Scholarly article on topic 'Ecosystem model parameterization and adaptation for sustainable cellulosic biofuel landscape design'

Ecosystem model parameterization and adaptation for sustainable cellulosic biofuel landscape design Academic research paper on "Earth and related environmental sciences"

Share paper
Academic journal
GCB Bioenergy

Academic research paper on topic "Ecosystem model parameterization and adaptation for sustainable cellulosic biofuel landscape design"

GCB Bioenergy (2016), doi: 10.1111/gcbb.12316

Ecosystem model parameterization and adaptation for sustainable cellulosic biofuel landscape design


1Natural Resource Ecology Laboratory, Colorado State University, 1499 Campus Delivery, Fort Collins, CO 80523, USA, 2Department of Mechanical Engineering, Colorado State University, 1374 Campus Delivery, Fort Collins, CO 80523, USA, 3Pasture Systems and Watershed Management Research Unit, United States Department of Agriculture-Agricultural Research Service, University Park, PA 16802, USA, 4Department of Soil and Crop Science, Colorado State University, 1170 Campus Delivery, Fort Collins, CO 80523, USA

Renewable fuel standards in the US and elsewhere mandate the production of large quantities of cellulosic biofuels with low greenhouse gas (GHG) footprints, a requirement which will likely entail extensive cultivation of dedicated bioenergy feedstock crops on marginal agricultural lands. Performance data for such systems is sparse, and non-linear interactions between the feedstock species, agronomic management intensity, and underlying soil and land characteristics complicate the development of sustainable landscape design strategies for low-impact commercial-scale feedstock production. Process-based ecosystem models are valuable for extrapolating field trial results and making predictions of productivity and associated environmental impacts that integrate the effects of spatially variable environmental factors across diverse production landscapes. However, there are few examples of ecosystem model parameterization against field trials on both prime and marginal lands or of conducting landscape-scale analyses at sufficient resolution to capture interactions between soil type, land use, and management intensity. In this work we used a data-diverse, multi-criteria approach to parameterize and validate the DayCent biogeochemistry model for upland and lowland switchgrass using data on yields, soil carbon changes, and soil nitrous oxide emissions from US field trials spanning a range of climates, soil types, and management conditions. We then conducted a high-resolution case study analysis of a real-world cellulosic biofuel landscape in Kansas in order to estimate feedstock production potential and associated direct biogenic GHG emissions footprint. Our results suggest that switchgrass yields and emissions balance can vary greatly across a landscape large enough to supply a biorefinery in response to variations in soil type and land-use history, but that within a given land base both of these performance factors can be widely modulated by changing management intensity. This in turn implies a large sustainable cellulosic biofuel landscape design space within which a system can be optimized to meet economic or environmental objectives.

Keywords: biogeochemistry, carbon sequestration, cellulosic biofuel, ecosystem model parameterization, ecosystem modeling, feedstock, greenhouse gas emissions, landscape design, marginal land, nitrous oxide

Received 3 September 2015; accepted 7 October 2015


Correspondence: John L. Field, tel. +1 (970) 491 1547, fax +1 (970) 491 1965, e-mail:

The Energy Independence and Security Act of 2007 expanded the US renewable fuel standard to require the use of large quantities of 'advanced' and 'cellulosic' biofuels with lifecycle greenhouse gas (GHG) emissions reductions of 50% and 60%, respectively, relative to a conventional gasoline baseline (110th Congress of the United States, 2007). This mandate is predicated on the wide availability of biomass feedstocks with low direct


environmental impacts and causing minimal disruption to agricultural commodity markets, which could lead to indirect leakage effects (Searchinger et al., 2008). A variety of low-impact feedstock provisioning strategies have been envisioned including the collection of agricultural residues, forestry residues, and municipal wastes, as well as the cultivation of dedicated woody and herbaceous crops on marginal or non-agricultural lands (Campbell et al, 2008; Robertson et al., 2008; Tilman et al., 2009). In particular, perennial grasses such as switchgrass, Miscanthus, bioenergy-optimized sugarcane, and mixed prairie species have been identified as promising cellulosic feedstock crops due to their high

© 2015 The Authors. Global Change Biology Bioenergy Published byJohn Wiley & Sons Ltd.

This is an open access article under the terms of the Creative Commons Attribution License,

which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

yield potential, relatively low input requirements, high nitrogen use efficiency, and ability to sequester carbon through soil organic matter formation (Vogel et al., 2002; Tilman et al., 2006; Heaton et al, 2008; Walter et al., 2014). A comprehensive review of US feedstock availability by the Department of Energy suggests that widespread cultivation of high-yielding dedicated perennial grasses will be necessary to achieve the target of displacing 30% of U.S. petroleum consumption with biofuels (U.S. Department of Energy, 2011), and it has been estimated that biofuel supply chains based on such feedstocks will have highly favorable lifecycle GHG impacts as compared to first-generation biofuel technologies (Schmer et al., 2008; Davis et al., 2012; Wang et al., 2012).

Despite their promise, agronomic experience with perennial grass feedstock crops is still relatively limited, and questions around the best management practices to balance the often-competing goals of maintaining high yields while minimizing environmental impacts have not yet been resolved. The debate between the relative merits of low-intensity cultivation over large areas (often referred to as 'land-sharing') vs. more intensive production on a more limited land base ('land-sparing') is still being waged (Anderson-Teixeira et al., 2012). Process-based ecosystem modeling can play an important role in extrapolating limited existing perennial grass field trial results to make more general estimates of productivity, environmental impacts, and optimal management and landscape design strategies.

Modeling cellulosic feedstock yields & environmental impacts

The use of crop models for assessing managementenvironment interactions and predicting bioenergy feedstock crop productivity was thoroughly reviewed by Nair et al. (2012). Crop models such as APSIM, BioCro, and ALMANAC have been applied at regional or national scales to assess the productivity of first- and second-generation feedstock crops as affected by broad-scale climate-soil associations (Bryan et al., 2010; Miguez et al., 2012; Behrman et al., 2014). Bioenergy system design is not just a question of yield, however, and understanding the biogeochemical cycling of carbon, nitrogen, and water through these systems is essential for quantitatively evaluating their sustainability (Robertson et al., 2011). Perennial grass feedstock crops are often associated with a high potential for soil carbon sequestration. One recent meta-analysis suggests that switchgrass increases soil organic carbon (SOC) levels at a median rate of ca. 0.7 t C ha-1 yr-1 when cultivated on carbon-depleted agricultural soils, though performance is more neutral on pastureland or areas that

were not previously cropped (Qin et al., 2015a). Application of nitrogen fertilizers is typically required to replace losses during harvest and maintain yield levels, but emissions of nitrous oxide (N2O, a byproduct of soil microbial activity and a potent GHG) increase non-line-arly with increasing N rate (Hoben et al., 2011; Shcher-bak et al., 2014) and can threaten the overall lifecycle GHG footprint of any bioenergy system based on feedstocks with inefficient nitrogen cycling (Crutzen et al., 2008). The biogeochemical cycling of C, N, and H2O and associated fluxes of biogenic GHGs are tightly linked in all agroecosystems by fundamental mechanisms including plant tissue stoichiometry, photosyn-thetic pathway (C3 vs. C4 vs. CAM), stomatal conductance, and microbial mineralization/immobilization, and are sensitive to local climate, soil type, and land-use history.

Detailed reviews of process-based biogeochemical model use to capture these interactions in the context of bioenergy system sustainability assessment are provided by Thomas et al. (2013) and Robertson et al. (2015). The CENTURY model was among the first to be applied to bioenergy sustainability assessment, and it and its derivative DayCent model have been widely used to evaluate corn grain production, corn stover removal, and the dedicated cultivation of switchgrass and Miscanthus from the level of individual sites to national scales (Sheehan et al., 2003; Kim & Dale, 2005; Chamberlain et al., 2011; Davis et al., 2012; Lee et al., 2012; Duval et al., 2013). The Environmental Policy Integrated Climate model has also been applied extensively to bioenergy feedstocks in the context of economic analyses (Jain et al., 2010; Egbendewe-Mondzozo et al., 2011) and environmental sustainability assessments (Gelfand et al., 2013) at scales from regional (Zhang et al., 2010) to global (Kang et al, 2014).

Experience with using models to assess the productivity and biogeochemical implications of fine-scale variations in environmental factors such as soil type or topography across landscapes or individual farms is more limited. Studies of switchgrass cultivation in the southeastern US using crop production models have come to different conclusions as to whether yield is sensitive (Woli, 2012) or insensitive (Persson et al., 2011) to underlying soil type. At the finest spatial scale, a group at Idaho National Laboratory has used corn yield data from a precision agriculture system in Iowa to drive DayCent and models of soil erosion and identify areas of low nitrogen use efficiency to target for conversion to switchgrass (Abodeely et al., 2013). Many studies conduct large numbers of fine-scale simulations to make regional-scale estimates of feedstock productivity and environmental performance, though they typically do not report on soil-climate-management interactions

explicitly but rather emphasize more aggregate descriptions of landscape performance (e.g. Davis et al., 2012; Gelfand et al., 2013).

Challenges in bioenergy ecosystem modeling

The use of ecosystem models to assess different bioen-ergy landscape design strategies is complicated by challenges around the representation of marginal lands, adequate bioenergy crop parameterization, and selection of the most appropriate spatial resolution and agronomic management practices for simulation. While biogeochemical process models are increasingly used to simulate conversion of marginal agricultural lands to bioenergy feedstock cultivation (Bandaru et al., 2013; Qin et al., 2015b), these scenarios are particularly challenging from a modeling perspective. The definition of 'marginal' land itself is not straightforward or consistent across studies, and the term is applied to both land with low productivity potential and to land with ample productivity but vulnerable to long-term degradation (erosion, loss of soil organic matter, etc.) under conventional cropping systems. In many bioenergy assessment studies the marginal designation has been based on unfavorable biophysical properties as judged using land suitability ratings (Gelfand et al., 2013) or remote sensing techniques (Cai et al., 2011). Another more direct basis for the designation considers past transitions in an out of agricultural production as inferred from land-use datasets (Campbell et al., 2008), remote sensing (Wright & Wimberly, 2013), or sector-level economic modeling (Swinton et al., 2011). When considering lands designated as marginal based on their low productivity potential, bioenergy feedstock crops are often not immune to the factors that make such lands challenging for conventional crops. Recent perennial grass field trials purposely conducted on marginal sites indicate reduced productivity relative to performance on the prime lands typically encountered at many agricultural field stations (Mooney et al., 2009; Shield et al., 2012; Boyer et al., 2013). From an ecosystem modeling perspective, accurate assessment is only possible to the extent that underlying biophysical limitations on productivity (e.g. unfavorable climates, soil texture extremes, shallow soils, low soil organic matter levels, site drainage problems, slope, vulnerability to erosion, etc.) are represented directly or indirectly in model data inputs and the processes simulated.

It is well understood that process-based ecosystem models require proper parameterization specific to the agroecosystems being simulated in order to achieve good performance, and that such models have limited predictive power when extrapolated far beyond their

parameterization scope (Thomas et al., 2013). However, many bioenergy assessment studies are based on models parameterized under prime conditions and then extrapolated to highly marginal sites, or lacking an explicit independent validation of performance (e.g. Gelfand et al., 2013; Kang et al., 2014). In the case of the DayCent model, parameterization typically involves adjusting study site and crop parameters by hand in order to match observed real-world performance for a small number of field trial cases for which extensive data are available (Del Grosso et al., 2011). While this approach can often yield a high degree of fidelity across a range of performance criteria for the sites in question (Hudiburg et al., 2015), the very large number of empirically-determined crop and site parameters in the model makes the process vulnerable to over-parameterization. In such cases, model fit to the training dataset is improved via mechanisms that lack broader underlying ecological significance, reducing the generality of the resulting model for other geographic areas, environmental conditions, or management regimes (Necpálova et al., 2015). It is also possible to introduce bias with the selection of the parameterization cases themselves, if the researcher gravitates to focus on studies that confirm a priori assumptions of how a system 'should' perform.

There are additional challenges around the spatial resolution of landscape modeling and assumptions about crop agronomic management. Management factors including tillage intensity (Adler et al., 2007), fertilizer application rate (Davis et al., 2013), and rotation length (Pyorála et al., 2014) can potentially change the lifecycle GHG performance of a bioenergy system from positive to negative, an effect termed the 'management swing potential' (Davis et al., 2013). Management recommendations for bioenergy crops are not always well defined. For example, nitrogen fertilizer recommendations for Miscanthus have been widely debated (Arundale et al., 2014a), with important implications for the overall GHG footprint of production (Roth et al., 2015). To the extent that there are interactions between best management practices and sitelevel ecosystem properties (soil texture, land-use history, etc.), assumptions about management used within an assessment study should ideally be implemented at the level of management decision-making, that is, the field-scale. Some landscape modeling studies have started to endogenize management intensity questions by simulating productivity and environmental impacts at different rates of N application (Gelfand et al., 2013) or different levels of tillage (Zhang et al., 2010) for a given simulation run, for example, accounting for realistic variations in best management practices across the study landscape.

Study goals

Our study used the DayCent biogeochemical process model to assess perennial grass productivity and associated biogenic GHG emissions as a function of land quality and management intensity. Implications for bioenergy landscape design were explored through a case study of switchgrass production around a newly constructed commercial-scale cellulosic biorefinery in an area with substantial heterogeneity in soils and land use. This investigation expands on previous work in two main ways:

1. We conducted an extensive model parameterization and validation effort based on a data-diverse, multi-criteria approach, using a large parameterization dataset collected from the literature spanning wide gradients of climate, soil texture, and management intensity.

2. We evaluated the impacts of management intensification at the full spatial resolution of the assessment, estimating optimal levels of nitrogen fertilizer application for each point on the landscape in order to either maximize yield or minimize biogenic GHG emissions.

Our objective was to develop a rigorous, well-validated spatially explicit biogeochemical modeling capability that can serve as the basis for future integrated assessment and landscape optimization efforts.

Materials and methods

Case study introduction

We performed a landscape assessment case study simulating the cultivation of switchgrass (Panicum virgatum) to supply biomass to the Abengoa cellulosic biorefinery located outside the town of Hugoton in southwestern Kansas (Peplow, 2014), which began operations in fall 2014. While the plant will initially produce 25 million gallons of ethanol per year using corn stover as the primary lignocellulosic feedstock, switchgrass has been mentioned as an advanced cellulosic feedstock of interest and Biomass Crop Assistance Program Project Area 7 is sponsored by the company and targets the establishment of 20 000 acres of switchgrass production in the area (U.S. Department of Agriculture FSA, 2011). The case study region has long been at the center of agricultural sustainability and energy issues, having been deeply affected by the Dust Bowl in the 1930s (Kansas Historical Society, 2014) and being the site of the earliest hydraulic fracturing trials in the U.S. (Borowski, 2012).

Today, the surrounding Stevens County is a highly diverse and productive agricultural area. In 2012, 21.4% of the county was dedicated to highly-productive irrigated corn cultivation (average yield of 12.1 Mg ha-1, or 192 bushels acre-1) and 11% to dryland wheat (1.1 Mg ha-1, or 18 bu acre-1), with smaller fractions devoted to other crops and pasture/rangeland,

supporting an inventory of 45 500 head of cattle including calves (U.S. Department of Agriculture NASS, 2014). In this study we investigated the biogeochemical implications of converting non-irrigated cropland and rangeland in Stevens County and its six neighboring counties in southwestern Kansas and the Oklahoma panhandle to switchgrass cultivation (see Fig. 1), examining tradeoffs between productivity and associated biogenic GHG emissions as a function of underlying soil type and management intensity, specifically nitrogen fertilizer application rate. Issues of land ownership, conservation easement status, and other land-use policy limitations are excluded here, but explored in a subsequent publication dedicated to bioenergy landscape optimization.

The DayCent model

Productivity and net fluxes of biogenic CO and N2O from soils under switchgrass cultivation were modeled with the DayCent biogeochemistry model (Parton et al., 1998; Del Grosso et al, 2011). DayCent is a semi-empirical process-based model that simulates cycling of C, N, and water in natural and agroecosystems based on site-specific biophysical factors, land-use history, and management practices (e.g. tillage, fertilizer application, irrigation, etc.). The spatial and temporal scope of the model lies in between that of dedicated crop growth models (Miguez et al., 2012) and generalized earth climate system models (Anderson et al., 2013; Hallgren et al., 2013). DayCent has been used extensively to predict yields and environmental impacts of switchgrass cultivation (Adler et al, 2007; Chamberlain et al., 2011; Davis et al, 2012; Lee et al., 2012) and is also used to predict agricultural soil GHG emissions for the annual Inventory of U.S. Greenhouse Gas Emissions and Sinks (U.S. Environmental Protection Agency, 2014a).

DayCent computes soil temperature and moisture for different layers of the soil profile (resolved separately based on soil texture, bulk density, and pH) using daily climate data inputs. Crop growth (net primary productivity, or NPP) is simulated using species-specific parameters describing photosynthetic efficiency, tissue C/N ratio limits, above- and below-ground C partitioning, and phenology, many of which are determined empirically using model parameterization datasets as described previously. Daily biomass growth potential (NPPpot) is derived from top-of-atmosphere radiation (srad), corrected for atmospheric transmission losses and multiplied by a series of 0-1 factors that represent deviations from ideal temperature (ftemp) or soil moisture (fH2O) or limitations due to canopy immaturity or self-shading (fcanopy). This potential growth is then adjusted down if the available soil mineral nitrogen supply is limiting, as determined based on a maximum incremental biomass C : N ratio at the given level of plant maturity:

NPP = max(NPPpot, NPPNiim) where:

NPPpot = srad ■ transmission ■ ftemp ■ /h2O ■ fcanopy

NPPNiim = Navail ■ (C/N)n

Fig. 1 Map of all switchgrass field trial sites included in the full model parameterization and validation dataset prior to ecotype/lati-tude filtering. Ring size indicates the number of experimental treatments (e.g. different ecotypes or nitrogen fertilizer application rates) conducted at that site, and color represents the types of data available. The 7-county case study area is highlighted in pink, with a star designating the biorefinery location.

In addition to plant productivity DayCent also estimates soil carbon and nitrogen cycling including net changes in SOC levels and N2O emissions, the major constituents of the agricultural soil GHG balance. Carbon dynamics are simulated for soil surface pools and the top soil layer (0-20 cm by default), with organic matter represented by two litter pools (metabolic and structural) and three SOC pools- an 'active' pool representing microbial biomass and associated microbial products with a rapid turnover rates, and two others representing chemically/ physically stabilized carbon with decadal- ('slow' pool) and century-scale ('passive' pool) turnover times. Nitrogen mineralization and immobilization rates for each pool are controlled by the maximum and minimum permissible C : N ratios for each pool, soil temperature, soil moisture, and microbial efficiency as a function of soil texture (Parton et al., 1987). The soil nitrogen balance considers synthetic N fertilizer addition, manure and other organic N amendments, atmospheric deposition, volatilization, leaching, plant uptake, and N mineralization and immobilization associated with soil organic matter transformations. The model simulates nitrification of ammonium (NH4+) to nitrate (NO3~) including NOX and N2O by-products, as well as denitrification of nitrate to gaseous products (N2O, NOX, N2). Nitrification is simulated by multiplying available soil ammonium by a maximum potential nitrification rate adjusted based on soil temperature, water-filled pore space (WFPS), and pH limitations (Del Grosso et al., 2000; Parton et al., 2001). The overall rate of denitrification and the N2O/N2 ratio of its products are modeled based on the availability of nitrate and organic matter substrates (as inferred from heterotrophic respiration rate), and local soil micropore redox state and gas

diffusion rates as inferred from WFPS and heterotrophic respiration rate.

Ecosystem model parameterization & validation

We undertook an extensive parameterization and validation of the DayCent switchgrass growth model to improve overall simulation accuracy and verify the model response across gradients of land quality and management intensity. A large dataset of switchgrass field trials for the continental United States was assembled from the peer-reviewed literature. Studies were included in the dataset provided they (a) specified the underlying soil in sufficient detail that a corresponding Natural Resource Conservation Service Soil Survey Geographic database (SSURGO; Ernstrom & Lytle, 1993) map unit could be identified using the Web Soil Survey (U.S. Department of Agriculture NRCS); (b) specified key management variables (switchgrass ecotype planted, N fertilizer application rate, harvest date) in sufficient detail to define key DayCent simulation parameters; and (c) reported disaggregated annual yield and GHG results (studies reporting results averaged across multiple sites or N treatments were excluded). All modeled-vs.-measured performance comparisons were made on the basis of time-averaged results across the duration of the field trial, so studies that report either annual yield data or treatment-averaged data were included. If multiple cultivars of the same ecotype were included in a study, results were averaged across cultivars into a single representative value for the ecotype for simplicity. Additional details on the parameterization and validation dataset are provided in the Appendix S1.

A total of 24 appropriate studies were identified, covering 573 annual biomass yield points across 147 unique combinations of site (soil, climate) and management treatments (N rate, harvest date, etc.). Initial exploratory analysis confirmed the need to exclude the first two seasons of yield data before switchgrass yields stabilize (Lesur et al., 2013; Arundale et al., 2014b), and to filter out treatments for ecotypes grown outside their typical latitude range (up to 54°N for lowland varieties and down to 34°N for upland, as per Casler, 2012). The remaining dataset was then randomly split at the level of individual studies 70 : 30 into a parameterization dataset and an independent validation dataset (Table 1). Several of these field trials included more detailed information on biomass partitioning or nitrogen content (Frank et al., 2004; Dohleman et al., 2012; Anderson-Teixeira et al., 2013), long-term changes in SOC based on either repeat measurements or paired sites (Ma et al., 2001; Liebig et al., 2008; Follett et al., 2012; Anderson-Teixeira et al., 2013; Bonin & Lal, 2014), and/or time-resolved N2O emissions (Nikiema et al., 2011; Hong et al., 2012; Schmer et al., 2012; Smith et al., 2013). There was insufficient data available to perform an independent validation of these model performance criteria, so all of these studies were included in the parameterization dataset. While the raw SOC dataset was very noisy, eliminating individual data points that were not reported as

statistically significant yielded a final reduced modeling dataset that behaved more predictably and was consistent with recent switchgrass SOC meta-analysis results (i.e. Qin et al., 2015a).

The combined parameterization and validation dataset covers a wide range of latitudes and longitudes (Fig. 1) and temperature/precipitation regimes (Fig. S1), as well as a wide range of soil textures and Natural Resource Conservation Service land capability class (LCC) ratings (Helms, 1992; Fig. S2). Note that LCC ratings are reflective of a variety of land-use limitations, some of which are explicitly simulated in the Day-Cent model (e.g. dry climates, extreme textures, shallow soils) and some of which are not (e.g. erosion susceptibility, drainage class). The switchgrass crop parameterization was further informed with data from greenhouse or growth chamber experiments looking at productivity response across gradients of temperature (Balasko & Smith, 1971; Hsu et al., 1985; Reddy et al, 2008; Kandel et al., 2013; Wagle & Kakani, 2014) or moisture (Xu et al, 2006).

The parameterization process started with a default switchgrass crop parameter set based on previous work (Adler et al., 2007; Davis et al., 2012) and focused on refining parameters relating to productivity, temperature and moisture stress response, nitrogen management, shoot vs. root partitioning, and tissue death and turnover rates, with separate parameterizations

Table 1 DayCent switchgrass crop parameterization and validation and data sources

Parameter types

Data sources

# treatments

Parameterization Aboveground biomass yield

Soil organic carbon (SOC) changes

Soil nitrous oxide (N2O) emissions

Seasonal aboveground and belowground biomass accumulation, and/or C/N ratio Phenology

Productivity response to temperature

Productivity response to soil moisture Independent validation Yield

Ma et al. (2001), Fuentes & Taliaferro (2002), Vogel et al. (2002), Frank et al. (2004), Pearson (2004), Mulkey et al. (2006), Schmer et al. (2008, 2012), Nikiema et al. (2011), Dohleman et al. (2012), Follett et al. (2012), Hong et al. (2012), Kering et al. (2012), Anderson-Teixeira et al. (2013), Boyer et al. (2013), Bonin & Lal (2014) and Pedroso et al. (2014)

Ma et al. (2001), Liebig et al. (2008), Follett et al. (2012), Anderson-Teixeira et al. (2013) and Bonin & Lal (2014) Nikiema et al. (2011), Hong et al. (2012), Schmer et al. (2012) and Smith et al. (2013) Frank et al. (2004), Dohleman et al. (2012) and Anderson-Teixeira et al. (2013)

Sanderson (1992), Hopkins et al. (1995), Sanderson et al. (1997), Berdahl et al. (2005), Casler et al. (2007) and Wang et al. (2013) Balasko & Smith (1971), Hsu et al. (1985), Reddy et al. (2008), Kandel et al. (2013) and Wagle & Kakani (2014) Xu et al. (2006)

Staley et al. (1991), Muir et al. (2001), Cassida et al. (2005), Adler et al. (2006), Fike et al. (2006), Arundale et al. (2014b) and Wilson et al. (2014)

98 (67 after filtering)

18 (8 after filtering)


49 (44 after filtering)

for both upland and lowland ecotypes as appropriate. Initial exploratory analysis suggested that capturing differences in phenology between upland and lowland ecotypes was essential for accurate yield simulation, consistent with the current understanding of maturation based on photoperiod being a strong determinant of yield differences between different cultivars grown at a given latitude (Casler et al., 2004). We set green-up dates uniformly for both ecotypes as function of latitude based on a variety of literature sources (Sanderson, 1992; Sanderson et al., 1997; Wang et al., 2013) as illustrated in Fig. S5. Peak biomass dates were predicted as a function of both latitude and ecotype as inferred from a variety of sources (Sanderson, 1992; Hopkins et al, 1995; Sanderson et al., 1997; Vogel et al, 2002; Frank et al, 2004; Berdahl et al, 2005; Casler et al, 2007; Ander-son-Teixeira et al., 2013; see Figs S6 and S7) and used to trigger plant senescence events within DayCent.

After crop phenology was set other parameter adjustments were implemented manually, using the model automation routine described in the next section to rapidly evaluate parameter changes against the 67-point parameterization dataset. The parameterization process focused on maximizing modeled-vs.-measured fidelity (based on visual inspection of the plotted data and calculation of modeled-vs-measured regression parameters, Pearson correlation coefficient, and root mean squared error; e.g. Smith et al., 1996) for upland and lowland ecotype yields, changes in SOC, and growing season N2O emissions, but also took into account time-resolved shoot : root and C : N ratio data where available for a multi-criteria evaluation of parameter set performance (e.g. Fig. S13). Once the parameterization process was complete, independent validation of upland and lowland yield performance was conducted based on the data held in reserve (i.e. using the holdout validation method).

Spatial data inputs, model initialization & automation

A variety of spatially explicit data inputs are necessary to initialize and run the DayCent model for a large-scale parameterization or landscape analysis, including data on climate, soil type, and land-use history. Data sources used in this analysis are summarized in Table 2. Soil texture, rock fraction, and pH

Table 2 Summary of spatial data inputs

for different soil profile layers of the dominant soil component for each map unit were taken directly from the SSURGO database (Ernstrom & Lytle, 1993), and bulk density, field capacity, wilting point, and saturated hydraulic conductivity were computed using the Saxton equations (Saxton et al., 1986). Climate data on a 32 km grid was derived from the North American Regional Reanalysis database (NARR; Mesinger et al., 2006).

Land-use history and current land management practices were compiled from a variety of sources. Current land use was determined from the National Land Cover Database 2006 (NLCD; Wickham et al., 2013), re-sampled from the native 30 m resolution to 240 m for ease of use, and re-classified into the simplified categories of annual agricultural lands ('cultivated crops', 'pasture/hay'), rangeland ('dwarf shrub', 'shrub/ scrub', 'grassland/herbaceous', 'sedge/herbaceous'), and excluded (all other categories including forested and developed lands). Irrigated areas were identified using the MIrAD-US database (Pervez & Brown, 2010), and federally-owned lands were identified using the USGS Federal Lands of the United States data layer (U.S. Geological Survey, 2015) and excluded from further analysis (3.4% of the landscape, most part of Cimarron National Grassland). These five GIS layers were then intersected and small slivers were eliminated by merging all polygons smaller than 1 ha into the neighbor with which they shared the longest border. This yielded 39 320 polygons of a variety of sizes across the seven-county Hugoton case study area (Fig. S16), representing 3779 unique combinations of model inputs requiring individual simulation, which we refer to as DayCent modeling 'strata'.

For each strata, the DayCent model was pre-initialized using the same pre-settlement and historical land use assumptions as used in the EPA Inventory of U.S. Greenhouse Gas Emissions and Sinks (U.S. Environmental Protection Agency, 2014a) and described in detail by Ogle et al. (2010). Model initialization included an equilibration run of several thousand years duration reflecting the natural state of the land prior to conversion to agriculture in order for all soil C and N pools to achieve steady-state values. Historical management between initial plow-out and the modern period was simulated with crop rotations and management practices compiled at regional scale from a variety of historical and modern sources (Ogle et al.,

Spatial Data

database type Year Native resolution URL

SSURGO Soils 2012 1 : 12 000-1 : 63 360 soils/survey/geo/?cid=nrcs142p2_053627

NARR Daily weather 1979-2010 32 km

NLCD Land use 2006 30 m

MIrAD-US Irrigation extent 2007 250 m

Federal Lands Federal land 2005 640 acres/1 mi2/1 : 2 000 000

of the ownership

United States

MIrAD-US, MODIS Irrigated Agriculture Dataset for the United States; NARR, North American Regional Reanalysis; NLCD, National Land Cover Database; SSURGO, Soil Survey Geographic database.

2010). The future switchgrass simulations were then executed across part of a 29-node, 288-processor cluster computing system at the Colorado State University Natural Resource Ecology Laboratory. Parallel execution was implemented in Python ( using forking operations to take advantage of multiple cores within a given node.

Landscape design analysis scenarios, results processing & sensitivity analysis

For the landscape analysis case study we simulated conversion of all non-irrigated, non-federally owned polygons within the seven-county case study area to rain-fed lowland switchgrass cultivation. We conducted 30-year forward simulations to assess long-term productivity and trends in soil C and N cycling, recycling the full range of the NARR historic weather record to represent future weather conditions. In order to assess the response of crop productivity and GHG performance to management intensity, seven different rates of nitrogen fertilizer application were simulated for each strata (0150 kgN ha-1 in 25 kgN ha-1 increments). We assumed that switchgrass would be replanted every 10 years after field preparation consisting of chisel plow and field cultivator operations, and that the crop would be neither fertilized nor harvested the year of establishment in order to limit competition from weeds and ensure robust crop establishment, as per local

extension recommendations. These assumptions are highly conservative as switchgrass is often established in this region without tillage, the need for periodic replanting is widely debated, and first-year harvest can be possible if the crop achieves sufficient first-year productivity.

Switchgrass harvest yields, changes in SOC levels, and annual N2O emissions were then averaged over the 30-year simulation period for each strata. Average annual N2O emission values were converted into CO2 equivalents using a 100-year global warming potential (GWP100) value of 298 (Forster et al., 2007) then added to the CO2 flux values associated with average annual net SOC changes for an estimate of total direct biogenic emissions. To determine the biogenic GHG intensity of production (Mg CO2eq Mg-1 biomass harvested) total emissions per hectare were divided by the associated simulated switchgrass yield. Continuous functions of biomass yield and biogenic GHG intensity vs. nitrogen fertilizer application rate were developed for each strata by applying a cubic regression to the 30-year averaged simulation results for the different N rate simulations, and the yield-maximizing and GHG balance-minimizing N rates were then interpolated. The simulated yield and GHG intensity associated with these optimized N rates was then re-associated with the appropriate landscape polygons, and aggregated across the full landscape in order to develop curves illustrating total potential landscape productivity and biogenic GHG emissions balance when strata are managed for these different

Table З DayCent switchgrass crop parameter changes

Parameter Original Lowland Upland

Type of parameter change name value value value

Productivity potential and temperature response: PRDX(1) 2.V5 4* 3.5*

Increase in productivity to reflect updated solar PPDF(1) 30 30 30

radiation model and compensate for increased PPDF(2) 45 44 44

belowground partitioning; differentiation in PPDF(3) 1 0.75 0.75

productivity and temperature response between PPDF(4) 2.5 2 2

upland and lowland ecotypes

Growth response to moisture stress: Reduced CWSCOEFF(1,1) 0.38 0.35 0.35

sensitivity to soil moisture stress CWSCOEFF(1,2) 9 14 14

Belowground partitioning: Reduction in baseline CFRTCN(1) 0.5 0.7 0.7

BG partitioning rate. BG partitioning in response CFRTCN(2) 0.3 0.25 0.25

to moisture stress reduced, but response to CFRTCW(1) 0.6 0.4 0.4

nutrient stress increased CFRTCW(2) 0.3 0.25 0.25

Tissue N and lignin content: Root maximum PRBMX(1,1) 55 50 50

allowable C : N ratio lowered slightly. Root lignin FLIGNI(1,2) 0.26 0.06 0.06

content reduced. Small amount of N fixation FLIGNI(1,3) 0.26 0.13 0.13

added to make growth under no-fertilizer SNFXMX(1) 0 0.005 0.005

conditions more realistic

Tissue death rates: Death rate and fall rate for FSDETH(3) 0.05 0.075 0.075

shoots increased. Root maturation rate increased, FALLRT 0.01 0.1 0.1

and turnover rate of both juvenile and mature CMXTURN 0.12 0.3 0.3

roots increased RDRJ 0.4 0.72 0.72

RDRM 0.2 0.54 0.54

N conservation: Increased translocation of CRPRTF(1) 0.15 0.43 0.43

nitrogen from shoots to roots during senescence

*Change in the PRDX(1) values also reflects a recent change in the DayCent model to simulate atmospheric transmission losses of photosynthetically active radiation.

objectives. The sensitivity of these landscape results to key crop parameters, landscape characterization, and switchgrass cultivation scenario assumptions was assessed as detailed in Table 4 in order to determine the overall robustness of our conclusions. All results analysis routines were automated in Python through a combination of SQLite database operations ( ) via the sqlite3 module (, data manipulation in the native Python list data type, and figure generation using the matplotlib.pyplot module (


Ecosystem model parameterization & validation

A total of 79 different switchgrass parameter set iterations were ultimately developed and tested. Final upland and lowland ecotype parameter values that differ from the default DayCent switchgrass crop parameterization are detailed in Table 3. The most significant changes were:

• Increased plant potential NPP rate - The most recent version of DayCent explicitly models solar radiation atmospheric transmission losses, and the revised potential NPP parameter PRDX(1) must be adjusted higher relative to previously-published versions to reflect the new growth calculation on a canopy pho-tosynthetically active radiation (PAR) basis rather than a top-of-atmosphere PAR basis. Further fine adjustments were made to PRDX(1) to optimize the observed yield difference between the different eco-types and to offset slightly increased belowground C partitioning.

• Adjusted temperature and moisture stress response curves - Crop temperature (Fig. S8) and moisture stress (Fig. S9) response curves were set based on the greenhouse and growth chamber studies listed in Table 1. In the case of temperature response, fine adjustments to the edges of the curve where direct empirical data were lacking were implemented to improve overall modeled-vs.-measured yield performance across the full parameterization dataset. The same temperature and moisture response curves were used for both ecotypes. Comparison of measured and modeled yield ranges binned by site average growing degree day accumulation (Fig. S10) or annual precipitation (Fig. S11) verifies that the model accurately captures increasing switchgrass productivity at warmer, wetter sites.

• Increased belowground partitioning and root turnover - The default parameterization slightly underestimated belowground biomass, significantly underestimated observed SOC increases, and over-

predicted N2O emissions. A small increase in belowground partitioning coupled with a large increase in root turnover rates resulted in more carbon being cycled into the soil and more mineral N being taken up by the plant, improving model performance on all three criteria.

Model parameterization and validation results for yields and soil GHG fluxes are shown in Fig. 2. Sufficient data was available to perform holdout method independent validation of yield predictions for both the upland and lowland ecotypes (Fig. 2a). The out-of-sam-ple validation root mean square errors (RMSE) are 3.7 and 4.1 Mg ha-1 for the upland and lowland ecotypes, respectively, with minimal bias (mean difference of -0.2 and +2.3 Mg ha-1, respectively). When these data are binned by nitrogen fertilizer application rate (Fig. 2c) we see that the model is able to capture the general trend toward increased switchgrass productivity with increasing management intensity. In contrast, yield response to land quality was more ambiguous. Neither measured nor modeled yields exhibited a clear relationship with soil texture across the full parameterization and validation dataset (Fig. 2d), and the weak trend towards lower yields at sites with higher LCC rating (more marginal) was not replicated in our model (Fig. S12).

Field trial data on soil GHG balance were sparser, and the results presented in Fig. 2b reflect within-sample model performance against the parameterization dataset rather than an independent validation. Observed annual changes in SOC under switchgrass were much larger than measured growing season N2O emissions when compared in CO2-equivalent terms. The within-sample RMSE value for the combined SOC and N2O dataset is 0.69 MgCO2 eq ha-1 yr-1 across the wide range of sites, climates, and nitrogen application rates represented in the underlying studies (n = 19, Fig. 1). Bias calculations show that the model tends to err in the direction of overestimating N2O and underestimating SOC accumulation, so the resulting predictions of switchgrass GHG balance are somewhat conservative. Additional detail on SOC and N2O performance is available in Figs S14 and S15.

Landscape design case study

Simulated lowland switchgrass yields as a function of nitrogen fertilizer application rate across the 3779 unique DayCent strata in the Hugoton case study area are shown in Fig. 3 for both crop land and rangeland conversion, color-coded to the underlying soil texture of each strata. The maximum attainable yield under arbitrarily well-fertilized conditions shows significant

10 J. L. FIELD et al. (a):

o Low-param

• Low-valid

a Up-param

a Up-valid

3o ° P-

, A + O

A ' » A


5 10 15 20

Measured yields (Mg ha-1)

CT <D 0

o> s -1

• soc y y

a N20 y y y y - *

»III 1

-4 -3-2-10 1 2

Measured GHG flux (Mg C02 eq ha 1 yr_1)

n = 47

n = 42

n = 22

S 101-

Measured Modeled

0.0-7.4 7.5-14.9 >15.0

Nitrogen application rate (gN rrr2)

n = 40

n = 11

n = 21

n = 39

4 • 1 l-h

Measured Modeled

Clayey Loamy Sandy

Soil surface texture class

Fig. 2 Parameterization and validation results: (a) modeled vs. measured switchgrass biomass yield fits for lowland ecotype within sample ('Low-param', r = 0.10, root mean square error (RMSE) = 6.2 Mg ha-1, mean difference (MD) = —2.1 Mg ha-1) and out-of-sample ['Low-valid', r = 0.66, RMSE = 4.1, MD = +2.3] field trial results and upland ecotype within-sample ('Up-param', r = 0.38, RMSE = 4.0, MD = +2.1) and out-of-sample ('Up-valid', r = 0.26, RMSE = 3.7, MD = —0.2) results; (b) modeled vs. measured annualized changes in soil organic carbon (r = 0.08, RMSE = 1.05 MgCO2eq ha—1 yr—1, MD = +0.42 MgCO2eq ha—1 yr—1) and growing season N2O emissions (r = 0.54, RMSE = 0.22, MD = +0.10) on a CO2-equivalent basis (combined r = 0.79, RMSE = 0.69, MD = +0.24); (c) measured and modeled switchgrass yield ranges for different levels of nitrogen application across all parameterization and validation data points; and (d) measured and modeled switchgrass yield ranges for different soil surface texture classes across all parameterization and validation data points.

variation with soil texture, ranging from more than 10 Mg ha—1 in certain clay and sandy soils down to about 6 Mg ha—1 in the more moderate-textured silty soils. In this semi-arid climate of intermittent precipitation events soil moisture levels are often near wilting point, and simulated average yields reflect a tension between the greater total water holding capacity of finer-textured soils (an advantage during relatively wet years) vs. more effective infiltration and less surface soil evaporation in coarser soils (the so-called 'inverse texture effect', beneficial during dry years;

Noy-Meir, 1973; Epstein et al, 1997; Lane et al, 1998). Switchgrass yields on converted rangeland are generally less sensitive to fertilizer application rates than those on converted cropland, in some cases showing no response to increasing N rate. This is due to higher background mineral nitrogen levels from transient soil organic matter turnover following conversion in these areas. However, for most strata, full switchgrass yield potential is realized at N rates from 60 to 100 kgN ha—as indicated with circular markers in the upper panels of Fig. 3.

Fig. 3 Simulated yield and GHG intensity response to increasing nitrogen fertilizer rate for switchgrass production on former crop land and former rangeland across the 3779 distinct DayCent strata in the case study landscape. Interpolated yield maxima and GHG intensity minima are marked with solid markers for cropland conversion and open markers for rangeland conversion. The color of the lines and markers indicates the soil surface texture of the strata, with yellow = sand, red = clay, green = silt, and brown = loam.

When examining landscape assessment results under different levels of management intensity it is important to note a discrepancy between the metrics of total area GHG emissions or mitigation (MgCO2eq per hectare per year) and the GHG intensity of biomass production (MgCO2eq per Mg biomass grown). Biomass production GHG intensity results are shown as a function of nitrogen application rate across all simulation strata in the lower panels of Fig. 3. For most strata at most nitrogen fertilizer rates, 30 year-average soil carbon sequestration outweighs nitrous oxide emissions on a CO2-equivalent basis. Biomass has the lowest (most negative) direct biogenic GHG footprint when cultivated with no nitrogen fertilizer on previously cropped fine-textured soils due to their high potential for SOC accumulation. This sequestration value is increasingly offset at higher N rates due to marginal soil N2O emissions outpacing corresponding yield and SOC gains. In contrast, rangeland has higher initial SOC levels and thus less capacity for carbon sequestration after conversion to bioenergy feedstock cropping, which results in an overall net GHG intensity much closer to zero.

Total area biogenic GHG emissions are explored in relation to productivity in Fig. 4. Since switchgrass can

be managed either to maximize yields or to optimize soil GHG balance (i.e. to maximize soil carbon sequestration while minimizing N2O emissions), we selected a random 10% of the landscape polygons, determined the nitrogen application rate that maximizes yield and the rate that maximizes GHG mitigation for each polygon, and aggregated the results for these different management strategies across our landscape sub-sample. The difference between the two curves is representative of the degree to which system productivity and GHG performance can be modulated by adjusting management on the same limited land base. Converting 10% of the landscape to switchgrass managed under either strategy results in >0.75 Mt biomass feedstock production annually, enough material to supply approximately two facilities the size of the Abengoa biorefinery, with the associated net GHG impact of sequestering >0.06 Mt (60 000 metric tons) of CO2. Figure 5 illustrates that the optimal switchgrass management intensity for maximum GHG mitigation is related to the distribution of soil types and current land use across the case study area, factors that are also correlated with one another as conventional cropping tends to be concentrated on the moderate textured soils of this landscape.

Sensitivity analysis

Our modeled landscape productivity and GHG balance are highly sensitive to certain key crop model parameters, particularly optimal growth temperature, potential NPP, and response to moisture stress (Table 4), highlighting the importance of careful parameterization and validation. While the first two parameters were reasonably well constrained as a result of the switchgrass parameterization process, empirical data on switchgrass moisture stress in the literature is rare, and uncertainty around those parameters in the model should be considered large. Interestingly, the aggregate landscape simulation results show little sensitivity to the length of the switchgrass growing season in early spring and late fall, even though accurate characterization of the phenology proved essential for determining crop parameter sets with adequate out-of-sample validation performance. Accurate characterization of the landscape in terms of climate and land-use history is important to the integrity of simulation results as well, with GHG mitigation particularly sensitive to past land-use history. Landscape results were generally less sensitive to crop management scenario assumptions around tillage intensity and fertilizer timing.


Challenges of model development and data-diverse parameterization

This study was grounded in an extensive model parameterization and validation effort using a data-diverse multi-criteria approach enabled by automation of all simulation runs and results analyses. The diversity of studies included in our switchgrass field trial dataset

Fig. 4 Cumulative total soil GHG mitigation vs. cumulative switchgrass biomass production for a random 10% of the case study landscape under different management goals: managing each land parcel to maximize switchgrass yields (dashed line, highest-productivity sites aggregated first), or managing each land parcel to maximize ecosystem GHG mitigation (solid line, strongest mitigation sites aggregated first).

was intended to provide a highly general test of model performance independent of a single study, environment, or management practice, and the simultaneous consideration of different types of data (Table 1) was designed to ensure that accuracy of one model performance criteria was never improved at the expense of others. This approach proved useful for sorting out interrelated model responses, for example, the strong effects of belowground partitioning and root turnover rate parameters on harvested biomass yield, SOC changes, and N2O rates that must be balanced in a systematic manner during model parameterization.

Our resulting model explains approximately half of the observed variability in yield and GHG observations in our dataset, and realistically captures climate effects and responses to management intensity. While soil type has been observed to have an effect on bioenergy grass productivity in semi-arid climates outside the U.S. (Di Virgilio et al, 2007; o Di Nasso et al, 2015; Roncucci et al., 2015), we did not observe a strong texture signal in our domestic switchgrass productivity dataset, consistent with a similar previous large-scale model parameterization effort (Wullschleger et al, 2010). Future modeling work on soil-climate interactions in the U.S. would greatly benefit from additional field trails like Wilson et al. (2014) that include paired trails across multiple soil types on a landscape, in order to tease apart soil-precipitation interactions without the other larger sources of variability present in composite datasets.

While we believe the approach presented here represents an improvement over more limited model parameterization efforts, it is still possible to over-parameterize to a dataset of this size and achieve good fits via unrealistic mechanisms that do not translate well out-of-sample. In our experience, reliable yield performance was only achieved once crop phenology was adequately captured. This parameterization dependency is somewhat challenging, as there are a limited number of studies in the literature reporting detailed phenology in the form required for this type of generalized modeling effort (i.e. dates for green-up and peak biomass or senescence as a function of crop cultivar and site latitude). Additionally, since maturation dates can vary significantly even for cultivars within a single ecotype (Frank et al., 2004), focusing on only two ecotype groupings introduces additional errors. Accurate representation of phenology can be superseded to a certain extent by narrowing crop temperature response curves to truncate early and late-season productivity, but our initial parameterization efforts in this direction performed very poorly on validation.

After several dozen parameter set iterations we were approaching the limit of what could be accomplished with setting parameter values manually, as model

boil texti

Cla Silt

Fig. 5 Map illustrating (a) soil texture distribution, (b) current land use, and (c) the switchgrass nitrogen fertilizer application rates associated with maximum area GHG mitigation (as opposed to biomass GHG intensity explored in Fig. 3), for all non-federal non-irrigated cropland and rangeland across the Hugoton case study area. Aerial imagery is included as a background layer for scale, and the boundary of the 7-county case study area is shown in white with the biorefinery location marked with a yellow star.

responses became more subtle and antagonistic across the different performance criteria. This, coupled with the relatively large number of crop parameters in the DayCent model, suggests that future efforts could benefit greatly from systematic parameter optimization techniques such as inverse modeling and/or j-fold cross validation. Such techniques provide a more systematic and transparent approach, facilitate maximum extraction of information from a field trial dataset, and help to identify and avoid over-parameterization issues. While inverse modeling techniques have been demonstrated to improve model performance against small datasets (Necpalova et al., 2015), they are not commonly applied to crop growth and biogeochemistry models used in bioenergy sustainability analysis at the scale of a full parameterization and validation effort such as the one presented in this study.

More broadly, model performance is dependent on underlying model structure in addition to parameterization, and DayCent model development efforts to improve its performance in perennial grass systems are ongoing. For example, accurate yield estimates depend on capturing crop phenology and canopy dynamics, and efforts are underway to represent early-season leaf canopy closure and late-season senescence dynamics more realistically. Similarly, while past model development efforts have largely focused on the upper 0-20 cm soil layer most relevant for annual crops, representation of soil carbon changes under perennial grasses may be improved through more explicit simulation of deep

rooting, dissolved organic carbon movement, and other aspects of soil organic matter dynamics deeper in the soil profile (Campbell, 2015).

Landscape design case study interpretation and climate accounting issues

The Hugoton case study site was selected because it featured the most heterogeneity in soils and current land use among the first three commercial-scale cellulosic biorefinery sites in the U.S. (the other two being located in the Iowa corn belt; Peplow, 2014). However, the dry climate at this site proved challenging from a modeling perspective, with landscape simulation results showing high sensitivity to interactions between crop moisture stress parameterization and soil texture, and with N2O response to increasing N rate lower than what might be expected in a wetter climate. Overall, our landscape simulation results are similar to others in the literature that find the GHG balance of perennial grasses dominated by soil carbon sequestration immediately postestablishment, with only modest N2O emissions (e.g. Gelfand et al., 2013). In field trials where baseline N2O emission rates are reported and an Intergovernmental Panel on Climate Change (IPCC) Tier 1 estimate can be made, observed switchgrass N2O emissions rates are often near or below the lower bound of the IPCC N2O emissions factor range (0.3- 3% fertilizer N emitted as N2O-N; Fig. S15). This suggests that switchgrass is an efficient nitrogen cycler, and that the critique of biofuel

Table 4 Sensitivity analysis

Parameter or assumption


Change in total landscape productivity

Change in total landscape GHG mitigation

Baseline average landscape performance Yield = 7.24 Mg ha-1 yr-1 GHG mitigation = 0.76 Mg CO2eq ha-1 DayCent crop parameterization

Optimal temperature Productivity potential Moisture sensitivity

Root death rates


Site characterization Land-use history


Agronomic management Fertilizer date

Tillage intensity

Increase PPDF(1) by 3 degrees Decrease PRDX(1) value by 10% Switch to the default curve associated with annual crops Decrease RDRJ and RDRM by 10% Decrease growing season length by 15 days

Assume a uniform cropped history rather than a mix of cropped and grazed Assume the NARR weather for Hugoton for the entire 7-county study area

Assume fertilizer application 1 month earlier

Switch to no-till crop establishment

-21% -9.4% -1.1%



+0.11% +0.13%

-42% -18% +8.6%



+1.5% +0.9V%

GHG, greenhouse gas; NARR, North American Regional Reanalysis.

GHG mitigation benefits being outweighed by direct and indirect N2O emissions (Crutzen et al., 2008) is likely overstated for second-generation perennial grass feedstocks. Our predicted SOC sequestration rates after cropland conversion are very similar to median meta-analysis results in Qin et al. (2015a), while the finding of a small net positive sequestration with the conversion of grassland is consistent with, but on the optimistic edge of, their observed range.

Full quantification of the confidence interval around these landscape assessment results is highly challenging due to multiple levels of uncertainty in problem specification, landscape input data, model structure, and model parameterization (Walker et al., 2003). While previous work has shown that uncertainty in DayCent-esti-mated changes in SOC is dominated by model structural uncertainty (Ogle et al., 2010), repeating this empirical uncertainty assessment approach here is impractical as only one of the 145 treatments in our parameterization and validation dataset includes data for all three of the factors that determine biomass GHG intensity (yield, SOC change, and N2O; Fig. 1). Here we rely on sensitivity analysis to identify priorities for future model improvement efforts. When discussing model sensitivity and uncertainty issues it is also important to be cognizant of issues in the underlying field

trial datasets. Field measurements of N2O are very challenging due to emissions variability on extremely fine spatial (Li et al., 2013) and temporal (Jergensen et al., 2012; van der Weerden et al., 2013) scales, and studies based on sampling at weekly or every-other-week frequency (as was the case for all N2O studies in our dataset) are vulnerable to systematic biases of up to 20% and 60%, respectively (Parkin, 2008; Smemo et al., 2011). Additionally, small-scale field trials do not necessarily reflect imperfections in agronomic management (e.g. uneven fertilizer application) and real-world harvest losses, possibly introducing a systematic over-estimation of switchgrass productivity (Searle & Malins, 2014).

Finally, there are several issues around climate impact accounting relevant to this landscape study. While soil carbon sequestration will eventually attenuate even though N2O emissions will persist for as long as N fertilizer is being applied (Sheehan et al., 2003; Adler et al., 2007), our assessment for this semi-arid system suggests it will take 60-80 years for annual sequestration and N2O rates to reach parity (Fig. S17). A more dynamic climate impact accounting approach that takes transient forcing benefits into account (e.g. Holtsmark, 2015) would tend to further weight near-term SOC benefits against future N2O emissions, though no standardized accounting approach is yet widely accepted (U.S. Envi-

ronmental Protection Agency, 2014b). The current study is also somewhat limited in focusing on the climate impacts of biomass feedstock production from a purely biogeochemical perspective, ignoring potential biophysical impacts such as changes in surface albedo or evapotranspiration and water dynamics that are significant in some bioenergy production scenarios (Munoz et al., 2010; Georgescu et al., 2011; Cherubini et al., 2012; Caiazzo et al., 2014), and ignoring any potential broader impacts on other ecosystem services (Chamberlain & Miller, 2012). Additionally, future changes to atmospheric CO2 concentrations and climate are not considered here, though they may have large repercussions for landscape design (Bryan et al., 2010). The ability of current ecosystem models to accurately extrapolate to such future conditions is an active area of investigation (De Kauwe et al., 2013). Additional sensitivity analysis on these points would improve our understanding of bioenergy landscape performance potential, and highlight priorities for future research efforts.


This work was supported by a USDA/NIFA project 'Decision support tool for integrated biofuel greenhouse gas emission footprints' (grant # 2011-67009-30083), an NSF IGERT fellowship through the Multidisciplinary Approaches to Sustainable Bioenergy program at Colorado State University, and an NSF REU fellowship and graduate Chevron fellowship through the Colorado Center for Biorefining and Biofuels (C2B2). We would like to thank Jacqueline Marquez for her assistance in identifying and coding switchgrass field trial papers, Matthew Stermer for helping quality-controlling this dataset, Kendrick Killian and Ty Boyak for their advice on developing model automation and analysis code, Jill Schuler for her GIS contribution, Jeff Kent for help with DayCent growth submodel performance visualization, and Yao Zhang for his insights on crop model performance in dry climates.


110th Congress of the United States (2007) Energy Independence and Security Act of 2007.

Abodeely JM, Muth DJ, Koch JB, Bryden KM (2013) A model integration framework for assessing integrated landscape management strategies. In: Environmental Software Systems. Fostering Information Sharing, Vol 413 (eds Hrebícek J, Schimak G, Kubasek M, Rizzoli AE), pp. 121 128. Springer Berlin Heidelberg, Berlin, Heidelberg. Adler PR, Sanderson MA, Boateng AA, Weimer PJ, Jung H-JG (2006) Biomass yield and biofuel quality of switchgrass harvested in fall or spring. Agronomy Journal, 98, 1518 1525.

Adler PR, Del Grosso SJ, Parton WJ (2007) Life-cycle assessment of net greenhouse-

gas flux for bioenergy cropping systems. Ecological Applications, 17, 675 691. Anderson CJ, Anex RP, Arritt RW, Gelder BK, Khanal S, Herzmann DE, Gassman PW (2013) Regional climate impacts of a biofuels policy projection. Geophysical Research Letters, 40, 1217 1222. Anderson-Teixeira KJ, Duval BD, Long SP, DeLucia EH (2012) Biofuels on the landscape: is "land sharing" preferable to "land sparing"? Ecological Applications, 22, 2035 2048.

Anderson-Teixeira KJ, Masters MD, Black CK, Zeri M, Hussain MZ, Bernacchi CJ, DeLucia EH (2013) Altered belowground carbon cycling following land-use change to perennial bioenergy crops. Ecosystems, 16, 508 520.

Arundale RA, Dohleman FG, Voigt TB, Long SP (2014a) Nitrogen fertilization does significantly increase yields of stands of Miscanthus x giganteus and Panicum vir-gatum in multiyear trials in Illinois. BioEnergy Research, 7, 408 416.

Arundale RA, Dohleman FG, Heaton EA, Mcgrath JM, Voigt TB, Long SP (2014b) Yields of Miscanthus x giganteus and Panicum virgatum decline with stand age in the Midwestern USA. GCB Bioenergy, 6, 1 13.

Balasko JA, Smith D (1971) Influence of temperature and nitrogen fertilization on the growth and composition of switchgrass (Panicum virgatum L.) and timothy (Phleum pratense L.) at anthesis. Agronomy Journal, 63, 853 857.

Bandaru V, Izaurralde RC, Manowitz D, Link R, Zhang X, Post WM (2013) Soil carbon change and net energy associated with biofuel production on marginal lands: a regional modeling perspective. Journal of Environmental Quality, 42, 1802 1 814.

Behrman KD, Keitt TH, Kiniry JR (2014) Modeling differential growth in switchgrass cultivars across the Central and Southern Great Plains. BioEnergy Research, 7, 1165 1173.

Berdahl JD, Frank AB, Krupinsky JM, Carr PM, Hanson JD, Johnson HA (2005) Biomass yield, phenology, and survival of diverse switchgrass cultivars and experimental strains in western North Dakota. Agronomy Journal, 97, 549 555.

Bonin CL, Lal R (2014) Aboveground productivity and soil carbon storage of biofuel crops in Ohio. GCB Bioenergy, 6, 67 75.

Borowski S (2012) Idea for "fracking" came from Civil War battlefield. AAAS Mem-berCentral Available at: (accessed 17 November 2015).

Boyer CN, Roberts RK, English BC, Tyler DD, Larson JA, Mooney DF (2013) Effects of soil type and landscape on yield and profit maximizing nitrogen rates for switchgrass production. Biomass and Bioenergy, 48, 33 42.

Bryan BA, King D, Wang E (2010) Biofuels agriculture: landscape-scale trade-offs between fuel, economics, carbon, energy, food, and fiber. GCB Bioenergy, 2,330 345.

Cai X, Zhang X, Wang D (2011) Land availability for biofuel production. Environmental Science & Technology, 45, 334 339.

Caiazzo F, Malina R, Staples MD, Wolfe PJ, Yim SHL, Barrett SRH (2014) Quantifying the climate impacts of albedo changes due to biofuel production: a comparison with biogeochemical effects. Environmental Research Letters, 9, 024015.

Campbell EE (2015) Modeling soil organic matter: theory, development, and applications in bioenergy cropping systems. PhD Dissertation. Colorado State University, Fort Collins, CO, USA.

Campbell JE, Lobell DB, Genova RC, Field CB (2008) The global potential of bioen-ergy on abandoned agriculture lands. Environmental Science & Technology, 42, 5791 5794.

Casler MD (2012) Switchgrass breeding, genetics, and genomics. In: Green Energy and Technology (ed Monti A), pp. 29 53. Springer London, London.

Casler MD, Vogel KP, Taliaferro CM, Wynia RL (2004) Latitudinal adaptation of switchgrass populations. Crop Science, 44, 293 303.

Casler MD, Vogel KP, Taliaferro CM et al. (2007) Latitudinal and longitudinal adaptation of switchgrass populations. Crop Science, 47, 2249 2260.

Cassida KA, Muir JP, Hussey MA, Read JC, Venuto BC, Ocumpaugh WR (2005) Biomass yield and stand characteristics of switchgrass in South Central U.S. Environments. Crop Science, 45, 673 681.

Chamberlain JF, Miller SA (2012) Policy incentives for switchgrass production using valuation of non-market ecosystem services. Energy Policy, 48, 526 536.

Chamberlain JF, Miller SA, Frederick JR (2011) Using DAYCENT to quantify on-farm GHG emissions and N dynamics of land use conversion to N-managed switchgrass in the Southern U.S. Agriculture, Ecosystems & Environment, 141,332 341.

Cherubini F, Bright RM, Stromman AH (2012) Site-specific global warming potentials of biogenic CO2 for bioenergy: contributions from carbon fluxes and albedo dynamics. Environmental Research Letters, 7, 045902.

Crutzen PJ, Mosier AR, Smith KA, Winiwarter W (2008) N2O release from agro-biofuel production negates global warming reduction by replacing fossil fuels. Atmospheric Chemistry and Physics, 8, 389 395.

Davis SC, Parton WJ, Del Grosso SJ, Keough C, Marx E, Adler PR, DeLucia EH (2012) Impact of second-generation biofuel agriculture on greenhouse-gas emissions in the corn-growing regions of the US. Frontiers in Ecology and the Environment, 10, 69 74.

Davis SC, Boddey RM, Alves BJR et al. (2013) Management swing potential for bioenergy crops. GCB Bioenergy, 5, 623 638.

De Kauwe MG, Medlyn BE, Zaehle S et al. (2013) Forest water use and water use efficiency at elevated CO2: a model-data intercomparison at two contrasting temperate forest FACE sites. Global Change Biology, 19, 1759 1779.

Del Grosso SJ, Parton WJ, Mosier AR, Ojima DS, Kulmala AE, Phongpan S (2000) General model for N2O and N2 gas emissions from soils due to dentrification. Global Biogeochemical Cycles, 14, 1045 1060.

Del Grosso SJ, Parton WJ, Keough CA, Reyes-Fox M, Ahuja LR, Ma L (2011) Special features of the DayCent modeling package and additional procedures for parameterization, calibration, validation, and applications. In: Advances in Agricultural Systems Modeling (eds Ahuja LR, Ma L), pp. 155-176. American Society of Agronomy, Crop Science Society of America, Soil Science Society of America, Madison, WI.

o Di Nasso NN, Lasorella MV, Roncucci N, Bonari E (2015) Soil texture and crop management affect switchgrass (Panicum virgatum L.) productivity in the Mediterranean. Industrial Crops and Products, 65, 21 26.

Di Virgilio N, Monti A, Venturi G (2007) Spatial variability of switchgrass (Panicum virgatum L.) yield as related to soil parameters in a small field. Field Crops Research, 101, 232 239.

Dohleman FG, Heaton EA, Arundale RA, Long SP (2012) Seasonal dynamics of above- and below-ground biomass and nitrogen partitioning in Miscanthus x giganteus and Panicum virgatum across three growing seasons. GCB Bioenergy, 4, 534 544.

Duval BD, Anderson-Teixeira KJ, Davis SC, Keogh C, Long SP, Parton WJ, DeLucia EH (2013) Predicting greenhouse gas emissions and soil carbon from changing pasture to an energy crop. PLoS ONE, 8, e72019.

Egbendewe-Mondzozo A, Swinton SM, Izaurralde CR, Manowitz DH, Zhang X (2011) Biomass supply from alternative cellulosic crops and crop residues: a spatially explicit bioeconomic modeling approach. Biomass and Bioenergy, 35, 4636 4647.

Epstein HE, Lauenroth WK, Burke IC (1997) Effects of temperature and soil texture on ANPP in the U.S. Great Plains. Ecology, 78, 2628 2631.

Ernstrom DJ, Lytle D (1993) Enhanced soils information systems from advances in computer technology. Geoderma, 60, 327 341.

Fike JH, Parrish DJ, Wolf DD, Balasko JA, Green JT Jr, Rasnake M, Reynolds JH (2006) Switchgrass production for the upper southeastern USA: influence of cultivar and cutting frequency on biomass yields. Biomass and Bioenergy, 30, 207 213.

Follett RF, Vogel KP, Varvel GE, Mitchell RB, Kimble J (2012) Soil carbon sequestration by switchgrass and no-till maize grown for bioenergy. BioEnergy Research, 5, 866 875.

Forster P, Ramaswamy V, Artaxo P et al. (2007) Changes in atmospheric constituents and in radiative forcing. In: Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (eds Solomon S, Qin D, Manning M et al.). Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA.

Frank AB, Berdahl JD, Hanson JD, Liebig MA, Johnson HA (2004) Biomass and carbon partitioning in switchgrass. Crop Science, 44, 1391 1396.

Fuentes RG, Taliaferro CM (2002) Biomass yield stability of switchgrass cultivars. In: Trends in New Crop and New Uses (eds Janick J, Whipkey A), pp. 276 282. ASHS Press, Alexandria, Virginia.

Gelfand I, Sahajpal R, Zhang X, Izaurralde RC, Gross KL, Robertson GP (2013) Sustainable bioenergy production from marginal lands in the US Midwest. Nature, 493, 514 517.

Georgescu M, Lobell DB, Field CB (2011) Direct climate effects of perennial bioen-ergy crops in the United States. Proceedings of the National Academy of Sciences, 108, 4307 4312.

Hallgren W, Schlosser CA, Monier E, Kicklighter D, Sokolov A, Melillo J (2013) Climate impacts of a large-scale biofuels expansion. Geophysical Research Letters, 40, 1624 1630.

Heaton EA, Dohleman FG, Long SP (2008) Meeting US biofuel goals with less land: the potential of Miscanthus. Global Change Biology, 14, 2000 2014.

Helms D (1992) Readings in the History of the Soil Conservation Service. Soil Conservation Service, Washington, DC, pp. 60 73.

Hoben JP, Gehl RJ, Millar N, Grace PR, Robertson GP (2011) Nonlinear nitrous oxide (N2O) response to nitrogen fertilizer in on-farm corn crops of the US Midwest. Global Change Biology, 17, 1140 1152.

Holtsmark B (2015) Quantifying the global warming potential of CO2 emissions from wood fuels. GCB Bioenergy, 7, 195 206.

Hong CO, Owens VN, Schumacher T, Clay D, Osborne S, Lehman M, Schumacher J (2012) Nitrogen losses from switchgrass as affected by nitrogen fertilizer rate. In: Proceedings from Sun Grant National Conference: Science for Biomass Feedstock Production and Utilization, pp. 183 190. New Orleans, LA. Available at: (accessed 17 November 2015).

Hopkins AA, Vogel KP, Moore KJ, Johnson KD, Carlson IT (1995) Genotypic variability and genotype x environment Interactions among switchgrass accessions from the midwestern USA. Crop Science, 35, 565 571.

Hsu FH, Nelson CJ, Matches AG (1985) Temperature effects on seedling development of perennial warm-season forage grasses. Crop Science, 25, 249 255.

Hudiburg TW, Davis SC, Parton W, Delucia EH (2015) Bioenergy crop greenhouse gas mitigation potential under a range of management practices. GCB Bioenergy, 7, 366 374.

Jain AK, Khanna M, Erickson M, Huang H (2010) An integrated biogeochemical and economic analysis of bioenergy crops in the Midwestern United States. GCB

Bioenergy, 2, 217 234.

Jorgensen CJ, Struwe S, Elberling B (2012) Temporal trends in N2O flux dynamics in a Danish wetland effects of plant-mediated gas transport of N2O and O2 following changes in water level and soil mineral-N availability. Global Change Biology, 18, 210 222.

Kandel TP, Wu Y, Kakani VG (2013) Growth and yield responses of switchgrass eco-types to temperature. American Journal of Plant Sciences, 4, 1173 1180.

Kang S, Nair SS, Kline KL et al. (2014) Global simulation of bioenergy crop productivity: analytical framework and case study for switchgrass. GCB Bioenergy, 6,14 25.

Kansas Historical Society (2014) Dust clouds rolling over the prairies, Hugoton, Kansas. Kansas Memory.

Kering MK, Butler TJ, Biermacher JT, Guretzky JA (2012) Biomass yield and nutrient removal rates of perennial grasses under nitrogen fertilization. BioEnergy Research, 5, 61 70.

Kim S, Dale BE (2005) Environmental aspects of ethanol derived from no-tilled corn grain: nonrenewable energy consumption and greenhouse gas emissions. Biomass and Bioenergy, 28, 475 489.

Lane DR, Coffin DP, Lauenroth WK (1998) Effects of soil texture and precipitation on above-ground net primary productivity and vegetation structure across the Central Grassland region of the United States. Journal of Vegetation Science, 9, 239 250.

Lee J, Pedroso G, Linquist BA, Putnam D, van Kessel C, Six J (2012) Simulating switchgrass biomass production across ecoregions using the DAYCENT model. GCB Bioenergy, 4, 521 533.

Lesur C, Jeuffroy M-H, Makowski D et al. (2013) Modeling long-term yield trends of Miscanthus x giganteus using experimental data from across Europe. Field Crops Research, 149, 252 260.

Li Y, Fu X, Liu X et al. (2013) Spatial variability and distribution of N2O emissions from a tea field during the dry season in subtropical central China. Geoderma, 193-194, 1 12.

Liebig MA, Schmer MR, Vogel KP, Mitchell RB (2008) Soil carbon storage by switchgrass grown for bioenergy. BioEnergy Research, 1, 215 222.

Ma Z, Wood C, Bransby D (2001) Impact of row spacing, nitrogen rate, and time on carbon partitioning of switchgrass. Biomass and Bioenergy, 20, 413 419.

Mesinger F, DiMego G, Kalnay E et al. (2006) North American regional reanalysis. Bulletin of the American Meteorological Society, 87, 343 360.

Miguez FE, Maughan M, Bollero GA, Long SP (2012) Modeling spatial and dynamic variation in growth, yield, and yield stability of the bioenergy crops Miscant-hus x giganteus and Panicum virgatum across the conterminous United States. GCB Bioenergy, 4, 509 520.

Mooney DF, Roberts RK, English BC, Tyler DD, Larson JA (2009) Yield and breakeven price of "Alamo" switchgrass for biofuels in Tennessee. Agronomy Journal, 101, 1234 1242.

Muir JP, Sanderson MA, Ocumpaugh WR, Jones RM, Reed RL (2001) Biomass production of "Alamo" switchgrass in response to nitrogen, phosphorus, and row spacing. Agronomy Journal, 93, 896 901.

Mulkey VR, Owens VN, Lee DK (2006) Management of switchgrass-dominated Conservation Reserve Program lands for biomass production in South Dakota. Crop Science, 4б, 712 720.

Munñoz I, Campra P, Fernaández-Alba AR (2010) Including CO2-emission equivalence of changes in land surface albedo in life cycle assessment. Methodology and case study on greenhouse agriculture. The International Journal of Life Cycle Assessment, 15, 672 681.

Nair SS, Kang S, Zhang X et al. (2012) Bioenergy crop models: descriptions, data requirements, and future challenges. GCB Bioenergy, 4, 620 633.

Necpaálováa M, Anex RP, Fienen MN et al. (2015) Understanding the DayCent model: calibration, sensitivity, and identifiability through inverse modeling. Environmental Modelling & Software, бб, 110 130.

Nikiema P, Rothstein DE, Min D-H, Kapp CJ (2011) Nitrogen fertilization of switchgrass increases biomass yield and improves net greenhouse gas balance in northern Michigan, U.S.A. Biomass and Bioenergy, 35, 4356 4367.

Noy-Meir I (1973) Desert ecosystems: environment and producers. Annual Review of Ecology and Systematics, 4, 25 51.

Ogle SM, Breidt FJ, Easter M, Williams S, Killian K, Paustian K (2010) Scale and uncertainty in modeled soil organic carbon stock changes for US croplands using a process-based model. Global Change Biology, 16, 810 822.

Parkin TB (2008) Effect of sampling frequency on estimates of cumulative nitrous oxide emissions. Journal of Environmental Quality, 37, 1390 1395.

Parton WJ, Schimel DS, Cole CV, Ojima DS (1987) Analysis of factors controlling soil organic matter levels in Great Plains grasslands. Soil Science Society of America Journal, 51, 1173 1179.

Parton WJ, Hartman M, Ojima D, Schimel D (1998) DAYCENT and its land surface submodel: description and testing. Global and Planetary Change, 19, 35 48.

Parton WJ, Holland EA, Grosso SJD et al. (2001) Generalized model for NOx and N2O emissions from soils. Journal of Geophysical Research, 106, 17403 17419.

Pearson CH (2004) Pasture grass species evaluation at fruita 1995 2001. In: Colorado Forage Research 2003: Alfalfa, Irrigated Pastures, and Mountain Meadows (eds Brummer JE, Pearson CH), pp. 65 70. Agricultural Experiment Station and Cooperative Extension Technical Bulletin TB04-01, Colorado State University, Fort Collins, CO.

Pedroso GM, van Kessel C, Six J, Putnam DH, Linquist BA (2014) Productivity, 15N dynamics and water use efficiency in low- and high-input switchgrass systems. GCB Bioenergy, 6, 704 716.

Peplow M (2014) Cellulosic ethanol fights for life. Nature, 507, 152 153.

Persson T, Ortiz BV, Bransby DI, Wu W, Hoogenboom G (2011) Determining the impact of climate and soil variability on switchgrass (Panicum virgatum L.) production in the south-eastern USA; a simulation study. Biofuels, Bioproducts and Biorefining, 5, 505 518.

Pervez MS, Brown JF (2010) Mapping irrigated lands at 250-m scale by merging MODIS data and national agricultural statistics. Remote Sensing, 2, 2388 2412.

Pyörälä P, Peltola H, Strandman H, Antti K, Antti A, Jylhä K, Kellomäki S (2014) Effects of management on economic profitability of forest biomass production and carbon neutrality of bioenergy use in Norway spruce stands under the changing climate. BioEnergy Research, 7, 279 294.

Qin Z, Dunn JB, Kwon H, Mueller S, Wander MM (2015a) Soil carbon sequestration and land use change associated with biofuel production: empirical evidence. GCB Bioenergy, doi:10.1111/gcbb.12237.

Qin Z, Zhuang Q, Cai X (2015b) Bioenergy crop productivity and potential climate change mitigation from marginal lands in the United States: an ecosystem modeling perspective. GCB Bioenergy, 7, 1211 1221.

Reddy KR, Matcha SK, Singh SK, Brand D, Seepaul R (2008) Quantifying the effects of temperature and nitrogen on switchgrass growth and development. In: Presentation to the 38th Biological Systems Simulation Modeling Meeting Conference, Temple TX, USA. Available at (accessed 17 November 2015).

Robertson GP, Dale VH, Doering OC et al. (2008) Sustainable biofuels redux. Science, 322, 49 50.

Robertson GP, Hamilton SK, Del Grosso SJ, Parton WJ (2011) The biogeochemistry of bioenergy landscapes: carbon, nitrogen, and water considerations. Ecological Applications, 21, 1055 1067.

Robertson AD, Davies CA, Smith P, Dondini M, McNamara NP (2015) Modelling the carbon cycle of Miscanthus plantations: existing models and the potential for their improvement. GCB Bioenergy, 7, 405 421.

Roncucci N, o Di Nasso NN, Bonari E, Ragaglini G (2015) Influence of soil texture and crop management on the productivity of miscanthus (Miscanthus x giganteus Greef et Deu.) in the Mediterranean. GCB Bioenergy, 7, 998 1008.

Roth B, Finnan JM, Jones MB, Burke JI, Williams ML (2015) Are the benefits of yield responses to nitrogen fertilizer application in the bioenergy crop Miscant-hus x giganteus offset by increased soil emissions of nitrous oxide? GCB Bioen-ergy, 7, 145 152.

Sanderson MA (1992) Morphological development of switchgrass and kleingrass. Agronomy Journal, 84, 415 419.

Sanderson MA, West CP, Moore KJ, Stroup J, Moravec J (1997) Comparison of morphological development indexes for switchgrass and bermudagrass. Crop Science, 37, 871 878.

Saxton KE, Rawls WJ, Romberger JS, Papendick RI (1986) Estimating generalized soil-water characteristics from texture. Soil Science Society of America Journal, 50, 1031 1036.

Schmer MR, Vogel KP, Mitchell RB, Perrin RK (2008) Net energy of cellulosic etha-nol from switchgrass. Proceedings of the National Academy of Sciences, 105, 464 469.

Schmer MR, Liebig MA, Hendrickson JR, Tanaka DL, Phillips RL (2012) Growing season greenhouse gas flux from switchgrass in the northern great plains. Biomass and Bioenergy, 45, 315 319.

Searchinger T, Heimlich R, Houghton RA et al. (2008) Use of US croplands for biofuels increases greenhouse gases through emissions from land-use change. Science, 319, 1238 1240.

Searle SY, Malins CJ (2014) Will energy crop yields meet expectations? Biomass and Bioenergy, 65, 3 12.

Shcherbak I, Millar N, Robertson GP (2014) Global metaanalysis of the nonlinear response of soil nitrous oxide (N2O) emissions to fertilizer nitrogen. Proceedings of the National Academy of Sciences, 111, 9199 9204.

Sheehan J, Aden A, Paustian K, Killian K, Brenner J, Walsh M, Nelson R (2003) Energy and environmental aspects of using corn stover for fuel ethanol. Journal of Industrial Ecology, 7, 117 146.

Shield IF, Barraclough TJP, Riehe AB, Yates NE (2012) The yield response of the energy crops switchgrass and reed canary grass to fertiliser applications when grown on a low productivity sandy soil. Biomass and Bioenergy, 42, 86 96.

Smemo KA, Ostrom NE, Opdyke MR, Ostrom PH, Bohm S, Robertson GP (2011) Improving process-based estimates of N2O emissions from soil using temporally extensive chamber techniques and stable isotopes. Nutrient Cycling in Agroecosys-tems, 91, 145 154.

Smith J, Smith P, Addiscott T (1996) Quantitative methods to evaluate and compare Soil Organic Matter (SOM) models. In: Evaluation of Soil Organic Matter Models (eds Powlson DS, Smith P, Smith JU), pp. 181 199. Springer Berlin Heidelberg, Berlin, Heidelberg.

Smith CM, David MB, Mitchell CA, Masters MD, Anderson-Teixeira KJ, Bernacchi CJ, DeLucia EH (2013) Reduced nitrogen losses after conversion of row crop agriculture to perennial biofuel crops. Journal of Environmental Quality, 42, 219 228.

Staley TE, Stout WL, Jung GA (1991) Nitrogen use by tall fescue and switchgrass on acidic soils of varying water holding capacity. Agronomy Journal, 83, 732 738.

Swinton SM, Babcock BA, James LK, Bandaru V (2011) Higher US crop prices trigger little area expansion so marginal land for biofuel crops is limited. Energy Policy, 39, 5254 5258.

Thomas ARC, Bond AJ, Hiscock KM (2013) A multi-criteria based review of models that predict environmental impacts of land use-change for perennial energy crops on water, carbon and nitrogen cycling. GCB Bioenergy, 5, 227 242.

Tilman D, Hill J, Lehman C (2006) Carbon-negative biofuels from low-input high-diversity grassland biomass. Science, 314, 1598 1600.

Tilman D, Socolow R, Foley JA et al. (2009) Beneficial biofuels-the food, energy, and environment trilemma. Science, 325, 270 271.

U.S. Department of Agriculture FSA (2011) Program Fact Sheets. Available at: http:// pfs&newstype=prfactsheet&type=detail&item=pf_20110727_energ_en_bcap7.html (accessed 17 November 2015).

U.S. Department of Agriculture NASS (2014) Quick Stats Ad-hoc Query Tool. Available at: (accessed 17 November 2015).

U.S. Department of Agriculture NRCS. Web Soil Survey. Available at: (accessed 17 November 2015).

U.S. Department of Energy (2011) U.S. Billion-Ton Update: Biomass Supply for a Bioenergy and Bioproducts Industry (Leads Perlack RD, Stokes BJ), 227 pp. ORNL/TM-2011/224. Oak Ridge National Laboratory, Oak Ridge, TN.

U.S. Environmental Protection Agency (2014a) DRAFT Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990 2012. US Environmental Protection Agency, Washington, DC.

U.S. Environmental Protection Agency (2014b) Framework for Assessing Biogenic CO2 Emissions from Stationary Sources. Available at: climatechange/ghgemissions/biogenic-emissions.html (accessed 17 November 2015).

U.S. Geological Survey (2015) Federal Lands of the United States. Available at: (accessed 17 November 2015).

Vogel KP, Brejda JJ, Walters DT, Buxton DR (2002) Switchgrass biomass production in the Midwest USA. Agronomy Journal, 94, 413 420.

Wagle P, Kakani VG (2014) Environmental control of daytime net ecosystem exchange of carbon dioxide in switchgrass. Agriculture, Ecosystems & Environment, 186, 170 177.

Walker WE, Harremoes P, Rotmans J, van der Sluijs JP, van Asselt MBA, Janssen P, Krayer von Krauss MP (2003) Defining uncertainty: a conceptual basis for uncertainty management in model-based decision support. Integrated Assessment, 4, 5 17.

Walter A, Galdos MV, Scarpare FV et al. (2014) Brazilian sugarcane ethanol: developments so far and challenges for the future. Wiley Interdisciplinary Reviews: Energy and Environment, 3, 70 92.

Wang M, Han J, Dunn JB, Cai H, Elgowainy A (2012) Well-to-wheels energy use and greenhouse gas emissions of ethanol from corn, sugarcane and cellulosic biomass for US use. Environmental Research Letters, 7, 045905.

Wang C, Hunt ER Jr, Zhang L, Guo H (2013) Phenology-assisted classification of C3 and C4 grasses in the U.S. Great Plains and their climate dependency with MODIS time series. Remote Sensing of Environment, 138, 90 101.

van der Weerden T, Clough T, Styles T (2013) Using near-continuous measurements of N2O emission from urine-affected soil to guide manual gas sampling regimes. New Zealand Journal of Agricultural Research, 56, 60 76. Wickham JD, Stehman SV, Gass L, Dewitz J, Fry JA, Wade TG (2013) Accuracy assessment of NLCD 2006 land cover and impervious surface. Remote Sensing of Environment, 130, 294 304. Wilson DM, Heaton EA, Schulte LA et al. (2014) Establishment and short-term productivity of annual and perennial bioenergy crops across a landscape gradient. BioEnergy Research, 7, 885 898. Woli P (2012) Soil and variety effects on the energy and carbon balances of switchgrass-derived ethanol. Journal of Sustainable Bioenergy Systems, 02, 65 74.

Wright CK, Wimberly MC (2013) Recent land use change in the Western Corn Belt threatens grasslands and wetlands. Proceedings of the National Academy of Sciences, 110, 4134 4139.

Wullschleger SD, Davis EB, Borsuk ME, Gunderson CA, Lynd LR (2010) Biomass production in switchgrass across the United States: database description and determinants of yield. Agronomy Journal, 102, 1158 1168.

Xu B, Li F, Shan L, Ma Y, Ichizen N, Huang J (2006) Gas exchange, biomass partition, and water relationships of three grass seedlings under water stress. Weed Biology and Management, 6, 79 88.

Zhang X, Izaurralde RC, Manowitz D et al. (2010) An integrative modeling framework to evaluate the productivity and sustainability of biofuel crop production systems. GCB Bioenergy, 2, 258 277.

Supporting Information

Additional Supporting Information may be found in the online version of this article:

Appendix S1. Switchgrass parameterization and validation dataset development.

Appendix S2. Parameterization and validation detail.

Appendix S3. Landscape analysis detail.

Appendix S4. Dynamics of SOC changes and N2O emissions.

Figure S1. Climate range covered in the full switchgrass calibration & validation dataset as per the NARR database.

Figure S2. Classification of calibration and validation dataset field trial sites by soil surface texture and NRCS land capability class

(LCC) rating.

Figure S3. Scatter matrix of site location and climate parameters in the switchgrass parameterization & validation dataset.

Figure S4. Scatter matrix of soil parameters in the switchgrass parameterization & validation dataset.

Figure S5. Switchgrass green-up as a function of latitude only.

Figure S6. Heading dates estimated as a function of ecotype and site latitude.

Figure S7. Peak biomass as estimated to occur three weeks after heading, as compared to empirical observations for upland switchgrass at multiple sites.

Figure S8. Temperature stress response curve based on experimental data from a variety of sources.

Figure S9. Moisture stress response curve compared to normalized experimental data from Xu et al. (2006) and model defaults. Figure S10. Modeled and measured yield ranges binned by site average annual growing degree day accumulation (calculated for the range 12-30 °C).

Figure S11. Modeled and measured yield ranges binned by site average annual precipitation. Figure S12. Modeled and measured yield ranges binned by site NLCD land capability classification.

Figure S13. Example of an observed-modeled comparison for a study (Anderson-Teixeira et al., 2013) where time-resolved data is available for more detailed comparison, in this case for total aboveground carbon and belowground live carbon. Figure S14. Modeled vs. measured changes in soil organic carbon by study for all studies in the parameterization & validation dataset that include usable SOC measurements.

Figure S15. Modeled vs. measured cumulative annual emissions of nitrous oxide by study for all studies in the parameterization & validation dataset that include usable N2O measurements.

Figure S16. Histogram showing the area distribution of the polygons generated during the GIS intersect operation.

Figure S17. Annual fluxes, assuming moderate tilling during field preparation for initialswitchgrass crop establishment and every

8 years when the crop is replanted.