Biofuel feedstocks provide a renewable energy source that can reduce fossil fuel emissions; however, if produced on a large scale they can also impact local to regional water and carbon budgets. Simulation results for 2005–2014 from a regional weather model adapted to simulate the growth of two perennial grass biofuel feedstocks suggest that replacing at least half the current annual cropland with these grasses would increase water use efficiency and drive greater rainfall downwind of perturbed grid cells, but increased evapotranspiration (ET) might switch the Mississippi River basin from having a net warm-season surplus of water (precipitation minus ET) to a net deficit. While this scenario reduces land required for biofuel feedstock production relative to current use for maize grain ethanol production, it only offsets approximately one decade of projected anthropogenic warming and increased water vapor results in greater atmospheric heat content.
The potential to reduce net greenhouse gas emissions and diversify fuel sources has sparked considerable interest in locally grown fuels in the U.S., provoking substantial increases in maize grain ethanol production [Bagley et al., 2014]. However, the expanding use of maize for ethanol in the central U.S. may negatively impact food supplies [Hill et al., 2006] and may be a net carbon source from the amount of water input to the system [Chiu et al., 2009] or from the conversion of natural lands for agricultural expansion [Gibbs et al., 2008]. Perennial grasses such as miscanthus (Miscanthus × giganteus) and switchgrass (Panicum virgatum) have been proposed as “second generation” biofuel feedstocks to mitigate some of the problems with maize grain ethanol. Their high productivity could reduce net CO2 emissions [Stampfl et al., 2007], provide more energy per acre than maize [VanLoocke et al., 2010, 2012; Zhuang et al., 2013], increase below ground carbon storage [VanLoocke et al., 2012], and locally offset anthropogenic warming [Georgescu et al., 2011] while requiring fewer inputs of water [VanLoocke et al., 2012; Zhuang et al., 2013], fertilizer, and pesticides per unit output because of their higher water and nutrient use efficiency [Lewandowski et al., 2000].
Miscanthus, a perennial tropical grass, substantially increases evapotranspiration (ET) compared to current food and fiber crops in the region [Heaton et al., 2008; VanLoocke et al., 2010; Zhuang et al., 2013; Bagley et al., 2014; Chen et al., 2015], while switchgrass [VanLoocke et al., 2012], a native perennial grass, increases ET to a lesser extent. Widespread increases in ET upon conversion to biofuel grasses may enhance summer rainfall [DeAngelis et al., 2010; Harding and Snyder, 2012a, 2012b; Huber et al., 2014; Alter et al., 2015; Harding et al., 2015]; however, the lack of large-scale miscanthus and switchgrass cultivation necessitates the use of regional models to estimate hydrologic impacts that could possibly alter the perceived viability of biofuel grasses and limit their widespread adoption [VanLoocke et al., 2010]. In this study, we incorporate process-based algorithms that represent the dynamic response of miscanthus and switchgrass to climate in simulations of a regional climate model with varying coverage of these grasses (25%, 50%, and 75% of croplands replaced), thus representing two-way biosphere-climate interactions that have been unresolved until now.
2 Materials and Methods
2.1 WRF-CLM4crop-Biofuels Model
We developed the WRF-CLM4crop-Biofuels model by incorporating algorithms that explicitly simulate the growth and functioning of miscanthus and switchgrass [VanLoocke et al., 2010] into the dynamic crop module of WRF-CLM4crop [Lu et al., 2015], a fully coupled regional climate model with process-based dynamic crop growth. WRF-CLM4crop is a version of the Weather Research and Forecasting (WRF) model (version 3.3) [Skamarock et al., 2008] that has been coupled to the Community Land Model (CLM) version 4. WRF is a fully compressible, nonhydrostatic model that uses a terrain-following hydrostatic pressure coordinate and is typically used for operational and research applications. CLM4crop is a global dynamic land surface model, and the dynamic crop module contained within is based on the Integrated Biosphere Simulator, agricultural version (Agro-IBIS) [Kucharik, 2003; Kucharik and Brye, 2003; Levis et al., 2012].
WRF-CLM4crop calculates carbon and nitrogen in the leaf, stem, grain, and root of generic C3 and C4 crops based on carbon assimilation rates that are influenced by soil moisture, temperature, humidity, solar radiation, and atmospheric CO2 concentration. Shifts in carbon allocation among plant components occur as successive growth stages are reached throughout the growing season. Transitions between growth stages are determined through the accumulation of growing degree days (GDD). Climatological GDD totals were generated in a simulation of the outer model domain for 2004–2014 (see section 2.2), reducing the impact of simulated temperature biases on crop phenology. To represent switchgrass and miscanthus in WRF-CLM4crop, we added parameters and crop specific processes adapted from a biofuel version of Agro-IBIS [VanLoocke et al., 2010]. While miscanthus and switchgrass algorithms explicitly simulate the full carbon and nitrogen budgets, the representation of these biofuel grasses is more limited in WRF-CLM4crop-Biofuels (see supporting information Text S1).
2.2 WRF-CLM4crop-Biofuels Simulations
Model simulations were completed for 2004 to 2014, with 2004 discarded for model spin-up. Simulations were initialized using the National Centers for Environmental Prediction-Department of Energy (NCEP-DOE) Reanalysis II [Kanamitsu et al., 2002] and a two-way nested grid configuration with a 75 km resolution outer domain and a 15 km resolution inner domain (Figure S1). A control simulation with observed generic C3 and C4 croplands was completed to provide a baseline of climate conditions and maize grain ethanol production. Three biofuels simulations with 25%, 50%, and 75% of croplands replaced (C3 and C4 replaced equally) by biofuel grasses were completed. The CLM plant functional type (PFT) database was used to determine cropland and natural vegetation grid cells [Lawrence and Chase, 2007]. Generic C3 and C4 croplands were placed based on the ratio of observed [Monfreda et al., 2008] C3 crops (soybean, wheat, and cotton) to C4 crops (maize and sorghum) in each grid cell with a cropland PFT. Miscanthus and switchgrass were placed over the major agricultural regions of the Midwest and Great Plains, following proposed biofuel grass distributions from the United States Department of Energy [Department of Energy, 2006] (Figure 1a). Biofuel grasses were planted within the U.S. predominantly south of 45°N (with a linear decay in replacement percentages between 45°N and 46°N) based on cold temperature thresholds for miscanthus with 5 cm of residue [Kucharik et al., 2013]. Irrigation was applied to the irrigated portion of grid cells determined by a Moderate Resolution Imaging Spectroradiometer-derived irrigation data set of current irrigation (Figure S2) at the rate of typical center-pivot systems (0.0002 mm s−1) whenever plant water was low (based on a root stress function) or when leaf temperature exceeded 35°C [Lu et al., 2015]. Additional model configuration details are provided in the supporting information Text S2.
2.3 Data Analysis and Model Evaluation
We calculated the precipitation recycling ratio, defined as the amount of rainfall derived from ET within the basin divided by the total rainfall, for each model simulation scenario (see supporting information Text S3), and used these calculations to evaluate potential impacts to water recharge and availability upon land conversion for biofuel feedstock production. We also calculated the water use efficiency (WUE) of the control and replacement scenario simulations to quantify whether the perennial grasses use less water per unit of ethanol yield compared with current estimates of maize grown for ethanol production, as well as the land use efficiency to compare amount of land used to obtain the same amount of ethanol (see supporting information Text S4). Control simulation maize grain ethanol yields assume that 40% of all maize production contributes to ethanol production based on U.S. Department of Agriculture estimates for 2014 (http://www.ers.usda.gov; see supporting information Text S5). Although simulations were run continuously from January 2004 through December 2014, we focus our analysis on the May–September period as impacts to hydrology will mainly occur during the growing season.
Because land conversion for perennial grass production can alter local surface temperature as well as humidity, we evaluated both the change in simulated 2 m air temperature and the near-surface heat content. Heat content is defined as the equivalent potential temperature and accounts for heat associated with air temperature as well as the phase change of water (see supporting information Text S6).
We evaluated simulated precipitation and temperature from the control simulation with observations from the PRISM data set [Daly et al., 2002] and simulated precipitation recycling results with relevant studies. We evaluated biological variables from simulated grass production with observations from three sites in Illinois and compared them with previous modeling studies (see supporting information Text S7).
3.1 Model Evaluation
This version of CLM4crop has previously been found to compare reasonably well with observations of simulated leaf area index (LAI) and latent heat fluxes for C3 and C4 crops [Harding et al., 2015; Lu et al., 2015]. WRF-CLM4crop-Biofuels realistically simulates miscanthus (R2 = 0.90 and 0.92) and switchgrass (R2 = 0.94 and 0.93) LAI compared to field observations [Heaton et al., 2008] from three Illinois sites for 2005 and 2006 (Figure S3). Simulated miscanthus dry yield (~30 Mg hectare−1) is similar to observations [Heaton et al., 2004a]; however, switchgrass yield (~20 Mg hectare−1) is overestimated by 30–50% compared with observations [Heaton et al., 2004a; Wullschleger et al., 2010] but is similar to that simulated by Miguez et al. . Comparison of WRF to observations of latent heat flux over miscanthus fields demonstrates that WRF accurately simulates miscanthus ET (−2.6% error for June–August), with significant improvement compared to VanLoocke et al.  (Figure S4). Estimates of land and water use efficiencies for miscanthus are slightly higher than previous estimates [Zhuang et al., 2013], with larger biases for switchgrass. The simulated land use efficiency for maize grain ethanol closely matches previous estimates [Zhuang et al., 2013] (see supporting information Text S8), resulting in reasonable estimates of total maize grain ethanol production according to Renewable Fuels Association 2015 data sourced at http://www.ethanolrfa.org/pages/statistics (Figure S5).
WRF closely approximates the observed average temperature and precipitation over the Mississippi River basin, with a slight warm (0.50°C) and dry (−3.5%) bias (Figure S6). The intensity and frequency of observed precipitation is accurately simulated (Figure S7), but regional precipitation recycling is slightly greater compared to previous estimates (see supporting information Text S8). Because WRF-CLM4crop-Biofuels realistically simulates the warm-season climate over the Mississippi River basin and credibly represents miscanthus and switchgrass growth and productivity, biofuel replacement scenarios in this study provide plausible estimates of the impacts of second-generation biofuel feedstocks on the warm-season climate of the central U.S. Based on a comparison of model outputs with observations and previous publications, the largest uncertainty in model results in this study likely is found in estimates of total ethanol production from biofuel replacement scenarios due to potential overestimates of switchgrass yield and potential changes in future biomass to biofuel conversion rates.
3.2 Enhanced Regional Precipitation Recycling Causes Biofuel Grasses to Use Water More Efficiently
In this section, our primary focus is on the 50% replacement scenario, which would fuel all (113%) U.S. cars with E-85 fuel (Table S1). Widespread cultivation of biofuel grasses substantially enhances simulated ET throughout the region (Figure 1b) due to greater leaf area (Figure S8), a deeper root structure, and a longer growing period, especially for miscanthus. Evapotranspired water is drawn from infiltrated precipitation and applied irrigation, both of which vary between the control and scenario simulations. In the 50% replacement scenario, a significant percentage (67%) of the increased ET is offset by enhanced precipitation (Table 1) throughout the region (Figure 1c) because of enhanced convective available potential energy and precipitable water (Figure S9). However, greater increases in ET than precipitation shift the region from having a net surplus of water (i.e., more precipitation than ET) without biofuel grasses to having a net deficit (slightly more ET than precipitation). Much of the regional precipitation increase is derived from locally enhanced ET in the form of recycled precipitation, with most precipitation and recycled precipitation increases occurring in the Upper Midwest, downwind of replacement areas (Figure 1d). Overall, changes in ET explain 49% and 72% of the variability in the precipitation and recycled precipitation responses, respectively, to biofuel cultivation over the Mississippi River basin (Figure S10). Approximately half of the increased rainfall is from recycled precipitation, suggesting that significant increases in precipitation of nonlocal (i.e., outside the basin) moisture sources occur as well. Changes in recycled precipitation primarily occur during the summer months (Figure S11) when differences in LAI and ET are the greatest between current crop cover and the biofuel grass scenarios (not shown). In addition, recycled precipitation is enhanced more over miscanthus than switchgrass grid cells (Figure 2). Precipitation increases more with greater regional miscanthus and switchgrass coverage, with much of the Upper Midwest experiencing large rainfall increases in the 75% replacement scenario (Table 1 and Figure S12). Statistically significant increases in heavy rainfall events occur on average over the basin, but changes are much smaller than late 21st century estimates [Harding and Snyder, 2014] from anthropogenic climate change (Table S2).
Table 1. The 2005–2014 Averages Over Mississippi River Basin for Control Simulation and Differences Between Biofuel Replacement Simulations and Controla
25% Biofuels Change
50% Biofuels Change
75% Biofuels Change
aSignificance values for paired t tests are as follows: *p < 0.1, **p < 0.05, and ***p < 0.01. Statistical testing conducted on bdifference between control and biofuels simulations, cvalue from each biofuels simulation compared with control, and dnetWUE for each simulation compared with WUE for the same simulation.
+24.62 ± 3.08 (+5.6%)***
+43.68 ± 4.99 (+9.9%)***
+57.31 ± 8.61 (+13.0%)***
+13.61 ± 8.01 (3.1%)**
+29.28 ± 7.66 (+6.6%)**
+42.10 ± 13.51 (+9.5%)***
+8.00 ± 2.73 (+4.5%)***
+14.88 ± 2.90 (+8.4%)***
+21.64 ± 4.15 (+12.2%)***
Water use efficiency (WUE)c
L ethanol m−3
0.703 ± 0.051
+0.373 ± 0.042 (+51.5%)***
+0.422 ± 0.043 (+58.4%)***
+0.481 ± 0.036 (+66.8%)***
Net WUE ( )d
L ethanol m−3
0.703 ± 0.051
+0.398 ± 0.057 (+55.1%)
+0.477 ± 0.057 (+66.2%)
+0.583 ± 0.050 (+81.3%)***
Conversion from current crops to miscanthus and switchgrass drives increases in both ET and WUE. The WUE of biofuel grasses increases with greater biofuel coverage (Table 1) because enhanced low-level moisture from biofuel grasses (Figure S13) reduces the vapor pressure gradient, thereby limiting ET increases. Such two-way feedbacks have not previously been resolved by offline simulations that only capture one-way interactions. In addition, conventional estimates of WUE assume that any water used for ET is consumptive even though land use changes that enhance ET also enhance local rainfall [Sacks et al., 2008; DeAngelis et al., 2010; Harding and Snyder, 2012b; Huber et al., 2014; Harding et al., 2015], which effectively recovers some of the water lost to ET and more properly maintains mass balance. Therefore, changes in precipitation should be included when considering the full hydrologic impacts and WUE of biofuel grasses. When accounting for enhanced precipitation with biofuel grasses compared with maize production (see supporting information Text S4), biofuel grasses have even larger increases in Net Water Use Efficiency (netWUE; Table 1). NetWUE increases substantially in the aggressive biofuel replacement scenarios as the increased biofuel coverage further enhances precipitation, resulting in greater recovery of ET. For example, because of higher recycling ratios after conversion to biofuel grass production (Figure 2), in the 75% replacement scenario netWUE is 89% higher over miscanthus and 60% higher over switchgrass compared with maize.
Such improvements in WUE are achieved while producing no statistically significant change in irrigation water used on average over the Mississippi River basin or within the Ogallala Aquifer. While several miscanthus grid cells experience declines in irrigation water use (Figure S14) due to enhanced precipitation with greater biofuel coverage (Figure S12), numerous individual switchgrass grid cells within the Ogallala experience significant increases in irrigation water use. Upon conversion to biofuel crops, deeper roots significantly deplete deep soil moisture (Figure S15), similar to model results [Georgescu et al., 2011] that only include one-way feedbacks. Depletion of deep soil moisture increases with time in switchgrass but recovers after the first few years in miscanthus (Figure S16).
3.3 Biofuel Grasses Offset Some Summertime Warming From Climate Change
Much is known about the potential biogeochemical impacts of second-generation biofuels, as the increased belowground biomass, compared with maize [VanLoocke et al., 2012], can enhance carbon sequestration and mitigate anthropogenic warming. However, the potential biophysical effects of miscanthus and switchgrass on the regional climate are less well understood. Substantial warm-season regional cooling has been predicted as a significant benefit of the widespread cultivation of second-generation biofuel grasses, as the regional cooling could mitigate a large portion of warming from climate change. A warm-season WRF simulation with modified crop parameters that mimic some aspects of miscanthus predicted a 0.9°C cooling from latent cooling and a higher albedo [Georgescu et al., 2011], which locally offset several decades of anthropogenic warming.
Our simulation results, which include two-way biosphere-atmosphere interactions, show that the increased ET from biofuel grass cultivation reduces the partitioning of net radiation into sensible heating (Figure S17), causing significant near-surface cooling over much of the central U.S. (Figure S13). Over grid cells with biofuel grasses, the average May–September 2 m air temperature decreases by 0.63°C in the 50% replacement scenario (Table S3). However, this local cooling only offsets 10.1 years of the average 21st century warming trend in the high emissions (Representative Concentration Pathway, RCP8.5) scenario for grid cells with biofuel grasses (Figure 3 and Table S3). Over the Mississippi River basin as a whole, cooling in the 50% scenario offsets 7.7 years and 15.7 years of RCP8.5 and RCP4.5 warming, respectively (Table S3). Simulated cooling over miscanthus grid cells is much greater than switchgrass grid cells (Figure 2d and Table S3).
Greater near-surface humidity associated with biofuel grass cultivation could significantly reduce the benefits of warm-season cooling because higher summertime humidity increases near-surface heat content. Summertime cooling substantially reduces the number of hot days (≥90°F), but the frequency of oppressive humidity events (dew point ≥70°F) substantially increases (Figure S18). The equivalent potential temperature, an estimate of near-surface heat content that more effectively estimates vegetative impacts on atmospheric heating [Davey et al., 2006] (see supporting information Text S6), increases with biofuel grass cultivation (Figure S13 and Table S3). Increases in moisture more than offset the simulated cooling, as near-surface heat content in the 50% replacement simulation increases by 0.87°C over perturbed grid cells within the basin and by 0.99°C over miscanthus grid cells alone. The fact that biofuel grasses could increase near-surface heat content and the frequency of high humidity events suggests that limited benefits exist for regionally mitigating the warm-season effects of anthropogenic climate change through biofuel grass production.
Results from our novel simulations of a regional climate model with dynamic biofuel grasses showed that most of the increased ET from two biofuel grasses—miscanthus and switchgrass—was recovered over the Mississippi River basin as precipitation, with no significant change in irrigation water applied. Larger increases in recycled precipitation with greater biofuel grass coverage further enhanced the water use efficiency (WUE) of biofuel grasses compared with maize grain ethanol, but ET increased across the basin more than precipitation and resulted in a net loss of water to the basin. Simulated warm-season cooling from biofuel grass cultivation only offsets a decade of projected 21st century warming, and enhanced humidity from biofuel grasses increased atmospheric heat content, suggesting little to no regional mitigation of anthropogenic warming from biofuels.
While our results alleviate concerns regarding the impact of large-scale conversion to second-generation biofuel feedstocks on the warm-season hydrologic cycle of the central U.S., substantial problems remain when considering impacts on the global food supply and indirect effects on the global carbon cycle from nonlocal land use change. However, reductions in crop acreage and yield from ethanol production are much lower with miscanthus and switchgrass compared with maize grain ethanol. With global food demand expected to double by the middle of the century due to a growing population and dietary changes [Tilman et al., 2011], such substantial reductions in global food supply and the potential negative impacts on the global carbon budget from replacing lost food production pose significant challenges for biofuel production.
Support for this project was provided by the United States Department of Energy under Award DE-EE0004397 and U.S. Department of Agriculture's Agriculture and Food Research Initiative Competitive Grant 2011-68005-30411. This work was carried out in part using computing resources at the University of Minnesota Supercomputing Institute. The WRF model used herein can be acquired from the WRF home page online at http://www2.mmm.ucar.edu/wrf/users/download/get_source.html. The model was modified to include dynamic biofuel feedstock based on references cited within the text. All other data and programs used to replicate the results in this study not sourced from references cited within the text are available upon request from the corresponding author at email@example.com. We thank Mutlu Ozdogan for providing the fractional irrigation data set.