Mechanistic Modeling of Microtopographic Impacts on CO2 and CH4 Fluxes in an Alaskan Tundra Ecosystem Using the CLM‐Microbe Model

Spatial heterogeneities in soil hydrology have been confirmed as a key control on CO2 and CH4 fluxes in the Arctic tundra ecosystem. In this study, we applied a mechanistic ecosystem model, CLM‐Microbe, to examine the microtopographic impacts on CO2 and CH4 fluxes across seven landscape types in Utqiaġvik, Alaska: trough, low‐centered polygon (LCP) center, LCP transition, LCP rim, high‐centered polygon (HCP) center, HCP transition, and HCP rim. We first validated the CLM‐Microbe model against static‐chamber measured CO2 and CH4 fluxes in 2013 for three landscape types: trough, LCP center, and LCP rim. Model application showed that low‐elevation and thus wetter landscape types (i.e., trough, transitions, and LCP center) had larger CH4 emissions rates with greater seasonal variations than high‐elevation and drier landscape types (rims and HCP center). Sensitivity analysis indicated that substrate availability for methanogenesis (acetate, CO2 + H2) is the most important factor determining CH4 emission, and vegetation physiological properties largely affect the net ecosystem carbon exchange and ecosystem respiration in Arctic tundra ecosystems. Modeled CH4 emissions for different microtopographic features were upscaled to the eddy covariance (EC) domain with an area‐weighted approach before validation against EC‐measured CH4 fluxes. The model underestimated the EC‐measured CH4 flux by 20% and 25% at daily and hourly time steps, suggesting the importance of the time step in reporting CH4 flux. The strong microtopographic impacts on CO2 and CH4 fluxes call for a model‐data integration framework for better understanding and predicting carbon flux in the highly heterogeneous Arctic landscape.

Previous studies have found that Arctic tundra ecosystems have shifted from net CO 2 sinks to sources (Belshe et al., 2013;Oechel et al., 1993); whether they behave as sinks or sources of atmospheric CH 4 largely depends on local microtopography (Jørgensen et al., 2014;Nauta et al., 2015;Oh et al., 2016;Tan et al., 2015). Minor changes in surface elevation might shift Arctic soils from sinks to sources of atmospheric CH 4 (Olivas et al., 2010;Zona et al., 2011). CH 4 is produced in the poorly drained low-elevation ground with anoxic conditions, whereas atmospheric CH 4 can be oxidized in well-drained high-elevation ground (Atchley et al., 2016;Grant, Mekonnen, Riley, Arora, & Torn, 2017a;Lipson et al., 2012;von Fischer et al., 2010;Zona et al., 2011). Microtopography also affects the CO 2 flux by altering the soil water content and O 2 concentration (Olivas et al., 2010). In addition, soil hydrological conditions affect vegetation growth and substrate availability, further influencing ecosystem C input and microbial community structure and altering the transport and production of CH 4 , root respiration, and microbial respiration (Davidson et al., 2016;von Fischer et al., 2010;Wagner et al., 2017). Thus, accurate quantification of the strength of C sinks or sources requires explicit consideration of the microtopographic effects on C cycling in Arctic tundra ecosystems (Ebrahimi & Or, 2017).
To more accurately capture the fine-scale variations in CH 4 and CO 2 fluxes in Arctic tundra, microtopographic effects need to be considered by ecosystem models as microbial functions such as fermentation, C mineralization, methanogenesis, and methanotrophy dramatically differ between wet and dry polygons (Tas et al., 2018). A number of mechanistic CH 4 models have incorporated the mechanisms of CH 4 production, consumption, and transport pathways , such as the ecosys model (Grant, Mekonnen, Riley, Arora, & Torn, 2017a), CLM-Microbe (Xu et al., 2015), CLM4Me (Riley et al., 2011), LPJ-WHyMe (Wania et al., 2010), and NEST-DNDC (Zhang et al., 2012). However, few CH 4 models explicitly simulate microtopography and are capable of investigating the microtopographic impacts on CO 2 and CH 4 fluxes in Arctic tundra (Grant, Mekonnen, Riley, Arora, & Torn, 2017a;Kaiser et al., 2017). For example, the ecosys model indicates that microtopography determines CO 2 and CH 4 emissions by regulating soil water content, active layer depth, and O 2 availability (Grant, Mekonnen, Riley, Arora, & Torn, 2017a). The CLM-Microbe model simulates fine-scale thermal and hydrological dynamics and microbial mechanisms for CH 4 production and oxidation (Xu et al., 2015), which allows investigation of Arctic CO 2 and CH 4 fluxes across multiple scales from a microbial perspective.
In this study, we used the CLM-Microbe model to simulate the microtopographic effects on CO 2 and CH 4 flux. We aimed to address three questions: (1) How do different microtopographic types affect CO 2 and CH 4 fluxes in Arctic tundra ecosystems? (2) Which processes are more important in controlling CO 2 and CH 4 fluxes among the microtopographic types? (3) How do annual estimates of CO 2 and CH 4 fluxes differ under the microtopographic impacts in the Arctic? The field observational data from the U.S. Department of Energy's Office of Science Next Generation Ecosystem Experiments (NGEE)-Arctic project were integrated with the CLM-Microbe model to understand microtopographic impacts on land surface CO 2 and CH 4 fluxes in an Arctic tundra landscape.
HCP rim (12.2%), and HCP transition (6.2%). Soil organic matter density and plant cover used in model simulations for each landscape type are shown in Figure 1.

Data Availability Statement
The CH 4 and CO 2 fluxes were monitored using the static chamber approach on several dates during June-September 2012 and 2013 (Torn, 2016). Transparent and opaque surface chambers were placed within the study area in the trough, LCP center, and LCP rim. The CO 2 fluxes from the transparent chambers were considered to be the net ecosystem exchange of CO 2 (NEE), and those from the opaque chambers were considered the ecosystem respiration (ER). NEE (i.e., ER − GPP) is negative when CO 2 uptake via photosynthesis is greater than CO 2 release from ER. The concentrations of soil dissolved organic carbon (DOC), CH 4 , and CO 2 were measured for the trough, LCP center, and LCP rim in 2013-2014 (Herndon et al., 2015a;Herndon et al., 2015). An eddy covariance (EC) tower was installed in the center of the study area to measure CO 2 and CH 4 fluxes in 2012; those data are available from the NGEE Arctic project website (Raz-Yaseef et al., 2013). It is noted that this data set has been updated to 2012-2016 although we used the previous data set for 2012-2013. Daily and hourly fluxes of CH 4 and CO 2 were calculated based on the half-hourly EC data. Detailed information about the measurement protocols is posted in the NGEE Arctic archives (http://ngee-arctic.ornl.gov/).

Journal of Advances in Modeling Earth Systems
In a previous study, the CH 4 module in the CLM-Microbe model was validated for simulating the dynamics of CO 2 and CH 4 emissions from incubation experiments on Arctic soils with invariant soil temperature and soil water content (Xu et al., 2015). In this study, we focused on the fully incorporated CLM-Microbe model, created separate model runs for each landscape type, and modified the soil hydrological setup according to landscapes' unique soil conditions. Three soil hydrological parameters, largely affected by microtopography, were selected: soil water content (h2osoi_vol), surface runoff (qflx_surf), and the inundated fraction (finundated). Because the low-elevation polygonal features, such as trough, LCP center, LCP transition, and HCP transition are poorly drained and oversaturated in summer, the parameters for soil water content were changed to a maximum of 1.0, and those for the inundated fraction and surface runoff were changed to 0.99 and 0, respectively. In contrast, the high-elevation features, such as LCP rim, HCP rim, and HCP center, are welldrained in summer. For these features, the parameters for soil water content were set at a maximum of 0.3, which is the baseline allowing the gas transport in soil profiles due to model structures; and those for surface runoff and inundated fraction were set in the default processes with dynamic changes.
The model driving data included meteorological, plant, and soil data. The meteorological data included shortwave and longwave radiation, air temperature, relative humidity, wind speed, and precipitation from 1 January 1991 to 31 December 2014, derived by Xu and Yuan (2016) from the Utqiaġvik, AK, station of NOAA/Earth System Laboratory, Global Monitoring Division (http://www.esrl. noaa.gov/gmd/obop/brw/ ). The data set is gap-filled and at a half-hour time step. The observed plant and soil data for each landscape type-including the composition of plant functional types, plant cover, and soil organic carbon densitywere supplied by the NGEE Arctic project.

Model Implementation
To identify the role of microbial functions in CH 4 dynamics, we set up model simulations using the default CLM4.5 and the CLM-Microbe models separately for each landscape type. The model implementation was carried out in three stages. First, the accelerated model spin-up was set up for 2,000 years to allow the system to accumulate C. Then a final spin-up for 50 years allowed the modeled system to reach a relatively steady state. After the final spin-up, the transient model simulation was set up to cover the period of 1850-2014.
For the default CLM4.5 model simulations, the parameters were set to be the default values for each landscape type. For the CLM-Microbe model simulations, the model parameterization was initialized with the default parameters in Xu et al. (2015); it was performed within their ranges to determine the optimal values of parameters in the microbial module for simulating the observational CO 2 and CH 4 fluxes for each landscape type. For the trough, LCP center, and LCP rim, the observed CO 2 and CH 4 fluxes in 2012 were used for model parameterization, and the fluxes in 2013 were used for model validation. Based on the current knowledge of mechanisms of CO 2 and CH 4 cycling, we primarily focused on the parameters for decomposition and substrate availability for methanogenesis (e.g., k_bacteria, k_fungi, KAce, and AceProdACmax) for the CH 4 cycling. For CO 2 cycling, we focused on the parameters for plant growth respiration (grperc), maintenance respiration (br_mr), and C allocation within biogeochemical cascade (e.g., fatm_f). According to the values of parameters reported in the previous studies, the parameters were calibrated empirically in the model parameterization (Table 1). Because of the lack of the observational data, most of the parameters for LCP transition, HCP center, HCP rim, and HCP

Journal of Advances in Modeling Earth Systems
transition, were set as the default values; and some parameters were modified according to the setup for the trough, LCP center, and LCP rim ( Table 1). The transient simulations of 1850-2014 produced output at both daily and hourly time steps. Simple linear regression was conducted to evaluate the modeled CO 2 and CH 4 fluxes compared with measured fluxes. The error statistics were used to distinguish the difference between the modeled and measured fluxes on the platform of R Studio platform (version 1.1.456), such as the coefficient of determination (R 2 ).

Uncertainty Analysis
The uncertainties of the CO 2 and CH 4 fluxes for each landscape type were quantified using the Markov Chain Monte Carlo method based on Bayesian statistics (Gilks et al., 1998) and were determined by a large ensemble of model simulations with different parameter settings. In this study, a total of 100 model simulations with different settings of 15 key parameters were set up for each landscape type, separately. These 15 key parameters determine the decomposition of organic carbon, methanogenesis, microbial growth, and plant photosynthesis and respiration and therefore control the CO 2 and CH 4 fluxes ( Table 2). They varied within a range of 30% of their optimal values (Xu et al., 2015). Model simulations were conducted from the transient simulation to cover 1850-2014 at a daily time step.

Area-Weighted Upscaling
The modeled CO 2 and CH 4 fluxes in 2013 for each landscape type were further used for upscaling to the EC tower domain based on an area-weighted average approach. Because of the limitation of the landscape classification data, the EC domain was confined to an area of 100 m × 100 m. The area-weighted average approach includes information for landscape heterogeneity in the upscaling process. The upscaled flux was calculated by the following equation: where F is the upscaled plot flux for the EC domain, f i is the plot-level CH 4 or CO 2 fluxes for a given landscape type for a given time period, area i is the fraction of each major landscape type within the EC domain (Davidson et al., 2016). In addition, the average fluxes were calculated based on the CH 4 and CO 2 fluxes from the seven landscape types for comparison with the upscaled fluxes.

Sensitivity Analysis
To identify the most important process and the most sensitive parameters for CH 4 and CO 2 dynamics in Arctic tundra, a global sensitivity analysis was conducted for each microtopographic type. It focused on the 15 parameters related to plant and microbial processes that were used in the uncertainty analysis (Table 2). For each parameter, we set up model simulations with +20% and − 20% changes and investigated the responses of the modeled CH 4 and CO 2 fluxes in 2013. 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 flowing equation (Xu et al., 2015): Table 2 Key Parameters for Uncertainty Analysis and Sensitivity Analysis.

KAce
Half-saturation coefficient of available carbon mineralization ACminQ10 Temperature sensitivity of available carbon mineralization AceProdACmax Maximum rate of acetate production from available carbon H2ProdAcemax Maximum rate of H 2 production from available acetate KH2ProdAce Half-saturation coefficient of conversion of H 2 and CO 2 to acetate KCO2ProdAce Assuming it is half of that for H 2 based on stoichiometry theory KCO2ProdCH4 Half coefficient of CO 2 for methane production from H 2 GrowRAceMethanogens Growth rate of acetoclastic methanogens YAceMethanogens Growth

Journal of Advances in Modeling Earth Systems
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, and P a and P n are the altered and nominal parameters, respectively. S is negative if the direction of model response opposes the direction of parameter change (Xu et al., 2015).

Model-Simulated CH 4 and CO 2 Fluxes
The CLM-Microbe model was more accurate than the default CLM4.5 in simulating the CH 4 and CO 2 fluxes during the summer of 2013 ( Figure 2). The dynamics of CH 4 was captured well by the CLM-Microbe model for the trough, LCP center, and LCP rim, but those sites were simulated as a small CH 4 sink in summer by the default CLM4.5 (Figures 2a-2c and Table 3). NEE and ER were used to represent the dynamics of CO 2 flux. In the summer months, NEE and ER had the similar patterns for the trough and LCP center but showed significant differences for LCP rim in the default CLM4.5 and CLM-Microbe models (Figures 2d-2i). For the trough and LCP center, the CLM-Microbe model simulated the NEE well, whereas the default model underestimated it by 47-50%; however, both of them overestimated the ER by 28-47% ( Figure 3 and Table 3). The variations of NEE and ER were not captured by either the default or the CLM-Microbe models for LCP rim, but the CLM-microbe model performed slightly better than the default model ( Figure 3 and Table 3).

Journal of Advances in Modeling Earth Systems
For the simulations of the CLM-Microbe model, larger CH 4 fluxes with greater variations were observed for the trough than that for the LCP center and LCP rim (Figures 2a-2c). Additionally, modeled CH 4 fluxes were more consistent with the observational fluxes for the trough (R 2 = 0.7381, p = 0.0001) than for the LCP center (R 2 = 0.2820, p = 0.0507) and LCP rim (R 2 = 0.3432, p = 0.0277; Table 3). For the dynamics of CO 2 fluxes, similar patterns of NEE and ER were simulated for the trough and LCP center (Figures 2d and 2e). Modeled NEE was highly consistent with the observed data for the trough (R 2 = 0.9569, p = 0.0007) and LCP center (R 2 = 0.9194, p = 0.0025); whereas ER was overestimated by 47.3% for the trough (R 2 = 0.8316, p < 0.0001) and 39.2% for the LCP center (R 2 = 0.9188, p < 0.0001; Table 3). Compared with the trough and LCP center, a slightly higher NEE (lower CO 2 uptake) was modeled for the LCP rim and it was underestimated by 78.4% (R 2 = 0.4080, p = 0.1757; Table 3). However, ER for the LCP rim was overestimated by 64.5% (R 2 = 0.88, p < 0.0001; Table 3).

Variability and Seasonality of CH 4 and CO 2 Fluxes Across the Arctic Landscape Types
Modeled CH 4 and CO 2 fluxes exhibited large variabilities among all seven landscape types. In the summer months, high CH 4 emissions were generally simulated, associated with low NEE (i.e., high CO 2 uptake) and high Note. LCP = low-centered polygon. *The significant level < 0.01. ER for all the landscape types ( Figure 4). Low-elevation landscape types, such as trough, LCP transition, and HCP transition, had higher CH 4 emissions with greater variation compared with high-elevation types, such as LCP rim, HCP rim, and HCP center (Figure 4a). However, low-elevation types, including trough, LCP transition, and LCP center showed lower NEE (i.e., higher CO 2 uptake) than high-elevation types, including LCP rim and HCP rim (Figure 4b). ER was roughly higher in the HCP center and lower in the trough and LCP center in the summer months ( Figure 4c).
Larger seasonal variations of CH 4 dynamics were modeled in low-elevation landscape types (e.g., trough and transitions) than high-elevation types (e.g., rim; Figure 4a). In the early spring, most of the landscape types showed a burst release of CH 4 flux, corresponding to the early Spring thaw. During the growing seasons, the lower ground of the trough and LCP transition had similar seasonality of CH 4 and CO 2 fluxes, with the highest CH 4 emission and highest CO 2 uptake ( Figure 4). The higher ground of rims tended to have smaller variations of CH 4 and NEE fluxes (Figure 4). A sudden rise was simulated in NEE for the trough, LCP transition, LCP center and HCP center, during the late growing seasons; a similar rise simultaneously was simulated in ER ( Figure 4).

Annual Estimates of CH 4 and CO 2 Fluxes Across the Landscape Types
Annual CH 4 fluxes were estimated for all landscape types, resulting in a range of 0.66 to 3.97 g Cm −2 ( Table 4). The HCP transition, as the largest CH 4 source, has released 5.06 times more annual CH 4 fluxes more than the smallest CH 4 sources (i.e., LCP rim). Low-elevation types, including trough and transition, contributed a larger proportion of CH 4 emissions than high-elevation types in Arctic tundra ecosystems. Based on the areal fractions, the HCPs and LCPs were estimated to have comparable annual CH 4 fluxes of 2.12 and 2.19 g C m −2 year −1 , respectively, both of which were smaller than the annual CH 4 fluxes of 3.63 g C m −2 year −1 from the trough ( Table 4).
The landscape types were found to be net sources of CO 2 , except for the trough, in which 21.08 g C·m −2 ·year −1 of CO 2 was sequestered (Table 4). Large variations existed in the annual NEE, which ranged from −21.08 to 248.22 g C·m −2 ·year −1 ( Table 4). The HCP transition was the largest CO 2 source to the atmosphere among the landscape types. All the landscape types had comparable ERs-ranging from 582.72 to 776.31 g C·m −2 ·year −1 -except the HCP center, which exhibited the greatest ER of 927.98 g C·m −2 ·year −1 . The trough, LCP center, and LCP rim with smaller NEE estimates were also estimated to have smallest ER budget.

Upscaling CH 4 and CO 2 Fluxes to the EC Domain
Modeled CH 4 and CO 2 fluxes were upscaled to the EC domain based on the area fraction of each landscape type. High consistency was shown between the upscaled fluxes and EC measurements (Figures 5 and 6). The regression analysis showed that more accurate simulations of CH 4 and NEE fluxes for the EC domain were obtained at daily time steps than at hourly time steps (Figure 6). CH 4 fluxes were underestimated at daily (R 2 = 0.7931, p < 0.0001) and hourly (R 2 = 0.6135, p < 0.0001) time steps by 20.08% and 24.95%, respectively. NEE was underestimated at the daily time steps by 21.44% (R 2 = 0.2843, p < 0.0001), but was overestimated at the hourly time steps by 30.01% (R 2 = 0.3464, p < 0.0001; Table 5).
Annual CH 4 and CO 2 fluxes were estimated for the EC domain using the arithmetic average and areaweighted average approaches. When the proportions of different landscape types were not considered, the annual CH 4 flux was underestimated at 2.39 g C·m −2 ·year −1 , compared with an annual estimate of Note. LCP = low-centered polygon; HCP = high-centered polygon.

Journal of Advances in Modeling Earth Systems
2.67 g C·m −2 ·year −1 based on the areal fractions. However, NEE and ER were overestimated at 104.13 g C·m −2 ·year −1 and 716.81 g C·m −2 ·year −1 , respectively, compared with the annual estimates of 73.83 g C·m −2 ·year −1 and 692.86 g C·m −2 ·year −1 for NEE and ER, respectively, that considered the heterogeneity of landscapes (Table 4).

Sensitivity Analysis
CH 4 and CO 2 fluxes were sensitive to a portion of the key parameters that are related to available carbon mineralization, CH 4 production, growth of methanogens, decomposition, photosynthesis, plant growth respiration, and maintenance respiration. Specifically, the CH 4 flux was strongly sensitive to the parameters of AceProdAcemax and ACMinQ10, followed by YAceMethanogens, GrowRAceMethanogens, KAce, and k_dom for all landscape types (Figure 7a), which suggested that acetate production and available C mineralization were the key controls on CH 4 dynamics in Arctic tundra ecosystems. Growth of methanogens also regulated CH 4 flux by influencing CH 4 production. Changes in the decomposition rate of the dissolved organic matter (DOM) had a positive influence on CH 4 flux for all the landscapes except the LCP rim ( Figure 7a). In the high-elevation features LCP rim, HCP rim, and HCP center, the CH 4 flux was sensitive to the fraction of leaf nitrogen in the Rubisco enzyme functioning in photosynthesis (flnr). In the HCP rim, CH 4 dynamics also responded to changes in the decomposition rate of fungi biomass (k_fungi) and plant growth respiration (grcep; Figure 7a). Autotrophic respiration (e.g., plant growth respiration and maintenance respiration) and the flnr were also the key controlling factors for CH 4 flux in low-elevation features but exhibited opposite effects (Figure 7a).

Journal of Advances in Modeling Earth Systems
The most important processes of CO 2 dynamics are related to photosynthesis and respiration, which control CO 2 uptake and release across the Arctic landscape types. The flnr was identified as the primary factor for NEE and ER (Figure 7b and 7c). An increase in the flnr lead to a rise in NEE in the trough, LCP rim, and HCP rim but a NEE reduction in the LCP center, LCP transition, HCP transition, and HCP center. There was a significant decrease in ER for all landscapes with increased flnr (Figure 7b and 7c). For high-elevation features, including LCP rim, HCP rim, and HCP center, NEE dynamics showed negative responses to ACMinQ10 and positive responses to AceProdACmax, suggesting that acetate production is also important for CO 2 uptake from the atmosphere (Figure 7b). In the trough, NEE was sensitive to many other parameters related to acetate production, decomposition, and respiration ( Figure 7b). Beside the flnr, ER dynamics was sensible to maintenance respiration (br_mr) in both high-elevation and low-elevation landscape types ( Figure 7c). Additionally, changes in the decomposition rates of bacteria and fungi biomass for the LCP center could also result in changes in ER (Figure 7c).

Microtopographic Impacts on CH 4 and CO 2 Fluxes
Microtopography determines CH 4 and CO 2 dynamics in Arctic polygonal tundra by affecting the hydrological processes and thereby the soil water content, active layer depth, vegetation, and microbial functional groups in ecosystem C exchange (Davidson et al., 2016;Grant, Mekonnen, Riley, Arora, & Torn, 2017a;Lipson et al., 2012;Throckmorton et al., 2015;. Soil water content was greater in the low-elevation ground of trough, LCP transition, LCP center, and HCP transition than in the high-elevation ground of the LCP rim, HCP rim, and HCP center; this discrepancy largely explained the variability in CH 4 fluxes among the heterogeneous landscape types in the Arctic (Grant, Mekonnen, Riley, Wainwright, et al., 2017b;Lu & Zhuang, 2012). In summer, larger CH 4 emissions were observed and modeled in the trough than in higher-elevation rims and centers because of the trough's higher soil water content and oversaturated soils, which is consistent with previous studies (Grant, Mekonnen, Riley, Arora, & Torn, 2017a;Schrier-Uijl et al., 2010). Saturated and oversaturated soils create an anoxic condition facilitating methanogenesis to produce CH 4 (von Fischer et al., 2010;Von Fischer & Hedin, 2007). Higher CH 4 emissions were also modeled for other low-elevation types, including LCP center, LCP transition, and HCP transition, a finding that supported the promotion effects of high soil water content on CH 4 production.
CH 4 flux is strictly produced by methanogens at very low O 2 concentration in soils, mainly converted from acetate and CO 2 + H 2 (Nazaries et al., 2013). The sensitivity analysis suggested that the substrate availability for methanogenesis resulting from acetate production and DOM decomposition was the key constraint for CH 4 dynamics in Arctic polygonal landscapes (Xu et al., 2015). In summer months, higher CO 2 uptake (i.e., lower NEE) and stronger photosynthesis (i.e., increased flnr) were modeled associated with the higher CH 4 emissions in the low-elevation ground. The results implied that a positive correlation existed between the CO 2 uptake and CH 4 emission. Stronger CO 2 uptake refers to a higher amount of plant biomass, which facilitates the emission of large CH 4 fluxes emitting to the atmosphere via plant-mediated transport (von Fischer et al., 2010). Additionally, high CO 2 uptake can provide abundant C input into soils as litter for microbial decomposition, which in turn produces a high amount of available C for methanogenesis. However, the field study reported that no relationship was observed between instantaneous GPP (i.e., CO 2 uptake) and CH 4 fluxes (Davidson et al., 2016), suggesting that the processes from decomposition of organic matter to substrates for methanogenesis are very complicated and need to be considered cautiously in CH 4 models.
Microtopographic effects on CO 2 dynamics are caused primarily by changes in soil hydrological conditions and hence O 2 diffusion (Davidson et al., 2016;Olivas et al., 2010). Low-elevation landscape types, especially the trough and LCP center, were shown by modeling to have the largest CO 2 uptake in summer. Additionally, the ER via roots and microbial communities was suppressed by the low dissolved O 2 concentration in the saturated soils (Olivas et al., 2010;Zona et al., 2011). These results suggested that high soil moisture in Arctic tundra promotes the plant growth and suppresses the ER, eventually increasing the strength of the CO 2 sink in low-elevation landscapes. In contrast, lower CO 2 uptake and greater ER in summer were simulated for the rims and HCP center, confirming that the high-elevation ground in the Arctic acts as a smaller CO 2 sink . The ER variations in low-elevation and high-elevation landscape types in the summer months were caused largely by the difference in the availability of soil O 2 for microbial respiration. The strength of the CO 2 sink in Arctic tundra can be biased by the effects of microtopography on soil water and O 2 conditions. However, most landscape models have not incorporated microtopographic effects in simulating CO 2 fluxes. Not considering those effects might cause large biases; therefore, better simulation of microtopographic impacts is critical for model applications to C cycling in the Arctic.

NEE and CH 4 Flux at Daily and Hourly Time Steps
Biological processes occur instantaneously, on a time scale inconsistent with field measurements normally undertaken at hourly or daily scales. Ecosystem functions are more apparent at the hourly, daily, monthly, and annual scales, at which the CLM-Microbe performs. The model performance was more consistent with observed CH 4 and CO 2 fluxes for the EC domain on the daily than on the hourly scale, indicating that the model did not perform well in capturing some pulse fluxes on an hourly scale. Since the "CH 4 pulse" in the spring season has been widely recognized as an important component for ecosystem models in recent decades (Lu & Zhuang, 2012;Song et al., 2012;Tokida et al., 2007), an improvement of the CLM-Microbe model for better simulation of these outbreak events is needed.
In general, upscaled CH 4 and CO 2 fluxes based on modeled plot-level fluxes were able to capture most variations of measured EC fluxes at both daily and hourly time steps. We conclude that the CLM-Microbe model can be used to estimate CO 2 and CH 4 fluxes at landscape scale if fluxes are scaled by different landscape types (Schrier-Uijl et al., 2010). Moreover, the dynamics of CH 4 and CO 2 fluxes was modeled more accurately at daily than at hourly time steps. This is probably because the key factors or processes controlling CH 4 and CO 2 dynamics are slightly different across the temporal scales; but they are well defined with stable priorities in the model according to the extant knowledge, usually from observations at long time scales. Another reason for the underestimation of CO 2 flux might be the unexplained CO 2 uptake during the nongrowing season (i.e., October) in the Alaska tundra ecosystem (Figures 5b and 5d). Until confirmed mechanisms are found for the underestimation, it has no clear implications for the model performance.

Model Implications
This study has three implications for model development and scientific understanding of the C dynamic in the Arctic. First, the CLM-Microbe model performed well in capturing the variabilities in CH 4 and CO 2 fluxes among primary polygonal landscapes in Arctic tundra, which emphasizes the importance of spatial heterogeneity in simulating CO 2 and CH 4 fluxes in ecosystem CH 4 models. The model simulations indicate that the trough and transitions had estimated CH 4 emissions of 3.6-4.0 g C·m −2 ·year −1 annually, and the rims had a smaller annual CH 4 emissions of 0.7-1.1 g C·m −2 ·year −1 . Differences in the annual estimates were likely due to the saturated and anoxic conditions in low-elevation ground that promote anaerobic methanogenesis, leading to a higher CH 4 emission. However, the annual CH 4 fluxes for the seven landscape types may be greatly underestimated because of the low estimates for the cold season in model simulations.
Many studies have reported that >50% of annual CH 4 emissions occur during the cold season in Alaskan tundra (September to May; Kittler et al., 2017;Zona et al., 2016). Moreover, our estimate of annual CH 4 fluxes for the entire study area was smaller compared with similar studies in Arctic tundra (Reeburgh et al., 1998;Wille et al., 2008). This discrepancy might be explained by the lower organic matter density or less plant cover in our study area. For example, the less abundant plant cover reduces the plant-mediated transport of CH 4 and therefore lowers CH 4 emissions (Bhullar et al., 2013). The variations in our estimates for seven landscape types were potentially biased because they ignored the lateral surface hydrologic and thermal processes. The spatial variability in soil moisture and soil temperature can be overpredicted if the lateral subsurface hydrologic and thermal processes are excluded (Bisht et al., 2018), and the same is true for the spatial variability in CH 4 emissions in Arctic polygonal landscapes. By incorporating these surface processes, the CH 4 models can improve the representation of lateral hydrologic and thermal transport, and thereby improve the accuracy of estimations (Aas et al., 2019).
Second, the potential shifts in Arctic tundra ecosystems as C sinks or sources is valuable information for climate projections. This study showed that the trough is the only net CO 2 sink among all landscape types and plays a key role in ecosystem C storage because of its 35% areal share of the entire study area. CO 2 dynamics in the trough were very sensitive to many processes related to photosynthesis, plant and soil respiration, and C mineralization and distribution. It is possible for trough shifting to being net C sources from net C sinks, even with a tiny change in CO 2 processes under climate changes. Annual estimates indicated that the HCPs were 310% greater CO 2 sources than the LCPs. Additionally, greater ER was estimated in the HCPs than in the LCPs. The HCP center, in particular, had an ER of 928.0 g C·m −2 , the highest among the landscape types. Because LCPs may eventually subside into HCPs, CO 2 emissions from Arctic soils tend to increase rapidly not only because of the effects of climate changes but also because of the changes in landscape patterns.
Finally, this study advocates the mechanistic modeling of C cycling to better estimate CO 2 and CH 4 fluxes across the Arctic tundra ecosystems. It is well known that differences in CH 4 and CO 2 emissions across the Arctic landscapes are directly led by the mechanisms and dynamics of microbial activities in relation to C mineralization, decomposition, respiration, methanogenesis, and methanotrophy. By including these microbial processes, the CLM-Microbe model allows us to understand the mechanisms of Arctic C cycling according to the production and consumption processes of CO 2 and CH 4 . Soil DOC, CH 4 , and CO 2 concentrations were modeled and compared with the measured concentrations for the landscape types and soil depths for a few data points (Table 6). Modeled DOC concentrations were~1.6 times the measurements at the middle layer of soils in the trough on day of year (DOY) 183 and DOY 240 of 2013; these results suggested that the model could be useful for simulating the concentrations of C compounds. In Arctic ecosystems, high CH 4 emissions in saturated soils were modeled with high CH 4 production, which was consistent with the large amount of modeled acetoclastic methanogens. CH 4 oxidation was strengthened by high O 2 availability in the topsoil of the rims and HCP center compared with saturated trough and transitions. Moreover, differences in CH 4 transport via diffusion, ebullition, and plant-mediated transport were modeled with seasonal

Journal of Advances in Modeling Earth Systems
variations. Large CH 4 fluxes emitted from soils were associated with the fast plant growth in summer. Furthermore, soil microbial structure and biomass were simulated to understand the CO 2 and CH 4 dynamics, which suggested the importance of belowground microbial mechanisms in modeling surface CO 2 and CH 4 fluxes.

The Way Forward
The CLM-Microbe model can simulate the belowground microbial processes for surface CO 2 and CH 4 fluxes. Although promising results proved the robustness of the CLM-Microbe model in simulating surface CO 2 and CH 4 fluxes, a number of tasks were identified as follow-up efforts to this study. First, although the upscaling results with an area-weighted approach seems promising, the dominant roles of landscape types weakened the variations in C flux.
Upscaling with a mechanistic model should provide more accurate quantification of the C flux on a regional scale, as well as a finer-resolution C flux at both the spatial and the temporal scales (Watts et al., 2014). Second, belowground C dynamics-for example, DOC, acetate, CO 2 , and CH 4 concentrations-are important variables and precursors for observed surface gas fluxes. We call for a data-model integration approach to better integrate the observational data and better simulate belowground processes and surface flux. Third, hydrological dynamics is the key control for biogeochemical processes, particularly for the changing Arctic. Improving the model's ability to simulate hydrology is an important cornerstone for simulating soil biogeochemistry. Fourth, microbial genomic information is the most accurate information for microbial functions, yet it has not been well utilized for model parameterization. The CLM-Microbe model is capable of simulating the relative abundance of methanogenesis; thus, it is worthwhile to improve the model to better simulate the microbial functional groups responsible for CH 4 production and consumption. Fifth, although C flux data, particularly the CH 4 flux, have been scarce across the Arctic tundra ecosystem, recent projects and technical improvements have allowed year-round measurements. Those data can serve as a good constraint for the CLM-Microbe model at multiple scales; a multiscale "MODEX" (model-observation-experiment) framework to better integrate multiple observational data to quantify gas flux and understand its mechanisms in the Arctic is much needed.

Conclusions
This study reported the application of the CLM-Microbe model to seven microtopographic landscape types in the Arctic tundra near Utqiagvik, AK. The model results were promising and consistent with the observational gas fluxes. The modeled results showed that low-elevation landscape types (e.g. trough, transitions, and LCP center) have higher CH 4 emission with greater seasonal variations than high-elevation landscape types (e.g. rims and HCP center), as a result of the greater soil saturation in the low-elevation landscape types. Model sensitivity analysis determined that the substrate (e.g. acetate, CO 2 + H 2 ) availability for methanogens was the most important factor in controlling CH 4 emissions in Arctic ecosystems, and plant photosynthesis greatly affected the NEE and ER. The model performed more accurately in simulating the daily EC fluxes than hourly fluxes, indicating the importance of the time scale in simulating gas fluxes.
The large spatial heterogeneity in CO 2 and CH 4 fluxes across the Arctic landscape requires explicit consideration and modeling of microtopography, as well as of the mechanisms controlling C biogeochemistry in response to hydrology dynamics. As the climate continues to warm rapidly in the Arctic, large variations at both the spatial and the temporal scales are anticipated across the Arctic landscape. Dramatic changes in land surface CO 2 and CH 4 fluxes might alter the land-atmosphere feedback in the Arctic. An insightful regional-scale investigation of the thermal conditions, hydrology, and biogeochemistry across the pan-Arctic is urgently needed and will benefit the entire scientific community and the public.

Data Availability Statement
The data used in this study have been archived at the NGEE-Arctic data repository, including gas flux data (DOI: 10.5440/1362279), and meteorological data (https://ngee-arctic.ornl.gov/data/).