‐AL	65.15°N, 147.5°W	US	PLFA & FAME	0–10	2009	Docherty et al. (2015)
Tropical/subtropical forest	TSF‐OS	29.69°N, 81.99°W	US	PLFA & FAME	0–10	2009–2010	Docherty et al. (2015)
Grassland	GRS‐BC	49.23°N, 121.77°W	Canada	DMO	0–10	1997–1999	Bittman et al. (2005)
Temperate broadleaf forest	TBF‐SH	36.02°N, 106.47°E	China	PLFA	0–11	2012–2013	Liu et al. (2018)
Temperate broadleaf forest	TBF‐MS	37.38°N, 80.52°W	US	PLFA	0–30	2018	National Ecological Observatory Network (2020)
Temperate broadleaf forest	TBF‐TL	32.95°N, 87.39°W	US	PLFA	0–30	2017	National Ecological Observatory Network (2020)
Shrub	SHB‐AC	30.63°S, 71.67°W	Chile	CFU	0–20	1997–2006	Aguilera et al. (2016)
Wetland	WET‐EF	51.87°N, 12.39°E	Germany	PLFA	0–20	1998–2000	Moche et al. (2015)

Abbreviations: CF, chloroform fumigation; DMF, direct count using fluorescence microscope; DMO, direct microscopy using optical microscope; FAME, fatty acid methyl ester; PC, plate count; PLFA, phospholipid fatty acid.

The FBC and BBC were derived from multiple approaches, such as direct microscopy using optical microscope (DMO) or fluorescence microscope (DMF), plate count (PC), chloroform fumigation (CF), fatty acid methyl ester (FAME), and phospholipid fatty acid (PLFA). Based on our previous study, large variations exist in measured fungal and bacterial biomass among different approaches (He et al., 2020). The PLFA was the most widely used in field observed data (Table 2), and likely the most appropriate approach for estimating FBC and BBC simultaneously (He et al., 2020; Waring et al., 2013). To reduce the biases introduced by various approaches, we converted the reported FBC and BBC measured using DMO (Balser et al., 2005; Olsson & Wallander, 1998), DMF (Frostegård & Bååth, 1996; Stahl & Parkin, 1996), PC (Bai et al., 2013; Priha et al., 1999), CF (Bailey et al., 2002), and FAME (Miura et al., 2017) to PLFA measured values using the conversion factors reported by previous studies. Then, we performed the conversion from the measured fungal and bacterial PLFA concentration to FBC and BBC using conversion factors reported by Frostegård and Bååth, 1996; Klamer and Bååth, 2004.

2 TableKey Model Parameters in Processes Involving Fungal and Bacterial Biomass

Symbol	Rangea	Unit	Description	Reference
k_dom	0.0025–0.5	d−1	Decomposition rate constant of DOM	Cherrier et al. (1996), Kirchman et al. (1991), Wheeler et al. (1996)
k_bacteria	0.00143–2	d−1	Lysis rate constant of bacteria	Moore et al. (2005), Rousk and Bååth (2007, 2011), Schippers et al. (2005)
k_fungi	0.00027–0.05	d−1	Lysis rate constant of fungi	Moore et al. (2005), Rousk and Bååth (2011), Thornton and Rosenbloom (2005), Wallander et al. (2004)
m_rf_s1m	0–1		Fraction factor quantifying carbon from SOM1 to microbes	Calibrated
m_rf_s2m	0–1		Fraction factor quantifying carbon from SOM2 to microbes	Calibrated
m_rf_s3m	0–1		Fraction factor quantifying carbon from SOM3 to microbes	Calibrated
m_rf_s4m	0–1		Fraction factor quantifying carbon from SOM4 to microbes	Calibrated
m_batm_f	0–1		Fraction factor quantifying carbon respired by bacteria	Calibrated
m_bdom_f	0–1		Fraction factor quantifying carbon from DOM to bacteria	Calibrated
m_bs1_f	0–1		Fraction factor quantifying carbon from bacteria to SOM1	Calibrated
m_bs2_f	0–1		Fraction factor quantifying carbon from bacteria to SOM2	Calibrated
m_bs3_f	0–1		Fraction factor quantifying carbon from bacteria to SOM3	Calibrated
m_fatm_f	0–1		Fraction factor quantifying carbon respired by fungi	Calibrated
m_fdom_f	0–1		Fraction factor quantifying carbon from DOM to fungi	Calibrated
m_fs1_f	0–1		Fraction factor quantifying carbon from fungi to SOM1	Calibrated
m_fs2_f	0–1		Fraction factor quantifying carbon from fungi to SOM2	Calibrated
m_fs3_f	0–1		Fraction factor quantifying carbon from fungi to SOM3	Calibrated
m_domb_f	0–1		Fraction factor quantifying carbon from DOM to bacteria	Calibrated
m_domf_f	0–1		Fraction factor quantifying carbon from DOM to fungi	Calibrated
m_doms1_f	0–1		Fraction factor quantifying carbon from DOM to SOM1	Calibrated
m_doms2_f	0–1		Fraction factor quantifying carbon from DOM to SOM2	Calibrated
m_doms3_f	0–1		Fraction factor quantifying carbon from DOM to SOM3	Calibrated
cn_bacteria	3–12		C:N ratio of bacteria	Strickland and Rousk, (2010)
cn_fungi	3–60		C:N ratio of fungi	Strickland and Rousk (2010)
cn_dom	4.2–185		C:N ratio of DOM	Sinsabaugh et al. (2016)
CUEmax	0.46–0.9		Maximum carbon use efficiency of microbes	Gommers et al. (1988), Sinsabaugh et al. (2013, 2016)

Abbreviation: CUE, carbon use efficiency.

^aThe values may not be the same as those from literature sources due to unit conversion.

Model Improvements

The CLM‐Microbe model was developed based on the default CLM4.5 with vertical profiles of biogeochemistry, and we specifically incorporated soil microbial processes into the decomposition subroutines (Koven et al., 2013; Thornton et al., 2007; Thornton & Rosenbloom, 2005). The key algorithm for simulating microbial controls on C processes is based on the model framework in Xu et al. (2014). The CLM4.5 classified litter into three pools, that is, litter 1 (labile), litter 2 (cellulose), and litter 3 (lignin), and soil organic matter (SOM) into four pools, that is, SOM 1, SOM 2, SOM 3, and SOM 4. The three litter and four SOM pools differ in base decomposition rate (τ), with turnover time of litter pools ranging from 20 h to 71 days and SOM pools ranging from 14 days to 27 years (Figure 1). Coarse woody debris is fragmented, decomposed, and gradually transferred into litter pools, and further from litter to SOM pools (Koven et al., 2013).

One critical improvement in the CLM‐Microbe model is the representation of the pools of dissolved organic matter (DOM), fungal and bacterial biomass into the biogeochemistry cascade in the default CLM4.5 (Figure 1). The DOM pool is further linked with a microbial functional group‐based methane module (Wang et al., 2019; Xu et al., 2015). In the decomposition subroutine, we changed the original transfers from litter to SOM to mechanisms that were mediated by soil fungi and bacteria. Specifically, we added the C transfer from litter and SOM pools to fungal and bacterial biomass pools and DOM, from DOM pool to fungal and bacterial biomass and SOM pools, and from fungal and bacterial biomass pools to DOM and SOM pools. A certain proportion of C, defined by fraction factors in fungal and bacterial biomass pools, will be respired as CO₂ into the atmosphere. In total, the CLM‐Microbe model included 41 transitions mediated by fungi and bacteria, which largely increased the accuracy of simulating the complex soil processes relative to nine transitions in the default CLM4.5. In each soil layer, these transitions are regulated by environmental factors (e.g., temperature, moisture, and oxygen) in the soil. We defined 26 parameters related to fungi and bacteria related processes in the CLM‐Microbe model, with the range and description of each parameter to be found in Table 2. The code for the CLM‐Microbe model has been archived at https://github.com/email-clm/clm-microbe since 2015. The model version used in this study was checked out from GitHub on June 18, 2018.

Vegetation Effects on Soil Microbial Community

Vegetation also has a significant influence on soil microbial growth through litter input and root exudation (Blagodatskaya & Kuzyakov, 2008). Labile C from litter and root exudates, in the form of DOM, can be readily used to enhance fungal and bacterial growth (Göttlicher et al., 2006). Therefore, in addition to the slow breakdown of SOM and litter, the DOM pool is refreshed by a rapid release from living roots and fresh litter, playing an important role in soil microbial activity (Sulman et al., 2014). In the CLM‐Microbe model, we incorporated the DOM input from fine roots and litter, and the quantity of DOM input from these pools are determined by a parameter quantifying the labile C release from pools of fine roots and litter and their pool size. The incorporation of DOM input from litter and fine roots represents the vegetation effects on soil microbial community.

Decomposition

The decomposition of SOM, DOM, and litter was controlled by both their potential decomposition rates and environmental conditions. The decomposition processes in the CLM‐Microbe model were defined following the equations as below, [Image Omitted. See PDF] [Image Omitted. See PDF] [Image Omitted. See PDF] [Image Omitted. See PDF]where D_C is the rate of substrate (e.g., SOM, DOM, and litter) breakdown; k is the potential decomposition rate; r_oxygen represents environmental modifier determined by soil oxygen concentration, which is set as 1 for the single layer model; r_depth is environmental modifier determined by soil depth, which is set as 1 for the single layer model; r_water is environmental modifier determined by soil moisture; r_tsoil means environmental modifier determined by soil temperature; z means soil depth; $z_{\tau }$ is the e‐folding depth for decomposition; T_{soil, j} is soil temperature at layer j; T_ref is the reference temperature for decomposition, which is set as 25°C; Q₁₀ indicates the temperature dependence of decomposition, it is the ratio of the rate at a specific temperature to that at 10°C lower; Ψ_j is the soil water potential in layer j; Ψ_min is a lower limit for soil water potential control on decomposition rate (set to −10 MPa), r_water will be set as 0 if Ψ_j is lower than Ψ_min; Ψ_max is the upper limit for soil water potential control on decomposition, which equals to the saturated soil matric potential, r_water will be set as 1 if Ψ_j is higher than Ψ_max; w_{soil, j} means soil water content in layer j.

Although there are variations in Q₁₀ of substrate mineralization under various land use types, nutrient concentrations, moisture contents, property of substrates, and temperature gradients for measurement (Fierer et al., 2003; Hopkins et al., 2006; Larionova et al., 2007), the Q₁₀ value is confined close to 1.5 at ecosystem‐level, which is set as default Q₁₀ value in the CLM4.5. There is no difference in decomposition between aboveground and belowground substrate, Q₁₀ values of the decomposition of three litter pools (Litter 1, Litter 2, and Litter 3) and two less stable SOM pools (SOM 1 and SOM 2) were set as 1.5 in the CLM‐Microbe model, which is consistent with the default CLM4.5. Stable SOM in deep soils is believed to have higher Q₁₀ value than in surface soils (Fierer et al., 2003; von Lützow & Kögel‐Knabner, 2009), indicating that the decomposition of stable SOM in subsurface soil is more sensitive to temperature change than that in surface soil. Therefore, to differentiate the Q₁₀ of SOM decomposition in different soil depths, Q₁₀ values of third SOM pool and fourth SOM pool are set as 2 and 2.5, respectively, in the CLM‐Microbe model. Due to the simple chemical structure and low activation energy of DOM, Q₁₀ value of DOM is expected to be lower than SOM and litter (Davidson & Janssens, 2006). Consequently, we set the Q₁₀ value of DOM as 1.25 in the CLM‐Microbe model.

Microbial Lysis

The microbial biomass turnover is closely associated with the SOM formation, while the contribution of microbial biomass residues to the formation of SOM has been largely underestimated (Liang et al., 2019). However, growing evidence showed that the soil microbial community made a relatively high contribution to soil organic carbon (SOC) due to its large pool size (Xu et al., 2013) and fast turnover rate (Glaser et al., 2004). Sinsabaugh et al. (2016) estimated a global mean biomass turnover time of 67 ± 22 days based on a negative linear relationship between CUE and microbial biomass turnover time, with the mean microbial CUE estimated as 0.25–0.30. Xu et al. (2017) also quantified the microbial biomass turnover time as 23 and 28 days based on the area‐weighted global average of the metabolic quotient in soils (1.8 μmol C·mmol microbial biomass C⁻¹·h⁻¹) and reference metabolic quotient (1.5 μmol C·mmol microbial biomass C⁻¹·h⁻¹), respectively, from a global microbial metabolic quotient data set. These estimates in soil microbial biomass turnover are generally in the same order and vary slightly; however, the turnover rates of different soil microbial groups (e.g., fungi and bacteria) were distinct and in a wide range of variation, with fungal and bacterial biomass turnover rate reported as 0.00143 to 2 d⁻¹ (Moore et al., 2005; Rousk & Bååth, 2007, 2011) and 0.00027 to 0.05 d⁻¹ (Moore et al., 2005; Rousk & Bååth, 2011; Strickland & Rousk, 2010; Wallander et al., 2004), respectively.

In addition, bacterial and fungal growth are highly sensitive to environmental conditions, such as soil moisture and temperature. As a result, in the CLM‐Microbe model, fungal and bacterial biomass lysis process is mechanistically represented as the interactive effects of lysis rate constant and environmental factors, that is, r_oxygen, r_water, r_tsoil, and r_depth, described above.

Soil Microbial Respiration

Bacteria and fungi assimilate DOM, SOM, and litter to form their biomass, and a proportion of the assimilated C is respired (Figure 1). The proportion of C used for fungal and bacterial respiration is determined by the factors indicated in Tables 2 and 3. In addition, heterotrophic respiration (HR) is widely affected by multiple abiotic and biotic factors, such as substrate concentration and availability, soil moisture, and soil temperature (Gomez‐Casanovas et al., 2012; Zhang et al., 2013). Therefore, fungal and bacterial respirations in the CLM‐Microbe model are defined as the interactive effects of substrate (i.e., DOM, SOM, and litter), fraction factors quantifying C being respired by fungi and bacteria, and environmental factors (i.e., r_oxygen, r_water, r_tsoil, and r_depth) regulating the respiration process.

3 TableKey Parameters for the Different Biomes

Parameters	Biomes
BRF	DST	GRS	SHB	TBF	TCF	TSF	TUN	WET
m_bdom_f	0.15	0.1	0.15	0.1	0.15	0.15	0.1	0.1	0.08
m_bs1_f	0.1	0.03	0.05	0.05	0.05	0.05	0.03	0.03	0.02
m_bs2_f	0.12	0.06	0.1	0.1	0.1	0.1	0.06	0.06	0.04
m_bs3_f	0.18	0.12	0.15	0.15	0.15	0.15	0.12	0.12	0.08
m_fdom_f	0.15	0.1	0.15	0.1	0.15	0.15	0.1	0.1	0.08
m_fs1_f	0.1	0.03	0.05	0.05	0.05	0.05	0.03	0.03	0.02
m_fs2_f	0.12	0.06	0.1	0.1	0.1	0.1	0.06	0.06	0.04
m_fs3_f	0.18	0.12	0.15	0.15	0.15	0.15	0.12	0.12	0.08
k_dom	0.007	0.007	0.007	0.007	0.008	0.007	0.0005	0.007	0.007
k_bacteria	0.008	0.0178	0.005	0.0036	0.008	0.004	0.0085	0.0032	0.072
k_fungi	0.004	0.009	0.0045	0.002	0.0018	0.002	0.01	0.0012	0.032
m_rf_s1m	0.4	0.4	0.4	0.4	0.4	0.4	0.4	0.4	0.4
m_rf_s2m	0.6	0.6	0.6	0.6	0.6	0.6	0.6	0.6	0.6
m_rf_s3m	0.7	0.7	0.7	0.7	0.7	0.7	0.7	0.7	0.7
m_rf_s4m	0.8	0.8	0.8	0.8	0.8	0.8	0.8	0.8	0.8
m_batm_f	0.2	0.08	0.12	0.08	0.12	0.12	0.12	0.12	0.1
m_fatm_f	0.1	0.04	0.08	0.04	0.08	0.08	0.08	0.08	0.06
m_domb_f	0.008	0.12	0.04	0.16	0.86	0.045	0.05	0.24	0.27
m_domf_f	0.001	0.18	0.9	0.64	0.04	0.005	0.89	0.56	0.45
m_doms1_f	0.32	0.24	0.03	0.1	0.06	0.32	0.03	0.1	0.14
m_doms2_f	0.27	0.2	0.02	0.06	0.03	0.28	0.02	0.06	0.08
m_doms3_f	0.22	0.15	0.01	0.03	0.01	0.2	0.01	0.03	0.04
cn_bacteria	5	4	5	5	6	5	5	4	6
cn_fungi	15	15	15	15	12	15	15	16	12
CUEmax	0.8	0.8	0.8	0.8	0.8	0.8	0.8	0.8	0.8

Abbreviations: BRF, boreal forest; CUE, carbon use efficiency; DST, desert; GRS, grassland; SHB, shrub; TBF, temperate broadleaf forest; TCF, temperate coniferous forest; TSF, tropical/subtropical forest; TUN, tundra; WET, wetland.

Carbon Use Efficiency

The CUE of soil microbes for three litter pools in the CLM‐Microbe model are determined following the equation in Sinsabaugh et al. (2013). In addition, CUE is reported to vary with temperature, showing a coefficient of −0.012 with increasing temperature (Devêvre & Horwáth, 2000; Xu et al., 2014). Therefore, we assumed that CUE decreased compared with the ambient thermal regime of microbes’ habitats following the equation as below, [Image Omitted. See PDF]where CUE is C use efficiency, which is defined as the growth‐to‐assimilation ratio for soil microbes; CUE_max is the maximum value of C use efficiency; CUE_T is the coefficient indicating the dependence of C use efficiency on temperature; T_CUEref is the reference temperature of C use efficiency, which is defined as 15°C in the CLM‐Microbe model; M_C:N means the C:N ratio of soil microbial biomass, which is defined as 8 in the CLM‐Microbe model; S_C:N represents C:N ratio of substrate (e.g., litter).

The C flow from litter and SOM pools to soil microbes will be partitioned by fungal and bacterial biomass pools based on the C:N ratio of fungal and bacterial biomass. The fraction factor quantifying bacteria C gain from litter and SOM is calculated based on the weighted average of assimilation efficiency of fungi and bacteria following the equation as below, [Image Omitted. See PDF] [Image Omitted. See PDF]where fb is the fraction of C flowing into bacteria; ff is the fraction of C flowing into fungi; B_C:N means the C:N ratio of bacteria; F_C:N means the C:N ratio of fungi; S_C:N represents C:N ratio of substrates (e.g., litter and SOM).

Model Forcing Data

The forcing data for this model include meteorological variables such as air temperature, relative humidity, incoming solar radiation, longwave radiation, precipitation rate, surface pressure, and surface winds. Since the sampling year of the sites spans from 1973 to 2018, which was not fully covered by any commonly used forcing datasets of CLM. After examining the data distribution, we found that sites sampling later than 2014 are in North America (Table 1). Therefore, for sites sampled before 2014, we extracted the forcing data during January 1, 1971 through December 31, 2014 from CRUNCEP Version 4 provided by Climate Data Gateway at the National Center for Atmospheric Research (https://www.earthsystemgrid.org). The forcing data for sites sampled later than 2014 were extracted from the Global Land Data Assimilation System Version 2 (https://ldas.gsfc.nasa.gov). Using the latitude and longitude information of study sites in each biome (Table 2), we extracted the meteorological variables for all sites. Since the standardized input forcing data are in half‐hourly time steps, the extracted 6‐hourly data for each study site was interpolated to half‐hourly step using linear interpolation via na.approx function in R (R for Mac OS X version 3.5.3).

Model Implementation

We used default parameters for the subroutine of soil hydrological properties and methane module (Koven et al., 2013; Xu et al., 2015), and we focused on the parameterization for soil microbial mechanisms related to FBC and BBC dynamics (Table 2). To parameterize the model, we set up model simulations separately for nine sites (TSF‐HS, TCF‐NJ, TBF‐VA, BRF‐WC, SHB‐OB, GRS‐IA, DST‐CH, TUN‐MH, and WET‐EM) in the phase of calibration and 12 sites (TUN‐ES, TCF‐NT, DST‐GB, DST‐JN, BRF‐AL, TSF‐OS, GRS‐BC, TBF‐SH, TBF‐MS, TBF‐TL, SHB‐AC, and WET‐EF) in the phase of validation (Table 1).

The model implementation for all sites was carried out in three stages. First, we ran the accelerated decomposition spin‐up to make the system reach steady state (Thornton & Rosenbloom, 2005). Due to the differences in the length of time to reach steady state among biomes, we set the model run as 1,500 years for tropical and temperate biomes (i.e., tropical/subtropical forest, temperate coniferous forest, temperate broadleaf forest, shrub, grassland, and desert), 2,000 years for boreal and arctic biomes (i.e., boreal forest and tundra), and 3,000 years for wetlands. Then, we ran a final spin‐up of 100 years to make the system ready for transient simulations during 1850–2018. Since observational FBC and BBC are reported at daily scale, we set the output resolution of transient simulations as a daily time step.

To guarantee the reasonable soil and vegetation conditions in each site at the same standard, we extracted the SOC of the top 1 m soil profile from the Harmonized World Soil Database (HWSD, https://daac.ornl.gov/cgi-bin/dsviewer.pl?ds_id=1247) and net primary productivity (NPP) from MODIS gridded data set with a spatial resolution of 30 s during 2000–2015 (http://files.ntsg.umt.edu/data/NTSG_Products/) for each site using their latitude and longitude information. Before parameterizing the CLM‐Microbe model, we adjusted the parameters related to plant photosynthesis (e.g., flnr), C allocation (e.g., froot_leaf), and e‐folding depth for decomposition (e.g., decomp_depth_efolding) to make all the sites have soil and vegetation conditions reported by global datasets. Since soil and vegetation are in high spatial heterogeneity, the global datasets may not be able to capture the variation at fine scale. If the SOC and NPP extracted from global datasets were extremely high or low for the biome, we used the values reported in the literature. If the SOC and NPP were not available from literature, we used the biome‐level averages instead (Chapin et al., 2011; Jobbágy & Jackson, 2000). For tropical/subtropical forest, we had a site specified as needleleaf trees, whereas tropical needleleaf tree is not available in default plant function types. Therefore, we modified the parameters for needleleaf_evergreen_temperate_tree following the parameters featuring tropical trees such as minimum and upper limit of temperature for growth and monthly temperature. Also, we altered the longevity for needleleaf_evergreen_temperate_tree following the trend of needle tree leaf longevity reported by Xiao (2003).

Based on the current knowledge of mechanisms for FBC and BBC dynamics, we primarily focused on parameters related to microbial lysis (m_bdom_f, m_bs1_f, m_bs2_f, m_bs3_f, m_fdom_f, m_fs1_f, m_fs2_f, m_fs3_f, k_bacteria, and k_fungi), microbial respiration (m_batm_f and m_fatm_f), decomposition of litter, DOM and SOM (k_dom, m_domb_f, m_domf_f, m_doms1_f, m_doms2_f, and m_doms3_f), and stoichiometric traits of fungal and bacterial biomass (cn_bacteria and cn_fungi). We first ran the model for each biome using the average values reported by previous studies, then we empirically calibrated the parameters based on the observed FBC and BBC of sites selected in calibration phase (Table 3). The calibration was based on the model behavior in capturing the seasonal variations in FBC and BBC. Next, we validated the model using the parameters in Table 3 to test the model simulation performance by plotting simulated FBC and BBC against the observed data.

Model Evaluation

We used three metrics to evaluate model performance, including:

1. Mean absolute error (MAE), a measure of model error, was computed as [Image Omitted. See PDF]where y_i is the simulated value; $\widehat{y}$ means the observed value; N is the number of data points. The MAE indicates the mean error of the model simulation, and thus lower MAE values are preferred.

2. Root mean square error (RMSE), indicating the model accuracy, was calculated as: [Image Omitted. See PDF]where y_i is the simulated value; $\widehat{y}$ means the observed value; N is the number of data points. Similar with MAE, RMSE also indicates the mean error of the model simulation, and the lower values indicate the higher model accuracy. The RMSE estimation is equal or larger than MAE estimation in most cases, and the degree to which RMSE estimation exceeds MAE estimation depend on the outliers in the simulated and observed data.

3. The coefficient of determination (R²), representing the variation in the observations explained by the model, was calculated following the equation as below, [Image Omitted. See PDF]where y_i is the ith simulated value; $\widehat{y_i}$ means the ith observed value; $\overline{y}$ is the mean of the observed values. Higher R² values indicate better performance of the model, while lower R² values mean the worse model performance and smaller proportion of variation being explained by the model. It is noteworthy that R² is not suitable for assessing the goodness‐of‐fit for a small number of data points due to the large bias in small samples.

Sensitivity Analysis and Uncertainty Analysis

To identify the key processes and parameters for FBC and BBC dynamics across biomes, we conducted sensitivity analysis using one site in each biome. We selected 25 parameters related to the decomposition of SOM, litter and DOM, fungal and bacterial respiration, CUE, and microbial lysis for identifying key processes and parameters in regulating FBC and BBC dynamics, which were also used for uncertainty analysis (Table 4). Eventually, nine sites, that is, DST‐CH, GRS‐IA, SHB‐OB, TUN‐MH, BRF‐WC, TBF‐VA, TCF‐NJ, TSF‐HS, WET‐EM, were used for sensitivity analysis, and these sites were also used for subsequent uncertainty analysis (Figure 7, Table 4). For each parameter, we set up model simulations with +20% or −20% change in the parameter, and thus we set up 50 model simulations in total for sensitivity analysis in each site. Next, we investigated the responses of the simulated FBC, BBC, F:B ratio, SOC, and HR in surface soil (0–30 cm) during sampling period of each site. The index S, comparing the change in the model output relative to the model response for a nominal set of parameters, was calculated based on the equation following Xu et al. (2015). [Image Omitted. See PDF]where S is the ratio of the standardized change in model response to the standardized change in parameter values; R_a and R_n are model responses for altered and nominal parameters, respectively; P_a and P_n are the altered and nominal parameters, respectively. Negative S values indicate the opposite direction of model response with the regards of the direction of parameter change, and vice versa. We visualized the sensitivity analysis results using the “ComplexHeatmap” package developed by Gu et al. (2016) in R 3.5.3 for Mac OS X (R Development Core Team, 2018).

4 TableSite‐Level Evaluation of the Goodness‐of‐Fit Criteria Computed for the Simulated Fungal and Bacterial Biomass Dynamics in the Calibration and Validation Phases

Phase	Site	Fungi	Bacteria
MAE	RMSE	R2	MAE	RMSE	R2
Calibration	DST‐CH	0.45	0.60	0.473	0.56	0.76	0.374
GRS‐IA	0.29	0.34	0.042	0.18	0.21	0.381
SHB‐OB	0.06	0.07	0.005	0.06	0.07	0.299
TUN‐MH	1.16	1.43	0.251	1.04	1.11	0.398
BRF‐WC	0.28	0.37	0.384	0.63	0.69	0.830
TBF‐VA	0.28	0.31	0.157	0.26	0.30	0.217
TCF‐NJ	0.23	0.24	0.066	0.37	0.38	0.067
TSF‐HS	0.19	0.24	0.156	0.14	0.17	0.002
WET‐EM	0.41	0.73	0.59	0.42	0.72	0.63
Validation	DST‐GB	0.54	0.65	0.513	0.32	0.33	0.269
DST‐JN	0.14	0.14	‐‐	0.15	0.16	‐‐
GRS‐BC	0.30	0.37	0.010	0.20	0.27	0.046
SHB‐AC	0.73	0.79	0.004	1.18	2.02	0.111
TUN‐ES	0.27	0.33	0.411	0.17	0.19	0.014
BRF‐AL	0.41	0.41	‐‐	0.04	0.04	‐‐
TBF‐TL	0.49	0.52	0.064	0.13	0.16	0.625
TBF‐SH	0.17	0.19	0.092	0.24	0.33	0.183
TBF‐MS	0.29	0.35	0.935	0.36	0.39	0.945
TCF‐NT	0.45	0.56	0.014	0.19	0.20	0.450
TSF‐OS	0.01	0.02	0.980	0.19	0.22	0.520
WET‐EF	0.25	0.31	0.071	0.30	0.37	0.052

Abbreviations: MAE, mean absolute error; RMSE, root mean square error; R², R square.

Notes. ‐‐ indicates not applicable. MAE and RMSE values indicate the mean error of the model, smaller values represent higher model performance. R² values mean the proportion of variation being explained by the mode, higher R² values indicate better model performance. Due to the difference in variations of simulated and observed fungal and bacterial biomass and our focus of estimating fungal and bacterial biomass dynamics, we did the evaluation using relative change in fungal and bacterial biomass instead, that is, the difference between simulated/observed fungal/bacterial biomass and the average of simulated/observed fungal/bacterial biomass over the average of simulated/observed fungal/bacterial biomass; R² is not suitable for assessing the goodness‐of‐fit for a small amount of data due to the large bias in small samples.

The uncertainties in FBC, BBC, F:B ratio, SOC, and HR in surface soil (0–30 cm) during the sampling period of each site were quantified using improved Latin Hypercube Sampling (LHS) approach. The LHS approach can randomly produce an ensemble of parameter combinations with high efficiency. This approach has been widely used in the modeling community to estimate uncertainties in model output (Haefner, 2005; Xu, 2010; Xu et al., 2014). First, we assumed that all parameters follow normal distribution, then we used LHS to randomly select an ensemble of 300 parameter sets using the function of “improvedLHS” in the R package “lhs” (Carnell & Carnell, 2019). Second, we computed the inverse of the standard normal cumulative distribution of 300 parameter sets using “norminv” function in MATLAB, 2018b (The MathWorks Inc., Natick, Massachusetts, USA), with the standard deviation of each parameter set as 20% of its optimal value. Third, we added the filter of setting parameters featuring fraction factors (m_rf_s1m, m_rf_s2m, m_rf_s3m, m_rf_s4m, m_batm_f, m_bdom_f, m_bs1_f, m_bs2_f, m_bs3_f, m_fatm_f, m_fdom_f, m_fs1_f, m_fs2_f, m_fs3_f, m_domb_f, m_domf_f, m_doms1_f, m_doms2_f, m_doms3_f) larger than 1 or smaller than 0 or the sum of an array of parameters (m_batm_f, m_bdom_f, m_bs1_f, m_bs2_f, and m_bs3_f, or m_fatm_f, m_fdom_f, m_fs1_f, m_fs2_f, and m_fs3_f, or m_domb_f, m_domf_f, m_doms1_f, m_doms2_f, and m_doms3_f) larger than 1 as their optimal values. Finally, we implemented the 300 model runs for each biome, and we then calculated the 95% confidence interval of FBC, BBC, F:B ratio, SOC, and SR in surface soil (0–30 cm) during experimental period of each site for reporting (Table 6).

Results

Model Parameterization and Validation Against Observational Data

The comparison between modeled and observed data showed that the CLM‐Microbe model can capture the average and seasonal variation of FBC and BBC across biomes (Figures 2–6, Table 5). On average, the simulated FBC and BBC were consistent with the observed values, with FBC and BBC showing R² values of 0.70 (P < 0.001) and 0.26 (P < 0.05), respectively (Figure 6). However, FBC and BBC were underestimated by the CLM‐Microbe model. For example, the simulated FBC was approximately 50% lower than the observed FBC in BRF‐AL (boreal forest, BRF‐) and SHB‐OB (shrub, SHB‐), and 40% lower than the observed FBC in DST‐GB (desert, DST‐) and TBF‐VA and TBF‐MS (temperate broadleaf forest, TBF‐). Simulated BBC was 55% lower than the observed BBC in TUN‐ES and TUN‐MH (tundra, TUN‐) and GRS‐BC (grassland, GRS‐), and 45% lower than the observed BBC in BRF‐AL and TCF‐NT (temperate coniferous forest, TCF‐). The CLM‐Microbe model overestimated FBC compared to the observed values in some sites; specifically, the simulated FBC was higher than the observed FBC in WET‐EF and WET‐EM (wetland, WET‐), DST‐JN, and GRS‐IA. The simulated BBC was higher than the observed BBC in WET‐EF, WET‐EM, DST‐CH, BRF‐WC, SHB‐AC, and TSF‐HS and TSF‐OS (tropical/subtropical forest, TSF‐) (Figures 2–6). To compare the seasonal dynamics between observed and simulated FBC and BBC, y axes in Figures 2–5 were adjusted to facilitate the visualization of modeling results due to the systematic underestimation of FBC and BBC.

2 Figure. Model calibration of fungal biomass for (a) desert, (b) grassland, (c) shrub, (d) tundra, (e) boreal forest, (f) temperate broadleaf forest, (g) temperate coniferous forest, (h) tropical/subtropical forest, and (i) wetland. The blue star indicates the observed fungal biomass, and the black filled circle represents simulated fungal biomass.

3 Figure. Model calibration of bacterial biomass for (a) desert, (b) grassland, (c) shrub, (d) tundra, (e) boreal forest, (f) temperate broadleaf forest, (g) temperate coniferous forest, (h) tropical/subtropical forest, and (i) wetland. The blue star indicates the observed bacterial biomass, and the black filled circle represents simulated bacterial biomass.

4 Figure. Model validation of fungal biomass for (a) and (b) desert, (c) grassland, (d) shrub, (e) tundra, (f) boreal forest, (g, h, and i) temperate broadleaf forest, (j) temperate coniferous forest, (k) tropical/subtropical forest, and l) wetland. The blue star indicates the observed fungal biomass, and the black filled circle represents simulated fungal biomass.

5 Figure. Model alidation of bacterial biomass for (a) and (b) desert, (c) grassland, (d) shrub, (e) tundra, (f) boreal forest, (g, h, and i) temperate broadleaf forest, (j) temperate coniferous forest, (k) tropical/subtropical forest, and l) wetland. The blue star indicates the observed bacterial biomass, and the black filled circle represents simulated bacterial biomass.

6 Figure. Comparison of the averaged observed and simulated (a) fungal and (b) bacterial biomass. The blue star indicates the (a) fungal or (b) bacterial biomass in calibration phase, and the black filled circle represents (a) fungal or (b) bacterial biomass in validation phase; vertical and horizontal error bars indicate standard error of simulated and observed values, respectively, for both (a) fungal and (b) bacterial biomass.

Additionally, we found relatively smaller variation (indicated as standard error, SE) in simulated FBC (SE ranging from 0.04 to 0.90 g C m⁻²) and BBC (SE ranging from 0.02 to 0.91 g C m⁻²) compared with the observed FBC (SE ranging from 0.26 to 99.11 g C m⁻²) and BBC (SE ranging from 0.07 to 12.48 g C m⁻²) (Figure 6). On average, the SE of simulated FBC and BBC was approximately 160 and 40 times smaller than the SE of observed FBC and BBC, respectively. The difference of SE in observed FBC and BBC among biomes was largely dependent on the magnitude of observed FBC and BBC (Figure 6). The highest SE of observed FBC was observed in TBF‐MS (99 g C m⁻²), BRF‐AL (55 g C m⁻²), and TUN‐MH (57 g C m⁻²). The lowest SE of observed FBC were observed in WET‐EF (0.93 g C m⁻²), WET‐EM (0.26 g C m⁻²), and DST‐JN (0.26 g C m⁻²). Compared with SE of observed FBC, SE of observed BBC was much smaller. However, SE of observed BBC were distinct among sites. We observed the highest SE of observed BBC in TUN‐MH (12 g C m⁻²), and TBF‐MS (12 g C m⁻²). The SE of observed BBC was lowest in TBF‐SH (0.40 g C m⁻²), WET‐EF (0.24 g C m⁻²), TSF‐HS (0.18 g C m⁻²), and WET‐EM (0.07 g C m⁻²).

Due to the large difference in variations of simulated and observed FBC and BBC, we estimated the simulated FBC and BBC dynamics using relative change in FBC and BBC, that is, the difference between individual and the average of biomass over the average of biomass during the study period (Table 4). In the calibration phase, MAE ranged from 0.06 to 1.16 for FBC and from 0.06 to 1.04 for BBC, while RMSE ranged from 0.07 to 1.43 for FBC and from 0.07 to 1.11 for BBC. In the validation phase, MAE ranged from 0.01 to 0.73 for FBC and from 0.04 to 1.18 for BBC, while RMSE ranged from 0.02 to 0.79 for FBC and from 0.04 to 2.02 for BBC. Although the model explained the FBC and BBC dynamics well in most sites, particularly TBF‐MS (R² = 0.94 for FBC and R² = 0.95 for BBC) and TSF‐OS (R² = 0.98 for FBC and R² = 0.52 for BBC), the simulation in some sites for FBC (GRS‐IA, SHB‐OB, GRS‐BC, SHB‐AC, TBF‐TL, TBF‐SH, TCF‐NT, and WET‐EF) and BBC (TSF‐HS, GRS‐BC, TUN‐ES, and WET‐EF) was relatively poor, with R² smaller than 0.1. It is noteworthy that the number of data points ranges from 2 to 13 for FBC and BBC data across sites, which is much smaller than 30, the threshold for a large sample. The R² may not be suitable for assessing the goodness‐of‐fit for a small number of data points due to the large bias in small samples, although it is widely used in model performance estimation. For example, R² was 0.005 for FBC and 0.299 for BBC in SHB‐OB, while MAE (0.06 for FBC and BBC) and RMSE (0.07 for FBC and BBC) were small, indicating small biases in the simulated FBC and BBC.

Sensitivity Analysis

We found high sensitivity of FBC and BBC dynamics to parameters that related to microbial turnover and C:N ratio of fungal and bacterial biomass across biomes (Figure 7). Fungal biomass turnover rate (k_fungi) had negative effects on FBC and F:B ratio across biomes, while bacterial biomass turnover rate (k_bacteria) had negative and positive effects on BBC and F:B ratio, respectively. The C:N ratio of bacterial biomass (cn_bacteria) had negative, positive, and negative effects on FBC, BBC, and F:B ratio, respectively. In contrast, C:N ratio of fungal biomass (cn_fungi) had positive, negative, and positive effects on FBC, BBC, and F:B ratio, respectively. A 20% increase or decrease in k_fungi and k_bacteria led to different magnitudes of negative effects on FBC and BBC, respectively, across biomes. While a 20% decrease in k_fungi and k_bacteria led to the S values around ‐1.25 and ‐1.30 for FBC and BBC, a 20% increase in k_fungi and k_bacteria resulted in S values around ‐0.85 and ‐0.84 for FBC and BBC, respectively. Changes in k_bacteria had similar magnitudes of positive effects on F:B ratio, with S values around 1.00 for both 20% increase and decrease in k_bacteria. While a 20% increase or decrease in k_fungi had different magnitudes of negative effects on F:B ratio, with a 20% decrease or increase in k_fungi leading to S values around −1.27 and −0.85, respectively, across biomes. The higher S values suggested that FBC and BBC were more sensitive to decreases in k_fungi and k_bacteria, which would induce larger increases in FBC and BBC; in particular, a 20% decrease in k_fungi led to higher sensitivity of F:B ratio.

7 Figure. Sensitivity analysis for model response of fungal biomass, bacterial biomass, F:B ratio, soil organic carbon, and soil microbial respiration in the top 30 cm to 25 parameters (m_bdom_f, m_bs1_f, m_bs2_f, m_bs3_f, m_fdom_f, m_fs1_f, m_fs2_f, m_fs3_f, k_dom, k_bacteria, k_fungi, m_rf_s1m, m_rf_s2m, m_rf_s3m, m_rf_s4m, m_batm_f, m_fatm_f, m_domb_f, m_domf_f, m_doms1_f, m_doms2_f, m_doms3_f, cn_bacteria, cn_fungi, and CUEmax) in (a) desert, (b) grassland, (c) shrub, (d) tundra, (e) boreal forest, (f) temperate broadleaf forest, (g) temperate coniferous forest, (h) tropical/subtropical forest, and (i) wetland. “+” and “‐” indicate 20% increase or 20% decrease of parameter values. CUE, carbon use efficiency; FBC, fungal biomass carbon; BBC, bacterial biomass carbon; F:B ratio, fungal:bacterial biomass carbon ratio; SOC, soil organic carbon; HR, heterotrophic respiration; Dark red and darker blue indicate a stronger positive or negative model response of the variable to parameter change. S is negative if the direction of model response opposes the direction of parameter change during the sampling years for all sites.

In addition, FBC and BBC were closely correlated with the decomposition of SOM and DOM and microbial respiration (Figure 7). Fraction factors quantifying C flow from SOM to soil microbes (m_rf_s1m, m_rf_s2m, m_rf_s3m, and m_rf_s4m) were positively correlated with FBC and BBC. However, the magnitude of the m_rf_s4m on FBC and BBC were different among biomes. The highest response was in TSF‐HS (S = 1.03 for FBC and S = 1.08 for BBC with ‐20% change vs. S = 1.42 for FBC and S = 1.49 for BBC with +20% change), while the lowest response was in BRF‐WC (S = 0.28 for FBC and S = 0.26 for BBC with −20% change vs. S = 0.28 for FBC and S = 0.24 for BBC with +20% change). In addition, we found a positive correlation between the fraction factor quantifying C flow from DOM to fungal biomass (m_domf_f) and FBC and F:B ratio and a positive correlation between fraction factor quantifying C flow from DOM to bacterial biomass (m_domb_f) and BBC, and a negative correlation between m_domb_f and F:B ratio. However, the magnitudes varied among biomes, with FBC being most sensitive to m_domf_f in GRS‐IA (S = 0.432), BBC being most sensitive to m_domb_f in TBF‐VA (S = 0.593), and F:B ratio being most sensitive to m_domf_f in GRS‐IA (S = 0.415) and m_domb_f in TBF‐VA (S = −0.630) with ‐20% change. In contrast, FBC, BBC, and F:B ratio were insensitive to m_domb_f and m_domf_f in BRF‐WC, with the absolute S values close to 0.001. Moreover, we found weak positive effects of fraction factors quantifying C flow from DOM to SOM (m_doms1_f, m_doms2_f, and m_doms3_f) on FBC and BBC across biomes. Across biomes, FBC, BBC, F:B ratio, SOC, and HR in surface soil (0–30 cm) were insensitive to maximum microbial CUE (CUE_max). We observed a higher negative correlation between fraction factor quantifying C being respired by fungi (m_fatm_f) and FBC and between the fraction factor quantifying C being respired by bacteria (m_batm_f) and BBC in GRS‐IA, TSF‐SH, and TCF‐NJ, with S values ranging from −0.104 to −0.044 of m_fatm_f for FBC and from −0.615 to −0.023 of m_batm_f for BBC, while FBC and BBC in other biomes were insensitive to m_batm_f and m_fatm_f change.

The SOC was positively correlated with fraction factors quantifying C flow from SOM to soil microbes (m_rf_s1m, m_rf_s2m, m_rf_s3m, or m_rf_s4m), while SOC was negatively correlated with fraction factor quantifying C being respired by bacteria (m_batm_f) and fungi (m_fatm_f), but the magnitudes varied among biomes (Figure 7). We found highly positive correlations between m_rf_s1m, m_rf_s2m, m_rf_s3m, or m_rf_s4m and SOC across biomes except for tundra and boreal forest. The parameters of m_batm_f and m_fatm_f were negatively correlated with SOC, but the magnitudes varied among biomes. Specifically, we found higher negative correlation between m_batm_f and SOC in TBF‐VA (S = −0.10 with −20% and +20% change), TCF‐NJ (S = −0.08 with ‐20% and +20% change), and TSF‐HS (S = −0.10 with −20% and +20% change). While m_fatm_f was more negatively correlated with SOC in GRS‐IA (S = −0.12 with −20% and +20% change), TCF‐NJ (S = −0.10 with −20% and +20% change), and TSF‐HS (S = −0.17 with −20% and +20% change). The SOC in other biomes were relatively insensitive to changes in m_batm_f (S values ranging from −0.06 to −0.02) and m_fatm_f (S values ranging from −0.05 to −0.03). Across biomes, we only found strong negative correlations between m_doms1_f, m_doms2_f, and m_doms3_f and SOC in TCF‐NJ (S = −0.08 for m_doms1, S = −0.12 for m_doms2, and S = −0.18 for m_doms3).

The HR widely responded of to all parameters listed for soil microbial mechanisms, but in low sensitivity (Figure 7). Specifically, the HR in sites such as DST‐CH, GRS‐IA, TBF‐VA, TCF‐NJ, and TSF‐HS showed a weak response to changes in all parameters. However, we also found relatively stronger negative correlations between m_batm_f and HR in BRF‐WC (S = −0.09 with −20% change and S = −0.12 with +20% change) and WET‐EM (S = −0.08 with −20% change and S = −0.11 with +20% change).

Simulated FBC, BBC, F:B Ratio, SOC, and HR at Annual Scale

Annual estimation of FBC, BBC, F:B ratio, SOC, and HR in the top 30 cm soils derived from the CLM‐Microbe model showed large variations among biomes (Table 5). The simulated FBC was the highest in TCF‐NJ (380 g C m⁻²), followed by TBF‐VA (259 g C m⁻²), BRF‐WC (205 g C m⁻²), and TUN‐MH (180 g C m⁻²), while it was lowest in WET‐EM (24 g C m⁻²) with a range of 14–40 g C m⁻². The simulated FBC in the top 30 cm of soils in DST‐CH, GRS‐IA, SHB‐OB, TUN‐MH, BRF‐WC, TBF‐VA, TCF‐NJ, TSF‐HS, and WET‐EM was 1.6 (0–15 cm), 5.9 (0–10 cm), 0.9 (0–10 cm), 1.7 (0–10 cm), 4.0 (5–20 cm), 2.1 (0–5 cm), 14.0 (0–2.3 cm), 1.7 (0–10 cm), and 22.6 (0–20 cm) times of the observed FBC (at varying soil depths), respectively. Compared with the global data set of FBC and BBC (He et al., 2020), the simulated FBC in the top 30 cm of soils was generally consistent with the biome‐averaged FBC in the top 30 cm of the soils. The model simulated FBC in DST‐CH, GRS‐IA, TUN‐MH, BRF‐WC, and TSF‐HS was similar with the biome‐averaged FBC in the top 30 cm of soils. However, we detected extreme FBC simulated by the model in some sites compared with the biome‐averaged FBC. The CLM‐Microbe model simulated higher FBC in SHB‐OB, TBF‐VA and TCF‐NJ and lower FBC in WET‐EM relative to their corresponding the biome‐averaged FBC in the top 30 cm soils.

5 TableKey Parameters Used in Model Sensitivity Analysis and Uncertainty Analysis

Parameters	Ecological meanings
k_dom	Decomposition rate constant of dissolved organic matter
k_bacteria	Lysis rate constant of bacteria
k_fungi	Lysis rate constant of fungi
m_rf_s1m	Fraction factor quantifying carbon from soil organic matter1 to microbes
m_rf_s2m	Fraction factor quantifying carbon from soil organic matter2 to microbes
m_rf_s3m	Fraction factor quantifying carbon from soil organic matter3 to microbes
m_rf_s4m	Fraction factor quantifying carbon from soil organic matter4 to microbes
m_batm_f	Fraction factor quantifying carbon respired by bacteria
m_bdom_f	Fraction factor quantifying carbon from dissolved organic matter to bacteria
m_bs1_f	Fraction factor quantifying carbon from bacteria to soil organic matter1
m_bs2_f	Fraction factor quantifying carbon from bacteria to soil organic matter2
m_bs3_f	Fraction factor quantifying carbon from bacteria to soil organic matter3
m_fatm_f	Fraction factor quantifying carbon respired by fungi
m_fdom_f	Fraction factor quantifying carbon from dissolved organic matter to fungi
m_fs1_f	Fraction factor quantifying carbon from fungi to soil organic matter1
m_fs2_f	Fraction factor quantifying carbon from fungi to soil organic matter2
m_fs3_f	Fraction factor quantifying carbon from fungi to soil organic matter3
m_domb_f	Fraction factor quantifying carbon from dissolved organic matter to bacteria
m_domf_f	Fraction factor quantifying carbon from dissolved organic matter to fungi
m_doms1_f	Fraction factor quantifying carbon from dissolved organic matter to soil organic matter1
m_doms2_f	Fraction factor quantifying carbon from dissolved organic matter to soil organic matter2
m_doms3_f	Fraction factor quantifying carbon from dissolved organic matter to soil organic matter3
cn_bacteria	C:N ratio of bacteria
cn_fungi	C:N ratio of fungi
CUEmax	Maximum carbon use efficiency of microbes

Abbreviation: CUE, carbon use efficiency.

The simulated BBC was the highest in TCF‐NJ (100 g C m⁻²), followed by TBF‐VA (70 g C m⁻²), BRF‐WC (40 g C m⁻²), and TUN‐MH (30 g C m⁻²), while the simulated BBC was the lowest in WET‐EM (7 g C m⁻²) and DST‐CH (9 g C m⁻²). The simulated BBC in the top 30 cm of soils in DST‐CH, GRS‐IA, SHB‐OB, TUN‐MH, BRF‐WC, TBF‐VA, TCF‐NJ, TSF‐HS, and WET‐EM was 5.0 (0–15 cm), 2.1 (0–10 cm), 2.2 (0–10 cm), 1.0 (0–10 cm), 4.1 (5–20 cm), 2.3 (0–5 cm), 13.5 (0–2.3 cm), 6.3 (0–10 cm), and 25.4 (0–20 cm) times of the observed BBC (at varying soil depths), respectively. Compared with the global data set of FBC and BBC (He et al., 2020), the simulated BBC in DST‐CH, GRS‐IA, TUN‐MH, and BRF‐WC was similar with their corresponding biome‐averaged BBC in the top 30 cm soils. However, the simulated BBC was higher in SHB‐OB, TBF‐VA, and TCF‐NJ and lower in TSF‐HS and WET‐relative to their corresponding biome‐averaged BBC in the top 30 cm soils.

The simulated F:B ratio was the highest in TUN‐MH (6.30), followed by BRF‐WC (5.39), DST‐CT (4.54), SHB‐OB (4.16), and TCF‐NJ (3.96), the F:B ratio was the lowest in TSF‐HS (2.31). The observed F:B ratio was highly variable, with the highest F:B ratio in DST‐CH (14.1, 0–15 cm), followed by SHB‐OB (9.1, 0–10 cm) and BRF‐WC (5.3, 5–20 cm), while GRS‐IA (1.1, 0–10 cm) featured the lowest F:B ratio among biomes. Compared with our recently compiled global data set of FBC and BBC (He et al., 2020), the CLM‐Microbe model simulated F:B ratio was generally consistent with the biome‐averaged F:B ratio in the top 30 cm soils. Similar as the CLM‐Microbe model simulated F:B ratio, the highest biome‐averaged F:B ratio was found in tundra (8.6), followed by boreal forests (5.0), temperate forests (4.9), and shrub (4.8), while the lowest biome‐averaged F:B ratio was found in tropical/subtropical forests (2.2).

Large variations were found in the simulated SOC of top 30 cm among biomes, with SOC highest in WET‐EM (13,204 g C m⁻²), which was 5.3 times of that in the site with the lowest values, that is, TSF‐HS (2,509 g C m⁻²). BRF‐WC (8,685 g C m⁻²) has the second largest SOC, followed by TCF‐NJ (7,990 g C m⁻²), TBF‐VA (5,873 g C m⁻²), and TUN‐MH (3,604 g C m⁻²) (Table 5). Similar as the CLM‐Microbe model, CLM4.5 simulated‐SOC in top 30 cm was high in WET‐EM (9,951 g C m⁻²) and low in SHB‐OB (1,598 g C m⁻²) and TUN‐MH (2086 g C m⁻²). In contrast, CLM4.5 simulated‐SOC in top 30 cm was much higher in GRS‐IA (11,991 g C m⁻²), TBF‐VA (11,239 g C m⁻²), and TSF‐HS (6,628 g C m⁻²) compared with that simulated by the CLM‐Microbe model. It is worthwhile to note that DST‐CH (0 g C m⁻²) featured the lowest SOC in the top 30 cm simulated by CLM4.5 among sites owing to the vegetation mortality. The simulated SOC in the top 30 cm was slightly lower than the average derived from a global data set of SOC (Jobbágy & Jackson, 2000). However, the simulated SOC in BRF‐WC was slightly higher than that of biome‐averaged SOC in the top 30 cm. Excluding wetlands due to lack of available data, the biome‐averaged SOC is consistent with the simulated SOC in the top 30 cm. The SOC is higher in temperate broadleaf forest (10,875 g C m⁻²), temperate coniferous forest (8,483 g C m⁻²), and tundra (7,739 g C m⁻²). In contrast to the lowest simulated SOC in the top 30 cm in TSF‐HS (2,509 g C m⁻²), the biome‐averaged SOC in the top 30 cm was lowest in the desert (2,728 g C m⁻²).

The fraction of SOC in microbial biomass (FBC and BBC combined) showed a large variation among biomes, with the proportion ranging from 0.2% to 6.0% (Table 5). The proportion was the highest in TCF‐NJ (6.0%), followed by TUN‐MH (5.80%), TBF‐VA (5.60%), SHB‐OB (5.50%), and GRS‐IA (5.40%), WET‐EM (0.23%) featured the lowest proportion between the sum of FBC and BBC and SOC among biomes. Similarly, the biome‐averaged proportion between the sum of FBC and BBC and SOC ranged from 1.1% to 6.2% among biomes. However, the rank of the biome‐averaged proportion between the sum of FBC and BBC and SOC was different with that simulated by the CLM‐Microbe model. The proportion between the sum of FBC and BBC and SOC was highest in boreal forests (6.2%) and lowest in temperate broadleaf forests (1.1%). Tundra had the second highest proportion between the sum of FBC and BBC and SOC (3.4%), followed by desert (2.7%), grassland (2.2%), shrub (1.5%), and then tropical/subtropical forest and temperate coniferous forest (1.4%).

The CLM‐Microbe model simulated annual estimation of HR was the highest in TCF‐NJ (811 g C m⁻² yr⁻¹), which was about nine times of that in the lowest site, that is, TUN‐MH (93 g C m⁻² yr⁻¹) (Table 5). We found second highest HR in TBF‐VA (560 g C m⁻² yr⁻¹), followed by WET‐EM (540 g C m⁻² yr⁻¹), BRF‐WC (420 g C m⁻² yr⁻¹), and TSF‐HS (401 g C m⁻² yr⁻¹). Compared with the CLM‐Microbe model, the CLM4.5 simulated highly consistent HR in TBF‐VA, WET‐EM, TSF‐HS, SHB‐OB, and TUN‐MH. However, we observed much higher HR in GRS‐IA simulated by the CLM4.5 (757 g C m⁻² yr⁻¹), which was 2.4 times of that simulated by the CLM‐Microbe model (313 g C m⁻² yr⁻¹).

Simulated Time‐Series C Flow into and out of FBC and BBC

The dynamics and magnitude of C flow related to the decomposition of litter, SOM, and DOM, HR, and microbial lysis for FBC and BBC differed among biomes, while the seasonal patterns of those C flows were similar for fungi and bacteria in each biome (Figures 8 and 9). In the CLM‐Microbe model, FBC and BBC assimilate C from the decomposition of litter, SOM, and DOM, the dominance of these C flows varied temporally and among biomes. The C flow from litter decomposition predominated fungal and bacterial C assimilation year‐round at a few sites such as DST‐CH (Figures 8a and 9a), TUN‐MH (Figures 8i and 9i), and TSF‐HS (Figures 8h and 9h). We also found the dominant role of litter decomposition on fungi and bacteria C assimilation in GRS‐IA (Figures 8b and 9b), TBF‐VA (Figures 8f and 9f), and WET‐EM during the nongrowing seasons (Figures 8i and 9i). The fungal and bacterial C assimilation at sites such as BRF‐WC (Figures 8e and 9e), SHB‐OB (Figures 8c and 9c), and TCF‐NJ (Figures 8g and 9g) were co‐dominated by the decomposition of litter and SOM. The decomposition of DOM is the least important pathway for fungal and bacterial C gain across biomes; however, we observed the predominant role of the DOM decomposition on fungal and bacterial C assimilation at TBF‐VA (Figures 8f and 9f) and WET‐EM (Figures 8i and 9i) during the whole year, and second largest C assimilation of DOM decomposition at TUN‐MH (Figures 8d and 9d), and temporarily dominant role of DOM decomposition during winter and spring at SHB‐OB (Figures 8c and 9c).

8 Figure. Time‐series of simulated carbon flow into and out of the bacterial biomass carbon pool.

9 Figure. Time‐series of simulated carbon flow into and out of the bacterial biomass carbon pool.

The C loss from fungal and bacterial biomass was primarily represented as microbial respiration and microbial lysis in the CLM‐Microbe model. The C flow from fungal and bacterial biomass to SOM during microbial lysis was the predominant mechanism of C loss at DST‐CH (Figures 8a and 9a), SHB‐OB (Figures 8c and 9c), BRF‐WC (Figures 8e and 9e), TBF‐VA (Figures 8f and 9f), and TCF‐NJ (Figures 8g and 9g). However, we also observed the co‐dominance of C flow from fungal and bacterial biomass to SOM and microbial respiration in controlling fungal and bacterial C loss at GRS‐IA (Figures 8b and 9b), and TUN‐MH (Figures 8d and 9d). The fungal and bacterial C loss were co‐determined by C flow from fungal and bacterial biomass to SOM and DOM and microbial respiration at WET‐EM (Figures 8i and 9i). While we found the predominant role of microbial respiration in regulating fungal and bacterial C loss in TSF‐HS, microbial lysis contributed less to fungal and bacterial C loss (Figures 8h and 9h).

Discussion

Model Performance and Comparison with Existing Models

The CLM‐Microbe model simulated FBC and BBC are consistent with the observed FBC and BBC, with a slight underestimation (Figures 2–6; Table 6). Similar as our study, Wang et al. (2015) reported that the MEND model can adequately capture the soil microbial biomass C dynamics with the representation of soil microbial processes such as microbial dormancy, microbial enzyme production, and enzyme catalyzing effects on decomposition. The TRIPLEX‐MICROBE model can also estimate the global‐ and biome‐level soil microbial biomass C with reasonable accuracy (Wang et al., 2017). Meanwhile, studies found that soil microbial traits play a key role in soil microbial biomass accumulation. Wang et al. (2015) compared the simulated soil microbial biomass C by MEND with and without dormancy, and they found that MEND model without dormancy largely underestimated the soil microbial biomass C. In the CLM‐Microbe model, the soil microbial community, represented by active fungi and bacteria, is directly related to many biogeochemical processes such as decompositions of litter, DOM, and SOM. The accumulations of FBC and BBC could largely affect soil respiration and soil C pools such as SOM. To ensure reasonable soil conditions, we finalized the parameters related to soil microbial processes by ensuring consistency between simulated SOC with the SOC reported in global datasets. Therefore, the missing representation of dormancy may be responsible for the slight underestimation of FBC and BBC in the CLM‐Microbe model.

6 TableAnnual Estimates of Fungal and Bacterial Biomass Carbon, Fungal: Bacterial (F:B) Ratio, Soil Organic Carbon (SOC), and Heterotrophic Respiration (HR) With the Uncertainties of Parameters During the Sampling Years for all Sites

Source	Site/Biome	Fungal biomass (g C m−2)	Bacterial biomass (g C m−2)	F:B ratio	SOC (g C m−2)	HR (g C m‐2 yr−1)
CLM‐Microbe simulateda	DST‐CH	38.65 (21.69∼62.97)	8.90 (5.21∼15.17)	4.54 (2.15∼7.94)	2567.62 (2011.82∼3148.15)	277.41 (269.87∼282.88)
GRS‐IA	110.25 (66.94∼182.63)	33.57 (21.20∼53.26)	3.44 (1.68∼5.94)	2677.50 (2225.43∼3104.13)	312.85 (307.67∼325.92)
SHB‐OB	148.74 (86.44∼253.85)	37.33 (22.44∼61.57)	4.16 (2.07∼7.47)	3363.62 (2908.50∼3829.85)	120.69 (117.80∼123.25)
TUN‐MH	180.25 (111.67∼283.71)	29.89 (19.44∼46.77)	6.30 (3.23∼10.70)	3603.51 (3460.15∼3741.94)	92.75 (90.15∼95.48)
BRF‐WC	204.76 (124.98∼328.30)	40.10 (27.41∼63.30)	5.39 (2.61∼9.45)	8685.33 (8133.04∼9341.41)	420.07 (367.18∼452.30)
TBF‐VA	258.79 (161.16∼429.01)	70.47 (45.63∼114.71)	3.83 (1.89∼6.49)	5872.70 (5221.21∼6540.73)	560.48 (538.93∼574.73)
TCF‐NJ	379.71 (213.91∼683.37)	100.15 (59.46∼166.58)	3.96 (1.86∼6.95)	7989.60 (6411.33∼9961.35)	811.32 (800.71∼818.67)
TSF‐HS	37.02 (19.70∼62.50)	16.66 (9.11∼29.09)	2.31 (1.14∼4.19)	2509.49 (1694.83∼3422.36)	400.69 (393.76∼404.51)
WET‐EM	23.62 (13.52∼39.54)	6.63 (4.13∼11.03)	3.75 (1.85∼6.67)	13203.92 (11732.67∼14385.34)	540.32 (524.52∼556.59)
CLM4.5 simulateda	DST‐CH	NA	NA	NA	0.00	0.00
GRS‐IA	NA	NA	NA	11990.54	756.86
SHB‐OB	NA	NA	NA	1598.15	63.07
TUN‐MH	NA	NA	NA	2085.97	94.61
BRF‐WC	NA	NA	NA	3465.00	157.68
TBF‐VA	NA	NA	NA	11238.67	599.18
TCF‐NJ	NA	NA	NA	4920.45	315.36
TSF‐HS	NA	NA	NA	6627.84	536.11
WET‐EM	NA	NA	NA	9950.83	567.65
Observedb	DST‐CHd	24.79 (13.78∼60.04)	1.76 (0.89∼4.95)	14.05 (12.12∼21.13)	NA	NA
GRS‐IAe	18.55 (9.32–28.53)	16.17 (12.07–22.99)	1.15 (0.57–2.22)	NA	NA
SHB‐OBe	156.65 (143.46–171.62)	17.13 (16.09–18.61)	9.14 (8.60–9.62)	3559.97e	NA
TUN‐MHe	105.89 (1.10–439.88)	29.63 (0.38–84.42)	3.57 (1.38–5.21)	1945.00e k	NA
BRF‐WCf	51.71 (27.07–100.96)	9.70 (1.29–20.55)	5.33 (2.92–36.93)	NA	NA
TBF‐VAg	122.78 (60.12–171.88)	31.01 (18.30–43.39)	3.96 (2.88–5.15)	NA	NA
TCF‐NJh	27.09 (18.05–35.35)	7.41 (3.82–10.77)	3.65 (3.04–5.08)	NA	NA
TSF‐HSe	22.33 (13.81–29.74)	2.65 (1.85–3.14)	8.43 (6.38–9.63)	4238.08e	NA
WET‐EMi	1.05 (0.25–3.74)	0.26 (0.08–0.97)	4.01 (3.02–6.33)	10819j	NA
Biome averagec	DST	59.04 (14.05–74.00)	15.28 (0.32–60.21)	3.14 (2.20–4.49)	2728.00 (2651.58–2804.42)	NA
GRS	88.69 (20.55–132.48)	46.14 (2.34–114.23)	4.03 (3.52–4.62)	6225.00 (5796.59–6653.41)	NA
SHB	48.06 (11.40–64.78)	17.31 (0.59–53.85)	4.82 (3.72–6.25)	4450.00 (4207.30–4692.70)	NA
TUN	226.96 (150.89–256.46)	32.65 (3.13–108.08)	8.60 (6.71–11.01)	7739.00 (6589.57–8888.43)	NA
BRF	304.44 (191.19–356.01)	58.66 (7.02–171.86)	5.03 (4.23–5.98)	5812.50 (5601.96–6023.04)	NA
TBF	88.89 (40.25–115.74)	29.88 (3.01–78.5)	4.92 (4.39–5.51)	10875.00 (9807.51–11942.49)	NA
TCF	88.89 (40.25–115.74)	29.88 (3.01–78.5)	4.92 (4.39–5.51)	8482.50 (7974.46–8990.54)	NA
TSF	64.42 (2.09–115.49)	51.58 (0.51–113.82)	2.22 (1.87–2.63)	8514.00 (7686.89–9341.11)	NA
WET	70.44 (30.9–99.91)	32.96 (3.47–72.44)	4.13 (3.50–4.86)	NA	NA

^aSimulated fungal and bacterial biomass, fungal:bacterial (F:B) ratio, soil organic carbon, and heterotrophic respiration of top 30 cm soil profile using either the CLM‐Microbe model or CLM4.5, values in the parentheses were 95% confidence interval.

^bObserved fungal and bacterial biomass, F:B ratio, soil organic carbon, and heterotrophic respiration from literature, values were expressed as average (range), and the values showed may not be the same as those from literature sources due to unit conversion.

^cBiome average of fungal and bacterial biomass and F:B ratio were from He et al. (2020) of top 30 cm soil profile, while soil organic carbon of top 30 cm were from Jobbágy and Jackson (2000) assuming that 20–30 cm and 30–40 cm contribute equally to the soil organic carbon in 20–40 cm layer; BRF, boreal forest; DST, desert; GRS, grassland; SHB, shrub; TBF, temperate broadleaf forest; TCF, temperate coniferous forest; TSF, tropical/subtropical forest; TUN, tundra; WET, wetland.

^dRepresent observed fungal and bacterial biomass, F:B ratio, soil organic carbon, and heterotrophic respiration of 0–15 cm.

^eRepresent observed fungal and bacterial biomass, F:B ratio, soil organic carbon, and heterotrophic respiration of 0–10 cm.

^fRepresent observed fungal and bacterial biomass, F:B ratio, soil organic carbon, and heterotrophic respiration of 5–20 cm.

^gRepresent observed fungal and bacterial biomass, F:B ratio, soil organic carbon, and heterotrophic respiration of 0–5 cm.

^hRepresent observed fungal and bacterial biomass, F:B ratio, soil organic carbon, and heterotrophic respiration of 0–2.3 cm.

ⁱRepresent observed fungal and bacterial biomass, F:B ratio, soil organic carbon, and heterotrophic respiration of 0–20 cm.

^jObserved soil organic carbon of 18–60 cm.

^kRecalculated SOC from Björk et al. (2007); NA means no data available.

Additionally, the CLM‐Microbe model simulated FBC and BBC showed smaller variation compared with the observed FBC and BBC, respectively (Figure 6). Soil microbial communities are not static, with microbial biomass showing temporal dynamics (Björk et al., 2008; Lipson et al., 2002; Lipson & Schmidt, 2004). This variation is highly associated with environmental factors such as soil temperature and soil moisture (Devi & Yadava, 2006). Meanwhile, plant‐produced C also control soil microbial growth, as confirmed by studies that reported positive correlation between soil microbial biomass and aboveground litter input (Feng et al., 2009) and root exudates (Göttlicher et al., 2006). Although we incorporated the vegetation effects in soil microbial biomass in the form of DOM released into soil, the decomposition of DOM will promote C availability for soil microbes, the stimulating effects of DOM on microbial activity was not included in the present CLM‐Microbe model. In addition, soil food web was not explicitly incorporated into the CLM‐Microbe model, even the turnover rate of soil microbial biomass was probably regulated by their predators such as nematodes, mites, and protozoa (CPD et al., 1995; Ingham et al., 1986). The predator activity can induce abrupt changes in soil microbial biomass; for example, Buckeridge et al. (2013) observed seasonal variation in soil microbial community structure, and the decline in FBC from winter to late winter, and then again in spring, was closely associated with the high abundance of protozoa. Furthermore, only a small proportion of the soil microbial community is active, while the majority is dormant, that is, a reversible state of low to zero metabolic activity (Cole, 1999). Soil microbes can determine whether the environmental conditions are suitable for microbes to remain viable within short time periods (Garcia‐Pichel & Pringault, 2001). Therefore, the rapid change in soil microbial state can lead to abrupt changes in soil microbial biomass (He et al., 2015). However, in the CLM‐Microbe model, we assumed that the activity of fungi and bacteria are regulated by soil environmental conditions such as soil temperature, soil moisture, and soil oxygen concentration (Section 2.2).

Controls on Soil Microbial Community Composition

Turnover rates of FBC and BBC are the most important factors regulating FBC and BBC dynamics across biomes, respectively, with increasing turnover rate of fungi decreased FBC and F:B ratio, and increasing turnover rate of bacteria decreased BBC and increased F:B ratio (Figure 7). Turnover rate, the inverse of lifespan, can be mathematically calculated by dividing the production by the biomass pool size. In the microbial world, biomass turnover is much faster relative to that of plants and animals in natural environments, microbial related biogeochemical fluxes are closely linked to turnover and succession of microbial communities (Schmidt et al., 2007). Higher estimates for biomass production consequently correspond to lower turnover times, and vice versa (Pritchard et al., 2008; Rousk & Bååth, 2007). Therefore, the increase in turnover rate of fungi and bacteria is expected to induce declining FBC and BBC, respectively. The increasing turnover rate of fungi will decrease the dominance of fungi, and thus narrowed F:B ratio. In contrast, the increasing bacterial turnover rate would suppress the bacterial dominance and thus a broadening F:B ratio.

In addition, we observed the important role of fungal and bacterial biomass C:N ratio in regulating FBC, BBC, and F:B ratio, with rising fungal biomass C:N ratio increased FBC and F:B ratio and rising bacterial biomass C:N ratio increased BBC and decreased F:B ratio (Figure 7). The C cycle is closely coupled with that of other essential elements, and the proportion of substrate C being respired by soil microbes is closely related to substrate C:N ratio, with more C being respired when substrate has high C:N ratio or low N concentration (Spohn, 2015). In addition, the ratio between the substrate and the microbial biomass C:N ratio determines the proportion of C being assimilated by soil microbes (Sinsabaugh et al., 2013), which was directly reflected as CUE and adopted into the CLM‐Microbe model. Fungi and bacteria have distinct C and nutrient compositions, with C:N ratio averaged around 5 for bacteria and 12 for fungi (Strickland & Rousk, 2010), their C:N ratios specifically determine the partitioning coefficient of C between fungi and bacteria in the CLM‐Microbe model. Increasing fungal biomass C:N ratio will thus increase the proportion of C assimilated by fungi and promote the dominance of fungi, while increasing bacterial biomass C:N ratio stimulates the C flow toward bacteria and suppresses the fungal C gain, and thereby decreasing F:B ratio.

The FBC, BBC, and F:B ratio were also positively affected by the C flow from SOM to soil microbes. We observed higher sensitivity of FBC and BBC in desert and tropical/subtropical sites to changes in C flow from SOM to soil microbes (Figure 7). The increase in microbial C gain from SOM will enhance the C and energy availability for soil microbial growth, which is thus reflected as an increase in soil microbial biomass. In the model, fungal and bacterial C gain from SOM is first expressed as the C input from SOM to soil microbes as a whole, then the C was partitioned by FBC and BBC pools based on the C:N ratio of their biomass, the biomass pool with higher C:N ratio is expected to gain higher proportion of C (Figure 1; Section 2.2). In other words, the C flow from SOM to soil microbes will determine the overall received C from SOM for both fungi and bacteria. Although there are large variations in fungal (3–60) and bacterial (3–12) biomass C:N ratio, indicating that a large proportion of fungi and bacteria overlap with regards to biomass C:N ratio, fungi tend to have higher mean C:N ratio (Strickland & Rousk, 2010). Accordingly, we assigned higher C:N ratio for fungal (cn_fungi) relative to bacterial (cn_bacteria) biomass, and the reasonability of these parameters were validated and tested by observed FBC and BBC dynamics (Table 3). As a result, the increase in microbial C gain from SOM will enhance FBC and BBC; however, FBC increase will be promoted more due to its higher C:N ratio, thereby exhibiting an increase in F:B ratio.

The FBC, BBC, and F:B ratio in DST‐CH and TSF‐SH showed relatively higher sensitivity to soil microbial C assimilation from SOM decomposition, which is likely result from the favorable soil moisture and temperature for decomposition. Tropical and subtropical forests are known to have higher decomposition rate due to the high annual precipitation and temperature, while decomposition in deserts is widely reported to be limited by soil moisture (Chapin et al., 2011). Despite the general recognition of water limitation in deserts, the desert site (DST‐CH) in this study has a mean annual precipitation of 380 mm, which is pretty high compared to “common” deserts (Bell et al., 2014). Given the high sensitivity of SOM decomposition to soil moisture condition, the higher water availability may enhance SOM decomposition in TSF‐SH and DST‐CH (Chapin et al., 2011). Furthermore, DST‐CH is vegetated by herbaceous plant species, the higher proportion of nonwoody components will improve the decomposability of substrates (Koven et al., 2013). In addition, although water limitation decreases the activity of soil microbes, primarily bacteria, fungi are more tolerant to drought due to their hyphal water uptake capability and dominate SOM decomposition in dry environmental conditions such as deserts (Yuste et al., 2011). Therefore, the decomposition of SOM plays an important role for fungal and bacterial C turnover, and fungal growth tends to be promoted more due to higher biomass C:N ratio and tolerance to water stress, leading to an increase in F:B ratio.

Controls on Soil Organic C Density and HR

The C flow from SOM to soil microbes enhanced SOC, but this enhancement is weak in tundra and boreal forest; soil microbial respiration strongly decreased SOC in forests except for boreal forest (Figure 7). In addition to processing SOM from other organic C forms, soil microbes contribute to the formation of persistent SOM via necromass (Gougoulias et al., 2014). Soil microbial necromass is about three orders of magnitude higher than soil microbial biomass (Glaser et al., 2004), and can make up more than half of SOC (Liang et al., 2019). Soil microbial necromass is directly related to soil microbial biomass turnover rate, the slow biomass turnover rate in boreal forests and tundra might be one of the important reasons for low SOC sensitivity to soil microbial biomass C gain from SOM (Table 3). Moreover, boreal forests and tundra are known to have low soil temperature, indicating smaller temperature effects on soil microbial community lysis. As a result, low biomass turnover rate of fungi and bacteria as well as low soil temperature may lead to the long persistence of organic C in soil microbial biomass, resulting in the lower contribution of soil microbial necromass to SOM. Compared with boreal forests, soil moisture and soil temperature in temperate and tropical forests are relatively desirable for soil microbial activity, and thus the decomposition process is more favorable in temperate and tropical forests. Higher decomposition in temperate and tropical forests is indicated by the C released as CO₂ by fungi and bacteria in the CLM‐Microbe model (Table 6). Therefore, in temperate and tropical forests, the higher proportion of C respired by fungi and bacteria will more prominently decrease C remaining in the ecosystem, leading to a reduction in SOC content.

The HR was generally responsive to all the parameters related to soil microbial processes but in low sensitivity (Figure 7). The HR is widely affected by multiple abiotic and biotic conditions, such as substrate concentration, soil moisture, and soil temperature (Gomez‐Casanovas et al., 2012; Zhang et al., 2013). In the CLM‐Microbe model, HR is explicitly represented as soil microbial respiration under the influences of environmental factors (i.e., soil moisture, soil temperature, and oxygen concentration). Meanwhile, fungal and bacterial respiration is not only related to C gain through the decomposition of DOM, SOM, and litter and microbial DOM uptake, but also to soil microbial lysis (Figure 1). Therefore, HR is directly determined by the microbial activities and substrate availability, and indirectly affected by a wide range of environmental factors and parameters. For example, parameters defining C transfer from litter and SOM pools to fungal and bacterial biomass pools and DOM, from DOM pool to fungal and bacterial biomass and SOM pools, and from fungal and bacterial biomass pools to DOM and SOM pools are closely related to soil microbial C gain and loss.

In general, the CLM‐Microbe model simulated comparable SOC and HR in grasslands, tundra, temperate broadleaf forests, tropical/subtropical forests, and wetlands, but much higher SOC and HR in shrub, boreal forests, and temperate coniferous forests compared with CLM4.5 (Table 6). This may be due to two reasons. First, to produce reasonable vegetation status, we adjusted the parameters related to plant photosynthesis (e.g., flnr) and C allocation (e.g., froot_leaf) in the CLM‐Microbe model to guarantee reasonable vegetation productivity indicated by MODIS data set. However, when running the CLM4.5, we used the default parameters for each plant functional type. Therefore, the difference in vegetation condition may induce the discrepancy in simulated SOC and HR. For example, we documented a SOC pool of 2,568 g C m⁻² and HR flux of 277 g C m⁻² yr⁻¹, but both SOC and HR were zeros in desert site (DST‐CH) because of vegetation mortality. Litterfall from vegetation serves as the C source for SOM formation (Thornton & Rosenbloom, 2005), the difference in vegetation condition may be one of the important reasons for the difference in simulated SOC and HR between CLM4.5 and the CLM‐Microbe model. Second, to produce comparable SOC in the CLM‐Microbe model with HSWD data set, we adjusted active soil depth for decomposition (decomp_depth_efolding) to reach the goal. While we used the default value of decomp_depth_efolding (0.5) for the CLM4.5 simulation in all sites. decomp_depth_efolding determines the vertical distribution of SOC in the soil profile, changes in decomp_depth_efolding may account for the differences in simulated SOC, and possibly HR, between CLM4.5 and the CLM‐Microbe model (Oleson et al., 2013).

Limitations and Improvements

The CLM‐Microbe model is capable of simulating FBC and BBC dynamics across biomes; a few improvements are identified as our future research needs. First, the dormant portion of fungi and bacteria were not explicitly considered in the CLM‐Microbe model. Dormant soil microbes can become viable in short time periods due to their capability of rapidly sensing limiting resources change (Garcia‐Pichel & Pringault, 2001). Therefore, dormant soil microbes are able to survive environmental stresses and serve as “seed banks” for the soil microbial community (Lennon & Jones, 2011). Representing the dormant portion of soil microbes would enhance the model capability in simulating microbial resilience to stressful conditions (Wang et al., 2015). Second, fresh C input‐induced priming effects is an important pathway affecting microbial activity, which needs to be considered in future studies. In addition to environmental factors such as temperature, soil moisture, and oxygen concentration, the addition of organic or mineral substances available for soil microorganisms may stimulate microbial activities, causing priming effects (Blagodatskaya & Kuzyakov, 2008). However, we did not test the model performance in simulating the priming impact within current model structure. Given the different physiology of bacteria and fungi, it would be worthwhile to robustly test the model behavior in simulating the priming effect, and further improve the model as needed.

Third, the soil food web regulates fungal and bacterial biomass dynamics, thus the inclusion of soil trophic interactions would help better understand the effects of soil food web on soil microbial biomass dynamics. Soil microbial growth is strongly shaped by predation, Buckeridge et al. (2013), for example, observed the seasonal variation in soil microbial community structure, and the decline in FBC from winter to late winter, and then again in spring, which was closely associated with the abundance of protozoa. Therefore, FBC and BBC may not only be controlled by abiotic factors such as soil temperature, soil moisture, oxygen concentration, and C availability, but the seasonal variation in their predator communities (Schadt et al., 2003). Fourth, the FBC and BBC data compiled were measured using a wide range of methods, while different methods may introduce a variety of biases (He et al., 2020). For example, direct microscopy was widely used in early stage of soil microbial studies, but the approach has inevitably included dead biomass of soil microbes, especially for fungi, into the estimated biomass (Buckeridge et al., 2013). Amino sugars such as glucosamine and muramic acid are used to estimate FBC and BBC, respectively; however, this method measures both living and dead microbial biomass due to the high stability of amino sugars in soil (Glaser et al., 2004). Therefore, comparing FBC and BBC data measured by multiple methods will suffer from uncertainties in data quality due to various biases introduced by the different methods. Although data‐model integration has been proposed for more than four decades, the intimate collaboration between experimentalists and modelers is still needed for model development. A standardized microbial data system that contains primary microbial variables with consistent measurement approach or after conversion is critical to reduce the bias associated with distinct methods (Xu et al., 2020). Last but not least, the observed data for bacterial and fungal biomass C commonly vary by more than five orders of magnitude (Guo et al., 2020; He et al., 2020; Sinsabaugh et al., 2016; Xu et al., 2013, 2017), while ecosystem‐level variables commonly vary by less than three orders of magnitude. This large discrepancy makes the validation approach applied to ecosystem‐level C pools and fluxes less robust in microbial models, as shown in this study of model validation (Section 3.1). The CLM‐Microbe is able to reasonably capture the seasonality of key microbial variables but less robust in simulating the magnitude of microbial variables. We call for community‐level efforts to develop a new model validation approach that is more applicable to microbial models.

Conclusions

This study reported the model parameterization, validation, uncertainty analysis, and sensitivity analysis of the CLM‐Microbe model in simulating fungal and bacterial biomass at the site level. The CLM‐Microbe model could simulate the seasonal variation of FBC and BBC, but the model tended to underestimate the magnitude of the observed biomass for most biomes. Sensitivity analysis showed that the turnover rates of funga and bacteria are the most important parameters regulating FBC and BBC, respectively. Meanwhile, C flow from SOM to soil microbes during decomposition and the C:N ratio of fungal and bacterial biomass are also important for FBC and BBC dynamics. We observed an enhancement of soil microbial C gain from SOM on SOC, but the enhancement is weak in tundra and boreal forests. The simulated HR was responsive to all parameters related to soil microbial processes across biomes but exhibited low sensitivity.

The CLM‐Microbe model represents the first attempt to simulate the soil microbial effects on C cycle by differentiating fungi and bacteria and their physiology in assimilating C in soils. Along with the emerging microbial macroecology (Xu et al., 2020), the improvements in modeling microbial mechanisms will likely bring more robust abilities to ESMs to better simulate and project the climate system. The explicit representation of soil microbial processes into the CLM‐Microbe model will improve our mechanistic understanding of ecosystem‐level C cycling and improve predictability of microbial community structure at regional to global scales, thereby reducing uncertainties in global C projection.

Acknowledgments

The authors are grateful to Dr. Yimei Huang from Northwest A&F University, and Drs. Qian Zhao and Weijun Shen from South China Botanical Garden, Chinese Academy of Sciences for providing the exact sampling date of fungal and bacterial biomass. We thank Dr. Jörg Rinklebe from University of Wuppertal for providing the time‐series data of fungal and bacterial biomass data. Liyuan He thanks Dr. Fenghui Yuan for the discussion about sensitivity analysis. This study is partially supported by San Diego State University and the CSU Program for Education & Research in Biotechnology. Partial support for this work was provided by an Early Career Award through the U.S. Department of Energy (DOE) Biological and Environmental Research Program ORNL is managed by UT‐Battelle, LLC, under contract DE‐AC05‐00OR22725 with the U.S. DOE. This manuscript has been authored by UT‐Battelle, LLC, under contract no. DE‐AC05‐00OR22725 with the US Department of Energy (DOE). The US government retains and the publisher, by accepting the article for publication, acknowledges that the US government retains a nonexclusive, paid‐up, irrevocable, worldwide license to publish or reproduce the published form of this manuscript, or allow others to do so, for US government purposes. DOE will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan (http://energy.gov/downloads/doe-publicaccess-plan).

Data Availability Statement

The data used for model parameterization and validation are obtained in published literature that have been clearly cited in the Table 1 of the manuscript.