Policy Analysis pubs.acs.org/est
Impacts of Variability in Cellulosic Biomass Yields on Energy Security Kimberley A. Mullins,†,‡,⊥ H. Scott Matthews,†,‡ W. Michael Griffin,*,†,§ and Robert Anex∥ †
Department of Engineering and Public Policy, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, United States Civil and Environmental Engineering Department, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, United States § Tepper School of Business, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, United States ∥ Biological Systems Engineering, University of Wisconsin-Madison, Madison, Wisconsin 53706, United States ‡
S Supporting Information *
ABSTRACT: The practice of modeling biomass yields on the basis of deterministic point values aggregated over space and time obscures important risks associated with large-scale biofuel use, particularly risks related to drought-induced yield reductions that may become increasingly frequent under a changing climate. Using switchgrass as a case study, this work quantifies the variability in expected yields over time and space through switchgrass growth modeling under historical and simulated future weather. The predicted switchgrass yields across the United States range from about 12 to 19 Mg/ha, and the 80% confidence intervals range from 20 to 60% of the mean. Average yields are predicted to decrease with increased temperatures and weather variability induced by climate change. Feedstock yield variability needs to be a central part of modeling to ensure that policy makers acknowledge risks to energy supplies and develop strategies or contingency plans that mitigate those risks.
■
INTRODUCTION The federal Energy Independence and Security Act of 2007 (EISA) addresses the United States’ undesired dependence on foreign sources of oil.1 The Renewable Fuel Standard (RFS) specified in this legislation is a liquid biofuel mandate designed in part to address energy security. Energy security can be approached from various perspectives, including fossil energy supply diversity, demand resiliency, and the impact of oil supply fluctuations on U.S. gross domestic product.2−6 Researchers and policy makers who advocate the use of biofuels as a way to decrease oil consumption and reduce the U.S. dependence on unstable foreign oil sources take as a given that a shift toward biofuels will reduce the risks associated with fuel availability or have simply chosen not to specifically define the term “energy security” in their context (e.g., see refs 4, 7, and 8). Discussions in the literature do not sufficiently acknowledge that a change in the domestic energy portfolio to include biomass does not necessarily translate to a reduced supply risk. An increase in renewable fuels in the supply portfolio brings different supply risks that may or may not be preferable to the risks associated with fossil fuel supply (e.g., supply disruptions due to fossil fuel infrastructure damage). Biomass availability is subject to natural temporal variation due to recurring periods of weather-induced crop water and temperature stress, and the resulting biomass yield fluctuations can create chronic bioenergy feedstock shortages. Of the two stresses, a lack of water is the more important one9 and is particularly relevant as cellulosic crops are usually assumed to be produced though dryland (rain-fed) farming. Drought is usually regional and yet can have national significance. Understanding the potential to supply a significant © 2014 American Chemical Society
amount of biomass-derived transportation fuel requires not only knowing the areas most likely to produce biomass and estimates of the yield under ideal conditions but also how the yield is impacted by weather-related risks, which vary across the country. In this study, switchgrass is used as a case study for the expected variability in cellulosic crop production. In this analysis, the United States is disaggregated to the state level, a politically relevant regional distinction, and weather data are used to estimate biofuel feedstock yields. Yield variability is assessed using historic and simulated future weather data and the switchgrass yield model developed by Grassini et al.10 Nair and colleagues have presented a review of many different crop models,11 highlighting that while model results are improving, there is still a need for better input data and better model calibration. Therefore, the overall yield trends observed in this article are reliable, but the specific yield values should not be taken as precise predictions. It is important to keep in mind that the annual yield time series output by this model vary as a result of changes in precipitation and temperature (which drive drought conditions) but not other factors such as an overabundance of local water or acute weather phenomena (e.g., hail) because the model is not equipped to deal with these situations (e.g., there is no “hail” module). The variability over time would be greater if these other causes of crop loss were included, and this could be a fruitful line of investigation for future work. Received: Revised: Accepted: Published: 7215
February 20, 2014 May 31, 2014 June 11, 2014 June 18, 2014 dx.doi.org/10.1021/es404174h | Environ. Sci. Technol. 2014, 48, 7215−7221
Environmental Science & Technology
■
Policy Analysis
Module 1: Crop Development Index. The crop development index is a dimensionless index that increases from 0 to 1 and is affected by the daily mean temperature in relation to cultivar-specific temperature parameters, including the optimal temperature and the temperatures above or below which no development occurs. Daily, incremental development is tracked and taken as input to various other modules. Module 2: Leaf Area Index (LAI) Expansion. The LAI is a dimensionless index that describes the fullness of the canopy. It affects rainfall partitioning in the soil water balance module and the rate of conversion of solar radiation to biomass (i.e., crop growth). The LAI value depends on the crop development index and a water stress factor. Module 3: Soil Water Balance. The soil water balance model is the most complex of the five modules. There are five submodules within this function. Rainfall can be intercepted by the canopy (a function of the LAI) and not be available to the soil layer. Heavy rainfall can result in water lost to surface runoff: the top layer can hold 110% of capacity right after rainfall, and the excess becomes runoff. Water evaporates from the upper layer of soil in a quantity that depends on temperature, solar radiation intensity, LAI, and previous water concentration in the upper soil layer. Water transpires through the plant during growth in a quantity determined by the existing biomass/development stage and temperature. Water for transpiration can come from either the top or bottom soil layer. Finally, excess water that may be in the top soil layer at the end of a day filters down to the bottom layer so that the top layer is not oversaturated. Water flow through these layers is illustrated in Figures S2 and S3 in the SI. Module 4: Crop Growth and Biomass Production. The mass of new biomass produced each day is a function of the amount of photosynthetically active radiation intercepted by the canopy (which is a function of the LAI), the rate at which the plant can convert radiation energy into biomass, and water and temperature stress factors. The growth rate parameter used is for the Blackwell switchgrass cultivar, which shows midrange performance. Module 5: Water and Temperature Stress Factors. Crop growth is retarded by insufficient water, or insufficiently high temperatures. Specific water stress factors are applied to the LAI function and the biomass growth function, and a specific temperature stress factor is applied to the biomass growth function. Supplemental calculation details that are necessary to complete the model but are not explicitly documented in the study by Grassini et al.10 or its references are taken from a publication on crop evapotranspiration published by the Food and Agriculture Organization of the United Nations (FAO).15 These include the necessary formulation and coefficient definitions for the Priestley−Taylor equation to calculate the daily reference evapotranspiration and a formula for calculating saturated vapor pressure. A key parameter assumed at the beginning of the model run is the fraction of available water-holding capacity filled at the date of growth initiation (FAWHCAGI), as an initial abundance of soil water can make up for some lack of rain during the growing season and a deficit of soil water can make that situation worse. This fraction varies from 0 to 1. The FAWHC AGI is assumed to decrease as the standard precipitation index (SPI) decreases. The SPI as a drought index is discussed by Guttman.16 A FAWHCAGI of 0.6 corresponds to the median precipitation scenario (the SPI
DATA AND METHODS Switchgrass Yield Model. The switchgrass growth model built for this study was defined by Grassini et al.10 and was published with the intention of generating annual crop yield values that could be used to inform energy policy models and discussions. The switchgrass growth model is composed of five interacting modules and operates at a daily time step in calculating biomass production. The key model parameters are summarized in Table 1. Inputs include daily rainfall; mean, Table 1. Summary of Key Input to the Switchgrass Growth Modela variable
magnitude
minimum growth temperature maximum growth temperature optimal growth temperature maximum LAI radiation use efficiency soil layer 1 (SL1) depth soil layer 2 (SL2) depth* soil available water-holding capacity, SL1 soil available water-holding capacity, SL2 extinction coefficient for incoming solar radiation fraction of available water-holding capacity at day of growth initiation, SL1* fraction of available water-holding capacity at day of growth initiation, SL2*
13 °C 42 °C 33 °C 10 4.7 g/MJ 150 mm 1450 mm 0.10 to 0.18 0.10 to 0.24 0.48 0.6 0.6
a Taken from Grassini et al.10 and modified if necessary (indicated by *) in order to better replicate empirical yield data.
high, and low temperatures; solar insolation; and information about soil characteristics and crop growth performance. Thirty years of daily data are used as the historical meteorological data set. Data from one city in an agricultural area from each state are taken, so one location is assumed to be representative of the whole state (as historical weather records are available for only one to a handful of locations in each state). The city selected from each state is listed in the Supporting Information (SI). The historical data are taken from the database assembled by the United States Department of Agriculture (USDA) for use in their weather generation model, named GEM (generation of weather elements for multiple applications).12 These data include daily observations for maximum and minimum temperature, precipitation, and solar radiation, and data covering 1961 to 1990 are available. Mean daily temperatures are calculated as averages of the high and low values. Soil characteristics for each state, specifically the available soil waterholding capacity, are calculated using data from the USDA’s Soil Survey Geographical Database (SSURGO).13 The available water-holding capacities for the top and bottom soil layers in this study are calculated using the 200 and 1500 mm depth layers in SSURGO, respectively, and averaging all of the data points from within each state at these depths. A complete table of values can be found in the SI. When daily biomass accumulation values (output in g/m2) are summed over an entire growing season, the result is an annual crop yield value. When many years are modeled, variability in the yield over time can be assessed. The Grassini model is not publically available as a stand-alone program or as code, so a model was built in MATLAB (version R2012b).14 The five interacting modules mentioned previously are defined as follows: 7216
dx.doi.org/10.1021/es404174h | Environ. Sci. Technol. 2014, 48, 7215−7221
Environmental Science & Technology
Policy Analysis
are added to these modified temperature and radiation residuals. The simulated data for this analysis are generated using methods for precipitation and temperature from Chen and colleagues,21 who implemented the Richardson model as described above (using a second-order Markov chain) and included a smoothing algorithm to reduce jaggedness in the mean temperature profiles. The jaggedness arises because the temperature means are calculated for two-week time periods, so each year is characterized by a set of 26 max/min temperatures. Without any modifications, there tends to be a jump between the temperature on the last day of one period and the first day of the next period. This particular implementation of the Richardson model was used because Chen and colleagues have graciously made the MATLAB code available online (the URL is listed in the SI). Means and standard deviations of solar radiation data are calculated using dry days, and a scaling factor is used to simulate radiation for wet days, following methods described by Zhang et al.22 In summary, the simulated weather data employed as input to the switchgrass yield model use the following: • a second-order Markov chain to generate precipitation occurrence patterns; • a two-parameter gamma distribution to generate precipitation magnitude data, smoothed to reduce jumps between successive two-week periods; • a conditional relationship between the maximum and minimum temperatures; • a scaling factor of 0.5 to scale the mean daily solar radiation on dry days for wet days. Given this weather simulation program from Chen and colleagues, certain parameters in the model are modified in order to produce a climate that is changed from the historical climate. The following scenarios are examined, with data presented in the form of box plots to illustrate key statistics: 1. Base-case weather, simulated using historical data (sources and locations described previously) and unmodified parameters (so that yields from simulated, modified climate data are not compared to historical data). 2. A higher-temperature case based on a scenario used in a future U.S. drought study.23 In the weather generation model, mean values for Tmin and Tmax time series (μtemp + 2) are modified as suggested in the study by Roy et al.23 3. A case with higher variability in temperature based on results shown in a study of temperature anomalies by Hansen et al.24 This is implemented though modified standard deviation values for Tmin and Tmax (1.2σtemp and 1.3σtemp) as shown in the Hansen study. 4. A scenario with higher dry- and wet-spell lengths. Rather than modification of precipitation amounts, which vary regionally (see ref 25), the impacts of increased dry- or wet-spell length are examined. This is accomplished by modifying the transition probabilities for precipitation occurrence in the Markov matrices, making it 10% (relative, not percentage points) more likely that there will be longer dry spells or wet spells. This is done not specifically following a procedure undertaken in another study or using empirical data found elsewhere but as a way to compare the yield sensitivity to spell length alongside temperature modifications.
median value is 0), and this is taken as the default value for the model runs in Grassini et al.10 It seems unlikely that even in a severe drought scenario the top two meters of soil would be completely dry, so the model was run assuming no precipitation at all over various FAWHCAGI values. For a range of FAWHCAGI values, the value for the lower soil layer at the time of growth does not drop below 0.1, while the top layer is constantly exhausted of moisture. As a result of this information, the FAWHCAGI is assumed to decrease linearly from 0.6 to 0.1 with a decrease in precipitation scenario likelihood and to increase from 0.6 to 1 with an increase in precipitation scenario likelihood. The validation of the model is presented in the SI. Weather Generation Model. To assess the impacts of a changing climate on switchgrass yields, this study simulated weather data using historical data for which certain parameters have been modified so that the output weather data are representative of a changed, future climate. As stated previously, this exercise is intended to show changing trends in yields, not to suggest exact magnitudes for the changes in variability or expected values for yields in any region. Weather simulation programs have been tools for modelers for the past 30 years. Many used for applications that requite daily data, such as this crop yield model, are built upon the stochastic simulation method for temperature, precipitation, and solar radiation suggested by C. W. Richardson, as described in refs 17 and 18 and more fully defined as the WGEN model by Richardson and Wright in ref 19. In most daily weather generation models, temperature and solar radiation data are conditioned on precipitation. Precipitation is modeled in two parts: whether there is any precipitation (dry or wet conditions; a binary random variable) and, when there is precipitation, how much falls. In this model, the wet or dry state is simulated as a Markov process; the status of the current day depends on the status of the previous day(s). In a first-order Markov model, precipitation status is conditioned only on the previous day; if it was dry yesterday, the likelihood that it will be dry today is P1 and the likelihood that it will be wet is (1 − P1), where P1 is estimated using historical data. Higher-order models have been used by more recent studies to more accurately reproduce long wet or dry spells (see the excellent discussion in ref 20). The quantity of precipitation on wet days is generally modeled using either an exponential function or a two-parameter gamma function, as both heavily weigh small amounts of precipitation but skew positively to allow for the low probability of much higher quantities of precipitation. With the precipitation time series generated, temperature and solar radiation data are then modeled using a first-order linear autoregressive model. Means and standard deviations for each temperature−time series, conditioned on precipitation, are calculated over some relevant time period (daily, weekly, monthly) across all years of historical data (i.e., two means for each two-week period are calculated: {μdry, σdry}week i and {μwet, σwet}week i). Time series of residuals are generated for maximum temperature, minimum temperature, and solar radiation using the calculated standard deviations. These residual values are modified by a lagged correlation matrix (modified on the basis of what temperatures and solar radiation happened the previous day) to which a simultaneous correlation matrix of normally distributed errors (modified on the basis of what the other meteorological values for the same day were) is added. Finally, the means and standard deviations calculated from the initial, historical data 7217
dx.doi.org/10.1021/es404174h | Environ. Sci. Technol. 2014, 48, 7215−7221
Environmental Science & Technology
Policy Analysis
Figure 1. Yield model variability results for the 19 states that are forecast to contribute at least 1% of national switchgrass production. Each dot is the mean yield over 30 years in one state, and error bars are the 10th and 90th percentile yields. Colors correspond to NOAA climate regions. The states are ordered by (A) decreasing crop production area, (B) increasing variance, and (C) decreasing mean.
terms of crop production area, as shown in Figure 1A, whereas other high-yield, low-variability regions (such as the Southeast) are not forecast to contribute as much to the switchgrass supply. Adding crop yield variability to the list of metrics by which cellulosic crop regions are evaluated might reorder the states in Figure 1A and could lead to a more stable expected energy crop yield. At present, the use of cellulosic feedstocks such as switchgrass is mandated at the national level through the RFS, so targets are met through an aggregated national cellulosic ethanol production volume. From this perspective, then, the national aggregate yield is of interest. Figure 2
By using the three modifications in some combination with each other, six scenarios (in addition to the base-case simulation results) are compared in the following section. The results ought only to be compared among themselves in order to get a sense of the sensitivity of crop yields to different changes in climate conditions and to see whether there are general trends to anticipate.
■
RESULTS Results using historical weather data for the 19 states forecast to contribute at least 1% of U.S. switchgrass production by mass are presented in Figure 1A (results for all of the continental states are presented the SI). The 19 states are ordered on the basis of how much crop production area is anticipated within the state and colored on the basis of climate region. Most switchgrass is expected to be grown in the South, Southeast, and Central regions of the United States (identified in the figure).26 The yield model results show mean yield values across the states ranging from about 12 to 19 Mg/ha. There is also substantial variability within each state; the range of yields representing the 80% confidence intervals (the 10th to 90th percentile range) plotted in Figure 1 cover at least 8 Mg/ha, and some span a range of more than 17 Mg/ha. For mean values in the range of 12 to 19 Mg/ha, this translates to a variability of 20 to 60% of the mean. A complete ordered list of states in Figure 1A is included in Table S1 in the SI. The maximum yields from this model are higher than those from another popular switchgrass yield model, Environmental Policy Integrated Climate (EPIC),27 because growth in the model from Grassini et al.10 is not limited by nutrient availability; if temperature conditions are optimal and water is available, biomass is produced at optimal rates, leading to high yields. These high yields are close to those reported from switchgrass test plots, some few of which had annual yields greater than 30 Mg/ha.28 On the basis of Figure 1B,C, which presents state yield distributions ordered by increasing variability and decreasing mean value, respectively, the Central states (which show high yields and comparatively low variability) are well-represented in
Figure 2. National average yields, an aggregation of state-by-state yields using the POLYSYS distribution of acreage. See Table S1 in the SI for the acreage per state.
presents the national aggregate switchgrass yield given historic weather data as well as curves showing the maximum and minimum yields from contributing states for comparison. Despite aggregation over many states, the fluctuation of the total yield over time is not so different from the yield variability in the individual states; the expected national yield over 30 years is 14.6 Mg/ha, and the 80% confidence interval of annual 7218
dx.doi.org/10.1021/es404174h | Environ. Sci. Technol. 2014, 48, 7215−7221
Environmental Science & Technology
Policy Analysis
the weather simulator do not represent the spatial correlation that is a common feature of many droughts.29 Correlations in yield are derived using the switchgrass model results from historical weather data, and the results for selected relevant states are presented in the SI. Some neighboring states demonstrate significant correlation, though not all do. All but one of the statistically significant correlation statistics are positive. The fact that most of the correlations are positive and are between about 0.4 and 0.8 suggests that there are not a lot of obvious opportunities to diversify planting locations in order to reduce the likelihood that reduced switchgrass yield is likely to occur regionally. This highlights the potential for widespread, simultaneous yield decreases and calls into question the ability of the U.S. biomass industry to consistently meet RFS biofuel targets.
yield values (11.2 to 17.8 Mg/ha) represents an uncertainty range of about 50% of the mean. The results for the base-case weather data (i.e., simulated weather using unmodified parameters calculated from historical data) are plotted in the top panel of Figure 3, with those for the
■
DISCUSSION The results in Figure 3 suggest that as the climate warms and becomes increasingly variable, switchgrass yields will tend to decrease in many states. Using historical average yield data in projections without adjusting for these climate changes will lead to overprediction of yields, and this shows cause for concern when past yield data are assumed to be representative of future yields in planning models. To ensure that sufficient biomass feedstock is available to supply biofuel conversion facilities, either large amounts of biomass must be kept in storage or additional acreage must be planted. Because of the low bulk density of biomass, very large scale multiyear storage is not practical.30 Storage does play an important role in eliminating negative price spikes in agricultural markets,31 and the limited ability to store biomass and the need to meet production volume mandates will tend to lead to high price volatility (another facet of energy security).32 Additional acreage does not address production variability but could be a strategy to raise total production amounts throughout the years, thereby reducing the likelihood of simply not having enough biomass for production facilities to operate at a profitable level. Of course, dealing with surplus production in high-yield years is a challenge, particularly with little storage capacity. Models combining stochastic yields by state (or other region) could be used to ensure that national feedstock yields are sufficient to meet EISA policy goals with some appropriate degree of confidence following the method outlined in ref 33. Consistency in dryland crop yields is likely to decrease in the future, as a number of studies have reported that drought severity and frequency is predicted to increase under the most likely future climate scenarios (see refs 25 and 34 for two examples). Tied to this are expectations of increasing and increasingly variable temperatures.24 Crop water stress will increase as a result of both meteorological changes and increases in water demand due to economic growth.23 This could lead to chronic multiyear biomass shortages, which would be very costly. Biomass energy forecasts are developed from models based on temporally and spatially aggregated yields.35−37 Policy recommendations informed by modelers’ predictions based on average productivity do not sufficiently address how policy should accommodate yield variation such as seen in the severe drought of the 1987−1988 crop year or how the developing biomass industry in, for example, Tennessee might deal with the resulting 90% reduction from the mean. For decisions that hinge on biomass yield, the models that predict yields must include stochastic analysis that accounts for expected changes in
Figure 3. Comparison of means and standard deviations for yields for all states for each of the seven simulated weather scenarios. The y-axis values are yields (Mg/ha), and the x-axis values are standard deviations. Solid red dots are the centroids of the data. The hollow blue dots show the centroid of the base case, which has been plotted on the other scenarios for comparison. Each gray point represents one of the 19 states.
modified climate scenarios shown in the six panels below. The centroid on each plot is shown relative to the centroid for the base case. Detailed figures similar to Figure 3 in which the data are labeled with the corresponding states are included in the SI. Increasing the standard deviation of temperature by 20% (“1.2sigma”) results in the smallest decrease in the mean yield. Increasing the standard deviation of temperature by 30% (“1.3sigma”) has a greater impact, reducing some of the highest mean yield values. Modifying the precipitation patterns by increasing the likelihood of multiday wet or dry spells by 10% (“Precip.”) tends to decrease the mean yields as well, though it also produces the highest single-state average yield value, and it tends to increase the standard deviation. Increasing the mean temperature by 2 °C (“T+2”) also tends to lower the yield. When the temperature is lower and more variable (“1.3sigma, T +2”), the points show a slightly reduced range of mean values and a slightly wider range of standard deviation values, though the centroid is not much changed from when either of those scenarios is run individually. Finally, modifying the precipitation patterns and increasing the expected temperatures (“Precip., T+2”) has the greatest effect on the mean and standard deviation statistics; the overall mean yield (centroid y value) is about 20% lower than for the base case, and the points are increasingly concentrated around the mean of the yield standard deviation (centroid x value). An important component of historical yields that is not maintained in yields estimated using simulated weather data is regional correlation. The meteorological variables generated in 7219
dx.doi.org/10.1021/es404174h | Environ. Sci. Technol. 2014, 48, 7215−7221
Environmental Science & Technology
Policy Analysis
(11) Nair, S. S.; Kang, S.; Zhang, X.; Miguez, F. E.; Izaurralde, R. C.; Post, W. M.; Dietze, M. C.; Lynd, L. R.; Wullschleger, S. D. Bioenergy crop models: Descriptions, data requirements, and future challenges. GCB Bioenergy 2012, 4, 620−633. (12) Natural Resources Conservation Service. Weather Generation Technology (GEM); USDA: Washington, DC. (13) Soil Survey Staff, Natural Resources Conservation Service. Gridded Soil Survey Geographic (SSURGO) Database; USDA: Washington, DC, 2012. (14) MATLAB, version R2012b; The MathWorks, Inc: Natick, MA, 2012. (15) Allen, R. G.; Pereira, L. S.; Raes, D.; Smith, M. Crop Evapotranspiration: Guidelines for Computing Crop Water Requirements; FAO Irrigation and Drainage Paper 56; Food and Agriculture Organization of the United Nations (FAO): Rome, 1998. (16) Guttman, N. B. Accepting the standardized precipitation index: A calculation algorithm. J. Am. Water Resour. Assoc. 1999, 35, 311−322. (17) Richardson, C. W. Stochastic simulation of daily precipitation, temperature, and solar radiation. Water Resour. Res. 1981, 17, 182− 190. (18) Larsen, G.; Pense, R. Stochastic Simulation of Daily Climate Data; Statistical Reporting Service (SRS) Staff Report No. AGES810831; USDA: Washington, DC, 1981; pp 1−64. (19) Richardson, C. W.; Wright, D. A. WGEN: A Model for Generating Daily Weather Variables; Publication ARS-8; USDA Agricultural Research Service: Washington, DC, 1984. (20) Wilks, D. S.; Wilby, R. L. The weather generation game: A review of stochastic weather models. Prog. Phys. Geog. 1999, 23, 329− 357. (21) Chen, J.; Brissette, P. F.; Leconte, R. A daily stochastic weather generator for preserving low-frequency of climate variability. J. Hydrol. 2010, 388, 480−490. (22) Zhang, Q.; Singh, B.; Gagnon, S.; Rousselle, J.; Evora, N.; Weyman, S. The application of WGEN to simulate daily climatic data for several canadian stations. Can. Water Resour. J. 2004, 29, 59−72. (23) Roy, S. B.; Chen, L.; Girvetz, E. H.; Maurer, E. P.; Mills, W. B.; Grieb, T. M. Projecting water withdrawal and supply for future decades in the U.S. under climate change scenarios. Environ. Sci. Technol. 2012, 46, 2545−2556. (24) Hansen, J.; Sato, M.; Ruedy, R. Perception of climate change. Proc. Natl. Acad. Sci. U.S.A. 2012, 109, E2415−E2423. (25) Strzepek, K.; Yohe, G.; Neumann, J.; Boehlert, B. Characterizing changes in drought risk for the United States from climate change. Environ. Res. Lett. 2010, 5, No. 044012. (26) Agricultural Policy Analysis Center. The POLYSYS Modeling Framework: An Overview; Institute of Agriculture, University of Tennessee: Knoxville, TN, 2010. (27) Thomson, A. M.; Izarrualde, R. C.; West, T. O.; Parrish, D. J.; Tyler, D. D.; Williams, J. R. Simulating Potential Switchgrass Production in the United States; Report PNNL-19072; Pacific Northwest National Laboratory: Richland, WA, 2009; pp 1−22. (28) Kszos, L. A.; Downing, M. E.; Wright, L. L.; Cushman, J. H.; McLaughlin, S. B.; Tolbert, V. R.; Tuskan, G. A.; Walsh, M. E. Bioenergy Feedstock Development Program Status Report; Oak Ridge National Laboratory: Oak Ridge, TN, 2000. (29) Douglas, E. M.; Vogel, R. M.; Kroll, C. N. Trends in floods and low flows in the United States: Impact of spatial correlation. J. Hydrol. 2000, 240, 90−105. (30) Hess, J. R.; Wright, C. T.; Kenney, K. L. Cellulosic biomass feedstocks and logistics for ethanol production. Biofuels, Bioprod. Biorefin. 2007, 1, 181−190. (31) Wright, B. D. The economics of grain price volatility. Appl. Econ. Perspect. Policy 2011, 33, 32−58. (32) Diffenbaugh, N. S.; Hertel, T. W.; Scherer, M.; Verma, M. Response of corn markets to climate volatility under alternative energy futures. Nat. Clim. Change 2012, 2, 514−518. (33) Mullins, K. A.; Griffin, W. M.; Matthews, H. S. Policy implications of uncertainty in modeled life-cycle greenhouse gas emissions of biofuels. Environ. Sci. Technol. 2011, 45, 132−138.
not only the mean yield but also the yield variability. A probabilistic approach is central to assessing the consequences of yield variability on model recommendations. Without an acknowledgment of variability, there is no basis upon which the emerging biofuel industry and biofuel policy makers can begin to consider and evaluate mitigation options and contingency plans.
■
ASSOCIATED CONTENT
S Supporting Information *
Further details of model validation and detailed statistics on state-by-state variability. This material is available free of charge via the Internet at http://pubs.acs.org.
■
AUTHOR INFORMATION
Corresponding Author
*Phone: (412) 268-2299; fax: (412) 268-3757; e-mail: wmichaelgriffi[email protected]. Present Address ⊥
K.A.M: Department of Bioproducts and Biosystems Engineering, University of Minnesota, St. Paul, MN 55108. Notes
The authors declare no competing financial interest.
■
ACKNOWLEDGMENTS The authors thank Chris Hendrikson for his assistance in study design and Ryan Noe for his assistance in extracting data from SSURGO. Funding was provided by the Center for Climate and Energy Decision Making through a cooperative agreement between the National Science Foundation and Carnegie Mellon University (SES-0949710), the USDA (NIFA Grant 201367009-20377), and the US DOE (EERE Grant DEEE0004397).
■
REFERENCES
(1) Rahall, N. Energy Independence and Security Act of 2007. Public Law 110-140, 2007. (2) Jansen, J. C.; Seebregts, A. J. Long-term energy services security: What is it and how can it be measured and valued? Energy Policy 2010, 38, 1654−1664. (3) Brown, S. P. A.; Huntington, H. G. Reassessing the Oil Security Premium; RFF Discussion Paper 10-05; Resources for the Future: Washington, DC, 2010. (4) Angelis-Dimakis, A.; Biberacher, M.; Dominguez, J.; Fiorese, G.; Gadocha, S.; Gnansounou, E.; Guariso, G.; Kartalidis, A.; Panichelli, L.; Pinedo, I.; Robba, M. Methods and tools to evaluate the availability of renewable energy sources. Renewable Sustainable Energy Rev. 2011, 15, 1182−1200. (5) Hughes, L. A generic framework for the description and analysis of energy security in an energy system. Energy Policy 2012, 42, 221− 231. (6) Greene, D. L.; Leiby, P. N.; Patterson, P. D.; Plotkin, S. E.; Singh, M. K. Oil independence: Achievable national goal or empty slogan? Transp. Res. Rec. 2008, 2017, 47−53. (7) Sperling, D.; Yeh, S. Toward a global low carbon fuel standard. Transp. Policy 2009, 17, 47−49. (8) Dale, B. E. Cellulosic biofuels and the road to energy security. Environ. Sci. Technol. 2011, 45, 9823. (9) McLaughlin, S. B.; Kszos, L. A. Development of switchgrass (Panicum virgatum) as a bioenergy feedstock in the United States. Biomass Bioenergy 2005, 28, 515−535. (10) Grassini, P.; Hunt, E.; Mitchell, R. B.; Weiss, A. Simulating switchgrass growth and development under potential and waterlimiting conditions. Agron. J. 2009, 101, 564−571. 7220
dx.doi.org/10.1021/es404174h | Environ. Sci. Technol. 2014, 48, 7215−7221
Environmental Science & Technology
Policy Analysis
(34) Dai, A. Increasing drought under global warming in observations and models. Nat. Clim. Change 2012, 3, 52−58. (35) U.S. Department of Energy. U.S. Billion-Ton Update: Biomass Supply for a Bioenergy and Bioproducts Industry; Perlack, R. D., Stokes, B. J., Leads; ORNL/TM-2011/224; Oak Ridge National Laboratory: Oak Ridge, TN, 2011. (36) A USDA Regional Roadmap to Meeting the Biofuels Goals of the Renewable Fuels Standard by 2022; USDA: Washington, DC, 2010; pp 1−21. (37) Renewable Fuel Standard Program (RFS2) Regulatory Impact Analysis; EPA-420-R-10-006; U.S. Environmental Protection Agency: Washington, DC, 2010.
7221
dx.doi.org/10.1021/es404174h | Environ. Sci. Technol. 2014, 48, 7215−7221
Impacts of Variability in Cellulosic Biomass Yields on Energy Security Kimberley A. Mullins,†,‡,⊥ H. Scott Matthews,†,‡ W. Michael Griffin,*,†,§ and Robert Anex∥ †
Department of Engineering and Public Policy, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, United States Civil and Environmental Engineering Department, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, United States § Tepper School of Business, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, United States ∥ Biological Systems Engineering, University of Wisconsin-Madison, Madison, Wisconsin 53706, United States ‡
S Supporting Information *
ABSTRACT: The practice of modeling biomass yields on the basis of deterministic point values aggregated over space and time obscures important risks associated with large-scale biofuel use, particularly risks related to drought-induced yield reductions that may become increasingly frequent under a changing climate. Using switchgrass as a case study, this work quantifies the variability in expected yields over time and space through switchgrass growth modeling under historical and simulated future weather. The predicted switchgrass yields across the United States range from about 12 to 19 Mg/ha, and the 80% confidence intervals range from 20 to 60% of the mean. Average yields are predicted to decrease with increased temperatures and weather variability induced by climate change. Feedstock yield variability needs to be a central part of modeling to ensure that policy makers acknowledge risks to energy supplies and develop strategies or contingency plans that mitigate those risks.
■
INTRODUCTION The federal Energy Independence and Security Act of 2007 (EISA) addresses the United States’ undesired dependence on foreign sources of oil.1 The Renewable Fuel Standard (RFS) specified in this legislation is a liquid biofuel mandate designed in part to address energy security. Energy security can be approached from various perspectives, including fossil energy supply diversity, demand resiliency, and the impact of oil supply fluctuations on U.S. gross domestic product.2−6 Researchers and policy makers who advocate the use of biofuels as a way to decrease oil consumption and reduce the U.S. dependence on unstable foreign oil sources take as a given that a shift toward biofuels will reduce the risks associated with fuel availability or have simply chosen not to specifically define the term “energy security” in their context (e.g., see refs 4, 7, and 8). Discussions in the literature do not sufficiently acknowledge that a change in the domestic energy portfolio to include biomass does not necessarily translate to a reduced supply risk. An increase in renewable fuels in the supply portfolio brings different supply risks that may or may not be preferable to the risks associated with fossil fuel supply (e.g., supply disruptions due to fossil fuel infrastructure damage). Biomass availability is subject to natural temporal variation due to recurring periods of weather-induced crop water and temperature stress, and the resulting biomass yield fluctuations can create chronic bioenergy feedstock shortages. Of the two stresses, a lack of water is the more important one9 and is particularly relevant as cellulosic crops are usually assumed to be produced though dryland (rain-fed) farming. Drought is usually regional and yet can have national significance. Understanding the potential to supply a significant © 2014 American Chemical Society
amount of biomass-derived transportation fuel requires not only knowing the areas most likely to produce biomass and estimates of the yield under ideal conditions but also how the yield is impacted by weather-related risks, which vary across the country. In this study, switchgrass is used as a case study for the expected variability in cellulosic crop production. In this analysis, the United States is disaggregated to the state level, a politically relevant regional distinction, and weather data are used to estimate biofuel feedstock yields. Yield variability is assessed using historic and simulated future weather data and the switchgrass yield model developed by Grassini et al.10 Nair and colleagues have presented a review of many different crop models,11 highlighting that while model results are improving, there is still a need for better input data and better model calibration. Therefore, the overall yield trends observed in this article are reliable, but the specific yield values should not be taken as precise predictions. It is important to keep in mind that the annual yield time series output by this model vary as a result of changes in precipitation and temperature (which drive drought conditions) but not other factors such as an overabundance of local water or acute weather phenomena (e.g., hail) because the model is not equipped to deal with these situations (e.g., there is no “hail” module). The variability over time would be greater if these other causes of crop loss were included, and this could be a fruitful line of investigation for future work. Received: Revised: Accepted: Published: 7215
February 20, 2014 May 31, 2014 June 11, 2014 June 18, 2014 dx.doi.org/10.1021/es404174h | Environ. Sci. Technol. 2014, 48, 7215−7221
Environmental Science & Technology
■
Policy Analysis
Module 1: Crop Development Index. The crop development index is a dimensionless index that increases from 0 to 1 and is affected by the daily mean temperature in relation to cultivar-specific temperature parameters, including the optimal temperature and the temperatures above or below which no development occurs. Daily, incremental development is tracked and taken as input to various other modules. Module 2: Leaf Area Index (LAI) Expansion. The LAI is a dimensionless index that describes the fullness of the canopy. It affects rainfall partitioning in the soil water balance module and the rate of conversion of solar radiation to biomass (i.e., crop growth). The LAI value depends on the crop development index and a water stress factor. Module 3: Soil Water Balance. The soil water balance model is the most complex of the five modules. There are five submodules within this function. Rainfall can be intercepted by the canopy (a function of the LAI) and not be available to the soil layer. Heavy rainfall can result in water lost to surface runoff: the top layer can hold 110% of capacity right after rainfall, and the excess becomes runoff. Water evaporates from the upper layer of soil in a quantity that depends on temperature, solar radiation intensity, LAI, and previous water concentration in the upper soil layer. Water transpires through the plant during growth in a quantity determined by the existing biomass/development stage and temperature. Water for transpiration can come from either the top or bottom soil layer. Finally, excess water that may be in the top soil layer at the end of a day filters down to the bottom layer so that the top layer is not oversaturated. Water flow through these layers is illustrated in Figures S2 and S3 in the SI. Module 4: Crop Growth and Biomass Production. The mass of new biomass produced each day is a function of the amount of photosynthetically active radiation intercepted by the canopy (which is a function of the LAI), the rate at which the plant can convert radiation energy into biomass, and water and temperature stress factors. The growth rate parameter used is for the Blackwell switchgrass cultivar, which shows midrange performance. Module 5: Water and Temperature Stress Factors. Crop growth is retarded by insufficient water, or insufficiently high temperatures. Specific water stress factors are applied to the LAI function and the biomass growth function, and a specific temperature stress factor is applied to the biomass growth function. Supplemental calculation details that are necessary to complete the model but are not explicitly documented in the study by Grassini et al.10 or its references are taken from a publication on crop evapotranspiration published by the Food and Agriculture Organization of the United Nations (FAO).15 These include the necessary formulation and coefficient definitions for the Priestley−Taylor equation to calculate the daily reference evapotranspiration and a formula for calculating saturated vapor pressure. A key parameter assumed at the beginning of the model run is the fraction of available water-holding capacity filled at the date of growth initiation (FAWHCAGI), as an initial abundance of soil water can make up for some lack of rain during the growing season and a deficit of soil water can make that situation worse. This fraction varies from 0 to 1. The FAWHC AGI is assumed to decrease as the standard precipitation index (SPI) decreases. The SPI as a drought index is discussed by Guttman.16 A FAWHCAGI of 0.6 corresponds to the median precipitation scenario (the SPI
DATA AND METHODS Switchgrass Yield Model. The switchgrass growth model built for this study was defined by Grassini et al.10 and was published with the intention of generating annual crop yield values that could be used to inform energy policy models and discussions. The switchgrass growth model is composed of five interacting modules and operates at a daily time step in calculating biomass production. The key model parameters are summarized in Table 1. Inputs include daily rainfall; mean, Table 1. Summary of Key Input to the Switchgrass Growth Modela variable
magnitude
minimum growth temperature maximum growth temperature optimal growth temperature maximum LAI radiation use efficiency soil layer 1 (SL1) depth soil layer 2 (SL2) depth* soil available water-holding capacity, SL1 soil available water-holding capacity, SL2 extinction coefficient for incoming solar radiation fraction of available water-holding capacity at day of growth initiation, SL1* fraction of available water-holding capacity at day of growth initiation, SL2*
13 °C 42 °C 33 °C 10 4.7 g/MJ 150 mm 1450 mm 0.10 to 0.18 0.10 to 0.24 0.48 0.6 0.6
a Taken from Grassini et al.10 and modified if necessary (indicated by *) in order to better replicate empirical yield data.
high, and low temperatures; solar insolation; and information about soil characteristics and crop growth performance. Thirty years of daily data are used as the historical meteorological data set. Data from one city in an agricultural area from each state are taken, so one location is assumed to be representative of the whole state (as historical weather records are available for only one to a handful of locations in each state). The city selected from each state is listed in the Supporting Information (SI). The historical data are taken from the database assembled by the United States Department of Agriculture (USDA) for use in their weather generation model, named GEM (generation of weather elements for multiple applications).12 These data include daily observations for maximum and minimum temperature, precipitation, and solar radiation, and data covering 1961 to 1990 are available. Mean daily temperatures are calculated as averages of the high and low values. Soil characteristics for each state, specifically the available soil waterholding capacity, are calculated using data from the USDA’s Soil Survey Geographical Database (SSURGO).13 The available water-holding capacities for the top and bottom soil layers in this study are calculated using the 200 and 1500 mm depth layers in SSURGO, respectively, and averaging all of the data points from within each state at these depths. A complete table of values can be found in the SI. When daily biomass accumulation values (output in g/m2) are summed over an entire growing season, the result is an annual crop yield value. When many years are modeled, variability in the yield over time can be assessed. The Grassini model is not publically available as a stand-alone program or as code, so a model was built in MATLAB (version R2012b).14 The five interacting modules mentioned previously are defined as follows: 7216
dx.doi.org/10.1021/es404174h | Environ. Sci. Technol. 2014, 48, 7215−7221
Environmental Science & Technology
Policy Analysis
are added to these modified temperature and radiation residuals. The simulated data for this analysis are generated using methods for precipitation and temperature from Chen and colleagues,21 who implemented the Richardson model as described above (using a second-order Markov chain) and included a smoothing algorithm to reduce jaggedness in the mean temperature profiles. The jaggedness arises because the temperature means are calculated for two-week time periods, so each year is characterized by a set of 26 max/min temperatures. Without any modifications, there tends to be a jump between the temperature on the last day of one period and the first day of the next period. This particular implementation of the Richardson model was used because Chen and colleagues have graciously made the MATLAB code available online (the URL is listed in the SI). Means and standard deviations of solar radiation data are calculated using dry days, and a scaling factor is used to simulate radiation for wet days, following methods described by Zhang et al.22 In summary, the simulated weather data employed as input to the switchgrass yield model use the following: • a second-order Markov chain to generate precipitation occurrence patterns; • a two-parameter gamma distribution to generate precipitation magnitude data, smoothed to reduce jumps between successive two-week periods; • a conditional relationship between the maximum and minimum temperatures; • a scaling factor of 0.5 to scale the mean daily solar radiation on dry days for wet days. Given this weather simulation program from Chen and colleagues, certain parameters in the model are modified in order to produce a climate that is changed from the historical climate. The following scenarios are examined, with data presented in the form of box plots to illustrate key statistics: 1. Base-case weather, simulated using historical data (sources and locations described previously) and unmodified parameters (so that yields from simulated, modified climate data are not compared to historical data). 2. A higher-temperature case based on a scenario used in a future U.S. drought study.23 In the weather generation model, mean values for Tmin and Tmax time series (μtemp + 2) are modified as suggested in the study by Roy et al.23 3. A case with higher variability in temperature based on results shown in a study of temperature anomalies by Hansen et al.24 This is implemented though modified standard deviation values for Tmin and Tmax (1.2σtemp and 1.3σtemp) as shown in the Hansen study. 4. A scenario with higher dry- and wet-spell lengths. Rather than modification of precipitation amounts, which vary regionally (see ref 25), the impacts of increased dry- or wet-spell length are examined. This is accomplished by modifying the transition probabilities for precipitation occurrence in the Markov matrices, making it 10% (relative, not percentage points) more likely that there will be longer dry spells or wet spells. This is done not specifically following a procedure undertaken in another study or using empirical data found elsewhere but as a way to compare the yield sensitivity to spell length alongside temperature modifications.
median value is 0), and this is taken as the default value for the model runs in Grassini et al.10 It seems unlikely that even in a severe drought scenario the top two meters of soil would be completely dry, so the model was run assuming no precipitation at all over various FAWHCAGI values. For a range of FAWHCAGI values, the value for the lower soil layer at the time of growth does not drop below 0.1, while the top layer is constantly exhausted of moisture. As a result of this information, the FAWHCAGI is assumed to decrease linearly from 0.6 to 0.1 with a decrease in precipitation scenario likelihood and to increase from 0.6 to 1 with an increase in precipitation scenario likelihood. The validation of the model is presented in the SI. Weather Generation Model. To assess the impacts of a changing climate on switchgrass yields, this study simulated weather data using historical data for which certain parameters have been modified so that the output weather data are representative of a changed, future climate. As stated previously, this exercise is intended to show changing trends in yields, not to suggest exact magnitudes for the changes in variability or expected values for yields in any region. Weather simulation programs have been tools for modelers for the past 30 years. Many used for applications that requite daily data, such as this crop yield model, are built upon the stochastic simulation method for temperature, precipitation, and solar radiation suggested by C. W. Richardson, as described in refs 17 and 18 and more fully defined as the WGEN model by Richardson and Wright in ref 19. In most daily weather generation models, temperature and solar radiation data are conditioned on precipitation. Precipitation is modeled in two parts: whether there is any precipitation (dry or wet conditions; a binary random variable) and, when there is precipitation, how much falls. In this model, the wet or dry state is simulated as a Markov process; the status of the current day depends on the status of the previous day(s). In a first-order Markov model, precipitation status is conditioned only on the previous day; if it was dry yesterday, the likelihood that it will be dry today is P1 and the likelihood that it will be wet is (1 − P1), where P1 is estimated using historical data. Higher-order models have been used by more recent studies to more accurately reproduce long wet or dry spells (see the excellent discussion in ref 20). The quantity of precipitation on wet days is generally modeled using either an exponential function or a two-parameter gamma function, as both heavily weigh small amounts of precipitation but skew positively to allow for the low probability of much higher quantities of precipitation. With the precipitation time series generated, temperature and solar radiation data are then modeled using a first-order linear autoregressive model. Means and standard deviations for each temperature−time series, conditioned on precipitation, are calculated over some relevant time period (daily, weekly, monthly) across all years of historical data (i.e., two means for each two-week period are calculated: {μdry, σdry}week i and {μwet, σwet}week i). Time series of residuals are generated for maximum temperature, minimum temperature, and solar radiation using the calculated standard deviations. These residual values are modified by a lagged correlation matrix (modified on the basis of what temperatures and solar radiation happened the previous day) to which a simultaneous correlation matrix of normally distributed errors (modified on the basis of what the other meteorological values for the same day were) is added. Finally, the means and standard deviations calculated from the initial, historical data 7217
dx.doi.org/10.1021/es404174h | Environ. Sci. Technol. 2014, 48, 7215−7221
Environmental Science & Technology
Policy Analysis
Figure 1. Yield model variability results for the 19 states that are forecast to contribute at least 1% of national switchgrass production. Each dot is the mean yield over 30 years in one state, and error bars are the 10th and 90th percentile yields. Colors correspond to NOAA climate regions. The states are ordered by (A) decreasing crop production area, (B) increasing variance, and (C) decreasing mean.
terms of crop production area, as shown in Figure 1A, whereas other high-yield, low-variability regions (such as the Southeast) are not forecast to contribute as much to the switchgrass supply. Adding crop yield variability to the list of metrics by which cellulosic crop regions are evaluated might reorder the states in Figure 1A and could lead to a more stable expected energy crop yield. At present, the use of cellulosic feedstocks such as switchgrass is mandated at the national level through the RFS, so targets are met through an aggregated national cellulosic ethanol production volume. From this perspective, then, the national aggregate yield is of interest. Figure 2
By using the three modifications in some combination with each other, six scenarios (in addition to the base-case simulation results) are compared in the following section. The results ought only to be compared among themselves in order to get a sense of the sensitivity of crop yields to different changes in climate conditions and to see whether there are general trends to anticipate.
■
RESULTS Results using historical weather data for the 19 states forecast to contribute at least 1% of U.S. switchgrass production by mass are presented in Figure 1A (results for all of the continental states are presented the SI). The 19 states are ordered on the basis of how much crop production area is anticipated within the state and colored on the basis of climate region. Most switchgrass is expected to be grown in the South, Southeast, and Central regions of the United States (identified in the figure).26 The yield model results show mean yield values across the states ranging from about 12 to 19 Mg/ha. There is also substantial variability within each state; the range of yields representing the 80% confidence intervals (the 10th to 90th percentile range) plotted in Figure 1 cover at least 8 Mg/ha, and some span a range of more than 17 Mg/ha. For mean values in the range of 12 to 19 Mg/ha, this translates to a variability of 20 to 60% of the mean. A complete ordered list of states in Figure 1A is included in Table S1 in the SI. The maximum yields from this model are higher than those from another popular switchgrass yield model, Environmental Policy Integrated Climate (EPIC),27 because growth in the model from Grassini et al.10 is not limited by nutrient availability; if temperature conditions are optimal and water is available, biomass is produced at optimal rates, leading to high yields. These high yields are close to those reported from switchgrass test plots, some few of which had annual yields greater than 30 Mg/ha.28 On the basis of Figure 1B,C, which presents state yield distributions ordered by increasing variability and decreasing mean value, respectively, the Central states (which show high yields and comparatively low variability) are well-represented in
Figure 2. National average yields, an aggregation of state-by-state yields using the POLYSYS distribution of acreage. See Table S1 in the SI for the acreage per state.
presents the national aggregate switchgrass yield given historic weather data as well as curves showing the maximum and minimum yields from contributing states for comparison. Despite aggregation over many states, the fluctuation of the total yield over time is not so different from the yield variability in the individual states; the expected national yield over 30 years is 14.6 Mg/ha, and the 80% confidence interval of annual 7218
dx.doi.org/10.1021/es404174h | Environ. Sci. Technol. 2014, 48, 7215−7221
Environmental Science & Technology
Policy Analysis
the weather simulator do not represent the spatial correlation that is a common feature of many droughts.29 Correlations in yield are derived using the switchgrass model results from historical weather data, and the results for selected relevant states are presented in the SI. Some neighboring states demonstrate significant correlation, though not all do. All but one of the statistically significant correlation statistics are positive. The fact that most of the correlations are positive and are between about 0.4 and 0.8 suggests that there are not a lot of obvious opportunities to diversify planting locations in order to reduce the likelihood that reduced switchgrass yield is likely to occur regionally. This highlights the potential for widespread, simultaneous yield decreases and calls into question the ability of the U.S. biomass industry to consistently meet RFS biofuel targets.
yield values (11.2 to 17.8 Mg/ha) represents an uncertainty range of about 50% of the mean. The results for the base-case weather data (i.e., simulated weather using unmodified parameters calculated from historical data) are plotted in the top panel of Figure 3, with those for the
■
DISCUSSION The results in Figure 3 suggest that as the climate warms and becomes increasingly variable, switchgrass yields will tend to decrease in many states. Using historical average yield data in projections without adjusting for these climate changes will lead to overprediction of yields, and this shows cause for concern when past yield data are assumed to be representative of future yields in planning models. To ensure that sufficient biomass feedstock is available to supply biofuel conversion facilities, either large amounts of biomass must be kept in storage or additional acreage must be planted. Because of the low bulk density of biomass, very large scale multiyear storage is not practical.30 Storage does play an important role in eliminating negative price spikes in agricultural markets,31 and the limited ability to store biomass and the need to meet production volume mandates will tend to lead to high price volatility (another facet of energy security).32 Additional acreage does not address production variability but could be a strategy to raise total production amounts throughout the years, thereby reducing the likelihood of simply not having enough biomass for production facilities to operate at a profitable level. Of course, dealing with surplus production in high-yield years is a challenge, particularly with little storage capacity. Models combining stochastic yields by state (or other region) could be used to ensure that national feedstock yields are sufficient to meet EISA policy goals with some appropriate degree of confidence following the method outlined in ref 33. Consistency in dryland crop yields is likely to decrease in the future, as a number of studies have reported that drought severity and frequency is predicted to increase under the most likely future climate scenarios (see refs 25 and 34 for two examples). Tied to this are expectations of increasing and increasingly variable temperatures.24 Crop water stress will increase as a result of both meteorological changes and increases in water demand due to economic growth.23 This could lead to chronic multiyear biomass shortages, which would be very costly. Biomass energy forecasts are developed from models based on temporally and spatially aggregated yields.35−37 Policy recommendations informed by modelers’ predictions based on average productivity do not sufficiently address how policy should accommodate yield variation such as seen in the severe drought of the 1987−1988 crop year or how the developing biomass industry in, for example, Tennessee might deal with the resulting 90% reduction from the mean. For decisions that hinge on biomass yield, the models that predict yields must include stochastic analysis that accounts for expected changes in
Figure 3. Comparison of means and standard deviations for yields for all states for each of the seven simulated weather scenarios. The y-axis values are yields (Mg/ha), and the x-axis values are standard deviations. Solid red dots are the centroids of the data. The hollow blue dots show the centroid of the base case, which has been plotted on the other scenarios for comparison. Each gray point represents one of the 19 states.
modified climate scenarios shown in the six panels below. The centroid on each plot is shown relative to the centroid for the base case. Detailed figures similar to Figure 3 in which the data are labeled with the corresponding states are included in the SI. Increasing the standard deviation of temperature by 20% (“1.2sigma”) results in the smallest decrease in the mean yield. Increasing the standard deviation of temperature by 30% (“1.3sigma”) has a greater impact, reducing some of the highest mean yield values. Modifying the precipitation patterns by increasing the likelihood of multiday wet or dry spells by 10% (“Precip.”) tends to decrease the mean yields as well, though it also produces the highest single-state average yield value, and it tends to increase the standard deviation. Increasing the mean temperature by 2 °C (“T+2”) also tends to lower the yield. When the temperature is lower and more variable (“1.3sigma, T +2”), the points show a slightly reduced range of mean values and a slightly wider range of standard deviation values, though the centroid is not much changed from when either of those scenarios is run individually. Finally, modifying the precipitation patterns and increasing the expected temperatures (“Precip., T+2”) has the greatest effect on the mean and standard deviation statistics; the overall mean yield (centroid y value) is about 20% lower than for the base case, and the points are increasingly concentrated around the mean of the yield standard deviation (centroid x value). An important component of historical yields that is not maintained in yields estimated using simulated weather data is regional correlation. The meteorological variables generated in 7219
dx.doi.org/10.1021/es404174h | Environ. Sci. Technol. 2014, 48, 7215−7221
Environmental Science & Technology
Policy Analysis
(11) Nair, S. S.; Kang, S.; Zhang, X.; Miguez, F. E.; Izaurralde, R. C.; Post, W. M.; Dietze, M. C.; Lynd, L. R.; Wullschleger, S. D. Bioenergy crop models: Descriptions, data requirements, and future challenges. GCB Bioenergy 2012, 4, 620−633. (12) Natural Resources Conservation Service. Weather Generation Technology (GEM); USDA: Washington, DC. (13) Soil Survey Staff, Natural Resources Conservation Service. Gridded Soil Survey Geographic (SSURGO) Database; USDA: Washington, DC, 2012. (14) MATLAB, version R2012b; The MathWorks, Inc: Natick, MA, 2012. (15) Allen, R. G.; Pereira, L. S.; Raes, D.; Smith, M. Crop Evapotranspiration: Guidelines for Computing Crop Water Requirements; FAO Irrigation and Drainage Paper 56; Food and Agriculture Organization of the United Nations (FAO): Rome, 1998. (16) Guttman, N. B. Accepting the standardized precipitation index: A calculation algorithm. J. Am. Water Resour. Assoc. 1999, 35, 311−322. (17) Richardson, C. W. Stochastic simulation of daily precipitation, temperature, and solar radiation. Water Resour. Res. 1981, 17, 182− 190. (18) Larsen, G.; Pense, R. Stochastic Simulation of Daily Climate Data; Statistical Reporting Service (SRS) Staff Report No. AGES810831; USDA: Washington, DC, 1981; pp 1−64. (19) Richardson, C. W.; Wright, D. A. WGEN: A Model for Generating Daily Weather Variables; Publication ARS-8; USDA Agricultural Research Service: Washington, DC, 1984. (20) Wilks, D. S.; Wilby, R. L. The weather generation game: A review of stochastic weather models. Prog. Phys. Geog. 1999, 23, 329− 357. (21) Chen, J.; Brissette, P. F.; Leconte, R. A daily stochastic weather generator for preserving low-frequency of climate variability. J. Hydrol. 2010, 388, 480−490. (22) Zhang, Q.; Singh, B.; Gagnon, S.; Rousselle, J.; Evora, N.; Weyman, S. The application of WGEN to simulate daily climatic data for several canadian stations. Can. Water Resour. J. 2004, 29, 59−72. (23) Roy, S. B.; Chen, L.; Girvetz, E. H.; Maurer, E. P.; Mills, W. B.; Grieb, T. M. Projecting water withdrawal and supply for future decades in the U.S. under climate change scenarios. Environ. Sci. Technol. 2012, 46, 2545−2556. (24) Hansen, J.; Sato, M.; Ruedy, R. Perception of climate change. Proc. Natl. Acad. Sci. U.S.A. 2012, 109, E2415−E2423. (25) Strzepek, K.; Yohe, G.; Neumann, J.; Boehlert, B. Characterizing changes in drought risk for the United States from climate change. Environ. Res. Lett. 2010, 5, No. 044012. (26) Agricultural Policy Analysis Center. The POLYSYS Modeling Framework: An Overview; Institute of Agriculture, University of Tennessee: Knoxville, TN, 2010. (27) Thomson, A. M.; Izarrualde, R. C.; West, T. O.; Parrish, D. J.; Tyler, D. D.; Williams, J. R. Simulating Potential Switchgrass Production in the United States; Report PNNL-19072; Pacific Northwest National Laboratory: Richland, WA, 2009; pp 1−22. (28) Kszos, L. A.; Downing, M. E.; Wright, L. L.; Cushman, J. H.; McLaughlin, S. B.; Tolbert, V. R.; Tuskan, G. A.; Walsh, M. E. Bioenergy Feedstock Development Program Status Report; Oak Ridge National Laboratory: Oak Ridge, TN, 2000. (29) Douglas, E. M.; Vogel, R. M.; Kroll, C. N. Trends in floods and low flows in the United States: Impact of spatial correlation. J. Hydrol. 2000, 240, 90−105. (30) Hess, J. R.; Wright, C. T.; Kenney, K. L. Cellulosic biomass feedstocks and logistics for ethanol production. Biofuels, Bioprod. Biorefin. 2007, 1, 181−190. (31) Wright, B. D. The economics of grain price volatility. Appl. Econ. Perspect. Policy 2011, 33, 32−58. (32) Diffenbaugh, N. S.; Hertel, T. W.; Scherer, M.; Verma, M. Response of corn markets to climate volatility under alternative energy futures. Nat. Clim. Change 2012, 2, 514−518. (33) Mullins, K. A.; Griffin, W. M.; Matthews, H. S. Policy implications of uncertainty in modeled life-cycle greenhouse gas emissions of biofuels. Environ. Sci. Technol. 2011, 45, 132−138.
not only the mean yield but also the yield variability. A probabilistic approach is central to assessing the consequences of yield variability on model recommendations. Without an acknowledgment of variability, there is no basis upon which the emerging biofuel industry and biofuel policy makers can begin to consider and evaluate mitigation options and contingency plans.
■
ASSOCIATED CONTENT
S Supporting Information *
Further details of model validation and detailed statistics on state-by-state variability. This material is available free of charge via the Internet at http://pubs.acs.org.
■
AUTHOR INFORMATION
Corresponding Author
*Phone: (412) 268-2299; fax: (412) 268-3757; e-mail: wmichaelgriffi[email protected]. Present Address ⊥
K.A.M: Department of Bioproducts and Biosystems Engineering, University of Minnesota, St. Paul, MN 55108. Notes
The authors declare no competing financial interest.
■
ACKNOWLEDGMENTS The authors thank Chris Hendrikson for his assistance in study design and Ryan Noe for his assistance in extracting data from SSURGO. Funding was provided by the Center for Climate and Energy Decision Making through a cooperative agreement between the National Science Foundation and Carnegie Mellon University (SES-0949710), the USDA (NIFA Grant 201367009-20377), and the US DOE (EERE Grant DEEE0004397).
■
REFERENCES
(1) Rahall, N. Energy Independence and Security Act of 2007. Public Law 110-140, 2007. (2) Jansen, J. C.; Seebregts, A. J. Long-term energy services security: What is it and how can it be measured and valued? Energy Policy 2010, 38, 1654−1664. (3) Brown, S. P. A.; Huntington, H. G. Reassessing the Oil Security Premium; RFF Discussion Paper 10-05; Resources for the Future: Washington, DC, 2010. (4) Angelis-Dimakis, A.; Biberacher, M.; Dominguez, J.; Fiorese, G.; Gadocha, S.; Gnansounou, E.; Guariso, G.; Kartalidis, A.; Panichelli, L.; Pinedo, I.; Robba, M. Methods and tools to evaluate the availability of renewable energy sources. Renewable Sustainable Energy Rev. 2011, 15, 1182−1200. (5) Hughes, L. A generic framework for the description and analysis of energy security in an energy system. Energy Policy 2012, 42, 221− 231. (6) Greene, D. L.; Leiby, P. N.; Patterson, P. D.; Plotkin, S. E.; Singh, M. K. Oil independence: Achievable national goal or empty slogan? Transp. Res. Rec. 2008, 2017, 47−53. (7) Sperling, D.; Yeh, S. Toward a global low carbon fuel standard. Transp. Policy 2009, 17, 47−49. (8) Dale, B. E. Cellulosic biofuels and the road to energy security. Environ. Sci. Technol. 2011, 45, 9823. (9) McLaughlin, S. B.; Kszos, L. A. Development of switchgrass (Panicum virgatum) as a bioenergy feedstock in the United States. Biomass Bioenergy 2005, 28, 515−535. (10) Grassini, P.; Hunt, E.; Mitchell, R. B.; Weiss, A. Simulating switchgrass growth and development under potential and waterlimiting conditions. Agron. J. 2009, 101, 564−571. 7220
dx.doi.org/10.1021/es404174h | Environ. Sci. Technol. 2014, 48, 7215−7221
Environmental Science & Technology
Policy Analysis
(34) Dai, A. Increasing drought under global warming in observations and models. Nat. Clim. Change 2012, 3, 52−58. (35) U.S. Department of Energy. U.S. Billion-Ton Update: Biomass Supply for a Bioenergy and Bioproducts Industry; Perlack, R. D., Stokes, B. J., Leads; ORNL/TM-2011/224; Oak Ridge National Laboratory: Oak Ridge, TN, 2011. (36) A USDA Regional Roadmap to Meeting the Biofuels Goals of the Renewable Fuels Standard by 2022; USDA: Washington, DC, 2010; pp 1−21. (37) Renewable Fuel Standard Program (RFS2) Regulatory Impact Analysis; EPA-420-R-10-006; U.S. Environmental Protection Agency: Washington, DC, 2010.
7221
dx.doi.org/10.1021/es404174h | Environ. Sci. Technol. 2014, 48, 7215−7221
