Nitrogen fertilization challenges the climate benefit of cellulosic biofuels

Cellulosic biofuels are intended to improve future energy and climate security. Nitrogen (N) fertilizer is commonly recommended to stimulate yields but can increase losses of the greenhouse gas nitrous oxide (N2O) and other forms of reactive N, including nitrate. We measured soil N2O emissions and nitrate leaching along a switchgrass (Panicum virgatum) high resolution N-fertilizer gradient for three years post-establishment. Results revealed an exponential increase in annual N2O emissions that each year became stronger (R2 > 0.9, P < 0.001) and deviated further from the fixed percentage assumed for IPCC Tier 1 emission factors. Concomitantly, switchgrass yields became less responsive each year to N fertilizer. Nitrate leaching (and calculated indirect N2O emissions) also increased exponentially in response to N inputs, but neither methane (CH4) uptake nor soil organic carbon changed detectably. Overall, N fertilizer inputs at rates greater than crop need curtailed the climate benefit of ethanol production almost two-fold, from a maximum mitigation capacity of −5.71 ± 0.22 Mg CO2e ha−1 yr−1 in switchgrass fertilized at 56 kg N ha−1 to only −2.97 ± 0.18 Mg CO2e ha−1 yr−1 in switchgrass fertilized at 196 kg N ha−1. Minimizing N fertilizer use will be an important strategy for fully realizing the climate benefits of cellulosic biofuel production.


Introduction
The global production of biofuels has increased dramatically in response to calls for greater energy security and climate change mitigation. In the US, legislation mandates production of 136 billion liters of ethanol biofuel by 2022 with a growing fraction from cellulosic sources (US Congress 2007). Cellulosic biofuels offer the potential for greater environmental benefits compared to grain based biofuels (Tilman et al 2006, Robertson et al 2008, 2011. Switchgrass (Panicum virgatum), a perennial grass native to North America, is among the most promising cellulosic biofuel crops due to its ability to grow on marginal and erosive lands, sequester soil carbon (Liska and Cassman 2008), reduce nitrogen (N) leaching (Smith et al 2013), and be grown with relatively little fossil fuel input (McLaughlin and Adams Kszos 2005).
Switchgrass is often considered an inherently N-thrifty plant, especially when managed for biomass production (Parrish and Fike 2005). Nevertheless, multiple studies have documented a productivity response to added N, with most reporting maximum yields at N rates between 56 and 202 kg N ha −1 (Vogel et al 2002, Mulkey et al 2006, Mooney et al 2009, Nikièma et al 2011. In a recent on-farm experiment (Schmer et al 2008), farmers fertilized switchgrass at rates up to 212 kg N ha −1 .
Although N fertilizer can increase biomass production, added N increases the greenhouse gas (GHG) contributions of biofuel production substantially: not only through the production, transportation, and distribution of the fertilizer itself, but also through fertilizer-induced microbial emissions of nitrous oxide (N 2 O), a GHG with a global warming potential ∼300 times that of carbon dioxide (CO 2 ) (Solomon et al 2007) and a major cause of stratospheric ozone depletion (Portmann et al 2012). Moreover, fertilizer N lost to the environment as nitrate ( ) -NO 3 leads to indirect emissions of N 2 O elsewhere in downstream surface waters (Beaulieu et al 2011) as well as to degraded water quality (Robertson and Vitousek 2009). Additionally, well-aerated soils are a globally significant sink for atmospheric methane (CH 4 ), and ammonium ( ) + NH 4 from N fertilizers can competitively inhibit microbial CH 4 oxidation in soils (Gulledge andSchimel 1998, Le Mer andRoger 2001).
Earlier studies have noted the potential for N fertilizer inputs to substantially reduce and even eliminate the climate benefit of food crops grown for biofuels (Crutzen et al 2007, Mosier et al 2009, Smith et al 2012. While Erisman et al (2010) further noted that this is unlikely to be the case for purpose-grown cellulosic crops because of their lower N fertilizer needs, perennial nature, and higher C:N ratios in harvested biomass, this has not yet been empirically tested.
Here we test the potential for N fertilization to significantly reduce the climate change mitigation benefit of cellulosic biofuels. We present results from a 3 yr experiment to investigate direct and indirect N 2 O emissions, CH 4 uptake, -NO 3 leaching, soil organic carbon (SOC) accumulation, and biomass production in recently established switchgrass under eight different N fertilizer rates. Our analysis allows an evaluation of the impact of N fertilization on the net GHG balance of switchgrass grown as a cellulosic biofuel feedstock.

Methods
The experiment was conducted at a site in southwest Michigan USA, in the northeastern portion of the US Corn Belt. The Switchgrass N Rate Experiment is part of the Great Lakes Bioenergy Research Center (GLBRC) and located at the Kellogg Biological Station Long-term Ecological Research Site (www.lter.kbs. msu.edu; 42°23′ N, 85°22′ W, elevation 284 m asl). Precipitation averages 1005 mm yr −1 with an average snowfall of ∼1.3 m. Mean annual temperature is 10.1°C ranging from a monthly mean of −3.8°C in January to 22.9°C in July (NCDC 2013). Soils are mesic Typic Hapludalfs of Kalamazoo loam developed on glacial outwash (Robertson and Hamilton 2015). Prior to establishing the experiments, soil pH (0-25 cm depth) was 7.47±0.04 (mean±standard error, n=12 plots), bulk density (BD) was 1.24±0.04 g cm −3 , total N was 1.25±0.09 g kg −1 soil, and SOC was 10.2±0.74 g kg −1 soil (http:// data.sustainability.glbrc.org/). Switchgrass (variety Cave-in-Rock) was planted at a seeding rate of 7.84 kg ha −1 on 11 July, 2008, after tillage to a depth of 25 cm. Plots were established on land that had been in alfalfa, corn, and occasional soybean production for preceding decades. Eight fertilization treatments (0, 28, 56, 84, 112, 140, 168, and 196 kg N ha −1 ) were established in switchgrass plots (4.5×6 m) arranged in a randomized complete block design with four replicate blocks, for a total of 32 experimental plots. Nitrogen fertilizer was applied once per year in 2009-2011: granular 46% urea was broadcasted on 17 June, 2009, and liquid 28% urea ammonium nitrate was sprayed on 10 May, 2010, and 16 May, 2011. Biomass was harvested in late fall annually on each plot using a John Deere 7330 tractor with a plot harvester (Wintersteiger Inc., Salt Lake City, UT) and HarvestMaster HM800 Plot Harvest Data System (Juniper Systems Inc., Logan, UT). Dry matter percentage was determined by oven-drying subsamples from harvested plots at 50°C until a constant weight. Harvest height was ∼10 cm.
2.1. N 2 O and CH 4 sampling N 2 O and CH 4 fluxes were measured using a static chamber-gas chromatography approach (Ruan and Robertson 2013) from May to December in 2009-2011. We measured fluxes 2-3 times per week during the growing season to capture the temporal dynamics of N 2 O and CH 4 fluxes as influenced by fertilization and precipitation, and then measured fluxes every 2 weeks after mid-September. A vented chamber (28 cm diameter×26 cm height) equipped with a detachable lid and septum was installed in each treatment plot for a total of 32 chambers. Chamber bases were inserted into the soil ∼5 cm for the duration of the study. Vegetation inside (but not surrounding) chambers was clipped to maintain plant heights lower than chamber heights. During flux samplings, chambers were tightly sealed with the lid and then headspace gas samples were collected four times with a 10 ml syringe at approximately 15 min intervals. Samples were stored over-pressurized in 5.6 ml glass vials (Labco Ltd, High Wycombe, UK). Gases were analyzed within three days by gas chromatography (Hewlett Packard 5890 Series II, Rolling Meadows, IL, USA). Gases were separated on a Porapak Q column (1.8 m, 80/100 mesh) at 80°C; CH 4 was analyzed with a flame ionization detector at 300°C and N 2 O was analyzed with a 63 Ni electron capture detector at 350°C.

Soil water-filled pore space (WFPS %), inorganic N, and NO 3 − leaching
At each gas sampling event we measured soil temperature, gravimetric water content, and inorganic N ( + NH 4 and ) -NO 3 concentrations at 0-25 cm depth. Soil gravimetric water content (GWC, g water g −1 dry soil) was determined by oven-drying soil at 60°C for 48 h until constant mass. Soil BD was measured three times during each growing season using a fixedvolume soil core (123 cm 3 ) for each treatment plot. WFPS% was calculated as where soil porosity=1−[BD (g cm −3 )/particle density (g cm −3 )]. Soil particle density was assumed to be the standard 2.65 g cm −3 . For measuring + NH 4 and -NO , 3 three 10 g soil samples (4 mm sieved) were extracted with 100 ml of 1 M KCl. Filtrates from soil extracts were analyzed colorimetrically on a Flow Solution IV autoanalyzer (OI Analytical, College Station, TX, USA).
Nitrate leaching below the root zone (1.0 m depth) was determined by measuring concentrations in soil pore water and then multiplying concentrations by downward water percolation (drainage) from the overlying soil. The study site has no detectable overland runoff because of its highly permeable soils. Soil pore water was sampled at weekly to fortnightly intervals (except when the ground was frozen) using lowtension porous ceramic cup samplers (Eijkelkamp Agrisearch Equipment, California, USA) installed at a 45°angle from the soil surface. The collected and filtered (1 μm nominal pore size; Pall A/E) water samples were analyzed for -NO 3 using a Dionex 600 ion chromatograph. Previous work at this site (Syswerda et al 2012) has shown that -NO 3 dominates N leaching with negligible leaching of + NH 4 or dissolved organic N.
Percolation of water from the root zone was modeled at a daily time step using the systems approach for land use sustainability model well-calibrated for KBS soils (Basso and Ritchie 2015), which accounts for management practices, water balance, soil organic matter change, nitrogen and phosphorus dynamics, heat balance, and plant growth and development. The soil water balance module is based on CERES models (Ritchie et al 1998) with revisions for infiltration, soil water export (Suleiman and Ritchie 2004), evaporation (Suleiman and Ritchie 2003), and runoff. Daily leaching losses of nitrate were estimated from modeled water percolation plus linear interpolation of the measured nitrate concentrations.

SOC sampling
One intact soil core (7.6 cm diameter×100 cm depth) was taken from each of the 32 experimental plots in June 2008 and May 2013 with a hydraulic sampler (Geoprobe model 540MT, Salina, KS). Each core was then cut into three profile segments: 0-25 cm (to represent the Ap layer), 25-50 cm (E layer), and 50-100 cm (Bt layer) (Syswerda et al 2011). Each segment was sieved (4 mm), oven-dried, and weighed for BD. Dry soil samples were then finely ground in a roller mill and three 10 mg samples were analyzed for C using a Costech elemental combustion system (Costech Analytical Technologies, Valencia, California).

Net GHG balance
To estimate the global warming impact (GWI) of GHG fluxes, we multiplied fluxes of CH 4 and N 2 O by their 100 yr horizon global warming potential factors of 25 and 298, respectively, to yield CO 2 equivalents (CO 2 e) (Solomon et al 2007).
We assumed that all CO 2 taken up from the atmosphere as net primary production by switchgrass was stored in harvested biomass and SOC. SOC change in CO 2 e was calculated as the product of the difference of SOC (Mg ha −1 yr −1 ) over the 4 yr study and the conversion factor of C to CO 2 (44/12).
Fossil fuel offset credit (Mg CO 2 e ha −1 yr −1 ) is defined as the avoided CO 2 emissions due to the displacement of fossil fuel use by biofuels during production, transportation, distribution, combustion, and coproducts allocation (Plevin 2009). Avoided CO 2 e emissions were calculated from a comparison of life cycle analyses of petroleum gasoline versus ethanol from switchgrass. Gasoline releases 94 g CO 2 e per MJ petroleum gasoline produced, distributed, and combusted (Farrell et al 2006, Wang et al 2012. Net CO 2 e emissions per MJ of switchgrass ethanol were calculated as the product of net CO 2 e emissions (Mg CO 2 e ha −1 ) from ethanol production, transportation, distribution and combustion and the total energy equivalent of biomass yield (MJ ha −1 ). Net CO 2 e emissions were calculated using the GREET model (Huo et al 2009) to calculate fossil fuel offset credits for the fossil fuel CO 2 emissions that would be displaced by the production of both ethanol and biorefinery coproducts (Farrell et al 2006, Gelfand et al 2013, with all farming inputs equal to 0. Farming inputs were calculated separately using actual values from the study site as presented in table S3. Total energy equivalent (MJ ha −1 ) was calculated as the product of harvestable dry-weight biomass (Mg ha −1 ), biorefinery ethanol yield (380 l Mg −1 biomass) (Schmer et al 2008, Gelfand et al 2011 and ethanol energy content (21.1 MJ l −1 ; low heating value) (Gelfand et al 2011(Gelfand et al , 2013. Finally, the fossil fuel offset credit (Mg CO 2 e ha −1 ) was calculated as the product of the CO 2 e difference from life cycle analyses of petroleum gasoline and ethanol from switchgrass (g CO 2 e MJ −1 ) and the ethanol energy content.

Data analysis
Cumulative fluxes of gases over annual periods were calculated by linear interpolation of daily fluxes between sample days. Data were analyzed using the PROC MIXED procedure in SAS 9.2 (SAS Institute, Cary, NC, USA). Treatment means were compared for significance using t-tests at α=0.05 level. The relationships between daily N 2 O emissions and environmental factors such as soil temperature, soil moisture, and soil total N were assessed by multiple linear regressions (stepwise) using PROC REG. Quadratic-plateau curves for switchgrass yields versus N fertilization rates were calculated using PROC NLIN, with switchgrass yields at successive fertilizer rates weighted by the inverse of its rank order along the N gradient (1/1 to 1/8). To determine the relationship between annual N 2 O emissions or leached nitrogen and N fertilization rate we performed exponential regression using PROC NLIN and linear regression using PRPC REG. Likelihood ratio-based R 2 -values, the Akaike information criterion (AIC) and the Bayesian information criterion (BIC), were calculated for both linear and exponential models. Higher values of R 2 and lower values of AIC and BIC indicated better models. We observed an exponential increase in annual N 2 O emissions with increasing N fertilization rates in each year (figure 2). The AIC and the BIC values were consistently lower for the exponential model for each study year and as well for all 3 years together: for individual years AIC and BIC values for the linear model were 52%-121% and 76%-173% higher than those for the exponential model, respectively (table S1). Additionally, R 2 values for the exponential model (0.90-0.94) were consistently higher than were R 2 values for the linear model (0.84-0.88) (table S1).

Results and discussion
From 2009 to 2011, mean daily N 2 O emissions ranged from 1.28±0.14 g N ha −1 d −1 in the low (0 kg N ha −1 ) fertilization treatment to 25.8±1.9 g N ha −1 d −1 in the high (196 kg N ha −1 ) fertilization treatment (figure S1). The maximum daily N 2 O emission was 270±25 g N ha −1 d −1 in the highest N treatment and the minimum daily emission was undetectable in treatments that received less than 56 kg N ha −1 yr −1 . Most of the fertilizer-associated N 2 O emissions occurred within 40 days following fertilization, coincident with soil wetting by rainfall. N 2 O emissions were strongly correlated with soil inorganic N concentrations (mg N kg −1 ) and % soil WFPS (N 2 O emission=−34.8+0.83×inorganic N+81.9×WFPS, R 2 =0.48, n=2112, P<0.001). At all soil inorganic N levels, N 2 O emisisons were highly dependent on WFPS (figure S2).
Modeled soil water drainage was 275, 399 and 515 mm yr −1 in 2009, 2010 and 2011, respectively, representing 37%, 39% and 46% of annual precipitation. Annual -NO 3 leaching rates ranged from 2.65±1.29 kg N ha −1 yr −1 for unfertilized switchgrass to 56.0±2.7 kg N ha −1 yr −1 for switchgrass fertilized at 196 kg N ha −1 , and also increased exponentially (table S2) in response to increasing N inputs, with no significant difference in the increase among years ( figure 3). The exponential increases in N 2 O emissions and -NO 3 leaching are likely due to surplus soil N at levels above which plant N demands are met, resulting in more available N for the nitrifiers and denitrifiers that produce N 2 O, as well as for leaching. The exponential increase for N 2 O also implies that the N 2 O emission factor used for most national GHG inventories (De Klein et al 2006) varies with N fertilizer rate, in agreement with other recent studies for annual crops but yet untested for perennial crops (Shcherbak et al 2014). In this study the emission factor increased from 0.6% to 2.1% across the range of added N ( figure 4) year N 2 O emissions by 30% at lower levels of N fertilizer and up to 107% at higher levels.
Mean daily CH 4 uptake rates ranged from −1.49±0.31 to −0.82±0.27 g CH 4 -C ha −1 d −1 across all eight fertilizer levels ( figure S3). There were no significant N treatment differences detected (P>0.1), although mean CH 4 uptake rates in highly fertilized soils (>56 kg N added ha −1 ) were only 55%-74% of those in the unfertilized treatment. We also found no significant SOC accumulation in any of our N treatments over the 3 year study period ( figure S4). Likely this is due to lost soil C on conversion of the field to switchgrass in 2008, although spatial variability makes it difficult to detect SOC change in fewer than  10 years in many soils , including ours (Syswerda et al 2011).
We combined our field measurements with published carbon costs for agricultural inputs (Robertson et al 2000, Farrell et al 2006, Schmer et al 2008, Gelfand et al 2011 to estimate overall GWI in units of CO 2 equivalents (CO 2 e) for each of our N treatments (table S3). Measurements of -NO 3 loss allow us to include a major portion of indirect N 2 O production, missing from most empirical GWI assessments. Estimated indirect N 2 O emissions from the loss of leached -NO 3 ranged from 22.1±7.8 for unfertilized switchgrass to 157±12.3 kg CO 2 e ha −1 yr −1 for switchgrass fertilized at 196 kg N ha −1 . Calculated 3 year averages of fossil fuel offsets for our eight N treatments ranged from −4.84±0.12 Mg CO 2 e ha −1 yr -1 in our  unfertilized treatment to −6.42±0.38 Mg CO 2 e ha −1 yr −1 in our treatment with the highest yield per unit N added (56 kg N ha −1 ) ( figure 5(a)).
Including all GHG sources and credits in the GWI analysis, each N treatment shows net climate change mitigation (i.e., negative net CO 2 e), with maximum net mitigation capacities as high as −5.71±0.22 Mg CO 2 e ha −1 yr −1 in the treatment fertilized at 56 kg N ha −1 ( figure 5(b)). However, at fertilization rates above 56 kg N ha −1 net mitigation decreased monotonically with each increment of added N to only −2.97±0.18 Mg CO 2 e ha −1 yr −1 for the 196 kg N ha −1 treatment ( figure 5(b)).
Greater N 2 O emissions in N fertilized compared to non-fertilized switchgrass has been noted in Nova Scotia Canada ( (2014) included three fertilizer rates, and while N 2 O responses were in some site years consistent with an exponential response, three rates are insufficient to statistically test for nonlinearity. We are not familiar with any N leaching studies that have tested the effects of fertilizer rates on nitrate loss in switchgrass or any other perennial biofuel crop.
That the mitigation potential of switchgrass fertilized at high N rates is only about half of its mitigation potential at yield-optimizing N rates points to a significant challenge for realizing the environmental potential of cellulosic biofuels. Knowledge of and careful management for crop N needs appear to be crucial. In many cases, such as for the maturing switchgrass crops in this study, fertilizer needs may be close to nil: some varieties of switchgrass are known to be unresponsive to fertilizer N (Christian et al 2002), presumably because of a high N use efficiency and/or the presence of other N acquisition mechanisms, possibly including biological N fixation.

Conclusions
Breeding for low N needs, and then fertilizing only as needed-if at all-to meet these needs will be an important strategy for meeting the full climate mitigation benefits of cellulosic biofuels. In the meantime, incentives to grow N-conserving crop varieties and to apply as little fertilizer N as necessary will be needed to meet the climate benefit claims of this emerging industry (Robertson et al 2008). Incentives to reduce N fertilizer use would have an additional advantage of reducing unnecessary N pollution of ground-and surface waters and lowering the cost of biofuel crop production.