Geophysical Research Letters
  • Open Access

Regional climate impacts of a biofuels policy projection


Corresponding author: Christopher J. Anderson, 2104 Agronomy Hall, Ames, IA, 50011–1010. (


[1] The potential for regional climate change arising from adoption of policies to increase production of biofuel feedstock is explored using a regional climate model. Two simulations are performed using the same atmospheric forcing data for the period 1979–2004, one with present-day land use and monthly phenology and the other with land use specified from an agro-economic prediction of energy crop distribution and monthly phenology consistent with this land use change. In Kansas and Oklahoma, where the agro-economic model predicts 15-30% conversion to switchgrass, the regional climate model simulates locally lower temperature (especially in spring), slightly higher relative humidity in spring and slightly lower relative humidity in summer, and summer depletion of soil moisture. This shows the potential for climate impacts of biofuel policies and raises the question of whether soil water depletion may limit biomass crop productivity in agricultural areas that are responsive to the policies. We recommend the use of agronomic models to evaluate the possibility that soil moisture depletion could reduce productivity of biomass crops in this region. We conclude, therefore, that agro-economic and climate models should be used iteratively to examine an ensemble of agricultural land use and climate scenarios, thereby reducing the possibility of unforeseen consequences from rapid changes in agricultural production systems.

1 Introduction

[2] Assessments of enacted biofuel production targets predict large-scale production of dedicated energy crops in the central United States [U.S. Department of Energy, 2011]. The land surface of the central U.S. is a driver of its climate [Durre et al., 2000; Bonan, 2001; Pal and Eltahir, 2003; Koster et al., 2004; Pielke et al., 2007] through the terrestrial-atmospheric hydrological cycle [Georgescu et al., 2011, Vanloocke et al., 2010, Mishra et al., 2010]. Indeed, land use land cover (LULC) change has likely attenuated the region's recent surface warming trend [Bonan, 2001; Twine et al., 2004, Diffenbaugh, 2009]. The question then arises whether future LULC change in support of enacted biofuel policies may induce similar changes.

[3] We examine the hypothesis that LULC change necessary to support the Energy Independence and Security Act of 2007 (EISA) would alter regional climate. To our knowledge this is the first evaluation of regional climate impact resulting from enacted biofuel policy. The Department of Energy “Billion-Ton Study” [Perlack et al., 2005] and its 2011 update [U.S. Department of Energy, 2011] created scenarios of U. S. biofuel feedstock production that would displace approximately 30% of the present U. S. petroleum consumption. In the 2011 update, the Policy Analysis System (POLYSYS) [Ray and Moriak, 1976; Huang et al., 1988; Ugarte and Ray, 2000] predicted greatest conversion to switchgrass in the Great Plains: 20-30% in Kansas and 30-45% northern and northwest Oklahoma [Dicks et al., 2009]. We compare the land-atmosphere interaction under current land use with POLYSYS-generated land use for 2022 (the first year of full EISA implementation) using current climate conditions.

2 Methodology

2.1 Regional Climate Model

[4] Regional climate simulations were produced using Weather Research Forecasting model version 3.1.1 (WRF) [Skamarock and Klemp, 2008; Skamarock et al., 2008]. The model configuration is listed in Table S1. The simulation domain (Figure S1) covers the continental U.S. at a grid spacing of 24 km and is positioned to ensure low-level jet dynamics are not unduly influenced by lateral boundary conditions [Seth and Giorgi, 1998]. WRF is coupled to the Noah land surface model representing the interaction of soil and vegetation with the atmosphere [Ek et al., 2003]. Noah uses satellite-derived estimates of bulk vegetation parameters at each grid cell that are representative of land cover mixtures (see Table 1 for relevant parameters) rather than fractional coverage of vegetation classes.

Table 1. Values of land use parameters in the Noah land surface model as coupled within the WRF regional climate model. Parameters beneath the horizontal line are new land use categories used in the switchgrass scenario. Meanings of the parameters are listed below the table
  1. amin = minimum albedo (dimensionless)

  2. amax = maximum albedo (dimensionless)

  3. z0max = maximum aerodynamic roughness length (m)

  4. z0min = maximum aerodynamic roughness length (m)

  5. sh = maximum areal fraction of green vegetation (dimensionless)

  6. rt = number of root layers (dimensionless)

  7. rs = minimum canopy resistance (s m-1)

  8. rgl = sensitivity factor of canopy resistance due to solar radiation

Dryland crop0.
Irrigated crop0.
Mixed dryland / irrigated0.
Switchgrass /Grassland0.
Switchgrass /Cropland0.
Grassland /Switchgrass0.
Cropland /Switchgrass0.

[5] Transpiration is computed in Noah with a canopy resistance model. Resistance has inverse dependence on leaf area index (LAI), temperature, vapor pressure deficit, and solar radiation, using vegetation dependent minimum canopy resistance, rcmin, as a proportionality constant. Phenology (temporal evolution of LAI and albedo) is specified using satellite measurements of vegetation fraction for 1991–1995. This dataset has 0.144° spacing and a monthly interval. The monthly vegetation fraction (assumed to be valid at the 15th of each month) [Gutman and Ignatov, 1998; Csiszar and Gutman, 1999] is interpolated linearly in time to each climate model time step. LAI and albedo are computed as weighted average of minimum and maximum values, using time interpolated vegetation fraction. Finally, transpiration has threshold dependence on soil moisture specified by the following factor:

display math(1)

[6] where SMi is the volumetric soil moisture of soil layer i (i = 1,4), WP is the wilting point soil moisture, and RP is the reference point soil moisture (i.e., field capacity). As ((1)) approaches zero soil resistance to transpiration approaches infinity, preventing soil water decrease below WP.

[7] Noah has known limitations related to soil moisture. Accurate soil moisture simulation is reported for the central United States where soil is deep and vegetation is dense [Mitchell et al., 2004; Fan et al., 2006]. However, in the sparsely vegetated Great Plains, Noah overestimated latent heat flux and produced low soil moisture bias [Chen et al., 2007]. Guidance is mixed on optimal Noah parameters [LeMone et al., 2007; Alfieri et al., 2008]. Noah performance is least impacted by variability of soil moisture initial state and vegetation fraction and most impacted by parameter settings for the Zilitinkevich coefficient [LeMone et al., 2007] and rcmin [Alfieri et al., 2008]. Default Noah settings for rcmin produce results more consistent with observations of cropland and forests than grassland and shrubland [Kumar et al., 2011].

[8] Noah implicitly models some effects of irrigated vegetation through satellite-derived vegetation fraction and vegetation parameter settings for irrigated and non-irrigated land. This does not include, however, water removal from aquifers and water added to the surface. This crude treatment of irrigation did not impact our conclusions for reasons explained in Section 2.2.

2.2 Experimental Design

[9] Two simulations were performed for the period 1979–2004, one with the standard land use categories and monthly phenology in WRF-Noah and the other with identical atmospheric forcing data but alternative land use and monthly phenology. The NCEP-DOE Reanalysis-II (R2; Kanamitsu et al., 2002) data were used as initial (including soil moisture) and boundary conditions. The first two years (1979, 1980) of each simulation were discarded to allow for adjustment of the land surface with the atmosphere.

Control Simulation

The control simulation used default settings of land use and vegetation parameters. These include 24 vegetation classes, a vegetation parameter table, and satellite-based (1991–1995) monthly vegetation fraction from which LAI and albedo are derived.

Switchgrass Simulation

The switchgrass simulation included changes to land use, vegetation parameters, and vegetation phenology. POLYSYS switchgrass simulation for 2022 was the basis for the spatial distribution of LULC in the climate simulation. The POLYSYS simulation contained county-level switchgrass and crop production estimates for 2009 through 2022 using a $60/dry ton farmgate switchgrass price and the Billion-Ton Study baseline assumptions, including an extension to 2022 of the USDA 10-year yield forecast for major food and forage crops. An area weighting method was used to resample POLYSYS county-level estimates of switchgrass conversion to the WRF grid. The switchgrass conversion fraction for each grid cell is shown in Figure S1.

[10] The POLYSYS scenarios assumed land converted to switchgrass is non-irrigated, because irrigated crops are high value and would not be displaced by a lower value biofuel feedstock crop. Additionally, expansion of irrigation for switchgrass production would add unsupportable production cost. Thus, land use conversion for switchgrass production was emulated in WRF by creating four new vegetation classes (Table 1) based upon Noah dryland categories. These classes were Switchgrass/Grassland Mosaic (control simulation Grassland with switchgrass conversion fraction > 0.3), Switchgrass/Cropland Mosaic (control simulation Dryland Crop with switchgrass conversion fraction > 0.3), Grassland/Switchgrass Mosaic (control simulation Grassland with switchgrass conversion fraction < 0.3), and Cropland/Switchgrass Mosaic (control simulation Dryland Crop Mosaic with switchgrass conversion fraction < 0.3). Figure 1 shows the spatial pattern of new land use categories.

Figure 1.

Default land use categories in WRFv3.1.1 and spatial distribution of new land use categories for switchgrass simulation. New land use categories that have switchgrass fraction ≥0.3 are labled Switchgrass/Grassland and Switchgrass/Cropland categories; whereas, Grassland/Switchgrass and Cropland/Switchgrass categories have switchgrass fraction ≤0.3. The box indicates the region over which spatial averages are computed for analysis of water balance components.

[11] Phenology for switchgrass vegetation classes was specified by adjusting the satellite-based monthly vegetation fraction. Vegetation fraction was increased during February-October to emulate earlier greening, denser foliage at peak LAI, and later senescence of switchgrass (a perennial grass) compared to annual crops and rangeland. Increase of vegetation fraction ranges 10-20%, except in north central OK where it is increased by 90% to offset low LAI due to winter wheat harvest in April and May. Monthly average LAI for Kansas illustrates the modified vegetation (Figure 2). We note LAI >6 as measured in field trials of managed switchgrass is used in simulations of switchgrass production [Jain et al., 2010; Mitchell and Schmer, 2012; Miguez et al., 2012]. We set LAI <6 reflecting a regional vegetation mixture.

Figure 2.

Annual cycle of Kansas Leaf Area Index.

3 Results and Discussion

[12] Comparison of growing season variables with observations is provided in Figures S2 – S4. We average the change (i.e., switchgrass minus control) in 2-m temperature and relative humidity over Kansas and Oklahoma.

[13] Change in 2-m temperature in Kansas and Oklahoma follows a similar seasonal pattern of reduced temperature during February – June and little or no temperature change during July – October, while relative humidity increases during February – June and decreases during July – October (Figure 3). Temperature changes in February – June are outside the range of interannual variability. In Noah, increase in LAImax will increase latent heat flux through the inverse relation of canopy resistance to LAI. Since maximum and minimum albedo are unchanged, modified vegetation fraction will impact primarily seasonality of LAI, and net radiation will be relatively unchanged (absent a change in cloudiness). Thus, latent heat increase will be compensated primarily by reduction in sensible heat flux. Subsequently, reduction of 2-m maximum temperature is expected, though this was not observed in July-October.

Figure 3.

Change in average (1981–2004) monthly 2-m temperature (solid lines) and specific humidity (dashed lines) for Kansas (blue lines) and Oklahoma (red lines). The spatial average is over the boxed region in Figure 1. Whiskers are the range of monthly change for each of the 24 simulation years.

[14] We further explore the seasonal variations of temperature change through evaluation of the terrestrial hydrological cycle for the Kansas-Oklahoma region (box in Figure 1), for which seasonality of changes are similar. Change in evapotranspiration (ET) in the Kansas-Oklahoma region is positive in March-June and negative in August-November (Table 2). ET decreases during the broad LAI peak of the switchgrass simulation in August-November, when an increase rather than zero change is expected. Note that reduction in ET lags that of soil moisture (Table 2). This implies that ET is at a higher rate under modified land use until soil moisture nears wilting point, eliminating transpiration (Eq (1)) and inhibiting further soil moisture decrease.

Table 2. Difference (switchgrass scenario minus control simulation) of surface and sub-surface water balance (mm) and soil moisture (mm). P – Precipitation; ET – Evapotranspiration; R – Surface Runoff; SMcolumn – Total Column Soil Moisture; SMroot – Root Zone Soil Moisture
MonthΔ PΔ ETΔ RΔ(P-R-ET)Δ SMrootΔ SMcolumn
  • *

    indicates null hypothesis of equal mean value is rejected at 95% confidence interval. One-way ANOVA tests were performed in JMP software to test the null hypotheses of equal mean values for the monthly surface water budget components.


[15] Dependence of ET on soil moisture emerges from the limited impact of switchgrass on precipitation. Annual precipitation for Kansas in both simulations is ~740 mm with peak in May and slow decline from June through October. Thus, the soil moisture reservoir provides a larger fraction of water supply as the growing season progresses. Monthly water budget components show precipitation change is insufficient to offset increased ET (Table 2). Consequently, monthly difference in root zone soil moisture exceeds 30 mm in July and August (Table 2). For each LULC type, the difference in soil moisture, expressed as a percent of the control simulation (Figure 4), is smallest for no change in LULC category; whereas with switchgrass conversion the range of annual percent change does not overlap zero. These results indicate permanent depletion of the soil moisture reservoir with conversion to switchgrass.

Figure 4.

Percent difference in annual mean soil moisture for three soil layers: (a) 0–10 cm, (b) 10–40 cm, and (c) 40–100 cm averaged over the boxed region in Figure 1. Box top and bottom edges are the interquartile range of percent difference for each year, and whiskers are maximum and minimum annual values. X-axis labels are land use categories: NC (No Change), S/G (Swithgrass/Grassland), S/C (Swithgrass/Cropland), G/S (Grassland/Switchgrass), C/S (Cropland/Switchgrass), Avg (average over all categories).

[16] Sensitivity to rcmin was tested by using the switchgrass simulation as an initial condition for an alternative simulation of March-August 1987 in which a higher value of rcmin was specified for the switchgrass classes (96 s m-1 compared to 40 s m-1) [Alfieri et al., 2008]. Increase of rcmin will decrease transpiration and increase soil moisture. In the rcmin test, smaller decline in July to August ET is evident (Table S2). Thus, higher rcmin could mean soil moisture would approach wilting point less often, and higher ET and lower summer temperature would be expected. However, soil moisture depletion found in the switchgrass simulation is broadly consistent with results from other modeling studies [Miguez et al., 2012; Vanloocke et al., 2010]. Our simulations used a slightly larger land use conversion fraction but smaller change of LAI.

4 Conclusion

[17] We conclude that surface climate change would likely be an outcome of enacted biofuel policy should production mandates be achieved through widespread land conversion in the Great Plains. Notable changes for regions where switchgrass has been predicted by agro-economic models to replace current vegetation include lower temperature in spring, slightly higher relative humidity in spring, slightly lower relative humidity in summer, and depletion of summer soil moisture. The magnitude and seasonality of these climate effects is dependent upon soil moisture, and the predicted change is sensitive to design of interactive soil-vegetation models.

[18] Climate and agricultural systems are coupled, but are generally analyzed separately in making projections. This may introduce two errors. First, feedback from land conversion through regional climate back into the converted land's agricultural productivity is not considered in socio-economic and policy scenarios, possibly biasing these projections toward expectations of yield that cannot be realized. Our results suggest that in some locations and some years, depletion of soil moisture could reduce productivity of the biofuel crop. This possibility should be considered within socio-economic and policy scenarios and should be explored further with agronomic models (such as the methodology of Miguez et al., 2012). Second, to be accurate, simulations of future climate must consider agricultural LULC change whether it is exogenous or an adaptation to climate change. We conclude, therefore, that agro-economic and climate models should be used iteratively, or possibly coupled dynamically, to examine an ensemble of agricultural land use and climate scenarios. This approach would reduce the possibility of unforeseen consequences from rapid changes in agricultural production systems.


[19] The authors thank Dr. De La Torre Ugarte for providing access to and technical assistance with the POLYSYS model projections. This research was supported in part by subcontract NFT-8-88540-01 from the National Renewable Energy Laboratory and by grant 20116800230190 from the U.S. Department of Agriculture. This material is based upon work supported by the National Science Foundation under Grants No. CBET-1137677. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.