Parameterizing Perennial Bioenergy Crops in Version 5 of the Community Land Model Based on Site‐Level Observations in the Central Midwestern United States

With projected expansion of biofuel production at a global scale, there is a pressing need to develop adequate representation of bioenergy crops in land surface models to help effectively quantify the biogeophysical and biogeochemical effects of its associated land use changes. This study implements two new perennial bioenergy crops, Miscanthus and switchgrass, into the Community Land Model Version 5 based on site‐level observations from the midwestern United States by modifying parameters associated with photosynthesis, phenology, allocation, decomposition, and carbon cost of nitrogen uptake and integrating concomitantly land management practices. Sensitivity analyses indicate that carbon and energy fluxes of the perennial crops are most sensitive to photosynthesis and phenology parameters. Validation of simulated fluxes against site‐level measurements demonstrates that the model is capable of capturing the overall patterns of energy and carbon fluxes, as well as physiological transitions from leaf emergence to senescence. Compared to annual crops, perennial crops feature longer growing season, greater leaf areas, and higher productivity, leading to increased transpiration, lower annual runoff, and larger carbon uptake. The model simulations suggest that with higher CO2 assimilation rates and lower demands for nutrients and water, high‐yielding perennial crops are promising alternatives of bioenergy feedstocks compared to traditional annual crops not only for mitigating climate change but also for environmental conservation purposes by reducing fertilizer application and therefore alleviating surface‐ and ground‐water contaminations. Although the local‐scale simulations shed light on potential benefits of using perennial grasses as bioenergy feedstocks, quantifying consequences of their plantations at larger scales warrants additional investigation.


Introduction
The expansion of clean and renewable bioenergy production is one suggested pathway and a critical component of future land use scenarios designed to mitigate environmental and global changes (Hickman et al., 2010;Nakada et al., 2014;Robertson et al., 2017;Sims et al., 2006;Tan et al., 2008). By the end of the 21st century, increases in bioenergy crop planting areas range from 3 to 8 million km 2 relative to the 2005 baseline in three Shared Socioeconomic Pathways (SSPs) scenarios (O'Neill et al., 2016), as part of Coupled Model Intercomparison Project Phase 6 (CMIP6). Various studies suggest that substituting fossil fuels with biofuels can potentially lead to both greenhouse gas mitigation and air pollution reduction Calvin et al., 2017;Popp et al., 2014;Qin et al., 2018;Thornton et al., 2017).
While traditional row crops, such as corn and soybean, can be used as biofuel feedstocks, their potential unfavored consequences for food security (Msangi et al., 2007) and environmental sustainability Searchinger et al., 2008) limit their use as biofuel feedstocks. Perennial grasses, such as Panicum virgatum (switchgrass) and Miscanthus x giganteus (Miscanthus), are better alternatives to traditional food crops (Smith & Searchinger, 2012;Stokstad, 2019) because they have high water-and nutrientuse efficiencies. For example, Zhuang et al. (2013) reported that the substitution of Miscanthus for maize on currently available croplands in the United States would save half of the land and one third of the water. Agricultural pollutants in groundwater and surface water as a result of excessive application of nitrogen fertilizer can be significantly reduced by planting perennial bioenergy crops at two watersheds in the Midwest United States (Cibin et al., 2016;VanLoocke et al., 2017). In addition, perennial grasses can be planted in degraded land abandoned from agricultural use, namely marginal land, where most food crops could not survive due to poor soil condition (Aragon et al., 2017;Bandaru et al., 2013). However, much of the technology and consequences are still unproven for achieving large-scale biofuel production (Stokstad, 2019).
Models with comprehensive carbon, water, and energy parameterizations, such as the land model in Earth system models (ESMs), are effective tools to study mitigation effects and implications on water, carbon, and nitrogen dynamics of land use changes at large scales (Surendran Nair et al., 2012). In order to accurately quantify the potential of future land use and management decisions that deployed increasing biofuel plantation to aid the mitigation of global change (Lawrence et al., 2016), it is important that these models have adequate representations of key bioenergy crops. Bioenergy crops have been implemented into land models and ecosystem models in a few studies to investigate the effect of conversion from food crops to bioenergy crops on ecosystem dynamics and associated socio-economic consequences. Jain et al. (2010) integrated a biophysical model, Integrated Science Assessment Model (ISAM), with a county-specific economic analysis of breakeven prices of bioenergy crop production to assess the biophysical and economic potential of biofuel production in the Midwestern United States. Song et al. (2015) integrated dynamic crop growth processes (Song et al., 2013) of bioenergy crops into ISAM to estimate the spatial and temporal variations of biomass yields in the eastern United States. Qin et al. (2015) applied a biogeochemical model to estimate potential greenhouse gas emissions from switchgrass and Miscanthus grown on marginal lands in the United States. They reported that the ecosystem models were still limited by model structure and parameterization deficiencies, which leaded to insufficient predictions of carbon dynamics altered by land use changes. Zhu et al. (2017) parameterized two perennial bioenergy crops into the Version 4.5 of the Community Land Model (CLM) and recognized that neglecting biophysical effects of altered water and energy fluxes in land surface modeling could lead to underestimates in the carbon storage through bioenergy crop conversion by up to 50% at regional scales. Li, Yue, et al. (2018) developed the ORCHIDEE-MICT-BIOENERGY model by adding four new plant functional types into ORCHIDEE to represent four major lignocellulosic bioenergy crops and demonstrated the importance of explicit representation of bioenergy crops in a global vegetation model to reproduce the observation-based biomass yields. Such recent progress in land models together with long-term observations of carbon, water, and energy fluxes at the site levels facilitate better representations of these key bioenergy crops in ESMs to help effectively quantify the biogeophysical and biogeochemical effects of alternative land use change scenarios projected by multisector dynamics models on the Earth system .
The objective of this study is to parameterize two perennial bioenergy crop types, switchgrass, and Miscanthus, into the latest version of CLM (CLM5), and validate the model simulated water, energy, and carbon fluxes in five continuous growing seasons (2009)(2010)(2011)(2012)(2013) at a traditional annual crop rotation site (i.e., maize and soybean rotation) and two additional sites with perennial bioenergy crop plantations (i.e., switchgrass and Miscanthus) at the University of Illinois Energy Farm (Zeri et al., 2011;Zeri et al., 2013). Sensitivity analyses are performed to examine the influence of crucial parameters and fertilizer supply strategy on model performance and in altering biogeochemical and biogeophysical cycles to potentially guide future field characterization efforts to reduce model prediction uncertainty. We hypothesize that compared to conventional annual crops, the physiological, morphological, and phenological characteristics of perennial grasses support higher evapotranspiration (ET) and therefore will result in lower total runoff and total water storage throughout the growing season. However, the higher above-ground biomass production and longer growing season of perennial bioenergy crops, achievable under modest fertilization applications, will contribute to higher carbon-, water-and fertilizer-use efficiencies for Miscanthus and switchgrass.

Study Area and Data
The observation data were collected at University of Illinois Energy Farm (UIEF) located in central Illinois (40.064°N, 88.197°W,~220 m above sea level; Figure 1a). The historical mean annual temperature was 11°C, while the mean annual rainfall was 1,042 mm (Joo et al., 2016(Joo et al., , 2017Zeri et al., 2011Zeri et al., , 2013. The UIEF consists of several experiment plots, 4 ha (200 m × 200 m) each in size ( Figure 1a). In 2008, three ecosystems, a rotational sequence of maize and soybean, a homogeneous Miscanthus, and a homogeneous switchgrass, were planted in UIEF (Figure 1a), each instrumented with an eddy covariance system at the center of the experiment plot to monitor surface energy, water, and carbon fluxes (Zeri et al., 2011). Eddy covariance fluxes at the three plots are available over the period of 2009-2013 at half-hour frequency. The annual total rainfall for the 5 years are 1,292, 1,047, 1,003, 777, and 1,027 mm, respectively. A full site description along with flux calculation and quality control can be found in Zeri et al. (2011).
The maize/soybean rotation plot consists of a cycle of two years of maize (2008-2009 and 2011-2012) followed by one year of soybean (2010 and 2013). The three ecosystems were initially planted in the spring of 2008 with maize/soybean rotation on 6 May 2008, Miscanthus on 26 May 2008, and switchgrass on 28 May 2008. The maize/soybean rotation was planted every year in May over 2009-2013. Except for the replantation of Miscanthus in 2010 due to high mortality after new establishment, there was no annual planting for perennial grasses. Maize received 202, 180, and 168 kg ha −1 nitrogen prior to the growing season in 2009, 2011, and 2012, respectively, while switchgrass was fertilized by 56 kg ha −1 nitrogen every year since 2010. No fertilizer was applied to soybean and Miscanthus. The maize/soybean rotation was harvested between late September and early November, whereas perennial grasses like Miscanthus and switchgrass were harvested in December or next spring following each growing season. Miscanthus and switchgrass have higher biomass yields than the maize/soybean rotation, especially during the last 3 years after establishment. A schematic illustration of the land management practices applied over 2009-2013 for the three  iod (2009-2013) for (1) maize-soybean rotation, (2) switchgrass, and (3) Miscanthus. Note that the size of the arrows in (b) represents for the relative magnitude of each variable. Photographs are not of the actual measurement fields. Photos of the University of Illinois Energy Farm, Switchgrass, and Miscanthus are from https://cabbi.bio/about/cabbi-facilities/. Adapted with permission from Center for Advanced Bioenergy and Bioproducts Innovation. agroecosystems is showed in Figure 1b. A detailed description of characteristics associated with productivities of these three ecosystems, such as leaf emergence, senescence, and harvest dates, is provided in Joo et al. (2016).

Model Description
The latest version of the Community Land Model, CLM5, is capable of simulating simultaneously the surface energy balance as well as hydrological and biogeochemical cycles driven by climate variability and land use change transitions. The model features multiple improvements in soil and plant hydrology, snow density, river modeling, coupled carbon and nitrogen cycling, and crop modeling compared to its previous versions (Lawrence et al., 2018).
CLM5 divides each grid cell into multiple land units to represent spatial land surface heterogeneity, and all crops reside in one land unit. The crop land unit is composed of multiple columns to further capture potential variability in water and energy state variables of different crop functional types (CFTs). Each existing CFT occupies up to two columns (irrigated and rainfed) within the crop land unit and represents a broad category of crops distinguishable from other crops in terms of functional characteristics. Maize and soybean were two CFTs that are already represented in CLM5 ( Figure 2).
An interactive crop module was implemented in CLM fostered by a need to explicitly modeling humanmanaged land use (Lawrence et al., 2016;Levis et al., 2012) aided by increasing availability of land management datasets at a global scale. The module explicitly simulates land management practices, including planting, fertilization, irrigation, and harvesting for each crop using crop-specific parameter values in phenology and allocation. Planting is determined by thresholds of temperature and growing degree days within the range of planting dates, while harvesting is initialized once the crop reaches maturity determined by maximum growing degree days or the days past planting. Fertilization rates are updated based on CFTs and geographic regions, and the application of irrigation dynamically responds to soil moisture conditions simulated in the model (Lawrence et al., 2018;Leng et al., 2013;Ozdogan et al., 2010).
CLM uses a fully prognostic treatment for terrestrial carbon and nitrogen cycling to predict all state variables in vegetation, litter, and soil organic matter (SOM) within each soil column. These prognostic variables are utilized by the biophysical model for coupled carbon, water, and energy processes. The Fixation and Uptake of Nitrogen (FUN) has been implemented into CLM5 to account for the cost of carbon for nitrogen uptake (Fisher et al., 2010;Shi et al., 2016).
The decomposition of SOM, the largest carbon reservoir in the Earth system (Tian et al., 2015), is a complicated process that involved litter and SOM turnover, plant-soil nutrient interaction, and soil respiration (Koven et al., 2013). The CENTURY model (Parton et al., 1988), which uses three state variables to represent litter pools and four state variables to represent SOM pools, is chosen to simulate decomposition of plant litter into SOM in CLM. The vertical dimensions of litter and SOM pools represent different decay rates during the decomposition process (Koven et al., 2013). The decomposition of litter leads to the formation of SOM, while a certain fraction of decomposed carbon flux is released as CO 2 during the transformations due to heterotrophic respiration.
In CLM, ET is partitioned into soil evaporation, canopy evaporation, and canopy transpiration. A transpiration wetness factor is used to evaluate the limitation of water supply through root absorption to sustain canopy transpiration, with a larger wetness factor representing a higher possibility to meet canopy transpiration requirements. Groundwater in the saturated zone is explicitly modeled to determine water table depth. This is enabled by the explicit treatment of soil thickness in this new version and solutions of the 1-D Richards equation from surface to bedrock. For a complete description of CLM5, especially its new features, interested readers are referred to Lawrence et al. (2018).

Model Improvements
In order to explicitly simulate the effects of bioenergy crops, it is necessary to develop a new scheme for perennial bioenergy crops, which are currently missing in CLM5 and have quite different physiological traits than annual crops . In this study, we added two new CFTs, Miscanthus and switchgrass, into CLM5 ( Figure 2) by representing the land management practices specific for these crops and tuning parameters using observed data at UIEF from 2009 to 2013. By performing sensitivity analyses, parameters related to photosynthesis capacity, plant physiology and phenology, carbon allocation, decomposition, and carbon cost of nitrogen uptake are the most sensitive ones to describe Miscanthus and switchgrass ( Table 1). The other parameter values are listed in Table A1.
The land management practices associated with Miscanthus and switchgrass during the study period such as planting, harvesting, fertilization, and irrigation, as described in the section 2 and illustrated in Figure 1b, are implemented in CLM5. Specifically, there is no annual planting for these two perennial grasses. Little fertilizer is applied to switchgrass, no fertilizer is applied to Miscanthus, and no irrigation is applied to either perennial crop during the growing seasons. The harvest frequency is once per year in December during the late winter. About 70% of the aboveground biomass is removed at the harvest time step to represent the harvest for lignocellulosic biofuel crops . Since we only focus on the impact of land management and plant characteristics, we decided to keep soil hydrologic parameters constant in the experiments.
Three single-point simulations are configured at the UIEF to represent maize-soybean rotation, switchgrass, and Miscanthus ecosystems. Meteorological forcing data are retrieved from the North American Land Data Assimilation System (NLDAS) forcing dataset (Xia et al., 2012) to drive model simulations. The model was spun-up for 1,000 years so that all the state variables in the model, especially total ecosystem soil carbon and groundwater table depth, reached equilibrium. The soil organic carbon (nitrogen) content to the typical depth of plowing (20-30 cm; Needelman et al., 1999;Yang & Wander, 1999) after spinning up varies from 4,200 (383) to 5,000 (457) gC m −2 (gN m −2 ), which is consistent with the observed soil organic carbon (nitrogen) content derived from U.S. Department of Agriculture, National Resources Conservation Service (https://websoilsurvey.nrcs.usda.gov/app/WebSoilSurvey.aspx) that ranges from 4,417 (404) to 6,339 (570) gC m −2 (gN m −2 ). The same initial condition was used for the three single-point simulations because uniform soil fertility was achieved before the plantation of the three ecosystems in 2008 (Zeri et al., 2011).

Sensitivity Analysis
Model sensitivity to each key parameter is examined through a linear sampling (i.e., 10 times with equal increment) of its value within a given range (Table 1) by varying one parameter at a time while fixing others as used in the benchmark simulation (Table 2). Additionally, the effect of fertilizer amount on biomass yields

Journal of Advances in Modeling Earth Systems
and carbon fixation is tested for perennial grasses (Table 1). The key parameters and their corresponding values used in the benchmark simulation for CFTs of corn, soybean, switchgrass, and Miscanthus in CLM5 are summarized in Table 2. These parameter values are consistent with previous studies (VanLoocke et al., 2012;Zhu et al., 2017).
The equation used to calculate sensitivity index (S) is where V s and V ref are the simulated gross primary production (GPP), net ecosystem exchange (NEE), ecosystem respiration (ER), or latent heat (LE) results for sensitivity tests (Table 1) and benchmark simulation (simulation using the parameters in Table 2), respectively, and P s and P ref are parameter values for sensitivity tests (Table 1) and reference simulation (Table 2), respectively.

Water Use Efficiency
Water use efficiency (WUE) is usually defined as the ratio of carbon accumulated over water used during a certain period of time. It represents the efficiency of crops in using water to produce biomass and is a key factor to evaluate bioenergy crop performance . Observational studies suggest that perennial crops have advantage over annual crops for enhanced WUE. The equation used to calculate exceeded water use efficiency (ExWUE) of perennial grasses than annual crops is ( 2) where GPP per and GPP ann are GPP for perennial grass and annual crop, respectively, and ET per and ET ann are ET for perennial grass and annual crop, respectively.
Perennial ecosystems differ from annual crops not only for physiology characteristics but also for phenological properties, such as a longer growing season, which could have large effects on water and carbon components (Zeri et al., 2011. The equation used to calculate the fraction of decrease/increase in a variable that explained by longer growing season (F) is where V per and V ann are mean annual values for perennial grass and annual crop, respectively, V per,w and V ann,w are mean values for perennial grass and annual crop after September, respectively.

Sensitivity Analysis
GPP, NEE, ER, and LE for Miscanthus ( Figure 3a) and Switchgrass (Figure 3b) are most sensitive to parameters associated with photosynthetic capacity (s_vcad, i_vcad, and slatop) and crop phenology (mxmat; equation (1)). The photosynthesis parameters control the light-use efficiency, and phenological parameters determine the transfer of carbon and nitrogen (CN) from storage pools to new growth and litter pools; both are important for crop production. Parameters for carbon cost of N uptake affect ER and thus have a large impact on NEE. Furthermore, ER is sensitive to CN allocation parameters (fleafi) as they control the partitioning of CN between above-and below-ground biomass. The differences in photosynthetic rates and crop phenology characteristics are potentially the dominant factors contributing to the physiological and morphological difference and environmental benefits of perennial crops with respect to annual crops .
While sensitivity patterns of carbon and energy fluxes to the key parameters are similar for Miscanthus and switchgrass, evident from Figure 3, switchgrass is relatively more sensitive to parameters related to 10.1029/2019MS001719

Journal of Advances in Modeling Earth Systems
photosynthesis, phenology, and CN allocation, while Miscanthus is more sensitive to parameters associated with decomposition (rf_s2s1_bgc) and carbon cost of nitrogen uptake (kn_nonmyc and fun_cn_flex_c; Figures 3 and A1 and A2). This is likely due to their difference in leaf-level photosynthesis . Given that photosynthetic rates and light-use efficiency appear to be key factors explaining the higher productivity of perennial grasses compared with annual crops, more field efforts with emphases on the estimation of these crucial parameters could help to better characterize perennial bioenergy crops and reduce uncertainty in model predictions.
In addition, our sensitivity analysis reveals that N supply has positive yield effects on both two perennial crops with a commensurate extent (Figures 3 and A2). A small amount of nutrient can help effectively establish the perennial cropping ecosystems ( Figure A2). Research indicates that the rooting systems of perennial grasses are very efficient in taking up nutrients (e.g., Smith et al., 2013) and retaining these nutrients from year to year (e.g., Heaton et al., 2009). Despite the uptake and within-plant translocation capabilities of perennial grasses, fertilization has been shown to increase establishment and growth of Miscanthus and switchgrass (Maughan et al., 2012) with addition of little to no leaching (Arundale et al., 2014;Smith et al., 2013).
Future field experiments with different land management practices, particularly different applications of fertilizers, deserve further investigation to better understand the effect of crop management on the biomass productivity of perennial bioenergy crops and potentially provide an optimum nitrogen fertilization rate (Cadoux et al., 2012;Song et al., 2016).
Based on the sensitivity analyses, we selected the simulation with highest coefficient of determination (R 2 ) and lowest relative differences between site-level measurements and simulations as the benchmark simulation for analyses in sections 4.2-4.4.

Seasonal Carbon, Water, and Energy Budgets
The overall patterns of GPP, including the general transitions from leaf emergence to senescence, the initial increase after leaf onset in spring, the timing of peak carbon uptake in summer, and the decline after leaf fading in winter, are well simulated at all three ecosystems at a daily time step over the study period (2009Figures 4a-4c). About 70%, 80%, and 75% of variances in observed GPP are explained by the model for maize-soybean rotation, switchgrass, and Miscanthus, respectively (Figures 5a-5c). Absolute   Figures 5d-5l).

Carbon Budget
The model captures the starts and ends of growing seasons of Miscanthus and switchgrass, which are about 50 days longer than the traditional maize/soybean rotation ecosystem (Figure 4). Switchgrass and Miscanthus become net carbon sink early in the growing season beginning in April, while maize/soybean rotation is still a carbon source because of its late leaf emergence ( Figure 4) and therefore does not become a carbon sink until June. In addition, perennial grasses still maintain high carbon assimilation rates into late fall (i.e., up to November), while traditional crops have already become a carbon source starting in October ( Figure 4). However, the maximum NEE values behave differently. The maize/soybean rotation has the  Table 1) and benchmark simulation (red line; highest rate of carbon uptake under average climate conditions (Figures 4d-4f), consistent with early findings that the maximum rate of photosynthesis for maize is greater than the other two perennial bioenergy crops (Dohleman, 2009;Zeri et al., 2011).
Although there are some biases between observed and simulated mean annual GPP and NEE, the model well simulates the relative magnitudes of these variables for the three ecosystems (Figures 6a and 6b).   Table 3), switchgrass and Miscanthus are more productive and have higher potentials for biomass production than maize/soybean rotation (Figure 6a) under the same climate and environmental conditions, as a result of their longer growing season for about 50 days (Figure 4; Zeri et al., 2011). In addition to higher productivity, perennial grasses can potentially provide ecosystem services such as reduce soil disturbance, improve soil carbon storage, and increase biological activity (Tolbert et al., 2002), associated with increased belowground biomass such as roots and rhizomes (7.6, and 10.5 kg C m −2 Miscanthus and switchgrass compared to 6.7 kg C m −2 for rotation) and fresh litter (1.56 and 0.65 kg C m −2 for Miscanthus and switchgrass compared to 0.66 kg C m −2 for rotation).
The two perennial grasses are larger net carbon sinks than traditional food crops, evident from higher mean annual NEE values of Miscanthus and switchgrass compared with maize/soybean rotation (Figure 6b), owing to their extended growing seasons. Note that while both perennial grasses can uptake more CO 2 than annual food crops, Miscanthus is a larger net carbon sink than switchgrass and has the highest carbon-use efficiency (Figure 6b). This is consistent with previous studies that perennial grasses, especially Miscanthus, could enhance carbon sequestration (Elshout et al., 2015;Qin et al., 2015). In addition, perennial grasses have extra beneficial effects on soil and environment because their high biomass yields and carbon sequestration can be achieved under modest fertilization needs, thereby can minimize direct carbon losses associated with planting and establishment and reduce nitrate leaching, consistent with many other studies (McIsaac et al., 2010;Housh et al., 2015;Song et al., 2016).
The simulated leaf area index (LAI) is also in good agreement with the observed data for the three crops (Figure 7). CLM5 well captured the seasonal development of LAIs and the longer growing season of switchgrass and Miscanthus than that of maize/soybean rotation. However, CLM5simulated LAI tends to reach peak values earlier than observations by about one month across the three crops. This is potentially due to the biased simulation of land management practice in CLM5. CLM5 assumes that it is warm enough for the crop to be planted when the air temperature reaches a threshold. Harvest is assumed to occur when the maximum growing degree days required for crop maturity is reached or the number of days past the planting date reaches a crop-specific maximum (Lawrence et al., 2018). This approach failed to capture the local planting and harvest dates at the study site (Sacks et al., 2010). As a result, the simulated peak

Journal of Advances in Modeling Earth Systems
time of LAI during the growing season was earlier than the observed peak time by almost one month (Figure 7). In summary, land management practices, especially planting and harvest, play a crucial role in simulating crop dynamics. Further efforts are needed to better represent these processes (e.g., in a spatialexplicit manner) in CLM5.

Energy Budget
The physiological characteristics of perennial grasses relative to traditional annual crops such as larger LAIs ( Figure 7) and above-ground biomass (e.g., leaf, stem, and grain; Table 3) also result in altered biophysical effects as evident in the surface energy budget terms. The daily LE for Miscanthus (LE=50 W m −2 ) and switchgrass (LE=49 Wm −2 ) are a little higher than maize/soybean rotation (LE=48 W m −2 ; Figures 4j-4l), in correspondence with the larger amount of annual cumulative ET for these two perennial grasses (see Figure 6c). The obvious greater LE values after September for perennial grasses also illustrate their longer growing season characteristics (Figure 4).
Simulated average albedo values of switchgrass (0.23) and Miscanthus (0.22) are slightly lower compared to that of maize/soybean rotation (0.24; Figure 8a), which is mainly attributed to their lower albedo values from January to March (Figure 8a), when snow is often present in the Midwestern United States. The perennial grasses retain higher above-ground biomass after harvesting and thereby smaller fractions of the land surface are snow covered compared to that of the maize/soybean rotation . The smaller area exposed to highly reflective snow results in the lower surface albedo of perennial grasses compared to that of maize/soybean rotation before March. Between the two perennial grasses, Miscanthus exhibits a lower simulated surface albedo because its LAI is larger than that of switchgrass ( Figure 7a).

Water Budget
Although the model slightly underestimates ET in general, it captured well the relative order of annual ETs across the three cropping systems (Figure 6c). Observed mean annual ET are 684, 706, and 722 mm for maize/soybean rotation, Miscanthus, and switchgrass, respectively, while simulated mean annual  ET are 617, 638, and 657 mm for the same three ecosystems, respectively (Figure 6c). Miscanthus and switchgrass evaporate and transpire more than the rotation ecosystem by approximately 3% and 5% at the annual scale, respectively. The order in magnitudes of the ET components for the three ecosystems is canopy transpiration followed by soil evaporation and canopy evaporation. The higher annual ET of Miscanthus and switchgrass can be attributed to their perennial nature and larger LAIs (Figure 7), consequently a longer duration and greater areas for canopy transpiration (460, 447, and 407 mm for Miscanthus, switchgrass, and maize/soybean rotation, respectively) and canopy evaporation (674, 645, and 530 mm for Miscanthus, switchgrass, and rotation, respectively; Figure 6c). The larger LAIs of these perennial crops effectively reduce soil exposure to incoming radiation, resulting in the smaller soil evaporation compared to that of the annual cropping system (128, 144, and 179 mm for Miscanthus, switchgrass, and rotation, respectively; Figure 6c).
The high water-utilization associated with perennial grasses as a result of their denser canopy and longer growing season leads to reduced annual total runoff (mean annual total runoff are 412, 399, and 482 mm for Miscanthus, switchgrass, and rotation, respectively; Table 3). The longer growing season accounts for roughly 14% and 16% for the net reduction in runoff for Miscanthus and switchgrass, respectively (equation (3); Figure A3). The remainders of runoff changes are induced by physiological and morphological differences of perennial grasses from annual food crops. These estimates in impacted runoff fractions are similar to those from previous studies that examining the effect of conversion from food crops to perennial crops on water cycle (Le et al., 2011). In addition, both surface and subsurface runoffs of perennial ecosystems are lower than that of annual crops (mean annual surface runoff are 113, 110, and 131 mm for Miscanthus, switchgrass, and rotation, respectively; mean annual subsurface runoff are 302, 289, and 351 mm for the same three crops, respectively).
Although annual ET of perennial ecosystems is higher, the production of above-ground biomass for switchgrass and Miscanthus is much more significant, leading to higher water-use efficiency by 35% and 28% than that of annual food crops, respectively (equation (2); Figure 6). The improved water-use efficiency of perennial grasses has been demonstrated in observational-based studies (Joo et al., 2016(Joo et al., , 2017Zeri et al., 2013). Furthermore, the difference in soil moisture (3,061, 2,980, and 2,968 mm for rotation, Miscanthus, and switchgrass, respectively) is not significant among the three ecosystems although perennial grasses have higher interception (674, 645, and 530 mm for Miscanthus, switchgrass, and rotation, respectively; Table 3) and transpiration (460, 447, and 407 mm for Miscanthus, switchgrass, and rotation, respectively; Figure 6c). The soil wetness factor of Miscanthus tend to be highest across the entire growing season, while that of switchgrass exceeds the maize/soybean rotation during the main growing period from August to October (0.923, 0.943, and 0.924 for rotation, Miscanthus, and switchgrass, respectively; Figure 8b). Evidently, perennial grasses are more efficient in supplying ET (Blanco-canqui, 2010;Tolbert et al., 2002). This is potentially associated with the more extended and penetrated root systems of switchgrass and Miscanthus (Song et al., 2016). For example, macropores formed by root channels under perennial grasses are reported to be twofold greater than under annual crops (Rachman et al., 2004), which can potentially contribute to decreased bulk density and increased soil porosity (Blanco-canqui, 2010) and thereby increased infiltration capacity (Cheng et al., 2017(Cheng et al., , 2018Zaibon et al., 2017).

Discussions on Study Scale Limitation
Many modeling and experimental studies have demonstrated that the productivity of switchgrass and Miscanthus at regional and global scales is controlled by various factors, such as spatial distribution of land cover, soil properties, climate conditions, and land management practices (Song et al., 2015(Song et al., , 2016. Song et al. (2015) identified four spatial zones characterized by their average yield amplitude and temporal yield stability over the period 2001-2012 in the United States by applying the ISAM. Song et al. (2016) demonstrated that the effect of production of biomass feedstocks on water quantity and nitrogen leaching differed among these four spatial yield zones. Therefore, the environmental benefits and climate change mitigation effects of widespread conversions from food crops to perennial crops would also highly depend on those climate and environmental factors (Blanco-canqui, 2010;Lemus & Lal, 2005;Leng & Huang, 2017), which vary region by region.
While this study demonstrates that perennial bioenergy crops are promising alternatives to traditional annual crops for biofuel feedstocks in terms of productivity as well as carbon-, water-, and nutrient-use 10.1029/2019MS001719

Journal of Advances in Modeling Earth Systems
efficiencies, these results are based on site-specific research in midwestern United States that under the same climate and environmental conditions. Questions remained to be answered about the potential climate change mitigation effects and environmental consequences of large-scale plantation of bioenergy crops.

Discussions on Model Structure Limitation
CLM5 introduces a new component for autotrophic respiration (Figure A4a) from the newly implemented FUN module, which takes into account of the carbon cost for nitrogen uptake (Fisher et al., 2010;Lawrence et al., 2018;Shi et al., 2016). The FUN model builds a reverse relationship between carbon expenditure and nitrogen content in soil. This carbon cost is a new reparation flux. Because little fertilizer was applied during the growing season and a subsequent low soil nitrogen content for the perennial crops, the FUN model would predict a remarkably big carbon cost for perennial crops. Consequently, the estimated nitrogen extraction using this cost would exceed the pool size for each source of nitrogen. After recalculating the estimated nitrogen uptake fluxes to take off the unmet demands, extra carbon cost still remains. One challenge we have met was that, rather than implementing an iteration round algorithm to recalculate a new carbon cost when this situation happens, the FUN model burns off the extra calculated carbon cost ( Figure A4b), which is eventually attributed to autotrophic respiration ( Figure A4c) and leads to a dramatic high simulated ER ( Figure A4d).
To remedy this problem, we tuned parameters associated with respiration fraction during decomposition (Tables 1 and 2) that is partially supported by field observations. Although the respiration fraction is fixed across all CFTs in CLM, observations show that perennial grasses can maintain similar or lower respiration levels as annual crops despite their much higher belowground biomass , which supports the lower respiration fraction of perennial grasses during the transformation process between SOMs (Table 2). Furthermore, before senescing for the year, perennial crops translocate much of the nutrients belowground, while annual crops translocate some to the seeds. Because of the translocation in perennial grasses, the C:N ratio is much higher for plant litter in perennial compared with annual ecosystems. Perennial ecosystems are also more efficient at taking up labile nutrients from the soil, which is supported by the lower simulated nitrogen content of litter and SOM for the perennials (Figures A4e and A4f and Table 3). The revised respiration fraction in decomposition would increase the formation of soil organic nitrogen ( Figure A4f) and resolve the issue of high carbon cost estimation ( Figure A4b).

Conclusions and Future Work
Areas planted with bioenergy crops are projected to expand in future emissions and land use change scenarios designed to mitigate climate change through enhancing terrestrial carbon uptake. The overarching objective of this study is to implement two perennial bioenergy crops (i.e., Miscanthus, and switchgrass) into CLM5 to enable explicit examination of the mitigation effects of projected land use change scenarios.
Our sensitivity analysis reveals that parameters associated with photosynthesis capacity and crop phenology play dominant roles in modulating simulated carbon and energy fluxes, while small amount of nitrogen fertilizers can greatly help the establishment and growth of perennial grasses. More field efforts on estimation of essential parameters could help to better characterize perennial bioenergy crops and reduce uncertainty in model predictions. Future field experiments with different land management practices, especially different fertilizer application strategies, deserve investigation to better understand the effect of crop management on the biomass yields and environmental benefits of perennial bioenergy crops.
We validate CLM5 against multiple in situ fluxes from study sites in central Illinois over five years of study from 2009 to 2013 for three ecosystems, namely, maize/soybean rotation, switchgrass, and Miscanthus. Variabilities in surface fluxes of water, carbon, and energy are well captured over the five continued growing seasons, including the transitions from leaf onset to peak productivity to leaf senescence as well as the longer growing season of perennial grasses relative to annual food crops. About 70%, 80%, and 75% of observed GPP is explained by the model for maize-soybean rotation, switchgrass, and Miscanthus, respectively.

Journal of Advances in Modeling Earth Systems
Miscanthus and switchgrass tend to be more productive than maize/soybean rotation and are larger net carbon sinks, which are primary attributed to their extended growing season for about 50 days and different canopy structural and biophysical characteristic such as larger LAIs and above-ground biomass (e.g., leaf and stem). These differences between annual food crops and perennial grasses lead to longer durations and greater leaf areas for canopy transpiration and interception, resulting in enhanced total ET and reduction in both surface and subsurface runoffs for sites with Miscanthus and switchgrass plantation. Our results reveal that by using more sustainable land management options, perennial bioenergy crops could produce higher harvestable biomass, sequestrate more carbon, and exhibit higher water-use efficiency compared to typical annual cropping systems, and thereby are promising alternatives to conventional food crops as biofuel feedstocks.
There was a one-month shift in the CLM5−simulated peak LAI toward early spring when it is compared to observations. Spatial-explicit planting and harvest dates should be implemented in CLM5. Our study also points to the need to improve model structure associated with nitrogen uptake in CLM5. Specifically, with limited fertilizer application during the growing season and a consequent low soil nitrogen content for the perennial crops, the FUN module in CLM5 that relies on a reverse relationship between carbon cost and soil nitrogen content to account for the carbon expenditure for nitrogen uptake would predict considerably high carbon expenses. To overcome the impact of such a structural deficiency on simulated fluxes, we tuned respiration fraction parameters in the Century-like decomposition cascade in CLM5. Additional investigation on the best approach to alleviate the problem is needed.
Given the model development feature of this paper, we performed sensitivity analysis rather than a formal calibration for the parameters used to represent the newly implemented bioenergy CFTs. Future work with comprehensive calibration may help to avoid the potential bias across different sites with perennial bioenergy plantations. Currently, the impact of bioenergy plantation on sustainable development has become an important scientific question as a result of a growing global demand for biofuels (Robledo-Abad et al., 2017). For example, Hejazi et al. (2015) showed that the climate change mitigation policies would actually increase water deficits if not fully considering their potential implication on water resources. While this study parameterizes CLM5 to explicitly simulate perennial bioenergy crops and validates simulation results against observations at site levels, the impact of extensive plantation of bioenergy crops on terrestrial hydrological and biogeochemical cycles as well as surface energy partitioning is complicated and yet to be examined. Such a study requires model configuration at regional scales driven by future climate and land use change scenarios that consider integrated roles of land policy and bioenergy availability in the context of global energy analyses . The Platform for Regional Integrated Modeling and Analysis (PRIMA) described in Kraucunas et al. (2015) that integrates an ESM with a multiscale dynamics model is a promising framework to tackle such problems. Future work on examining mitigation scenarios designed to represent different adaptation challenges  within this framework will be conducted to explicitly study the mitigation effect of large-scale plantation of perennial bioenergy crops. Table A1 displays the parameter values for maize, soybean, switchgrass, and Miscanthus used in CLM5. Figure A1 shows changes in GPP, NEE, LE, ET, and runoff from the sensitivity analysis for parameters associated with photosynthesis capacity and crop phenology for Miscanthus and switchgrass. It shows that switchgrass is more sensitive to parameters related to photosynthesis, phenology, and CN allocation, while Miscanthus is more sensitive to parameters associated with decomposition and carbon cost of nitrogen uptake. Figure A2 shows changes in GPP, NEE, LE, ET, and runoff from the sensitivity analysis for parameters associated with carbon cost of nitrogen uptake, CN allocation, decomposition, and fertilizer application for Miscanthus and switchgrass. It reveals that N supply has positive yield effects on both two perennial crops that a small amount of nutrient can help effectively establish the perennial cropping ecosystems. Figure A3 displays simulated seasonal total runoff for the three study ecosystems. It shows the longer growing season of perennial grasses accounts for about 14% and16% for the net reduction in runoff for Miscanthus and switchgrass, respectively. Figure A4 discusses the model structure limitation related to the FUN module.

10.1029/2019MS001719
Journal of Advances in Modeling Earth Systems Figure A1. Changes in (first column) gross primary production (GPP), (second column) net ecosystem exchange (NEE), (third column) latent heat (LE), (fourth column) evapotranspiration (ET), and (fifth column) runoff from the sensitivity analysis for parameters associated with photosynthesis capacity (first to third rows) and crop phenology (fourth and fifth rows) for Miscanthus (red solid line) and switchgrass (blue dash line). Detailed descriptions on the parameters are provided in Table 1.

10.1029/2019MS001719
Journal of Advances in Modeling Earth Systems Figure A2. Changes in (first column) gross primary production (GPP), (second column) net ecosystem exchange (NEE), (third column) latent heat (LE), (fourth column) evapotranspiration (ET), and (fifth column) runoff from the sensitivity analysis for parameters associated with carbon cost of nitrogen uptake (firstthird rows), CN Allocation (fourth row), decomposition (fifth row), and fertilizer application (sixth row) for Miscanthus (red solid line) and switchgrass (blue dash line). Detailed descriptions on the parameters are provided in Table 1