Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
Energy Biosciences Institute, University of Illinois, Urbana, Illinois 61801 USA
Global Change Solutions LLC, Urbana, Illinois 61801 USA
Institute of Genomic Biology, University of Illinois, Urbana, Illinois 61801 USA
Department of Plant Biology, University of Illinois, Urbana, Illinois 61801 USA
Corresponding author. Present address: Department of Plant Biology, University of Illinois at Urbana–Champaign, 265 Morrill Hall, 505 S. Goodwin Avenue, Urbana, Illinois 61801 USA. E-mail: firstname.lastname@example.org
Widespread land use changes, and ensuing effects on ecosystem services, are expected from expanding bioenergy production. Although most U.S. production of ethanol is from corn, it is envisaged that future ethanol production will also draw from cellulosic sources such as perennial grasses. In selecting optimal bioenergy crops, there is debate as to whether it is preferable from an environmental standpoint to cultivate bioenergy crops with high ecosystem services (a “land-sharing” strategy) or to grow crops with lower ecosystem services but higher yield, thereby requiring less land to meet bioenergy demand (a “land-sparing” strategy). Here, we develop a simple model to address this question. Assuming that bioenergy crops are competing with uncultivated land, our model calculates land requirements to meet a given bioenergy demand intensity based upon the yields of bioenergy crops. The model combines fractional land cover of each ecosystem type with its associated ecosystem services to determine whether land-sharing or land-sparing strategies maximize ecosystem services at the landscape level. We apply this model to a case in which climate protection through GHG regulation—an ecosystem's greenhouse gas value (GHGV)—is the ecosystem service of interest. Our results show that the relative advantages of land sparing and land sharing depend upon the type of ecosystem displaced by the bioenergy crop; as the GHGV of the unfarmed land increases, the preferable strategy shifts from land sharing to land sparing. Although it may be preferable to replace ecologically degraded land with high-GHGV, lower yielding bioenergy crops, average landscape GHGV will most often be maximized through high-yielding bioenergy crops that leave more land for uncultivated, high-GHGV ecosystems. Although our case study focuses on GHGV, the same principles will be applicable to any ecosystem service whose value does not depend upon the spatial configuration of the landscape. Whenever bioenergy crops have substantially lower ecosystem services than the ecosystems with which they are competing for land, the most effective strategy for meeting bioenergy demand while maximizing ecosystem services on a landscape level is one of land sparing: focusing simultaneously on maximizing the yield of bioenergy crops while preserving or restoring natural ecosystems.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Land use change by humans profoundly affects biodiversity and ecosystem services including climate regulation, regulation of water flow and quality, habitat provisioning, regulation of pests and diseases, and cultural services (Foley et al. 2005, Millennium Ecosystem Assessment 2005). Land use change is the number one factor responsible for losses of species diversity and ecosystem services (Sala et al. 2000, Foley et al. 2005). Land use change is also a major driver in the climate system; gross CO2 emissions from tropical deforestation equaled ∼40% of global fossil fuel emissions from 1990 to 2007 (Pan et al. 2011), and biophysical effects of land use change can significantly modify climate on local and global scales (Foley et al. 2003, Bala et al. 2007, Anderson-Teixeira et al. 2012). Land use pressures and associated threats to biodiversity and ecosystem services are growing because of a growing world population, changing diets, and bioenergy demand (Foley et al. 2005, Tilman et al. 2009, von Braun 2009). In the face of these pressures, it becomes increasingly important that limited land resources be optimally allocated among food production, energy production, and protection of natural ecosystems. One question is key to making wise land use decisions: Is it preferable to grow “eco-friendly” crops, intentionally sacrificing maximum yield to protect biodiversity and ecosystem services (“land sharing”), or to maximize crop yields, thereby using less land and sparing more land from cultivation (“land sparing”; Green et al. 2005, Godfray 2011, Phalan et al. 2011)?
Debate over the relative advantages of land sharing vs. land sparing takes on a new urgency in the face of expanding bioenergy production. Given that global food demand and associated land use pressures are increasing (Tilman et al. 2011), there is limited potential for widespread production of bioenergy crops on existing cropland without triggering indirect land use change (iLUC) elsewhere in the world (Searchinger et al. 2008, Plevin et al. 2010). This is the case in the United States, where a substantial proportion of existing cropland would need to be devoted to bioenergy crops to meet current biofuel targets (Fargione et al. 2010, Smith et al. 2012). Some current cropland could be made available for bioenergy without raising food prices by increasing yield on existing cropland or reducing allocation to animal feed; however, this same mechanism has also been identified as important to feeding a growing world population while reducing rates of deforestation (Foley et al. 2011, Tilman et al. 2011). Although current cropland that is not essential to food production (e.g., “marginal” or excessive cropland) may be ideally suited for bioenergy production (Campbell et al. 2008), an alternative use would be restoring natural vegetation or allowing this land to undergo natural succession (Righelato and Spracklen 2007, Kendall and Chang 2009), as occurs when land is abandoned or entered into conservation programs (e.g., the Conservation Reserve Program [CRP] in the United States). Thus, when considering land use for bioenergy, a precautionary approach to evaluating the environmental impacts of bioenergy production entails viewing any large-scale bioenergy production as competing primarily with nonfood land over the long run and on a global scale.
When it comes to selecting the most ecologically sustainable bioenergy crop, two leading arguments support ideals at the “land-sharing” and “land-sparing” ends of the spectrum. The first argument, representing the land-sharing ideal, is that bioenergy crops should maximize ecosystem services while maintaining satisfactory yield. Specifically, the ideal bioenergy cropping system would be one with high biodiversity and ecosystem services, such as a low-input high-diversity (LIHD) mix of native prairie species whose high biodiversity enhances productivity and reduces the need for chemical inputs (Tilman et al. 2006). The land-sharing approach favors ecosystems with high value in regulating greenhouse gases (GHGs); for example, crops with the potential to sequester soil carbon are valued for the climate mitigation potential (Anderson-Teixeira et al. 2009, DeLuca and Zabinski 2011). Unfortunately, there is often a trade-off between crop yield and protection of biodiversity and ecosystem services (Krebs et al. 1999, Green et al. 2005, Zhang et al. 2010, Hoben et al. 2011), and the land-sharing ideal would entail sacrificing yield when such a trade-off exists. Therefore, crops that represent the land-sharing ideal typically require more land to meet a given demand and may thereby place more pressure on natural ecosystems (Green et al. 2005, Heaton et al. 2008, Hodgson et al. 2010). A second major drawback to this approach is that lower yields are less profitable, making this approach less economically viable (Jiang and Swinton 2009).
At the other end of the spectrum, the land-sparing ideal values maximization of crop yield, often at the expense of biodiversity and ecosystem services. This approach typically implies more intensive cultivation with higher chemical inputs, secondary concern for GHG regulation and other ecosystem services, and lower support of species diversity. For example, relative to LIHD prairie, corn (Zea maize) or perennial grass monocultures such as switchgrass (Panicum virgatum L.) or miscanthus (Miscanthus × giganteus) may have higher yields but increased need for chemical inputs and decreased support of species diversity (Tilman et al. 2006, Heaton et al. 2008). Although the high-yield ideal places environmental factors such as GHGV, support of biodiversity, and need for chemical inputs secondary to yield, it does not necessarily imply that all ecosystem services are sacrificed; for example, miscanthus shows promise of both high yield and favorable biogeochemical cycling (Beale and Long 1997, Heaton et al. 2009, Davis et al. 2010), albeit sacrificing native plant species diversity. The approach of using high-yielding bioenergy crops has the advantage of being more economically viable for landowners and bioenergy producers and is therefore viewed as more feasible by economists (e.g., Jiang and Swinton 2009, Jain et al. 2010). Although ecologists often shy away from this option out of concerns for biodiversity and ecosystem services, this approach has major potential benefits for ecological sustainability on the landscape level, in that it allows a given demand to be met with less land, thereby placing less pressure on natural ecosystems. The key question, then, is whether, and under what circumstances, it is preferable to adopt a land-sharing vs. a land-sparing approach.
Here, we develop a simple mathematical framework to address the question of whether ecological sustainability is better promoted through land sharing or land sparing. Specifically, we present a framework to calculate landscape-level ecosystem services as a function of (1) bioenergy crop yield and (2) the ecosystem services associated with the bioenergy crop and the unfarmed land with which it is competing for space. We then apply this framework in the context of five potential bioenergy crops in North America, considering one specific ecosystem service of great importance—climate protection through GHG regulation. We use our model to determine whether, and under what circumstances, land sparing or land sharing is preferable from a GHG perspective.
We begin with a landscape (total land area: LT) where bioenergy production may occur (Fig. 1A). For simplicity, we assume that abiotic variables (e.g., climate, soil type) are uniform across this landscape. The land is divided among three major uses: food production (LF), bioenergy production (LB), and other, unfarmed land (LO). For simplicity we assume that LF is constant, such that decisions regarding the amount and type of bioenergy production will affect only the nonfood (NF) landscape (LT−LF). Although bioenergy production is likely to affect food prices and eating habits (Mitchell 2008, Searchinger et al. 2008), implying that LF, LB, and LO are all interdependent, our analysis makes the precautionary assumption that trade-offs will ultimately occur between bioenergy production and uncultivated land. Assuming that the ecosystem services associated with each land type are independent of LF, the question of whether land sharing or land sparing is preferable from a landscape ecosystem services perspective will depend only upon how the nonfood (NF) landscape is divided between LB and LO.
In this model, changes in LB are directly offset by equal and opposite changes in LO. That is, the fraction of the NF landscape allocated to bioenergy production, fB = LB/(LT − LF), and the fraction left unfarmed, fO = LO/(LT − LF), sum to one. These fractions will be determined by the amount of land required to meet a certain bioenergy demand (D; in liters per year)—which we assume to be independent of the bioenergy crop selected or the amount of land required to grow it—relative to the ethanol yield (Y; in liters per hectare per year) of the bioenergy crop:
When [LT − LF] and D are held constant, fB and fO vary solely with respect to Y. Specifically, land allocation to bioenergy is inversely related to yield, and the amount of land left unfarmed increases with yield (Fig. 1B). Of course, the likelihood that land managers will actually grow bioenergy crops will depend upon a number of agronomic and economic factors—in particular, biomass yield and ensuing profitability; however, our model is specifically designed to address ecological, not economic, trade-offs.
Once we know how land is divided among bioenergy crops and other land, we can combine this with metrics of the ecosystem services of bioenergy crops (ESB) and natural ecosystems (ESN) to quantify the total ecosystem services associated with the NF landscape (ESNF):
Thus, the total ecosystem services associated with the NF landscape depend on (1) the fraction of land required for bioenergy crops (Eq. 1a) and (2) the relative ES values of bioenergy crops and the ecosystem that it replaces (Fig. 2).
This framework can be used to understand whether the total ecosystem services of the NF landscape are higher under land-sharing or land-sparing strategies (Fig. 2). Its application is straightforward when there are no significant interactions among ES and landscape configuration, such as edge or patch size effects.
We apply this model to address whether landscape-level GHG regulation is better under land-sharing or land-sparing strategies for a range of bioenergy crop–unfarmed ecosystem type combinations. Here, the ES in Eq. 2 is GHGV, or the total GHG benefit of maintaining an ecosystem over a multiple year time frame—or, conversely, the cost of clearing it (Anderson-Teixeira and DeLucia 2011). We determine the extent to which bioenergy crops conform to the land-sharing or land-sparing ideal based on their yield and GHGV.
Our case study concerns potential bioethanol production in the north central and eastern United States. We consider five bioenergy crops and five other ecosystem types with which these bioenergy crops may compete (Table 1).
Table 1. Description of ecosystem types considered in this analysis.
Our model is not sensitive to the size of land area considered, but to the ethanol demand intensity relative to the size of the NF landscape (D/[LT − LF]; Eq. 1). Ethanol demand intensity will vary spatially, being concentrated around ethanol production plants. We considered the hypothetical case of a cellulosic ethanol plant with a production, D, of 136 million L ethanol per year, which is the expected capacity of a commercial-scale cellulosic ethanol plant currently under construction in Highlands County, Florida, USA (Graham-Rowe 2011). We assumed that the biofuel production facility would draw feedstocks from within 25 miles (LT = 510 000 ha) and that 75% of the land within this area is devoted to food production (LF = 382 500 ha). This implied that the 136 million L of ethanol per year must be obtained from the remaining 127 500 ha, resulting in an ethanol demand intensity, D/[LT − LF], of 1067 L·ha−1·yr−1 for the NF landscape surrounding an ethanol plant. We used this as a default value for our analysis, noting that the answer to the question addressed here—whether land-sharing or land-sparing strategies are preferable from an ecosystem services perspective—is independent of the value of D/[LT − LF] (see Appendix).
Potential ethanol yields for each crop type were calculated based upon biomass yields and ethanol conversion efficiencies (Table 2). Yield estimates were obtained from the USDA for corn grain (USDA National Agricultural Statistics Service 2011) and from published data on yields of cellulosic sources (sources detailed in Table 2). Yield estimates were consistent with the ranges given by the U.S. DOE (2011). We assumed a conversion efficiency of 310 L ethanol/Mg dry matter for cellulosic ethanol and 420 L ethanol/Mg grain for corn ethanol (Perrin et al. 2009, Somerville et al. 2010). Substantial variation in yield was driven by a range of factors including geographic location and management practices (Scurlock et al. 2002, Heaton et al. 2004, Miguez et al. 2008, Wang et al. 2010). Our analysis considered both average and maximum yield estimates for each bioenergy crop. We assumed that bioenergy crops are unlikely to be cultivated under conditions in which below-average yields would be obtained.
Table 2. Yield, GHGV, and ecosystem properties used to compute GHGV for ecosystem types considered in this analysis; uppercase letters indicate data sources.
Table 2. Extended.
Ecosystem properties needed to compute GHGV (Anderson-Teixeira and DeLucia 2011) were obtained from the literature (Table 2). These variables include organic matter storage in various pools; annual exchange of CO2, CH4, and N2O between the ecosystem and the atmosphere; and CO2 emissions from management (e.g., on-site CO2 emissions from farm machinery). In general, data on these ecosystem properties were limited, often being restricted to one or two locations considered being well suited to cultivation of the crop and meeting the criteria outlined in Table 1. Therefore, although it was possible to estimate GHGV for “typical” cultivation of each crop (Table 1), further research will be required to relate variation in the ecosystem properties that drive GHGV to geographic location, management practices, or yield. GHGV was calculated as in Anderson-Teixeira and DeLucia (2011; Table 2) by summing (1) the release of GHGs that would occur through the oxidation of stored organic material upon clearing of the ecosystem and (2) the annual GHG fluxes that would be displaced by clearing of the ecosystem (i.e., net ecosystem exchange of CO2, annual CH4 exchange, annual N2O release, associated management emissions). Ecosystem–atmosphere GHG exchanges over a 30 year “ecosystem” time frame (TE) were translated into changes in atmospheric GHG concentrations and were multiplied by the radiative efficiency of each GHG to obtain total radiative forcing from GHGs. Cumulative radiative forcing was translated into CO2-equivalents over a 100-year analytical time frame (TA). This is analogous to the commonly used approach for computing GHG global warming potentials (Forster et al. 2007), which typically use TA = 100, but differs in that TE > 1 for GHGV whereas TE = 1 for global warming potentials. Selection of TE and TA is inherently subjective (Anderson-Teixeira and DeLucia 2011); the time frames selected here (TE = 30, TA = 100) are typical of biofuels life cycle analyses (Anderson-Teixeira et al. 2011). Although GHGV varies with both TE and TA (Anderson-Teixeira and DeLucia 2011), a sensitivity analysis revealed that the general conclusions of our analysis are not sensitive to the time frames selected (data not shown).
Having calculated ethanol yield (Y) and GHGV for each ecosystem, we then used Eqs. 1 and 2 to calculate the average GHGV of the NF landscape. These calculations focused on a scenario in which D/(LT− LF) = 1067 L·ha−1·yr−1 (previously described). We also conducted a sensitivity analysis showing how results vary across the range of potential values for D/[ LT– LF] (see Appendix).
We note that our analysis does not consider the full life cycle GHG effects of bioenergy production, which includes GHG benefits of fossil fuel displacement and GHG emissions from the ethanol production process (Davis et al. 2009). However, GHG benefits from fossil fuel displacement will be equal for all bioenergy crops because ethanol production (D) is held constant. Differences may arise from differences between grain- and cellulosic-ethanol production costs, co-product crediting, or differences in transport costs. The first two, which may influence comparisons between grain- and cellulosic- based ethanol but do not systematically bias the analysis in favor of land sharing or land sparing, are beyond the scope of this analysis. The latter introduces a systematic bias in favor of the land-sharing strategy, because lower Y implies higher LB and therefore greater transport costs, which are not factored into this analysis; however, feedstock transport contributes so little to the net GHG balance of bioenergy production systems (U.S. EPA 2010) that this bias is unlikely to alter conclusions.
Yields and GHGVs
Yields of bioenergy crops varied substantially both across and within crop types (Table 2). Based upon average yields, the ethanol productivity of bioenergy crops was ranked as follows: LIHD prairie < switchgrass < corn grain < corn grain + stover < miscanthus (Table 2). There was, however, substantial variation in the range of yield estimates. In particular, there remains considerably uncertainty as to the potential yield of perennial grass bioenergy crops throughout the United States. When ranked according to maximum reported yields, switchgrass surpassed corn grain. The observed range of yields implies that the ranking of achieved yields is likely to vary across locations.
Perennial grasses had substantially higher GHGVs than corn bioenergy systems (Table 2). All three perennial grasses had a GHGV exceeding 70 Mg CO2-eq·ha−1·30 yr−1, whereas both corn grain and corn grain + stover had GHGVs of approximately −50 Mg CO2-eq·ha−1·30 yr−1, indicating that the corn-based systems were a net source of GHGs to the atmosphere. This difference was driven mainly by the GHG benefits of C storage in the root systems and soil C accumulation in the perennial grasses, and by substantially higher N2O emissions from corn (Table 2).
Non-cropland ecosystems generally had higher GHGVs than bioenergy agroecosystems (Table 2). Exceptions were grassland pasture, which had a lower GHGV than all three perennial grasses, and CRP grassland, which had a lower GHGV than miscanthus. Although the grassland ecosystems had GHGVs of 200 Mg CO2-eq·ha−1·30 yr−1 or less, regrowing or mature forests had high GHGVs of 420 and 1070 Mg CO2-eq·ha−1·30 yr−1, respectively.
In bioenergy crops, ethanol yield and GHGV of the agroecosystem did not covary strongly (Fig. 3). Negative covariation, indicative of a sparing–sharing trade-off, occurred across the sequence LIHD prairie–switchgrass–corn (either grain only or grain + stover). Corn grain and corn grain + stover had different yields but similar GHGVs, implying a consistent benefit of corn grain + stover over corn grain alone. Miscanthus had the highest yield and GHGV, indicating unilateral GHG benefits relative to the others.
Nonfood (NF) landscape division and GHGV for focal scenario
When D/(LT-LF) = 1067 L·ha−1·yr−1, the fraction of the nonfood (NF) landscape required to support bioenergy demand ranged from 0.15 (0.08 at max Y) for miscanthus to 0.82 (0.56 at max Y) for LIHD prairie (Fig. 4A). Combining this result with the GHGVs of bioenergy crops and other land types (Table 2), we calculated the percentage change in GHGV for the average NF landscape (GHGVNF) that would occur as a result of bioenergy production for each bioenergy crop–unfarmed land combination (Fig. 4B). It should be emphasized that this change in GHGVNF does not represent the net life cycle GHG effect of bioenergy production, only the GHG effects of land use change. The net effects of allocating land to bioenergy production depended upon the GHGV of the bioenergy crop relative to that of the unfarmed land; GHGVNF increased when the bioenergy crop had a higher GHGV than the unfarmed land, and vice versa. For example, displacing pasture with LIHD prairie, switchgrass, or miscanthus increased the GHGV of the combined biofuel and nonfarmed landscape, whereas displacing native forest with any biofuel crop reduced the GHGV of the combined biofuel and nonfarmed landscape.
The relative advantages of land sparing and land sharing shifted as a function of the GHGV of the unfarmed land (Figs. 4B and 5). This is best illustrated by focusing on the GHGVNF values achieved across the sparing–sharing trade-off continuum of LIDH prairie, switchgrass, and corn (either grain or grain + stover; Fig. 3). When the GHGV of the unfarmed land was relatively low (pasture, CRP grassland), there was a clear advantage to land sharing. That is, across the sparing–sharing trade-off continuum of LIDH prairie–switchgrass–corn, GHGVNF was greatest for LIHD prairie, intermediate for switchgrass, and lowest for corn (Figs. 4B and 5), indicating the success of the land-sharing approach. When the GHGV of the unfarmed land was intermediate (native grassland), land-sparing and land-sharing strategies were approximately equal. Specifically, GHGVNF was similar, with overlapping confidence intervals, for switchgrass, prairie, and the two corn systems (Fig. 4B), indicating that the GHGV benefits associated with LIDH prairie and switchgrass were approximately balanced by the benefits of lower land requirements associated with corn. Finally, when the GHGV of the unfarmed land was high (regrowing or mature forests), a land-sparing approach had strong benefits over a land-sharing approach. In these cases, GHGVNF was driven primarily by the division of the landscape (Fig. 4A). The differences in GHGV among bioenergy crops were dwarfed by the large differences between GHGVs of forests and bioenergy crops. When the unfarmed land was mature forest, GHGVNF was reduced by 772, 426, 306, 208, and 141 Mg CO2-eq·ha−1·30 yr−1 when LIHD prairie, switchgrass, corn grain, corn grain + stover, and miscanthus were the bioenergy crops, respectively.
Of course, when comparing bioenergy crops with positive covariation between yield and GHGV (Fig. 3), the high-Y, high-GHGV bioenergy crop was generally associated with a higher GHGVNF (Figs. 4B and 5). Specifically, corn grain + stover consistently had a higher GHGVNF than corn grain alone. Miscanthus typically had the most favorable GHGVNF, with the notable exception that a land-sharing strategy was favored when GHGVO was low (Fig. 5). When the unfarmed land was pasture, which had the lowest GHGV of all unfarmed land types considered here, LIHD prairie had a higher GHGVNF than miscanthus (Fig. 4B). This occurred because 82% of the pasture was replaced with a higher-GHGV bioenergy crop in the case of LIDH prairie, compared to only 15% in the case of miscanthus (Fig. 4A).
Under many situations, yield uncertainty resulted in overlapping confidence intervals for fB or GHGVNF (Figs. 4 and 5). For example, although average-yielding switchgrass performed worse than corn grain or corn grain + stover when GHGVO was relatively high, switchgrass attaining maximum yield would outperform both of these feedstocks (Figs. 4 and 5). Of course, maximally yielding switchgrass would no longer fall along the sparing–sharing trade-off illustrated in Fig. 3, as its yield would exceed that of corn grain (Table 2). Thus, although the relative performance of the various bioenergy crops is subject to yield uncertainty, the general finding that the optimal strategy shifted from land sharing at low GHGVO to land sparing at high GHGVO values is robust.
As illustrated by a case in which GHGV is the ecosystem service of interest, the relative advantages of land sparing vs. land sharing depend upon the type of ecosystem with which the bioenergy crop is competing for land and its associated ecosystem services (Figs. 4 and 5). When the unmanaged land had a relatively low GHGV, land sharing yielded higher landscape ecosystem services than land sparing. In instances where bioenergy crops had higher GHGVs than the unfarmed land, the best crop was typically the one with the lowest yield, because this replaced the greatest amount of low-GHGV land. As the GHGV of the unfarmed land increased, the advantage shifted from land sharing to land sparing. When the unfarmed land had a much higher GHGV than the bioenergy crops, the most influential driver of landscape ecosystem services was the fraction of land left unfarmed, and land sparing was categorically the best strategy. These results parallel findings of studies comparing land-sharing and land-sparing strategies in a food production context, where GHG emissions or species diversity losses are minimized through a land-sparing strategy when the unfarmed land (e.g., forest) has a much higher ES value than the cropland (Burney et al. 2010, Hodgson et al. 2010, Godfray 2011, Phalan et al. 2011).
The sparing–sharing trade-offs observed for GHGV would be fundamentally the same for any ecosystem service that does not depend upon the spatial configuration of the landscape, being driven by the same underlying mathematical principles (Eqs. 1 and 2). A more complex model would be necessary to address species diversity or ecosystem services that depend upon landscape characteristics such as patch size and connectivity (Hodgson et al. 2010). However, for any ecosystem service that is minimally affected by landscape configuration (e.g., GHGV, nutrient retention, soil formation), the mathematical behavior of the model depends only upon land division (a function of Y; Eq. 1) and the ES values of the bioenergy crop and the ecosystem with which it is competing (Eqs. 1 and 2). Eqs. 1 and 2 can be manipulated algebraically to define a general rule for comparing the ecosystem service benefits of various bioenergy options. A land-sharing strategy (“SH” subscript) will be preferable to a land-sparing strategy (“SP” subscript) when the following inequality is true:
From this, we can draw three general conclusions. First, land sharing is favored when the unfarmed land provides minimal ecosystem services (i.e., when ESO is low), whereas land sparing is favored when the ecosystem service value of the unfarmed land is high (Figs. 4B and 5). Second, the greater the ecosystem service value of a land-sharing bioenergy crop is relative to that of a land-sparing crop (i.e., the higher ESSH/ESSP), the more likely that the land-sharing strategy will be preferable. Third, the more the yield of a land-sharing bioenergy crop is reduced relative to that of a land-sparing crop (i.e., the lower YSH/YSP), the less likely that the land-sharing strategy will be preferable.
The relative advantages of land sharing and land sparing depend upon the relationship between ethanol yield and ecosystem services (Eq. 3; Green et al. 2005). The steeper the trade-off (i.e., the more ESB is decreased per incremental increase in Y), the more likely a land-sharing strategy is to maximize landscape ecosystem services. For example, there is a steeper GHGV–Y trade-off between LIHD prairie and corn grain than between LIHD prairie and corn grain + stover (Fig. 3). When the GHGVNF values for these bioenergy crops in combination with native grassland are compared (Fig. 4B), a land-sharing strategy (LIHD prairie) is slightly favored in the case of the LIHD prairie–corn grain trade-off, whereas a land-sparing strategy (corn grain + stover) is favored in the case of the LIHD prairie–corn grain + stover trade-off. Therefore, when optimizing land use to meet the interconnected goals of bioenergy production and ecosystem services, it is important to consider the relationship between ESB and Y (Zhang et al. 2010). This will be particularly relevant when selecting optimal management practices for a specific bioenergy crop, and further research will be required to delineate the relationship between ESB and Y for the various bioenergy crops (for an analysis of this relationship for corn, see Zhang et al. 2010). Both ethanol yield and ecosystem services depend upon management practices such as fertilization; for example, N fertilization may increase yield (Heaton et al. 2004, Wang et al. 2010), but it will also reduce GHGV by increasing N2O emissions (Zhang et al. 2010, Hoben et al. 2011). Although agricultural intensification has generally benefited the global GHG balance by averting deforestation (Burney et al. 2010), excessive intensification may disproportionately reduce ES values in relation to Y (e.g., Hoben et al. 2011), and reduce the benefits of the land-sparing strategy. However, it is important to appreciate that although nutrients such as N are an essential component of the product in the case of food, feed and herbage crops, biofuels require only C, H, and O. Therefore, nutrients such as N potentially can be recycled, thereby partially mitigating the costs of intensification in bioenergy crops. This might be achieved by using perennial feedstocks such as LIHD prairie, switchgrass, or miscanthus that recycle most of their nutrients to roots and rhizomes before harvest (Beale and Long 1997, Heaton et al. 2009), or by returning vinasse and ash to the soil in the case of sugarcane ethanol (Ometto et al. 2009).
The outcomes of this analysis are sensitive to estimates of yield and GHGV, both of which are subject to uncertainty. Although experimental trials have produced yield estimates for cellulosic bioenergy sources at a variety of locations (Heaton et al. 2004, Miguez et al. 2008, Wang et al. 2010) and models have been used to estimate yields across the United States (Davis et al. 2011; Miguez et al. in press), uncertainty remains as to the actual yields that will be achieved in agricultural-scale settings and how these vary regionally. In our analysis, yield uncertainty commonly resulted in overlapping confidence intervals for fB or GHGVNF (Figs. 4 and 5). Much of this variation is driven by regional differences in yield potential, and the optimal crop will vary regionally. For example, whereas miscanthus has the highest observed yield and most favorable GHGV under conditions that are favorable for its growth (Fig. 3; Heaton et al. 2008), its relative advantage over switchgrass is diminished under colder and drier conditions (Heaton et al. 2004). Similarly, both miscanthus and switchgrass are expected to perform poorly in the northern Great Plains (Miguez et al. 2012), whereas native prairies are well adapted to the cold climate of this region. Of course, ongoing development of crops, agronomic techniques, and biofuel production technology is likely to alter the ethanol yield potential of all bioenergy crops in the future. With regard to GHGV, further research will be required to understand how the parameters that drive GHGV vary regionally and with respect to yield. Thus, although the framework presented here is ideally suited for addressing the general question of whether a land-sharing or land-sparing approach will be most favorable in terms of ecosystem services, a more detailed analysis will be required to determine how the relative benefits of different crops vary regionally.
The model presented here assumes a closed system where a fixed amount of land is required to meet food demands (Fig. 1) and where land use changes within the landscape will not affect land use elsewhere. In theory, it is applicable at any scale, from the individual farm to the globe. In our simplified scenario, food cropland area is constant while land allocation to bioenergy trades off with uncultivated land including pastures, abandoned cropland, and native ecosystems within the same region (North America). In reality, global interconnectedness of food, energy, and economic systems implies that bioenergy production may trade off with food production in one region, altering food prices and triggering iLUC elsewhere in the world, particularly in tropical regions (Searchinger et al. 2008, Fargione et al. 2010). If bioenergy were to trade off with food production, replacement of low-GHGV crops (Table 2; Anderson-Teixeira and DeLucia 2011) with higher-GHGV perennial crops (Table 2) would result in a climate benefit, with the greatest benefit obtained from a land-sharing strategy (Fig. 5); however, this benefit would be counteracted by any GHG emissions from iLUC. There is a high degree of uncertainty regarding the extent and location of iLUC; estimates of the extent of indirect land displacement range from 0.28 to 0.89 ha cleared per 1 ha cropland converted to bioenergy, with 19–52% of cleared land being forest (Plevin et al. 2010). Depending upon the amount and type of ecosystems cleared, the costs of iLUC may or may not outweigh GHG benefits from bioenergy crops, but ΔGHGVNF will always be reduced—more so under land sharing than land sparing, because the iLUC cost will be proportional to LB. A preliminary analysis using tropical ecosystem GHGV data from Anderson-Teixeira and DeLucia (2011) indicates that the optimal strategy will depend upon the extent of land use change and types of ecosystems cleared, with land sharing being favored when iLUC costs are relatively low (within the above ranges of uncertainty) and land sparing favored when they are relatively high. Realistic application of our model over a global spatial scale would require a more complex analysis including multiple types of ecosystems and modeling of the complex socioeconomic drivers of land use, the latter of which remains a key uncertainty in biofuels life cycle analyses (Plevin et al. 2010). However, increasing model complexity will not fundamentally alter the emergent mathematical behavior of the model and the resulting conclusions.
Effective application of this framework also requires farsighted analysis of current and future conditions. Although the relative advantages of different bioenergy crops do not depend upon future changes in land allocated to food production or bioenergy demand (Eq. 1, Fig. A1), several considerations may affect the long-term trade-offs of various bioenergy strategies. First, the model only gives accurate results for the time frame over which yields and ecosystem services are sustainable. An intensively managed, high-yielding bioenergy crop may provide the greatest benefits over the short term, but intensive management may deplete soil resources and reduce long-term yields (Blanco-Canqui and Lal 2007). Moreover, climate change and agronomic advances may significantly alter crop yields (Lobell and Field 2007) and ecosystem services of both bioenergy crops and natural ecosystems (Field et al. 2007, Xue et al. 2011). Second, effective decision making must take a long-term view of abandoned or idle land, considering not only its current ecosystem services, but also its potential ecosystem services were it to be put back into production, restored, or allowed to undergo natural succession (Kendall and Chang 2009). Third, long-range economic feedbacks may alter the future benefits of either the land-sparing or land-sharing scheme. Land-sparing strategies may be ineffective in practice if agricultural subsidies incentivize surplus production or if the “spared” land is used to produce other agricultural commodities (Ewers et al. 2009). The land-sharing strategy may be vulnerable to the danger that, after putting a large fraction of available land into production of a crop with high ecosystem services, land managers may easily be tempted to intensify production to increase profits. Finally, effective implementation of either strategy requires consideration of local institutions, governance, and social dynamics (Bawa et al. 2004, Fischer et al. 2011). Appropriate analysis of long-term considerations such as these is crucial for creating effective strategies to maximize ecosystem services on bioenergy landscapes over the long run.
Our findings have several implications for the current debate on bioenergy production. Our results demonstrate that the relative advantages of land sparing and land sharing depend upon the type of ecosystem with which the bioenergy crop is competing for land; as the GHGV value of the unfarmed land increases, the preferable strategy shifts from land sharing to land sparing (Figs. 4 and 5). On one end of the spectrum is the proposal to grow high-ES bioenergy crops on marginal land or land unsuited for arable agriculture, thereby avoiding competition with food and improving ES values of the land (Tilman et al. 2006, Campbell et al. 2008, Fargione et al. 2010, Nijsen et al. 2012). Because degraded or abandoned land will generally have low ecosystem services, the land-sharing strategy may be preferable in this case (Fig. 4B). This strategy should, however, be weighed against an alternative strategy of growing the highest-yielding bioenergy crops on less of this land while restoring the remainder to high-ES unmanaged ecosystems (Kendall and Chang 2009). When considering productive, nondegraded land, the best strategy will generally be to select and develop bioenergy crops and management techniques so as to maximize yield while actively preserving or restoring natural land. A prime example of such a strategy is the “Midway Strategy” for the development of the sugarcane ethanol production system in Brazil, which involves improving the ethanol yield of sugarcane while preserving and restoring Atlantic Forest within the landscape (Buckeridge et al. 2012). On a policy level, GHG benefits of bioenergy production may best be ensured by coupling bioenergy production to mechanisms for forest protection (Thomson et al. 2010), such as REDD+ (United Nations mechanism to reduce deforestation and forest degradation; UNFCCC 2008). This conclusion is consistent with other research showing that land sparing in conjunction with protection of natural ecosystems is the most effective strategy for protecting biodiversity and ecosystem services (Burney et al. 2010, DeFries et al. 2010, Godfray 2011, Phalan et al. 2011). In sum, the economically favored strategy of maximizing yield is generally compatible with the ecologically preferable strategy of maximizing the yield of bioenergy crops while actively preserving or restoring natural land.
This research was supported by the Energy Biosciences Institute (University of California–Berkeley and University of Illinois) and by the BP (British Petroleum) Energy Sustainability Challenge.