Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

A comparative analysis of the carbon intensity of biofuels caused by land use changes

A comparative analysis of the carbon intensity of biofuels caused by land use changes Introduction As a result of concerns about energy security and climate change, interest has grown worldwide in renewable energy sources because they are not depletable and produce less greenhouse gas (GHG) emissions than fossil fuels. Biofuel (ethanol, diesel or kerosene) production from organic matter is seen as one of the strategies to reduce fossil fuel consumption and GHG emissions (Righelato & Spracklen, ). Many industrialized countries have established policies to increase the share of biofuels in the total energy production. For example, the European Union (EU) aims to achieve a minimum of 10% renewable fuels in the transport sector by 2020 as part of the overall EU renewable energy target of 20% of its final energy use by 2020 (EC, ). Biofuels can provide unique economic benefits compared with other alternative energy options (Farrell et al ., ). However, there have been concerns that the increased use of agrichemical products and the global land use changes (LUC) associated with the expansion of biofuel production could negatively impact the food supply and biodiversity (Harmon et al ., ; Righelato & Spracklen, ; Scharlemann & Laurance, ). Among these concerns, LUC carbon intensity of biofuels – defined as the amount of CO 2 emitted per unit of biofuel produced – has prompted an intense debate in the scientific and the political communities. LUC are changes in the areal extent of a particular type of land use over a given time period and within a given spatial entity. LUC has impacts (direct and indirect) on soil, water, biodiversity and climate change, and is thus related to many environmental issues of global importance. Direct land use change (dLUC) refers to the impact caused by switching a particular land area from some previous use (or disuse) to cultivated land for biofuel crop production. Indirect land use change (iLUC) impacts, also known as ‘leakages’, refer to impacts associated with LUC elsewhere as an unintended consequence of biofuel crop production. Recent studies have demonstrated that converting carbon rich ecosystems (e.g., forests, grasslands) to cultivated lands for biofuel crop production resulted in little GHG emission savings or even lead to a ‘carbon deficit’ that would take years or decades to be repaid (Fargione et al ., ; Gibbs et al ., ; Searchinger et al ., ). These and other studies have enhanced our understanding of the LUC implications of biofuels. However, the mechanisms leading to their conflicting outcomes have not been analysed. Also, the LUC carbon intensity of biofuels has not yet been inventoried or compared. This study analyses existing studies that assess the LUC‐GHG implication of biofuel production. Our aim is to provide a structured overview and a comparison of studies on LUC‐carbon intensities associated with biofuels that have emerged during the last 3 years. First, the techniques used to quantify the LUC‐carbon intensities are examined, summarized and compared. Second, the mechanisms that lead to diverging results are investigated. Finally, unanswered questions are identified and recommendations for future research are suggested. Literature selection of primary database We used a systematic review methodology to identify and to select studies included in this comparative analysis. Systematic reviews represent a rigorous, objective and transparent approach to synthesize and evaluate scientific evidence while minimizing bias (Oxman et al ., ). The systematic review methodology evolved within the medical field to support the development of clinical and public health practise guidelines and to formulate statements of scientific consensus (Briss et al ., ; Guirguis‐Blake et al ., ). Systematic reviews are now increasingly being performed and published for energy (Kubiszewski et al ., ) and bioenergy related topics (Whitaker et al ., ; Njakou Djomo et al ., ). The systematic review approach was chosen in this study to explain the reason for differences among studies and to identify priority areas for further research. Drawing on the principles and procedures of systematic reviews, we queried the ISI Web of Knowledge, Web of Science and Science Direct databases for studies published between 2008 and 2010 on the LUC implication of biofuels. The limitation to the last 3 years is due to the fact that the issue of LUC‐carbon intensity of biofuels is very recent and appeared in the scientific literature only since 2008. Our search did not yield any studies published on the issue before 2008. A pair‐combination of the keywords ‘land use change’, ‘greenhouse gas emissions’ and ‘payback time’ was used together with the term ‘biofuel’ for the search. The search was limited to studies published in English. Relevant publications were identified and their bibliographies were examined for additional articles. We also contacted key authors of most publications for additional information where necessary. All full‐length peer‐reviewed studies and reports that assessed LUC carbon intensities, as well as the carbon payback time of biofuels, and that provided numerical data were included in the analysis. Studies reporting only on LUC or on GHG emissions, as well as review studies were excluded. The titles and abstracts of 43 potential studies and reports were screened and their methodological quality and validity were assessed for eligibility. Among these, 16 studies were excluded because they were duplicates or they did not properly state the methodology used to obtain the data. The remaining 27 studies were examined in detail. Twelve studies were further excluded because the carbon intensity was not reported or could not be estimated, or the articles did not report original data. Finally, 15 relevant studies were retained for the data extraction Inventory of the reviewed studies Data extracted from the 15 primary studies were tabulated and presented in a descriptive form (Table ). Of the 15 studies on LUC of biofuels, 2 addressed the direct land use change (dLUC), 9 assessed the indirect land use change (iLUC) and the remaining 4 dealt with both dLUC and iLUC. While all studies reported on dLUC or iLUC carbon intensities, only five studies calculated the carbon payback time (Table ). The payback time is the period it takes the biofuel system to overcome the carbon debt incurred when land is converted and starts providing GHG benefits. Land use change ( LUC ) mechanisms, models, feedstocks, carbon pools and stocks of each of the reviewed studies LUC mechanisms Energy product Model/Approach Crop yield (t ha −1 yr −1 ) Area (10 6 ha) Land class and fraction (%) Carbon pool Carbon stock (t C ha −1 ) Geographic location References dLUC Bioethanol Biodiesel Excel (spreadsheet) M (10), SC (69) PO (20), SB (3) na F (100), P (100), G (100), C (100) AB, BB, SOC F: 192–201, PL: 943, G: 23–37, AC: 2–19 BR, ID, ML, USA Fargione et al . ( ) dLUC Biokerosene IPCC (simplified) JP (4) na F (100), P (100) AB, BB, SOC F: 168–245, G: 23–41, C: 22, G: 21–23 BR Bailis & Bake ( ) dLUC, iLUC Bioethanol EPPA‐TEM (GE model) M (7), SC (30) SG (10) na F, P, G na na Global Melillo et al . ( ) dLUC, iLUC Bioethanol DAYCENT (simplified) M (9), SB (3) na F (100), G (100) AB, BB, SOC F: 108–200, G: 51–87 USA Kim et al . ( ) dLUC, iLUC Bioethanol Biodiesel IMPACT (PE model) SC (68), SB (1) 16.5 F (73), O (27) AB, BB, SOC F: 201, R: 20, S: 45, O: 23 BR Lapola et al . ( ) dLUC, iLUC Bioethanol Biodiesel RISK‐ADDER (Simplified) M (8), W (4), SB (3) 15.9 F (3), G (55), S (5), C (33) AB, BB F: 266, G: 69, S: 134, AL 55 AR, BR, EU, IN, USA Fritsche et al . ( ) iLUC Bioethanol FAPRI‐CARD (PE model) M (11), SG (18) 10.8 F (52), G (48) AB, BB, SOC F: 165–313, G: 21–83 BR, CN, IN, RoW, USA Searchinger et al . ( ) iLUC Bioethanol GTAP‐Bio (GE model) M (9) 4.2 F (19), P (81) AB, BB, SOC F: 166, G: 27 Global Hertel et al . ( ) iLUC Bioethanol RFMI (Simplified) M (10) na F (15;50), G (85;48), W (0;2) AB, BB, F: 96–178, G: 20–55, WL: 273–819 Global Plevin et al . ( ) iLUC Bioethanol GTAP (GE model) M (9) 2.0 F (33), G (67) AB, BB, SOC F: 86–189, G: 12–57 AR, BR, EU, IN, USA Tyner et al . ( ) iLUC Bioethanol FASOM‐FAPRI (PE models) M (11), SB (3) 1.4 F (19), G (29), S (38), SL (14) AB, BB F, G, S, SL BR, RoW, USA EPA ( ) iLUC Bioethanol CARD‐GreenAgSIM M (11) 1.3–6.1 F (52), G (48) AB, BB, SOC F: 165–313, G: 21–83 BR, CN, IN, RoW, USA, Dumortier et al . ( ) iLUC Bioethanol Biodiesel MIRAGE (GE model) M, PO, SF, SB na F (41), S (33), G (12), C (14) AB, BB F: 151–225, S: 52–58, BR, CN, EU, IN, ML,RoW Al‐Riffai et al . ( ) iLUC Bioethanol Biodiesel IMAGE (Integrated) SC, B, SB, W, RS, PO na F (51), G (48) D (0.8), T (0.3) AB, BB, SOC F: 69–509, G: 30–162 T: 52–179, D: −21 to −3 AR, BR, EU, IN, ML, PK, USA Overmars et al . ( ) iLUC Bioethanol Biodiesel GLOBIOM (PE model) M, SC, W RS, SB 13 F, G, C, O AB, BB, SOC na EU, Global Havlik et al . ( ) AB, above‐ground biomass; AL, arable land; AR, Argentina; BB, below‐ground biomass; BR, Brazil; CN, China; C, cropland; dLUC, direct land use change; D, desert; EU, European Union; F, forest; G, grassland; iLUC, indirect land use change; ID, Indonesia; IN, India; JP, jatropha; M, corn; ML, Malaysia; na, not assessed; O, other vegetation; PL, peatland; PK, Pakistan; PO, palm oil; RoW, rest of the world; RS, rapeseed; S, savannah; SB, soybean; SC, sugarcane; SF, sunflower; SG, switchgrass; SL, shrubland; SOC, soil organic carbon; USA, United States of America; W, wheat; WL, wetland. This number represents the average of the three feedstocks (corn, wheat, soybean). Land use change ( LUC ) carbon intensities, carbon payback times and amortization periods of different biofuels reported in the reviewed studies. A positive LUC carbon intensity value suggests an increase in GHG emissions, whereas a negative LUC carbon intensity suggests a net carbon sequestration. Emissions from LCA are excluded from estimates of LUC carbon intensities, whereas they are included in estimates of the payback times Type of fuel Amortization period (years) LUC carbon intensity (g CO 2e MJ −1 ) Carbon payback time (years) References dLUC iLUC dLUC + iLUC dLUC iLUC dLUC+ iLUC Bioethanol 20 na 26–154 na na 12–70 na Overmars et al . ( ) na 16–79 na na na na Al‐Riffai et al . ( ) −27 to 27 0–34 −27 to 61 na na na Fritsche et al . ( ) 30 na 104–111 na na 52–167 na Searchinger et al . ( ) na 32–34 na na 22–27 na Havlik et al . ( ) na 28–45 na na na na EPA ( ) na 27 na na 28 na Hertel et al . ( ) na 21–118 na na 32–183 na Dumortier et al . ( ) na 21–142 na na na na Plevin et al . ( ) na 13–19 na na na na Tyner et al . ( ) −52 to 11 181–190 129–201 na na na Melillo et al ., ; 50 0–25 na na 0–93 na na Fargione et al . ( ) 34 327 361 4 40 44 Lapola et al . ( ) na 10–61 na na 12–70 na Overmars et al . ( ) −24 to 27 31–57 7–84 na na na Melillo et al . ( ) 100 −2 to −1.9 6–32 4–30 na na 12–31 Kim et al . ( ) −7 to 0 1–7 −6 to 7 na na na Melillo et al . ( ) Biodiesel 20 na 46–67 na na na na Al‐Riffai et al . ( ) na 30–204 na na 15–106 na Overmars et al . ( ) −98 to 31 0–41 −98 to 72 na na na Fritsche et al . ( ) 50 35–481 na na 37–423 na na Fargione et al . ( ) na 12–82 na na 15–106 na Overmars et al . ( ) 37–243 626–1434 663–1677 7–35 122–211 129–246 Lapola et al . ( ) Biokerosene 20 −27 to 101 na na na na na Bailis & Bake ( ) dLUC, direct land use change; iLUC, indirect land use change; na, not assessed. Values refer to the 25% of the theoretical potential. Land use change: Mechanisms, drivers, causes and consequences The dLUC occurs when a portion of a natural ecosystem (e.g., forest or grassland), a portion of cropland or a portion of unused land (e.g., degraded or abandoned land) is converted into cultivated land for biofuel crops production (Fig. , top panel). The dLUC occurs at the time of the energy crop production; it is certain and it is directly caused by the production of energy crops. It links the conversion of a specific portion of land in a given biofuel chain to resulting environmental impacts. Schematic diagrams describing the direct ( dLUC ) and indirect land use changes ( iLUC ). The panels illustrate the dLUC (top), local iLUC (middle) and global iLUC concepts (bottom). The symbol β represents the portion of additional land required due to expansion of biofuel crops, whereas the symbol Ω represents the fraction of natural ecosystems converted to cropland as a result of the conversion of cropland to land for the production of biofuels. The iLUC occurs when a portion of a natural ecosystem, or a portion of degraded or unused land, or a portion of other land dedicated to other uses is converted to new cropland to produce food, fibre or feed that were displaced by the expansion of biofuel crop production (Fig. , middle and bottom panels). It can occur locally when, for example, the expansion of biofuel crops to cropland induces the conversion of natural ecosystems elsewhere in the same country to cropland (Fig. , middle panel). It can also occur globally when the expansion of energy crops to cropland reduces food supplies in the near‐term future and induces – through market mechanisms – the conversion of natural ecosystems elsewhere in the world into cropland (Fig. , bottom panel). The iLUC can also be induced in situations where the dLUC for energy crop production is absent, but the use of existing crops is diverted to energy crops (Searchinger et al ., ; WBGU, ). The causal link in iLUC is weaker than in the dLUC case. The action of increased biofuel production and the LUC impacts in the iLUC case are mediated through prices. The location of iLUC cannot be predicted with certainty as it is the result of complex interactions of (international) market mechanisms, agricultural policies and other events that influence future patterns of land use. The time horizon, space, mediation, probability and causality characteristics distinguish iLUC from dLUC. Drivers of LUC can be grouped into: (i) proximate drivers, such as agricultural expansion, timber extraction and infrastructure development and (ii) ultimate drivers, such as economic, institutional, technological and demographic factors (Geist & Lambin, ; Lambin et al ., ). Taken alone, none of these drivers can adequately explain LUC at the proximate level, but all together, they explain more than 90% of LUC observed worldwide (Kim et al ., ). Land conversion for biofuel crop production can have significant impacts on the environment. Both dLUC and iLUC can affect the GHG balance of biofuel chains due to GHG emissions from changes in above‐ and below‐ground carbon, soil organic carbon, litter and dead wood (Searchinger et al ., ; Kim et al ., ; Plevin et al ., ). Land conversion can also affect air quality by altering emissions and changing conditions that influence reaction rates, transport and deposition. For example, tropospheric ozone is very dependent on changes in vegetation cover and in biogenic emissions (Foley et al ., ). LUC also affects the soil by altering the soil environment, which in turn affects microbial growth and decomposition processes that transform plant derived carbon into soil organic matter and carbon dioxide (Cotrufo et al ., ). Furthermore, LUC can disturb the surface water balance and the proportion of precipitation into evapotranspiration, runoff and groundwater flow. In fact, river discharge and surface runoff increase when a natural ecosystem is cleared (Sahin & Hall, ; Costa et al ., ). Finally, LUC can also reduce biodiversity through the modification, the fragmentation and the loss of habitats, the degradation of soil and water and the overexploitation of native species (Pimm & Raven, ). Modelling approaches for LUC Modelling dLUC The approaches or models used to estimate the LUC carbon intensity of biofuel production systems vary between both LUC mechanisms (i.e., dLUC and iLUC) and across studies. The dLUC carbon intensity can be estimated using mathematical (simulation) tools, such as the IPCC tool or the DAYCENT model (Table ). IPCC approach The IPCC Tier 1 method (IPCC, ) requires estimates of carbon in living biomass stocks prior to conversion, based on estimates of the area of land converted during the period between land use surveys. The difference between initial and final live biomass carbon pools is used to calculate changes in carbon stocks due to land use conversion (IPCC, ). The Tier 1 method is intended for national scale inventories in the absence of region specific data. This method requires no new data collection as it uses default data from the IPCC guidelines. However, many biofuel crops (e.g., sugarcane) are not represented as default value options. The method also assumes that N 2 O emissions are solely a function of nitrogen inputs to the soil. Furthermore, the method does not account for the carbon fluxes, and the simplification of assumptions involved in the method increases uncertainties. Despite these weaknesses, the IPCC Tier 1 approach is transparent and based on the best available science. The method is robust when used to obtain average values across a large agricultural landscape. Agroecosystem models The DAYCENT model is the daily time step version of the CENTURY model used to simulate the dynamics of soil organic carbon and related nitrogen pools (Del Grosso et al ., ; Parton et al ., ; Del Grosso et al ., ). One of the strengths of the DAYCENT model is that it can simulate ecological disturbance and management practices, such as fire, grazing, cultivation, organic matter addition and/or fertilizer addition (Kim & Dale, ). Unlike the IPCC Tier 1 approach, the DAYCENT model is able to predict the soil organic carbon along with the carbon in above‐ and below‐ground biomass, and nitrous oxide emissions from the soil (Kim et al ., ). The DAYCENT model also accounts for factors that influence emissions, such as weather, soil types and previous land history, thus making estimates more reliable. Moving from the Tier 1 method to DAYCENT improves the accuracy and reduces the uncertainty of assessments, but it also increases the complexity and the costs of monitoring. Modelling iLUC Unlike dLUC, which could be verified through experiments, the iLUC is a diffuse market response. The iLUC cannot be observed at a specific location or region, but can only be modelled using complex tools. Several models have been used in the analysed studies to estimate the iLUC carbon intensities of biofuels (Table ). These models can be grouped into agroeconomic models, combined models, biophysical models and deterministic (or simplified) models. They are briefly described, summarized and compared below. Agroeconomic models Agroeconomic models are indispensable tools in the preparation and negotiations of agricultural policy decisions. They include models, such as the general equilibrium (GE) and partial equilibrium (PE) models. A model is in general equilibrium when all markets are modelled explicitly and considered to be in equilibrium at every time step. In partial equilibrium, only a subset of the markets is considered and the remaining markets are parameterized. General equilibrium model ( GE ) The Global Trade Analysis Project (GTAP) model (Hertel et al ., ; Tyner et al ., ) is a static multiregion GE used to assess the iLUC created by biofuels production. The GTAP model utilizes economic data to estimate the potential GHG emissions, as well as other impacts of prospective biofuel technologies or policies. It takes into account interactions between the economic sectors of a country or of several regions. The main characteristic of the GTAP model is its economy wide coverage. The model contains a biofuel sector, includes conventional energy crops (e.g., corn) and is able to capture the relationship between changes in land area and yield increase. However, the GTAP model does not provide a detailed sectorial analysis of the economy because of its complex structure and aggregation. The GTAP model is data intensive; its specification as well as its related database does not include soil carbon stocks. Furthermore, biofuel coproducts are not yet specified in the GTAP model. Another model used to assess the iLUC effects of biofuel production is the Modelling International Relationship in Applied General Equilibrium (MIRAGE) model (Al‐Riffai et al ., ). The MIRAGE model is a dynamic, multisector, multiregion GE model that relies on the GTAP database. Similar to the GTAP model, the MIRAGE model uses economic data to simulate the iLUC associated with biofuel expansion. The model was recently modified to capture the interactions between biofuel expansion, agricultural markets and LUC. Key modifications incorporated into the model include the integration of biofuel sectors (e.g., biodiesel, bioethanol) and improvements in the modelling of the energy sector. The MIRAGE model uses most recently available data to model LUC. The MIRAGE model considers fertilizers used and coproducts that arise from biofuel production. However, the MIRAGE model does not include second generation biofuels or the non‐CO 2 GHGs. Also land classes, such as marginal and fallow lands are excluded from the MIRAGE model. A particular advantage of the GE models is that they incorporate all sectors of the economy. They are thus well suited to depict interactions between agriculture and other sectors of the economy. However, GE models do not capture all important characteristics of the agricultural economy. Moreover, GE models are very data intensive and the amount of data is determined by the level of disaggregation (e.g., countries or regions; activities or commodities) and the theoretical structure (homogenous/heterogeneous). Consequently, if data sets are not aggregated to a greater extent, the modelling task may become unmanageable. Furthermore, in GE models (e.g., GTAP), land data are usually expressed in monetary values, not in physical dimensions (e.g., hectare). Consequently, some assumptions need to be made to relate the physical quantity of land with a given volume of biofuels expressed in monetary units (e.g., dollars). Partial equilibrium models ( PE ) The Food and Agricultural Policy Research Institute (FAPRI) model (Searchinger et al ., ), the International Model for Policy Analysis of Agricultural Commodities and Trade (IMPACT) model (Lapola et al ., ) and the Global Biomass Optimization (GLOBIOM) model (Havlik et al ., ) are all PE models used in some of the analysed studies to assess the iLUC carbon intensity of biofuels. The FAPRI model is a recursive dynamic, multimarket and nonspatial PE model able to capture the technical, economic and biophysical relationships among key variables within a particular commodity and across commodities (Edwards et al ., ). The FAPRI model is often used to analyse the impacts of a policy proposal relative to a baseline scenario. Historical data from different sources (e.g., FAO, USDA) as well as current academic research data are used to calibrate the model. The FAPRI model can predict the location and the amount of LUC. The FAPRI model treats allocation of land as the result of the farmer's choices, assuming that farmers maximize their net return per hectare of land. The FAPRI model covers all major temperate crops for all crop producing and consuming countries. However, the modelling of biofuels in this model is restricted to first generation biofuels. Second generation biofuels are not included in the model. The IMPACT model is a recursive dynamic PE of the agricultural sector model designed to examine alternatives for global supply, demand and trade of food. The IMPACT model contains a water use module that balances water availability and uses within various economic sectors (e.g., agricultural, industrial and residential) at the regional and global scales. The integration of a water module into the IMPACT model allows it to explore the relationship between the increased biofuel crop production and the demand for irrigation water (Rosegrant et al ., ). Data from the UN and FAO statistics are often used to calibrate or validate the model. The GLOBIOM model is a bottom‐up and recursive dynamic PE model that integrates the agricultural, forestry and biofuel sectors. The general structure of this model is similar to that of the Agricultural Sector and Mitigation of Greenhouse Gas (ASMGHG) model. In the model the market is represented by implicit product supply functions based on detailed, geographically explicit, Leontief production functions and explicit, mostly constant elasticity product demand functions (Havlik et al ., ). A noticeable added‐value of the GLOBIOM model is its ability to model both carbon stocks and flows. The model also allows for accounting and taxing of the major GHG emissions related to agriculture and forestry. However, the model is restricted to agriculture and land use sectors and does not include downstream processes, such as biofuel conversion processes (De Vries, ). Therefore, there is no link between the agricultural and the energy sectors in the GLOBIOM model. The PE models use observed data to determine the quantity and the location of LUC due to the diversion of food crops to feedstock for biofuel production. One of the advantages of PE models is that they offer a more complex and accurate depiction of the agricultural sector. However, PE models focus on one specific economic sector only (i.e., the agricultural sector) and do not explicitly link to other sectors of the economy or to countries or regions other than the one(s) under investigation (Kretschmer & Peterson, ). Moreover, PE models do not capture production increases achieved by increasing agricultural yields or avoiding logistic losses and market distortion, such as taxes (Fritsche et al ., ). Compared with PE models, GE models allow a coherent representation of the economy, but the scale and traceability issues are less detailed than in PE models. Both GE and PE models do not consider factors, such as land price speculation, market information or land tenure in LUC decisions (Econometrica, ). Furthermore, GE and PE models do not always give a description of major biogeophysical constraints on agricultural production (e.g., soil conditions or climate) (WBGU, ). Finally, environmental aspects are insufficiently covered in both GE and PE models. Thus, combining those models with biophysical or other models is crucial and helps to overcome their individual weaknesses. Combined models The Emissions Prediction and Policy Analysis (EPPA) model (Paltsev et al ., ) has been combined with the Terrestrial Ecosystem Model (TEM) (Melillo et al ., ) to assess the iLUC carbon intensity of biofuel production. The EPPA model is a recursive dynamic multiregional GE model designed to analyse the impact of climate change policies on the global economy and on GHG emissions. In contrast, the TEM model is a process‐based ecosystem model used to examine patterns of land carbon dynamics across the globe. In the combined EPPA‐TEM model, land use estimates from EPPA are downscaled and organized for use in the TEM model (Melillo et al ., ). Competition in land use demand between recreation and biofuels is well examined in the EPPA‐TEM model. Both first and second generation biofuels are also included in the EPPA‐TEM model. Moreover, in the EPPA‐TEM model fossil fuels are treated as exhaustible resources, with diverging costs of exploitation and scarcity rents. This allows the EPPA‐TEM model to endogenously determine fossil fuel prices. However, the impacts of LUC are not fully assessed due to the high level aggregation of the agricultural sectors in the EPPA‐TEM model. Moreover, the EPPA‐TEM model does not include coproducts in the assessment of iLUC associated with biofuels. The US Environmental Protection Agency (EPA) also combined two PE models to estimate the iLUC implications for domestic and international commodity prices. In the EPA study the Forest and Agriculture Sector Optimization Model (FASOM) (Darius et al ., ), a dynamic PE agricultural sector model, is combined with FAPRI (FAPRI, ) to estimate the LUC carbon intensity of biofuel production (EPA, ). In the coupled FASOM‐FAPRI model, FASOM is used to estimate the domestic LUC, whereas FAPRI is used to estimate the international LUC. The outputs from the FASOM model are used as inputs into the FAPRI model. An advantage of such combination is the high resolution of the PE models as compared with the GE models. However, the combination of these two PE models obscures the link between the agricultural market and the rest of the economy, particularly the energy sector. An agricultural outlook model, such as the one developed by the Center for Agricultural and Rural Development (CARD) has also been combined with the Greenhouse gases from Agriculture Simulation (GreenAgSim) model. In this combined approach, the CARD model is used to assess the impact of biofuel policy changes and energy price increases on land conversion, whereas the GreenAgSim model is used to evaluate emissions from land conversion and agricultural production (Dumortier et al ., ). Although model combinations can in some cases help to overcome individual model weaknesses, they increase the model complexity and reduce its applicability. In fact, as combined models rely on historic time series, they cannot deal well with long‐term scenarios. Also, the underpinning causal mechanisms of LUC are insufficiently addressed or overlooked in combined models. Moreover, combined models are very data intensive, and it is very difficult to perform a proper calibration and validation. The failure of these models to include long‐term scenarios coupled to their inability to handle causal mechanisms and data issues reduces their utility. Bio‐physical models A bio‐physical model, such as the Integrated Model to Assess the Global Environment (IMAGE) has also been used to assess iLUC of biofuels (Overmars et al ., ). The IMAGE model is a non‐economic model that simulates the physical carbon flow on a global aggregate scale. It is used to explore the long‐term dynamics and interconnections of global change and to evaluate specific issues of sustainable development in the broader context of global socio‐ecological evolution. The strength of this model is that it allows an accurate assessment of the spatial structure of land use by describing the hierarchical organization of land use. However, the IMAGE model does not well represent the behaviour of individual sectors of the economy that are important for the understanding of the iLUC (e.g., the energy and agricultural sectors). Deterministic models A number of simplified approaches have also been developed and used in recent years to assess the LUC implications of biofuel production. One such approach is the ‘Risk‐Adder’ developed by the Institute for Applied Ecology (Öko‐Institut) in Germany (Fritsche et al ., ). The ‘Risk Adder’ approach is a simple and transparent methodology that uses statistical data to estimate the iLUC carbon intensity associated with biofuel production. In this approach, the modelling of agricultural markets is avoided by assuming that: (i) the global potential of GHG emissions from iLUC is a proxy for the current patterns of land use to produce traded agricultural commodities; (ii) for the near future, observed trade trends can be used to obtain the pattern of global trade in agricultural commodities. Another simple approach used to estimate the iLUC carbon intensity of biofuel systems is the Reduce‐Form Model of iLUC (RFMI) (Plevin et al ., ). Here, the model complexity is reduced by using the net displacement factor (NDF). The NDF is the ratio of hectares of land brought into cultivation anywhere in the world to the hectares of land dedicated directly to additional biofuel crops. The NDF includes the joined effects of (i) price induced yield increases, (ii) the relative productivity of land converted to cropping, (iii) price‐induced reductions in food consumption and (iv) the substitution of crop products by biofuel coproducts (Plevin et al ., ). The NDF is the most influential parameter in the RFMI and is calculated for a specific time period using crop yields for that period. However, as for GE and PE models, predictions of NDF have not yet been tested or verified, and there is no empirical evidence for choosing one number over another. Proponents of simplified approaches argue that they are simple, transparent and easy to implement in Excel spreadsheets. They also require less data compared with GE and PE, and provide a quick, rough estimation of the iLUC carbon intensity. However, many simplifications increase the inaccuracy. Furthermore, by simplifying the characterization of complex market links, these approaches miss some market feedbacks that drive the iLUC (Yeh & Witcover, ). Synthesis of results: a wide variation in LUC carbon intensities Among the 15 selected studies there is considerable variation in the estimates for both the dLUC and iLUC carbon intensities as well as in the carbon payback time associated with biofuel production. Depending on the amortization period (i.e., the time range over which the carbon cost and benefits will be considered), and the type of land and feedstock used, the dLUC carbon intensity ranged from −52 to 34 g CO 2 MJ −1 for bioethanol. Negative dLUC carbon intensity (i.e., net sequestration) values were associated with cellulosic crops (e.g., poplar) grown on cropland, whereas positive dLUC carbon intensity (i.e., net emissions) values were linked to sugarcane cultivated on rangeland in Brazil. For biodiesel, the carbon intensity values ranged from −98 to 481 g CO 2 MJ −1 . Here, the conversion of degraded land to palm derived biodiesel resulted in a net carbon sequestration in the soil, whereas the conversion of rainforest to palm derived biodiesel resulted in carbon emissions to the atmosphere. The dLUC carbon intensity of biokerosene ranged from −27 g CO 2 MJ −1 when Jatropha is planted on former pasture land to 101 g CO 2 MJ −1 when it is planted in cerrado woodlands (Table ). These results indicate that the dLUC‐carbon intensity depends on the type of biofuel feedstocks grown, and on the type of land converted. With regard to the iLUC carbon intensity, the values ranged from 0 to 327 g CO 2 MJ −1 for bioethanol, whereas for biodiesel the values ranged from 0 to 1434 g CO 2 MJ −1 depending on the feedstock used, on the type of land displaced and on the amortization period (Table ). For both bioethanol and biodiesel, zero iLUC carbon intensity values were reported when biofuel crops were grown on marginal or degraded lands as there was no new land brought into production. However, both bioethanol and biodiesel turned out to have very high carbon intensities when the displaced activities (e.g., rangelands) were relocated to forests. No studies reported on a negative iLUC carbon intensity. This is because most studies assumed that the displaced activity will move to carbon rich lands, such as forest and grassland. However, the iLUC carbon intensity is not necessarily positive: estimates of iLUC carbon intensity may be negative when, for example, the expansion of biofuel crops (e.g., palm oil) to grassland induces the conversion of marginal or degraded lands elsewhere in the world to grassland. However, such scenario was not investigated in the studies analysed here. The variation was large, suggesting that there is no consistency in the estimates of iLUC carbon intensity among the reviewed studies. Within each individual study and regardless of the feedstock used, the iLUC carbon intensity tended to be small with a longer amortization period. Estimates of both dLUC and iLUC for a given amortization period illustrated that in some cases the carbon intensity was larger for the iLUC than for the dLUC regardless of the type of biofuels (Table ). This means that in some cases, iLUC emissions could account for a significant part of the total LUC carbon intensity of biofuels. The comparison of bioethanol and biodiesel indicated that bioethanol incurred less carbon debt than biodiesel (Table ). However, this latest result should be interpreted with caution as it inherently depends on different assumptions and data used in the individual studies. Further research is needed to provide a clear answer on the land use carbon intensity of these two biofuels. The total LUC (i.e., dLUC + iLUC) carbon intensity ranged from −27 to 361 g CO 2 MJ −1 for bioethanol and from −98 to 1677 g CO 2 MJ −1 for biodiesel depending on the feedstock and amortization period (Table ). For bioethanol these values were about −29% (for sugarcane grown on degraded land) up to 384% (for sugarcane planted on cropland) of that of gasoline, which has a life cycle emission profile of about 94 g CO 2 MJ −1 (Farrell et al ., ). Similar results were obtained for biodiesel compared with conventional diesel. Thus, it was not clear from the 15 analysed studies whether bioethanol (or biodiesel) reduced or increased the GHG emissions relative to gasoline (or diesel) at the current state of knowledge. Note that the current estimate of gasoline's GHG performance is likely poorer than this baseline (i.e., 94 g CO 2 MJ −1 ) as it excludes, for example, emissions due to the military control over oil reserves in the Middle East (O'Rourke & Connolly, ). Liska & Perrin ( ) argued that the GHG intensity of gasoline would double if indirect military emissions were included in the gasoline's life cycle. Figure summarizes the life cycle emissions of bioethanol and biodiesel from different biofuel crops over different amortization periods (20, 30 and 50 years) and compares them with those of conventional gasoline and diesel. In most cases, adding life cycle GHG estimates to those of iLUC carbon intensity nearly cancels out the GHG benefits of bioethanol as compared with gasoline (Fig. , top panel) or of biodiesel relative to conventional diesel (Fig. , bottom panel). High yielding biofuel crops, such as oil palm, sugarcane and cellulosic crops have better overall GHG performances due to less land required than low yielding ones. Likewise, cellulosic bioethanol (or biodiesel) showed some advantages compared with other conventional biofuel crops, such as corn, wheat, sugarcane or oil palm. However, none of the biofuel crops was likely to achieve the 35% (increasing to 50% in 2017) reduction required by the EU directive by 2012 (EC, ) (Fig. , top and bottom panels). The substantial variation observed in Fig. reflected the assumptions made in the individual studies, the type of land displaced, the variation in the production and distribution methods of these biofuels and the different management conditions under which these biofuel crops can be grown. Life cycle greenhouse gas ( GHG ; on a logarithmic scale) emissions including estimates of indirect land use ( iLUC ), carbon intensities from different feedstock sources for bioethanol relative to gasoline (top panel) and for biodiesel relative to diesel (bottom panel). The bars represent the ranges of GHG emissions (in g CO 2 MJ −1 ) of each feedstock over a given amortization period. The numbers behind the biofuel feedstock species (20, 30, 50 years) refer to the amortization period. The horizontal lines represent the reference fuel (i.e., gasoline or diesel) and the European 35% reduction target by 2012. A cradle‐to‐plant approach was adopted for estimates of LCA . Values for GHG emissions per unit of energy for different feedstocks were derived from Fritsche et al . ( ). These values (all in g CO 2 MJ −1 ) are the following: corn: 65; sugarcane: 26; wheat: 45; soybean: 20; palm oil: 43; sunflower: 18; rapeseed: 40; and short‐rotation crops: 14. To put dLUC and iLUC intensities into perspective with respect to savings in emissions due to the substitution of biofuels, the carbon payback time has been adopted as a convenient indicator by several authors (e.g., Fargione et al ., ; Searchinger et al ., ). The payback time for bioethanol ranged from 0 to 93 years for the dLUC and from 12 to 183 years for the iLUC. For biodiesel, the payback time ranged from 7 to 423 years for dLUC and from 15 to 211 years for iLUC (Table ). As in the case of the dLUC and iLUC carbon intensities, significant variation was also observed in the estimates of the carbon payback times across the reviewed studies. Causes of the wide variation in LUC carbon intensity Land classes and proportion converted The differences in the estimates of LUC carbon intensity are caused by a number of assumptions. Different land classes have been considered in the analysed studies, but the most considered ones in all studies were: forest, grassland and cropland. Some studies assumed that the displaced land will be relocated in one land class, such as forest or grassland, whereas others assumed that this relocation will be distributed unequally over several land classes (Table ). These different assumptions on the land classes and on the apportioning partly explain the divergent outcomes in the estimates of carbon intensity among the analysed studies. For a robust modelling of iLUC, proper assumptions on land should account for all land classes available for agriculture and not merely some particular types of land as is the case in most analysed studies. Different techniques have been used to allocate land among land classes. In some models (e.g., GTAP), land allocation is based on decision theories and is governed by prices, whereas in other models, linear programming techniques are used as a land class allocation method. Linear programming allocation techniques provide the optimum area for land use, but do not provide information on the spatial distribution of results. In contrast, decision‐based theory allocation methods provide continuous land consideration maps and allow the consideration of socio‐economic factors. A spatial allocation method has also been tested and proposed (Heiderer et al ., ). Appreciating and adopting one allocation method could reduce inconsistencies in estimates of iLUC carbon intensity. However, whatever allocation method is chosen, sensitivity analyses should always be carried out to assess the effects of the chosen allocation method on the iLUC carbon intensity. Management practices Management practices are important for the estimates of dLUC and iLUC carbon intensity of biofuels. Good farming practices (e.g., no‐tillage) can result in carbon stored in organic matter in the soil, whereas poor farming practices (e.g., ploughing, subsoiling, harrowing) can result in significant emissions and loss of soil carbon. For example, no‐tillage practices increase the cumulative GHG benefits by 15% over 100 years in the grassland conversion case, and by 17% in the forest conversion case as compared with current tillage practices (Kim et al ., ). Considering which management practice is used is important to the outcomes of the LUC carbon intensity. Given that a variety of management practices is used by farmers, it may seem justifiable for the modelling of iLUC to compute the current tillage practices for the baseline scenario, and other management practices (e.g., no‐tillage and no‐tillage plus cover crops) for the projection scenario. In this way, effects of management practices on GHG emissions from iLUC could be quantified and policy measures to promote these management practices as a way to reduce the carbon footprint of crops could be suggested. Carbon stored in the harvested wood products The carbon stored in the harvested wood products at the time of land conversion also has a significant influence on the estimates of LUC carbon intensity. Some studies assumed that this carbon is emitted into the air as CO 2 (Searchinger et al ., ; Overmars et al ., ). This assumption holds if the land is cleared using fire. However, this is not always the case and land may be cleared using other techniques as well. In such cases, the harvested wood might be used as solid fuel in coal power systems (Kim et al ., ), or stored in short‐life wood products like paper and cardboard or long‐life wood products such as timber. In both cases, a carbon credit is gained and must be accounted for in the estimate of LUC carbon intensity. Allocation based on a system expansion could be used to estimate the carbon gained in the case of coal power systems. For the long‐life wood products, estimates of the net carbon gained should be based on stock differences and should account for the fraction of carbon in long‐life wood products in use, as well as in landfills after 100 years across wood product categories, such as softwood timber, softwood pulp, hardwood timber and hardwood pulp. No carbon benefits should be allocated to short‐life wood products as they are net emitters of GHG. Sources of uncertainty Crop yield Yields of energy crops are critical to land use considerations as they define the area of land needed to support the projected biofuel production targets. If yields are small, more land area will be needed to grow a certain amount of biofuel crops, thus releasing more carbon. In contrast, less land will be needed if yield improvements are large (Mathews & Tan, ). Historically, about 80% of the increase in crop production was attributed to improvements in crop yield, and 20% to the expansion of the harvested area. As a consequence, agricultural areas have expanded by only 5% since 1970 (Smith et al ., ). Projecting future land use is confounded by uncertainties about energy crop yields and about crop responses to future climate change. Higher yields can be expected under ideal climatic and agronomic conditions, through breeding or technological improvements including changes in agricultural practices and the use of fertilizers (Ewert et al ., ). Yield may also increase under future atmospheric CO 2 concentrations (Liberloo et al ., ). However, whether crop yields will increase, decrease or remain unaltered in the future depends on agronomic limits and on further progress in crop improvements. Data on crop yield are very crucial for the modelling of iLUC. They are also sources of inconsistencies among the analysed studies (Table ). One way to reduce these inconsistencies would be to use the normalized yield (i.e., the ratio of the actual yield to the regional average) in the baseline scenario assessment of the iLUC associated with biofuels. This approach helps to minimize the impact of different soil properties and farming practices, which vary across geographical regions. It also enables the identification of yield improvements. Coproducts Coproducts have a substantial impact on the requirement for land. The increased availability and the use of coproducts, such as Distillers Dried Grains with Solubles (DDGS) may reduce the pressure on land for animal fodder and for extensive grazing. Indeed, the land requirement for wheat ethanol is reduced from 0.40 ha tonne −1 ethanol to 0.03 ha tonne −1 ethanol (i.e., by 93%) when DDGS are used as a substitute for both soy meal and wheat in the production of animal feed (Lywood et al ., ). This means that more land will be needed for biofuel crops that have few or no coproducts, whereas less land will be needed for biofuel crops that yield large amounts of coproducts. However, the interaction between the production of coproducts and of biofuel production is changing and data on the actual substitution are scarce. Moreover, the feeding value of many coproducts remains uncertain, and there is a limit to the amount of coproducts that can be added to animal fodder. This is partly due to the high nutrient content (e.g., P, K and S) in coproducts relative to the original feedstock, which may have a negative effect on the livestock at high concentrations (Schauer et al ., ). These uncertainties may be exacerbated if new coproducts become available in the future. A better knowledge and a thorough assessment of the feeding value of co‐products will allow for more accurate estimates of the coproduct credits and reduce the uncertainty about the net land required for biofuel production. Many by‐products are deliberately generated in the biofuel production process. However, not all of these by‐products can be classified as coproducts. A market value could be used to differentiate the coproducts of biofuel production and processing from wastes. If a coproduct cannot find a market niche, it has no economic value and should therefore be classified as waste or residue. The economic value of coproducts should not be in any case larger than the main product (i.e., biofuel). Biofuel producers that are unable to provide proof of the sale of their coproducts to fodder industries, to farmers or to energy companies should not claim GHG benefits due to coproduct credits. This differentiation between co‐products and wastes could reduce the uncertainty associated with the estimates of iLUC and could help to only promote energy crops that have a well known market for their coproducts. Carbon stocks of different vegetations The amounts of carbon of various terrestrial carbon pools and land cover types are another source of uncertainty in the estimates of LUC‐carbon intensity. The carbon stock varies by type of vegetation and eco‐regions, and the reported values ranged from 69 to 509 t C ha −1 for forest, from 12 to 162 t C ha −1 for grassland and from 45 to 134 t C ha −1 for savannah (Table ). These variations are attributed to large errors in the spatial distribution of the vegetation biomass as well as to discrepancies in the estimates of land cover and land use change (Houghton et al ., ). More precise information on both changes in area and changes in carbon stocks can thus help to minimize the uncertainty in the LUC carbon intensity of biofuels. Yield elasticity on price The yield elasticity (σ) on the prices is an important parameter of uncertainty in the iLUC estimates. The yield elasticity is the ratio of the change in yield to the change in market prices for a given crop. A high yield elasticity means a reduction in the amount of land needed to cultivate a given biofuel crop, and thus, a reduction in iLUC. Assumptions about the yield elasticity vary considerably between models and studies. For example, the GTAP model assumes a much higher elasticity (σ = 0.25) compared with the FASOM model (σ = 0) or the FAPRI model (σ = 0.01). The yield elasticity also varies with time. For example, the long‐term yield elasticity in the FAPRI model is six times higher than the short‐term yield elasticity (i.e., σ = 0.01) assumed in the EPA analysis (EPA, ). Deciding on which yield elasticity value should be used is difficult as the literature on this issue is rather polarized. For example, in the USA some authors suggest a high response of σ = 0.76 (Houck & Gallagher, ), whereas others support with σ = 0.22 a low response (Lyons & Thompson, ) or even no response. The lack of data on values of yield elasticity for different countries or regions in the world further complicates the issue. As far as the models are concerned, we argue that it should not matter if the demand and price signals are driven by biofuels or other uses of the crop. For example, in the USA the amount of land dedicated to major crops has actually declined over time, whereas the demand for and the outputs of major crops have increased remarkably. So it appears – for the USA at least – that in the past the iLUC was negative: an increased demand resulted in a reduced land area in crop production. Lessons to be learned Models are needed for the assessment of the iLUC carbon intensity because iLUC cannot be determined by experiments. In the case of dLUC, the quantification of carbon intensity is straightforward if the amount of carbon in above‐ and below‐ground biomass as well as in the soil organic matter is known. Both the dLUC and iLUC can significantly alter the GHG benefits of biofuel production (Table ), although some studies (Fritsche et al ., ; Hertel et al ., ) suggest that their effects may be small. Several models and approaches to quantify the iLUC related GHG emissions of biofuels have been published over the last 3 years (Table ). The approaches used vary greatly, ranging from GE to PE models, and simplified models. However, very few models have examined past relations between land and biofuel production, or have examined whether there is any empirical evidence for or against iLUC from the historical data. Predictions of iLUC could be empirically tested by examining past relationships between biofuels production, export patterns from grain producing and importing countries and land use changes (Kim & Dale, ). Key elements for an accurate assessment of carbon gain or loss for an area due to its conversion into cultivated land for biofuel include: the amount of biomass above‐ and below‐ground before (and remaining after) the conversion; the carbon content in these biomass stocks; the different management techniques; the time path of the change in the soil carbon after conversion until a new equilibrium is reached; the influences of climate, temperature, soil quality and rainfall on these key elements. Unfortunately, they were not quantified in every study that estimated the iLUC carbon intensity. Existing iLUC models or approaches predict only a positive iLUC carbon intensity (i.e., increasing emissions). This may not always be the case. The models could also predict a negative iLUC carbon intensity (i.e., net sequestration) if assumptions, such as those that relocate the displaced activities (e.g., grazing) onto degraded lands, unused or marginal lands were investigated in previous studies. Likewise, a negative iLUC carbon intensity could be obtained in case increased yield due to biofuel demands frees some agricultural lands, which could later be returned either into grassland or into forest. This reversion will then increase the soil carbon and lead to a negative iLUC. Many LUC models or approaches assumed that LUC is driven by agricultural expansion and in particular biofuel expansion. However, the issue of LUC is quite complex and not solely driven by agricultural expansion, but rather by a multitude of processes and factors associated with development. In fact, the three proximate drivers (i.e., agricultural expansion, timber extraction and infrastructure development) are present in 25% of the observed cases of land use change worldwide (Geist & Lambin, ). Changes in carbon stock might also be due to historical factors and not a consequence of the more recent LUC. For example, of the 4.5 Pg C lost from Mato Grosso (Brazil) between 1901 and 2006, about 78% of these losses occurred between 1901 and 2001, largely due to land clearing for pasture and croplands (Galford et al ., ). It is thus desirable that future models or studies explicitly consider historical LUC, and capture the major drivers of domestic and global LUC along with those potentially associated with biofuels. Future studies should also investigate whether the presence or absence of a driving factor determines the changes. Without these developments, our understanding of the extent and implications of the LUC of biofuels will remain limited. When a constant yield over time and a high fraction of land from carbon rich pools, such as forests or peatlands are assumed, the resulting dLUC or iLUC carbon intensities are high, regardless of the type of energy crops. However, if yield increases are allowed to continue and the land cover is poor in carbon stock, the resulting dLUC or iLUC carbon intensities will be low. The dLUC (respectively, iLUC) carbon intensities may also be negative if the feedstocks (respectively, the displaced feedstocks) are grown (respectively, relocated) on degraded land. No study assessed the LUC induced GHG emissions of biofuel production on fallow land. When analysed over a short‐term period, biofuel production does not offer any GHG emission benefits as compared with the fossil energy they displace. However, biofuel can still pay back the carbon debt incurred during land conversion over a long‐term period. This suggests that the time frame over which GHG emissions are analysed and the use of a discount rate to value the short‐term period against long‐term emissions can have a significant impact on the net GHG balance of biofuel production systems. Models used in the reviewed studies suggested that most of the LUC impacts will occur in less industrialized and developing countries, such as Brazil, India, Malaysia, China and African countries which are the most competitive producers of biofuel crops (Table ). However, LUC models should not only identify countries or regions where LUC occurs but also the contributions of each ultimate driver to the total impacts. Failing to do this overestimates the LUC carbon intensities associated with biofuels. Finally, the recommendations that GHG emissions of biofuels should be quantified in a full life cycle assessment (LCA) raise the problem of which LCA methodology should be used. There are two basic types of LCA methodologies, an attributional LCA (aLCA) and a consequential LCA (cLCA). The attributional LCA assesses all environmentally relevant physical flows attributed to a production process, whereas the cLCA assesses the changes in environmentally relevant physical flows in response to a decision. These two types of LCA use different data and the choice of conducting either an aLCA or a cLCA depends on the stated goal of the study. If the chosen LCA method for biofuels is different from the one for the reference system, comparability is not guaranteed. For example, in the study of Searchinger et al . ( ), the cLCA used to quantify the GHG emissions of biofuels is wider in scope and includes indirect emissions from LUC, whereas the aLCA used to estimate the GHG emissions of the reference systems (i.e., fossil energy) is narrow in scope and excludes indirect GHG emissions due to the military control over oil reserves. Critical gaps and recommendations for further research The economic and the environmental viability of biofuels depend on productivity. Yield is a key component in biofuel production and in LUC discussions. The improvement in yield depends on breeding and agronomic practices. However, most second generation energy crops (e.g., poplar, willow) are essentially unimproved or have been bred only recently for biofuels, whereas conventional crops (e.g., corn, wheat) have undergone a substantial improvement in yield, in pest resistance and in other agronomic traits. A more complete understanding of the biological system coupled to management practices and to biotechnological advances will speed up energy crops with desirable attributes, such as increased yields and usability, optimal growth, better pest resistance, efficient water and nutrient use and greater resistance to stress. The land allocation module in existing models is restricted to only three land classes: forest, grassland and cropland. Other land classes are not included in this module. Further research is thus needed to extend the land class allocation problem to include other land classes, such as abandoned land, marginal land or degraded lands. Such an extension offers a basis for studying many empirical and policy questions like short‐ and long‐term biofuel demand, dynamic changes and allocation issues in forest conservation and deforestation. Future iLUC models should be improved to include sets of parameters and assumptions (e.g., relocation of displaced activities to degraded lands) that could enable models to predict a negative iLUC carbon intensity. Without this development, our understanding of the full impact of LUC of biofuels will be limited and the validity of existing models will be questioned. Also, a standard and transparent approach to depict the iLUC carbon intensity that can be added to an LCA needs to be developed to correctly assess the sustainability of biofuels. To the extent of our knowledge, no such standard methodology exists or is one under development. Ideally, such an approach should be based on LCA and include all drivers of LUC, all biomass sectors, all land classes and coproducts. It should also consider historical LUC and the fate of the cleared carbon, account for the effects of management and agricultural intensification, conversion efficiency, as well as improvements in yield. Finally, lack of data greatly contributes to the uncertainty about the LUC carbon intensity of biofuels. Datasets on biofuel crop production must be collected, synthesized and standardized to common data formats so as to minimize uncertainties in the input data for the iLUC models. Concluding remarks Models to deal with dLUC and iLUC exist, but each of the models gives different results. The variation in the estimate of the LUC carbon intensity of biofuels is due to model structures, to different data sets and to a number of assumptions made to model LUC. Despite substantial variations and uncertainties involved in the quantification of LUC carbon intensities, this study shows that in some cases, the LUC can potentially alter the GHG benefits of biofuels. Consequently, there is a substantial risk that current biofuel policies will lead to an increase in GHG emissions if emissions from LUC are not accounted for in the life cycle of biofuels. There is currently no way to determine which of the many models yields the most reliable overall LUC carbon intensity. Deciding how to estimate the LUC carbon intensity of biofuels will therefore have major implications for biofuel policies. Finally, key gaps not included in the LUC modelling at present need to be addressed to improve our understanding of biofuel LUC impacts. These include the potential use of coproducts to decrease the impacts of LUC, management practices and the inclusion of other land use classes, such as peatland and fallow land in existing modelling tools. Acknowledgements The research leading to these results has received funding from the European Research Council under the European Community's Seventh Framework Programme (FP7/2007–2013), ERC grant agreement nr. 233366 (POPFULL). We thank Florian Gahbauer for language corrections, four anonymous reviewers for their helpful comments and several authors for providing more detailed information on their published results. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png GCB Bioenergy Wiley

A comparative analysis of the carbon intensity of biofuels caused by land use changes

Loading next page...
 
/lp/wiley/a-comparative-analysis-of-the-carbon-intensity-of-biofuels-caused-by-O080LkwmMH

References (81)

Publisher
Wiley
Copyright
Copyright © 2012 Blackwell Publishing Ltd
ISSN
1757-1693
eISSN
1757-1707
DOI
10.1111/j.1757-1707.2012.01176.x
Publisher site
See Article on Publisher Site

Abstract

Introduction As a result of concerns about energy security and climate change, interest has grown worldwide in renewable energy sources because they are not depletable and produce less greenhouse gas (GHG) emissions than fossil fuels. Biofuel (ethanol, diesel or kerosene) production from organic matter is seen as one of the strategies to reduce fossil fuel consumption and GHG emissions (Righelato & Spracklen, ). Many industrialized countries have established policies to increase the share of biofuels in the total energy production. For example, the European Union (EU) aims to achieve a minimum of 10% renewable fuels in the transport sector by 2020 as part of the overall EU renewable energy target of 20% of its final energy use by 2020 (EC, ). Biofuels can provide unique economic benefits compared with other alternative energy options (Farrell et al ., ). However, there have been concerns that the increased use of agrichemical products and the global land use changes (LUC) associated with the expansion of biofuel production could negatively impact the food supply and biodiversity (Harmon et al ., ; Righelato & Spracklen, ; Scharlemann & Laurance, ). Among these concerns, LUC carbon intensity of biofuels – defined as the amount of CO 2 emitted per unit of biofuel produced – has prompted an intense debate in the scientific and the political communities. LUC are changes in the areal extent of a particular type of land use over a given time period and within a given spatial entity. LUC has impacts (direct and indirect) on soil, water, biodiversity and climate change, and is thus related to many environmental issues of global importance. Direct land use change (dLUC) refers to the impact caused by switching a particular land area from some previous use (or disuse) to cultivated land for biofuel crop production. Indirect land use change (iLUC) impacts, also known as ‘leakages’, refer to impacts associated with LUC elsewhere as an unintended consequence of biofuel crop production. Recent studies have demonstrated that converting carbon rich ecosystems (e.g., forests, grasslands) to cultivated lands for biofuel crop production resulted in little GHG emission savings or even lead to a ‘carbon deficit’ that would take years or decades to be repaid (Fargione et al ., ; Gibbs et al ., ; Searchinger et al ., ). These and other studies have enhanced our understanding of the LUC implications of biofuels. However, the mechanisms leading to their conflicting outcomes have not been analysed. Also, the LUC carbon intensity of biofuels has not yet been inventoried or compared. This study analyses existing studies that assess the LUC‐GHG implication of biofuel production. Our aim is to provide a structured overview and a comparison of studies on LUC‐carbon intensities associated with biofuels that have emerged during the last 3 years. First, the techniques used to quantify the LUC‐carbon intensities are examined, summarized and compared. Second, the mechanisms that lead to diverging results are investigated. Finally, unanswered questions are identified and recommendations for future research are suggested. Literature selection of primary database We used a systematic review methodology to identify and to select studies included in this comparative analysis. Systematic reviews represent a rigorous, objective and transparent approach to synthesize and evaluate scientific evidence while minimizing bias (Oxman et al ., ). The systematic review methodology evolved within the medical field to support the development of clinical and public health practise guidelines and to formulate statements of scientific consensus (Briss et al ., ; Guirguis‐Blake et al ., ). Systematic reviews are now increasingly being performed and published for energy (Kubiszewski et al ., ) and bioenergy related topics (Whitaker et al ., ; Njakou Djomo et al ., ). The systematic review approach was chosen in this study to explain the reason for differences among studies and to identify priority areas for further research. Drawing on the principles and procedures of systematic reviews, we queried the ISI Web of Knowledge, Web of Science and Science Direct databases for studies published between 2008 and 2010 on the LUC implication of biofuels. The limitation to the last 3 years is due to the fact that the issue of LUC‐carbon intensity of biofuels is very recent and appeared in the scientific literature only since 2008. Our search did not yield any studies published on the issue before 2008. A pair‐combination of the keywords ‘land use change’, ‘greenhouse gas emissions’ and ‘payback time’ was used together with the term ‘biofuel’ for the search. The search was limited to studies published in English. Relevant publications were identified and their bibliographies were examined for additional articles. We also contacted key authors of most publications for additional information where necessary. All full‐length peer‐reviewed studies and reports that assessed LUC carbon intensities, as well as the carbon payback time of biofuels, and that provided numerical data were included in the analysis. Studies reporting only on LUC or on GHG emissions, as well as review studies were excluded. The titles and abstracts of 43 potential studies and reports were screened and their methodological quality and validity were assessed for eligibility. Among these, 16 studies were excluded because they were duplicates or they did not properly state the methodology used to obtain the data. The remaining 27 studies were examined in detail. Twelve studies were further excluded because the carbon intensity was not reported or could not be estimated, or the articles did not report original data. Finally, 15 relevant studies were retained for the data extraction Inventory of the reviewed studies Data extracted from the 15 primary studies were tabulated and presented in a descriptive form (Table ). Of the 15 studies on LUC of biofuels, 2 addressed the direct land use change (dLUC), 9 assessed the indirect land use change (iLUC) and the remaining 4 dealt with both dLUC and iLUC. While all studies reported on dLUC or iLUC carbon intensities, only five studies calculated the carbon payback time (Table ). The payback time is the period it takes the biofuel system to overcome the carbon debt incurred when land is converted and starts providing GHG benefits. Land use change ( LUC ) mechanisms, models, feedstocks, carbon pools and stocks of each of the reviewed studies LUC mechanisms Energy product Model/Approach Crop yield (t ha −1 yr −1 ) Area (10 6 ha) Land class and fraction (%) Carbon pool Carbon stock (t C ha −1 ) Geographic location References dLUC Bioethanol Biodiesel Excel (spreadsheet) M (10), SC (69) PO (20), SB (3) na F (100), P (100), G (100), C (100) AB, BB, SOC F: 192–201, PL: 943, G: 23–37, AC: 2–19 BR, ID, ML, USA Fargione et al . ( ) dLUC Biokerosene IPCC (simplified) JP (4) na F (100), P (100) AB, BB, SOC F: 168–245, G: 23–41, C: 22, G: 21–23 BR Bailis & Bake ( ) dLUC, iLUC Bioethanol EPPA‐TEM (GE model) M (7), SC (30) SG (10) na F, P, G na na Global Melillo et al . ( ) dLUC, iLUC Bioethanol DAYCENT (simplified) M (9), SB (3) na F (100), G (100) AB, BB, SOC F: 108–200, G: 51–87 USA Kim et al . ( ) dLUC, iLUC Bioethanol Biodiesel IMPACT (PE model) SC (68), SB (1) 16.5 F (73), O (27) AB, BB, SOC F: 201, R: 20, S: 45, O: 23 BR Lapola et al . ( ) dLUC, iLUC Bioethanol Biodiesel RISK‐ADDER (Simplified) M (8), W (4), SB (3) 15.9 F (3), G (55), S (5), C (33) AB, BB F: 266, G: 69, S: 134, AL 55 AR, BR, EU, IN, USA Fritsche et al . ( ) iLUC Bioethanol FAPRI‐CARD (PE model) M (11), SG (18) 10.8 F (52), G (48) AB, BB, SOC F: 165–313, G: 21–83 BR, CN, IN, RoW, USA Searchinger et al . ( ) iLUC Bioethanol GTAP‐Bio (GE model) M (9) 4.2 F (19), P (81) AB, BB, SOC F: 166, G: 27 Global Hertel et al . ( ) iLUC Bioethanol RFMI (Simplified) M (10) na F (15;50), G (85;48), W (0;2) AB, BB, F: 96–178, G: 20–55, WL: 273–819 Global Plevin et al . ( ) iLUC Bioethanol GTAP (GE model) M (9) 2.0 F (33), G (67) AB, BB, SOC F: 86–189, G: 12–57 AR, BR, EU, IN, USA Tyner et al . ( ) iLUC Bioethanol FASOM‐FAPRI (PE models) M (11), SB (3) 1.4 F (19), G (29), S (38), SL (14) AB, BB F, G, S, SL BR, RoW, USA EPA ( ) iLUC Bioethanol CARD‐GreenAgSIM M (11) 1.3–6.1 F (52), G (48) AB, BB, SOC F: 165–313, G: 21–83 BR, CN, IN, RoW, USA, Dumortier et al . ( ) iLUC Bioethanol Biodiesel MIRAGE (GE model) M, PO, SF, SB na F (41), S (33), G (12), C (14) AB, BB F: 151–225, S: 52–58, BR, CN, EU, IN, ML,RoW Al‐Riffai et al . ( ) iLUC Bioethanol Biodiesel IMAGE (Integrated) SC, B, SB, W, RS, PO na F (51), G (48) D (0.8), T (0.3) AB, BB, SOC F: 69–509, G: 30–162 T: 52–179, D: −21 to −3 AR, BR, EU, IN, ML, PK, USA Overmars et al . ( ) iLUC Bioethanol Biodiesel GLOBIOM (PE model) M, SC, W RS, SB 13 F, G, C, O AB, BB, SOC na EU, Global Havlik et al . ( ) AB, above‐ground biomass; AL, arable land; AR, Argentina; BB, below‐ground biomass; BR, Brazil; CN, China; C, cropland; dLUC, direct land use change; D, desert; EU, European Union; F, forest; G, grassland; iLUC, indirect land use change; ID, Indonesia; IN, India; JP, jatropha; M, corn; ML, Malaysia; na, not assessed; O, other vegetation; PL, peatland; PK, Pakistan; PO, palm oil; RoW, rest of the world; RS, rapeseed; S, savannah; SB, soybean; SC, sugarcane; SF, sunflower; SG, switchgrass; SL, shrubland; SOC, soil organic carbon; USA, United States of America; W, wheat; WL, wetland. This number represents the average of the three feedstocks (corn, wheat, soybean). Land use change ( LUC ) carbon intensities, carbon payback times and amortization periods of different biofuels reported in the reviewed studies. A positive LUC carbon intensity value suggests an increase in GHG emissions, whereas a negative LUC carbon intensity suggests a net carbon sequestration. Emissions from LCA are excluded from estimates of LUC carbon intensities, whereas they are included in estimates of the payback times Type of fuel Amortization period (years) LUC carbon intensity (g CO 2e MJ −1 ) Carbon payback time (years) References dLUC iLUC dLUC + iLUC dLUC iLUC dLUC+ iLUC Bioethanol 20 na 26–154 na na 12–70 na Overmars et al . ( ) na 16–79 na na na na Al‐Riffai et al . ( ) −27 to 27 0–34 −27 to 61 na na na Fritsche et al . ( ) 30 na 104–111 na na 52–167 na Searchinger et al . ( ) na 32–34 na na 22–27 na Havlik et al . ( ) na 28–45 na na na na EPA ( ) na 27 na na 28 na Hertel et al . ( ) na 21–118 na na 32–183 na Dumortier et al . ( ) na 21–142 na na na na Plevin et al . ( ) na 13–19 na na na na Tyner et al . ( ) −52 to 11 181–190 129–201 na na na Melillo et al ., ; 50 0–25 na na 0–93 na na Fargione et al . ( ) 34 327 361 4 40 44 Lapola et al . ( ) na 10–61 na na 12–70 na Overmars et al . ( ) −24 to 27 31–57 7–84 na na na Melillo et al . ( ) 100 −2 to −1.9 6–32 4–30 na na 12–31 Kim et al . ( ) −7 to 0 1–7 −6 to 7 na na na Melillo et al . ( ) Biodiesel 20 na 46–67 na na na na Al‐Riffai et al . ( ) na 30–204 na na 15–106 na Overmars et al . ( ) −98 to 31 0–41 −98 to 72 na na na Fritsche et al . ( ) 50 35–481 na na 37–423 na na Fargione et al . ( ) na 12–82 na na 15–106 na Overmars et al . ( ) 37–243 626–1434 663–1677 7–35 122–211 129–246 Lapola et al . ( ) Biokerosene 20 −27 to 101 na na na na na Bailis & Bake ( ) dLUC, direct land use change; iLUC, indirect land use change; na, not assessed. Values refer to the 25% of the theoretical potential. Land use change: Mechanisms, drivers, causes and consequences The dLUC occurs when a portion of a natural ecosystem (e.g., forest or grassland), a portion of cropland or a portion of unused land (e.g., degraded or abandoned land) is converted into cultivated land for biofuel crops production (Fig. , top panel). The dLUC occurs at the time of the energy crop production; it is certain and it is directly caused by the production of energy crops. It links the conversion of a specific portion of land in a given biofuel chain to resulting environmental impacts. Schematic diagrams describing the direct ( dLUC ) and indirect land use changes ( iLUC ). The panels illustrate the dLUC (top), local iLUC (middle) and global iLUC concepts (bottom). The symbol β represents the portion of additional land required due to expansion of biofuel crops, whereas the symbol Ω represents the fraction of natural ecosystems converted to cropland as a result of the conversion of cropland to land for the production of biofuels. The iLUC occurs when a portion of a natural ecosystem, or a portion of degraded or unused land, or a portion of other land dedicated to other uses is converted to new cropland to produce food, fibre or feed that were displaced by the expansion of biofuel crop production (Fig. , middle and bottom panels). It can occur locally when, for example, the expansion of biofuel crops to cropland induces the conversion of natural ecosystems elsewhere in the same country to cropland (Fig. , middle panel). It can also occur globally when the expansion of energy crops to cropland reduces food supplies in the near‐term future and induces – through market mechanisms – the conversion of natural ecosystems elsewhere in the world into cropland (Fig. , bottom panel). The iLUC can also be induced in situations where the dLUC for energy crop production is absent, but the use of existing crops is diverted to energy crops (Searchinger et al ., ; WBGU, ). The causal link in iLUC is weaker than in the dLUC case. The action of increased biofuel production and the LUC impacts in the iLUC case are mediated through prices. The location of iLUC cannot be predicted with certainty as it is the result of complex interactions of (international) market mechanisms, agricultural policies and other events that influence future patterns of land use. The time horizon, space, mediation, probability and causality characteristics distinguish iLUC from dLUC. Drivers of LUC can be grouped into: (i) proximate drivers, such as agricultural expansion, timber extraction and infrastructure development and (ii) ultimate drivers, such as economic, institutional, technological and demographic factors (Geist & Lambin, ; Lambin et al ., ). Taken alone, none of these drivers can adequately explain LUC at the proximate level, but all together, they explain more than 90% of LUC observed worldwide (Kim et al ., ). Land conversion for biofuel crop production can have significant impacts on the environment. Both dLUC and iLUC can affect the GHG balance of biofuel chains due to GHG emissions from changes in above‐ and below‐ground carbon, soil organic carbon, litter and dead wood (Searchinger et al ., ; Kim et al ., ; Plevin et al ., ). Land conversion can also affect air quality by altering emissions and changing conditions that influence reaction rates, transport and deposition. For example, tropospheric ozone is very dependent on changes in vegetation cover and in biogenic emissions (Foley et al ., ). LUC also affects the soil by altering the soil environment, which in turn affects microbial growth and decomposition processes that transform plant derived carbon into soil organic matter and carbon dioxide (Cotrufo et al ., ). Furthermore, LUC can disturb the surface water balance and the proportion of precipitation into evapotranspiration, runoff and groundwater flow. In fact, river discharge and surface runoff increase when a natural ecosystem is cleared (Sahin & Hall, ; Costa et al ., ). Finally, LUC can also reduce biodiversity through the modification, the fragmentation and the loss of habitats, the degradation of soil and water and the overexploitation of native species (Pimm & Raven, ). Modelling approaches for LUC Modelling dLUC The approaches or models used to estimate the LUC carbon intensity of biofuel production systems vary between both LUC mechanisms (i.e., dLUC and iLUC) and across studies. The dLUC carbon intensity can be estimated using mathematical (simulation) tools, such as the IPCC tool or the DAYCENT model (Table ). IPCC approach The IPCC Tier 1 method (IPCC, ) requires estimates of carbon in living biomass stocks prior to conversion, based on estimates of the area of land converted during the period between land use surveys. The difference between initial and final live biomass carbon pools is used to calculate changes in carbon stocks due to land use conversion (IPCC, ). The Tier 1 method is intended for national scale inventories in the absence of region specific data. This method requires no new data collection as it uses default data from the IPCC guidelines. However, many biofuel crops (e.g., sugarcane) are not represented as default value options. The method also assumes that N 2 O emissions are solely a function of nitrogen inputs to the soil. Furthermore, the method does not account for the carbon fluxes, and the simplification of assumptions involved in the method increases uncertainties. Despite these weaknesses, the IPCC Tier 1 approach is transparent and based on the best available science. The method is robust when used to obtain average values across a large agricultural landscape. Agroecosystem models The DAYCENT model is the daily time step version of the CENTURY model used to simulate the dynamics of soil organic carbon and related nitrogen pools (Del Grosso et al ., ; Parton et al ., ; Del Grosso et al ., ). One of the strengths of the DAYCENT model is that it can simulate ecological disturbance and management practices, such as fire, grazing, cultivation, organic matter addition and/or fertilizer addition (Kim & Dale, ). Unlike the IPCC Tier 1 approach, the DAYCENT model is able to predict the soil organic carbon along with the carbon in above‐ and below‐ground biomass, and nitrous oxide emissions from the soil (Kim et al ., ). The DAYCENT model also accounts for factors that influence emissions, such as weather, soil types and previous land history, thus making estimates more reliable. Moving from the Tier 1 method to DAYCENT improves the accuracy and reduces the uncertainty of assessments, but it also increases the complexity and the costs of monitoring. Modelling iLUC Unlike dLUC, which could be verified through experiments, the iLUC is a diffuse market response. The iLUC cannot be observed at a specific location or region, but can only be modelled using complex tools. Several models have been used in the analysed studies to estimate the iLUC carbon intensities of biofuels (Table ). These models can be grouped into agroeconomic models, combined models, biophysical models and deterministic (or simplified) models. They are briefly described, summarized and compared below. Agroeconomic models Agroeconomic models are indispensable tools in the preparation and negotiations of agricultural policy decisions. They include models, such as the general equilibrium (GE) and partial equilibrium (PE) models. A model is in general equilibrium when all markets are modelled explicitly and considered to be in equilibrium at every time step. In partial equilibrium, only a subset of the markets is considered and the remaining markets are parameterized. General equilibrium model ( GE ) The Global Trade Analysis Project (GTAP) model (Hertel et al ., ; Tyner et al ., ) is a static multiregion GE used to assess the iLUC created by biofuels production. The GTAP model utilizes economic data to estimate the potential GHG emissions, as well as other impacts of prospective biofuel technologies or policies. It takes into account interactions between the economic sectors of a country or of several regions. The main characteristic of the GTAP model is its economy wide coverage. The model contains a biofuel sector, includes conventional energy crops (e.g., corn) and is able to capture the relationship between changes in land area and yield increase. However, the GTAP model does not provide a detailed sectorial analysis of the economy because of its complex structure and aggregation. The GTAP model is data intensive; its specification as well as its related database does not include soil carbon stocks. Furthermore, biofuel coproducts are not yet specified in the GTAP model. Another model used to assess the iLUC effects of biofuel production is the Modelling International Relationship in Applied General Equilibrium (MIRAGE) model (Al‐Riffai et al ., ). The MIRAGE model is a dynamic, multisector, multiregion GE model that relies on the GTAP database. Similar to the GTAP model, the MIRAGE model uses economic data to simulate the iLUC associated with biofuel expansion. The model was recently modified to capture the interactions between biofuel expansion, agricultural markets and LUC. Key modifications incorporated into the model include the integration of biofuel sectors (e.g., biodiesel, bioethanol) and improvements in the modelling of the energy sector. The MIRAGE model uses most recently available data to model LUC. The MIRAGE model considers fertilizers used and coproducts that arise from biofuel production. However, the MIRAGE model does not include second generation biofuels or the non‐CO 2 GHGs. Also land classes, such as marginal and fallow lands are excluded from the MIRAGE model. A particular advantage of the GE models is that they incorporate all sectors of the economy. They are thus well suited to depict interactions between agriculture and other sectors of the economy. However, GE models do not capture all important characteristics of the agricultural economy. Moreover, GE models are very data intensive and the amount of data is determined by the level of disaggregation (e.g., countries or regions; activities or commodities) and the theoretical structure (homogenous/heterogeneous). Consequently, if data sets are not aggregated to a greater extent, the modelling task may become unmanageable. Furthermore, in GE models (e.g., GTAP), land data are usually expressed in monetary values, not in physical dimensions (e.g., hectare). Consequently, some assumptions need to be made to relate the physical quantity of land with a given volume of biofuels expressed in monetary units (e.g., dollars). Partial equilibrium models ( PE ) The Food and Agricultural Policy Research Institute (FAPRI) model (Searchinger et al ., ), the International Model for Policy Analysis of Agricultural Commodities and Trade (IMPACT) model (Lapola et al ., ) and the Global Biomass Optimization (GLOBIOM) model (Havlik et al ., ) are all PE models used in some of the analysed studies to assess the iLUC carbon intensity of biofuels. The FAPRI model is a recursive dynamic, multimarket and nonspatial PE model able to capture the technical, economic and biophysical relationships among key variables within a particular commodity and across commodities (Edwards et al ., ). The FAPRI model is often used to analyse the impacts of a policy proposal relative to a baseline scenario. Historical data from different sources (e.g., FAO, USDA) as well as current academic research data are used to calibrate the model. The FAPRI model can predict the location and the amount of LUC. The FAPRI model treats allocation of land as the result of the farmer's choices, assuming that farmers maximize their net return per hectare of land. The FAPRI model covers all major temperate crops for all crop producing and consuming countries. However, the modelling of biofuels in this model is restricted to first generation biofuels. Second generation biofuels are not included in the model. The IMPACT model is a recursive dynamic PE of the agricultural sector model designed to examine alternatives for global supply, demand and trade of food. The IMPACT model contains a water use module that balances water availability and uses within various economic sectors (e.g., agricultural, industrial and residential) at the regional and global scales. The integration of a water module into the IMPACT model allows it to explore the relationship between the increased biofuel crop production and the demand for irrigation water (Rosegrant et al ., ). Data from the UN and FAO statistics are often used to calibrate or validate the model. The GLOBIOM model is a bottom‐up and recursive dynamic PE model that integrates the agricultural, forestry and biofuel sectors. The general structure of this model is similar to that of the Agricultural Sector and Mitigation of Greenhouse Gas (ASMGHG) model. In the model the market is represented by implicit product supply functions based on detailed, geographically explicit, Leontief production functions and explicit, mostly constant elasticity product demand functions (Havlik et al ., ). A noticeable added‐value of the GLOBIOM model is its ability to model both carbon stocks and flows. The model also allows for accounting and taxing of the major GHG emissions related to agriculture and forestry. However, the model is restricted to agriculture and land use sectors and does not include downstream processes, such as biofuel conversion processes (De Vries, ). Therefore, there is no link between the agricultural and the energy sectors in the GLOBIOM model. The PE models use observed data to determine the quantity and the location of LUC due to the diversion of food crops to feedstock for biofuel production. One of the advantages of PE models is that they offer a more complex and accurate depiction of the agricultural sector. However, PE models focus on one specific economic sector only (i.e., the agricultural sector) and do not explicitly link to other sectors of the economy or to countries or regions other than the one(s) under investigation (Kretschmer & Peterson, ). Moreover, PE models do not capture production increases achieved by increasing agricultural yields or avoiding logistic losses and market distortion, such as taxes (Fritsche et al ., ). Compared with PE models, GE models allow a coherent representation of the economy, but the scale and traceability issues are less detailed than in PE models. Both GE and PE models do not consider factors, such as land price speculation, market information or land tenure in LUC decisions (Econometrica, ). Furthermore, GE and PE models do not always give a description of major biogeophysical constraints on agricultural production (e.g., soil conditions or climate) (WBGU, ). Finally, environmental aspects are insufficiently covered in both GE and PE models. Thus, combining those models with biophysical or other models is crucial and helps to overcome their individual weaknesses. Combined models The Emissions Prediction and Policy Analysis (EPPA) model (Paltsev et al ., ) has been combined with the Terrestrial Ecosystem Model (TEM) (Melillo et al ., ) to assess the iLUC carbon intensity of biofuel production. The EPPA model is a recursive dynamic multiregional GE model designed to analyse the impact of climate change policies on the global economy and on GHG emissions. In contrast, the TEM model is a process‐based ecosystem model used to examine patterns of land carbon dynamics across the globe. In the combined EPPA‐TEM model, land use estimates from EPPA are downscaled and organized for use in the TEM model (Melillo et al ., ). Competition in land use demand between recreation and biofuels is well examined in the EPPA‐TEM model. Both first and second generation biofuels are also included in the EPPA‐TEM model. Moreover, in the EPPA‐TEM model fossil fuels are treated as exhaustible resources, with diverging costs of exploitation and scarcity rents. This allows the EPPA‐TEM model to endogenously determine fossil fuel prices. However, the impacts of LUC are not fully assessed due to the high level aggregation of the agricultural sectors in the EPPA‐TEM model. Moreover, the EPPA‐TEM model does not include coproducts in the assessment of iLUC associated with biofuels. The US Environmental Protection Agency (EPA) also combined two PE models to estimate the iLUC implications for domestic and international commodity prices. In the EPA study the Forest and Agriculture Sector Optimization Model (FASOM) (Darius et al ., ), a dynamic PE agricultural sector model, is combined with FAPRI (FAPRI, ) to estimate the LUC carbon intensity of biofuel production (EPA, ). In the coupled FASOM‐FAPRI model, FASOM is used to estimate the domestic LUC, whereas FAPRI is used to estimate the international LUC. The outputs from the FASOM model are used as inputs into the FAPRI model. An advantage of such combination is the high resolution of the PE models as compared with the GE models. However, the combination of these two PE models obscures the link between the agricultural market and the rest of the economy, particularly the energy sector. An agricultural outlook model, such as the one developed by the Center for Agricultural and Rural Development (CARD) has also been combined with the Greenhouse gases from Agriculture Simulation (GreenAgSim) model. In this combined approach, the CARD model is used to assess the impact of biofuel policy changes and energy price increases on land conversion, whereas the GreenAgSim model is used to evaluate emissions from land conversion and agricultural production (Dumortier et al ., ). Although model combinations can in some cases help to overcome individual model weaknesses, they increase the model complexity and reduce its applicability. In fact, as combined models rely on historic time series, they cannot deal well with long‐term scenarios. Also, the underpinning causal mechanisms of LUC are insufficiently addressed or overlooked in combined models. Moreover, combined models are very data intensive, and it is very difficult to perform a proper calibration and validation. The failure of these models to include long‐term scenarios coupled to their inability to handle causal mechanisms and data issues reduces their utility. Bio‐physical models A bio‐physical model, such as the Integrated Model to Assess the Global Environment (IMAGE) has also been used to assess iLUC of biofuels (Overmars et al ., ). The IMAGE model is a non‐economic model that simulates the physical carbon flow on a global aggregate scale. It is used to explore the long‐term dynamics and interconnections of global change and to evaluate specific issues of sustainable development in the broader context of global socio‐ecological evolution. The strength of this model is that it allows an accurate assessment of the spatial structure of land use by describing the hierarchical organization of land use. However, the IMAGE model does not well represent the behaviour of individual sectors of the economy that are important for the understanding of the iLUC (e.g., the energy and agricultural sectors). Deterministic models A number of simplified approaches have also been developed and used in recent years to assess the LUC implications of biofuel production. One such approach is the ‘Risk‐Adder’ developed by the Institute for Applied Ecology (Öko‐Institut) in Germany (Fritsche et al ., ). The ‘Risk Adder’ approach is a simple and transparent methodology that uses statistical data to estimate the iLUC carbon intensity associated with biofuel production. In this approach, the modelling of agricultural markets is avoided by assuming that: (i) the global potential of GHG emissions from iLUC is a proxy for the current patterns of land use to produce traded agricultural commodities; (ii) for the near future, observed trade trends can be used to obtain the pattern of global trade in agricultural commodities. Another simple approach used to estimate the iLUC carbon intensity of biofuel systems is the Reduce‐Form Model of iLUC (RFMI) (Plevin et al ., ). Here, the model complexity is reduced by using the net displacement factor (NDF). The NDF is the ratio of hectares of land brought into cultivation anywhere in the world to the hectares of land dedicated directly to additional biofuel crops. The NDF includes the joined effects of (i) price induced yield increases, (ii) the relative productivity of land converted to cropping, (iii) price‐induced reductions in food consumption and (iv) the substitution of crop products by biofuel coproducts (Plevin et al ., ). The NDF is the most influential parameter in the RFMI and is calculated for a specific time period using crop yields for that period. However, as for GE and PE models, predictions of NDF have not yet been tested or verified, and there is no empirical evidence for choosing one number over another. Proponents of simplified approaches argue that they are simple, transparent and easy to implement in Excel spreadsheets. They also require less data compared with GE and PE, and provide a quick, rough estimation of the iLUC carbon intensity. However, many simplifications increase the inaccuracy. Furthermore, by simplifying the characterization of complex market links, these approaches miss some market feedbacks that drive the iLUC (Yeh & Witcover, ). Synthesis of results: a wide variation in LUC carbon intensities Among the 15 selected studies there is considerable variation in the estimates for both the dLUC and iLUC carbon intensities as well as in the carbon payback time associated with biofuel production. Depending on the amortization period (i.e., the time range over which the carbon cost and benefits will be considered), and the type of land and feedstock used, the dLUC carbon intensity ranged from −52 to 34 g CO 2 MJ −1 for bioethanol. Negative dLUC carbon intensity (i.e., net sequestration) values were associated with cellulosic crops (e.g., poplar) grown on cropland, whereas positive dLUC carbon intensity (i.e., net emissions) values were linked to sugarcane cultivated on rangeland in Brazil. For biodiesel, the carbon intensity values ranged from −98 to 481 g CO 2 MJ −1 . Here, the conversion of degraded land to palm derived biodiesel resulted in a net carbon sequestration in the soil, whereas the conversion of rainforest to palm derived biodiesel resulted in carbon emissions to the atmosphere. The dLUC carbon intensity of biokerosene ranged from −27 g CO 2 MJ −1 when Jatropha is planted on former pasture land to 101 g CO 2 MJ −1 when it is planted in cerrado woodlands (Table ). These results indicate that the dLUC‐carbon intensity depends on the type of biofuel feedstocks grown, and on the type of land converted. With regard to the iLUC carbon intensity, the values ranged from 0 to 327 g CO 2 MJ −1 for bioethanol, whereas for biodiesel the values ranged from 0 to 1434 g CO 2 MJ −1 depending on the feedstock used, on the type of land displaced and on the amortization period (Table ). For both bioethanol and biodiesel, zero iLUC carbon intensity values were reported when biofuel crops were grown on marginal or degraded lands as there was no new land brought into production. However, both bioethanol and biodiesel turned out to have very high carbon intensities when the displaced activities (e.g., rangelands) were relocated to forests. No studies reported on a negative iLUC carbon intensity. This is because most studies assumed that the displaced activity will move to carbon rich lands, such as forest and grassland. However, the iLUC carbon intensity is not necessarily positive: estimates of iLUC carbon intensity may be negative when, for example, the expansion of biofuel crops (e.g., palm oil) to grassland induces the conversion of marginal or degraded lands elsewhere in the world to grassland. However, such scenario was not investigated in the studies analysed here. The variation was large, suggesting that there is no consistency in the estimates of iLUC carbon intensity among the reviewed studies. Within each individual study and regardless of the feedstock used, the iLUC carbon intensity tended to be small with a longer amortization period. Estimates of both dLUC and iLUC for a given amortization period illustrated that in some cases the carbon intensity was larger for the iLUC than for the dLUC regardless of the type of biofuels (Table ). This means that in some cases, iLUC emissions could account for a significant part of the total LUC carbon intensity of biofuels. The comparison of bioethanol and biodiesel indicated that bioethanol incurred less carbon debt than biodiesel (Table ). However, this latest result should be interpreted with caution as it inherently depends on different assumptions and data used in the individual studies. Further research is needed to provide a clear answer on the land use carbon intensity of these two biofuels. The total LUC (i.e., dLUC + iLUC) carbon intensity ranged from −27 to 361 g CO 2 MJ −1 for bioethanol and from −98 to 1677 g CO 2 MJ −1 for biodiesel depending on the feedstock and amortization period (Table ). For bioethanol these values were about −29% (for sugarcane grown on degraded land) up to 384% (for sugarcane planted on cropland) of that of gasoline, which has a life cycle emission profile of about 94 g CO 2 MJ −1 (Farrell et al ., ). Similar results were obtained for biodiesel compared with conventional diesel. Thus, it was not clear from the 15 analysed studies whether bioethanol (or biodiesel) reduced or increased the GHG emissions relative to gasoline (or diesel) at the current state of knowledge. Note that the current estimate of gasoline's GHG performance is likely poorer than this baseline (i.e., 94 g CO 2 MJ −1 ) as it excludes, for example, emissions due to the military control over oil reserves in the Middle East (O'Rourke & Connolly, ). Liska & Perrin ( ) argued that the GHG intensity of gasoline would double if indirect military emissions were included in the gasoline's life cycle. Figure summarizes the life cycle emissions of bioethanol and biodiesel from different biofuel crops over different amortization periods (20, 30 and 50 years) and compares them with those of conventional gasoline and diesel. In most cases, adding life cycle GHG estimates to those of iLUC carbon intensity nearly cancels out the GHG benefits of bioethanol as compared with gasoline (Fig. , top panel) or of biodiesel relative to conventional diesel (Fig. , bottom panel). High yielding biofuel crops, such as oil palm, sugarcane and cellulosic crops have better overall GHG performances due to less land required than low yielding ones. Likewise, cellulosic bioethanol (or biodiesel) showed some advantages compared with other conventional biofuel crops, such as corn, wheat, sugarcane or oil palm. However, none of the biofuel crops was likely to achieve the 35% (increasing to 50% in 2017) reduction required by the EU directive by 2012 (EC, ) (Fig. , top and bottom panels). The substantial variation observed in Fig. reflected the assumptions made in the individual studies, the type of land displaced, the variation in the production and distribution methods of these biofuels and the different management conditions under which these biofuel crops can be grown. Life cycle greenhouse gas ( GHG ; on a logarithmic scale) emissions including estimates of indirect land use ( iLUC ), carbon intensities from different feedstock sources for bioethanol relative to gasoline (top panel) and for biodiesel relative to diesel (bottom panel). The bars represent the ranges of GHG emissions (in g CO 2 MJ −1 ) of each feedstock over a given amortization period. The numbers behind the biofuel feedstock species (20, 30, 50 years) refer to the amortization period. The horizontal lines represent the reference fuel (i.e., gasoline or diesel) and the European 35% reduction target by 2012. A cradle‐to‐plant approach was adopted for estimates of LCA . Values for GHG emissions per unit of energy for different feedstocks were derived from Fritsche et al . ( ). These values (all in g CO 2 MJ −1 ) are the following: corn: 65; sugarcane: 26; wheat: 45; soybean: 20; palm oil: 43; sunflower: 18; rapeseed: 40; and short‐rotation crops: 14. To put dLUC and iLUC intensities into perspective with respect to savings in emissions due to the substitution of biofuels, the carbon payback time has been adopted as a convenient indicator by several authors (e.g., Fargione et al ., ; Searchinger et al ., ). The payback time for bioethanol ranged from 0 to 93 years for the dLUC and from 12 to 183 years for the iLUC. For biodiesel, the payback time ranged from 7 to 423 years for dLUC and from 15 to 211 years for iLUC (Table ). As in the case of the dLUC and iLUC carbon intensities, significant variation was also observed in the estimates of the carbon payback times across the reviewed studies. Causes of the wide variation in LUC carbon intensity Land classes and proportion converted The differences in the estimates of LUC carbon intensity are caused by a number of assumptions. Different land classes have been considered in the analysed studies, but the most considered ones in all studies were: forest, grassland and cropland. Some studies assumed that the displaced land will be relocated in one land class, such as forest or grassland, whereas others assumed that this relocation will be distributed unequally over several land classes (Table ). These different assumptions on the land classes and on the apportioning partly explain the divergent outcomes in the estimates of carbon intensity among the analysed studies. For a robust modelling of iLUC, proper assumptions on land should account for all land classes available for agriculture and not merely some particular types of land as is the case in most analysed studies. Different techniques have been used to allocate land among land classes. In some models (e.g., GTAP), land allocation is based on decision theories and is governed by prices, whereas in other models, linear programming techniques are used as a land class allocation method. Linear programming allocation techniques provide the optimum area for land use, but do not provide information on the spatial distribution of results. In contrast, decision‐based theory allocation methods provide continuous land consideration maps and allow the consideration of socio‐economic factors. A spatial allocation method has also been tested and proposed (Heiderer et al ., ). Appreciating and adopting one allocation method could reduce inconsistencies in estimates of iLUC carbon intensity. However, whatever allocation method is chosen, sensitivity analyses should always be carried out to assess the effects of the chosen allocation method on the iLUC carbon intensity. Management practices Management practices are important for the estimates of dLUC and iLUC carbon intensity of biofuels. Good farming practices (e.g., no‐tillage) can result in carbon stored in organic matter in the soil, whereas poor farming practices (e.g., ploughing, subsoiling, harrowing) can result in significant emissions and loss of soil carbon. For example, no‐tillage practices increase the cumulative GHG benefits by 15% over 100 years in the grassland conversion case, and by 17% in the forest conversion case as compared with current tillage practices (Kim et al ., ). Considering which management practice is used is important to the outcomes of the LUC carbon intensity. Given that a variety of management practices is used by farmers, it may seem justifiable for the modelling of iLUC to compute the current tillage practices for the baseline scenario, and other management practices (e.g., no‐tillage and no‐tillage plus cover crops) for the projection scenario. In this way, effects of management practices on GHG emissions from iLUC could be quantified and policy measures to promote these management practices as a way to reduce the carbon footprint of crops could be suggested. Carbon stored in the harvested wood products The carbon stored in the harvested wood products at the time of land conversion also has a significant influence on the estimates of LUC carbon intensity. Some studies assumed that this carbon is emitted into the air as CO 2 (Searchinger et al ., ; Overmars et al ., ). This assumption holds if the land is cleared using fire. However, this is not always the case and land may be cleared using other techniques as well. In such cases, the harvested wood might be used as solid fuel in coal power systems (Kim et al ., ), or stored in short‐life wood products like paper and cardboard or long‐life wood products such as timber. In both cases, a carbon credit is gained and must be accounted for in the estimate of LUC carbon intensity. Allocation based on a system expansion could be used to estimate the carbon gained in the case of coal power systems. For the long‐life wood products, estimates of the net carbon gained should be based on stock differences and should account for the fraction of carbon in long‐life wood products in use, as well as in landfills after 100 years across wood product categories, such as softwood timber, softwood pulp, hardwood timber and hardwood pulp. No carbon benefits should be allocated to short‐life wood products as they are net emitters of GHG. Sources of uncertainty Crop yield Yields of energy crops are critical to land use considerations as they define the area of land needed to support the projected biofuel production targets. If yields are small, more land area will be needed to grow a certain amount of biofuel crops, thus releasing more carbon. In contrast, less land will be needed if yield improvements are large (Mathews & Tan, ). Historically, about 80% of the increase in crop production was attributed to improvements in crop yield, and 20% to the expansion of the harvested area. As a consequence, agricultural areas have expanded by only 5% since 1970 (Smith et al ., ). Projecting future land use is confounded by uncertainties about energy crop yields and about crop responses to future climate change. Higher yields can be expected under ideal climatic and agronomic conditions, through breeding or technological improvements including changes in agricultural practices and the use of fertilizers (Ewert et al ., ). Yield may also increase under future atmospheric CO 2 concentrations (Liberloo et al ., ). However, whether crop yields will increase, decrease or remain unaltered in the future depends on agronomic limits and on further progress in crop improvements. Data on crop yield are very crucial for the modelling of iLUC. They are also sources of inconsistencies among the analysed studies (Table ). One way to reduce these inconsistencies would be to use the normalized yield (i.e., the ratio of the actual yield to the regional average) in the baseline scenario assessment of the iLUC associated with biofuels. This approach helps to minimize the impact of different soil properties and farming practices, which vary across geographical regions. It also enables the identification of yield improvements. Coproducts Coproducts have a substantial impact on the requirement for land. The increased availability and the use of coproducts, such as Distillers Dried Grains with Solubles (DDGS) may reduce the pressure on land for animal fodder and for extensive grazing. Indeed, the land requirement for wheat ethanol is reduced from 0.40 ha tonne −1 ethanol to 0.03 ha tonne −1 ethanol (i.e., by 93%) when DDGS are used as a substitute for both soy meal and wheat in the production of animal feed (Lywood et al ., ). This means that more land will be needed for biofuel crops that have few or no coproducts, whereas less land will be needed for biofuel crops that yield large amounts of coproducts. However, the interaction between the production of coproducts and of biofuel production is changing and data on the actual substitution are scarce. Moreover, the feeding value of many coproducts remains uncertain, and there is a limit to the amount of coproducts that can be added to animal fodder. This is partly due to the high nutrient content (e.g., P, K and S) in coproducts relative to the original feedstock, which may have a negative effect on the livestock at high concentrations (Schauer et al ., ). These uncertainties may be exacerbated if new coproducts become available in the future. A better knowledge and a thorough assessment of the feeding value of co‐products will allow for more accurate estimates of the coproduct credits and reduce the uncertainty about the net land required for biofuel production. Many by‐products are deliberately generated in the biofuel production process. However, not all of these by‐products can be classified as coproducts. A market value could be used to differentiate the coproducts of biofuel production and processing from wastes. If a coproduct cannot find a market niche, it has no economic value and should therefore be classified as waste or residue. The economic value of coproducts should not be in any case larger than the main product (i.e., biofuel). Biofuel producers that are unable to provide proof of the sale of their coproducts to fodder industries, to farmers or to energy companies should not claim GHG benefits due to coproduct credits. This differentiation between co‐products and wastes could reduce the uncertainty associated with the estimates of iLUC and could help to only promote energy crops that have a well known market for their coproducts. Carbon stocks of different vegetations The amounts of carbon of various terrestrial carbon pools and land cover types are another source of uncertainty in the estimates of LUC‐carbon intensity. The carbon stock varies by type of vegetation and eco‐regions, and the reported values ranged from 69 to 509 t C ha −1 for forest, from 12 to 162 t C ha −1 for grassland and from 45 to 134 t C ha −1 for savannah (Table ). These variations are attributed to large errors in the spatial distribution of the vegetation biomass as well as to discrepancies in the estimates of land cover and land use change (Houghton et al ., ). More precise information on both changes in area and changes in carbon stocks can thus help to minimize the uncertainty in the LUC carbon intensity of biofuels. Yield elasticity on price The yield elasticity (σ) on the prices is an important parameter of uncertainty in the iLUC estimates. The yield elasticity is the ratio of the change in yield to the change in market prices for a given crop. A high yield elasticity means a reduction in the amount of land needed to cultivate a given biofuel crop, and thus, a reduction in iLUC. Assumptions about the yield elasticity vary considerably between models and studies. For example, the GTAP model assumes a much higher elasticity (σ = 0.25) compared with the FASOM model (σ = 0) or the FAPRI model (σ = 0.01). The yield elasticity also varies with time. For example, the long‐term yield elasticity in the FAPRI model is six times higher than the short‐term yield elasticity (i.e., σ = 0.01) assumed in the EPA analysis (EPA, ). Deciding on which yield elasticity value should be used is difficult as the literature on this issue is rather polarized. For example, in the USA some authors suggest a high response of σ = 0.76 (Houck & Gallagher, ), whereas others support with σ = 0.22 a low response (Lyons & Thompson, ) or even no response. The lack of data on values of yield elasticity for different countries or regions in the world further complicates the issue. As far as the models are concerned, we argue that it should not matter if the demand and price signals are driven by biofuels or other uses of the crop. For example, in the USA the amount of land dedicated to major crops has actually declined over time, whereas the demand for and the outputs of major crops have increased remarkably. So it appears – for the USA at least – that in the past the iLUC was negative: an increased demand resulted in a reduced land area in crop production. Lessons to be learned Models are needed for the assessment of the iLUC carbon intensity because iLUC cannot be determined by experiments. In the case of dLUC, the quantification of carbon intensity is straightforward if the amount of carbon in above‐ and below‐ground biomass as well as in the soil organic matter is known. Both the dLUC and iLUC can significantly alter the GHG benefits of biofuel production (Table ), although some studies (Fritsche et al ., ; Hertel et al ., ) suggest that their effects may be small. Several models and approaches to quantify the iLUC related GHG emissions of biofuels have been published over the last 3 years (Table ). The approaches used vary greatly, ranging from GE to PE models, and simplified models. However, very few models have examined past relations between land and biofuel production, or have examined whether there is any empirical evidence for or against iLUC from the historical data. Predictions of iLUC could be empirically tested by examining past relationships between biofuels production, export patterns from grain producing and importing countries and land use changes (Kim & Dale, ). Key elements for an accurate assessment of carbon gain or loss for an area due to its conversion into cultivated land for biofuel include: the amount of biomass above‐ and below‐ground before (and remaining after) the conversion; the carbon content in these biomass stocks; the different management techniques; the time path of the change in the soil carbon after conversion until a new equilibrium is reached; the influences of climate, temperature, soil quality and rainfall on these key elements. Unfortunately, they were not quantified in every study that estimated the iLUC carbon intensity. Existing iLUC models or approaches predict only a positive iLUC carbon intensity (i.e., increasing emissions). This may not always be the case. The models could also predict a negative iLUC carbon intensity (i.e., net sequestration) if assumptions, such as those that relocate the displaced activities (e.g., grazing) onto degraded lands, unused or marginal lands were investigated in previous studies. Likewise, a negative iLUC carbon intensity could be obtained in case increased yield due to biofuel demands frees some agricultural lands, which could later be returned either into grassland or into forest. This reversion will then increase the soil carbon and lead to a negative iLUC. Many LUC models or approaches assumed that LUC is driven by agricultural expansion and in particular biofuel expansion. However, the issue of LUC is quite complex and not solely driven by agricultural expansion, but rather by a multitude of processes and factors associated with development. In fact, the three proximate drivers (i.e., agricultural expansion, timber extraction and infrastructure development) are present in 25% of the observed cases of land use change worldwide (Geist & Lambin, ). Changes in carbon stock might also be due to historical factors and not a consequence of the more recent LUC. For example, of the 4.5 Pg C lost from Mato Grosso (Brazil) between 1901 and 2006, about 78% of these losses occurred between 1901 and 2001, largely due to land clearing for pasture and croplands (Galford et al ., ). It is thus desirable that future models or studies explicitly consider historical LUC, and capture the major drivers of domestic and global LUC along with those potentially associated with biofuels. Future studies should also investigate whether the presence or absence of a driving factor determines the changes. Without these developments, our understanding of the extent and implications of the LUC of biofuels will remain limited. When a constant yield over time and a high fraction of land from carbon rich pools, such as forests or peatlands are assumed, the resulting dLUC or iLUC carbon intensities are high, regardless of the type of energy crops. However, if yield increases are allowed to continue and the land cover is poor in carbon stock, the resulting dLUC or iLUC carbon intensities will be low. The dLUC (respectively, iLUC) carbon intensities may also be negative if the feedstocks (respectively, the displaced feedstocks) are grown (respectively, relocated) on degraded land. No study assessed the LUC induced GHG emissions of biofuel production on fallow land. When analysed over a short‐term period, biofuel production does not offer any GHG emission benefits as compared with the fossil energy they displace. However, biofuel can still pay back the carbon debt incurred during land conversion over a long‐term period. This suggests that the time frame over which GHG emissions are analysed and the use of a discount rate to value the short‐term period against long‐term emissions can have a significant impact on the net GHG balance of biofuel production systems. Models used in the reviewed studies suggested that most of the LUC impacts will occur in less industrialized and developing countries, such as Brazil, India, Malaysia, China and African countries which are the most competitive producers of biofuel crops (Table ). However, LUC models should not only identify countries or regions where LUC occurs but also the contributions of each ultimate driver to the total impacts. Failing to do this overestimates the LUC carbon intensities associated with biofuels. Finally, the recommendations that GHG emissions of biofuels should be quantified in a full life cycle assessment (LCA) raise the problem of which LCA methodology should be used. There are two basic types of LCA methodologies, an attributional LCA (aLCA) and a consequential LCA (cLCA). The attributional LCA assesses all environmentally relevant physical flows attributed to a production process, whereas the cLCA assesses the changes in environmentally relevant physical flows in response to a decision. These two types of LCA use different data and the choice of conducting either an aLCA or a cLCA depends on the stated goal of the study. If the chosen LCA method for biofuels is different from the one for the reference system, comparability is not guaranteed. For example, in the study of Searchinger et al . ( ), the cLCA used to quantify the GHG emissions of biofuels is wider in scope and includes indirect emissions from LUC, whereas the aLCA used to estimate the GHG emissions of the reference systems (i.e., fossil energy) is narrow in scope and excludes indirect GHG emissions due to the military control over oil reserves. Critical gaps and recommendations for further research The economic and the environmental viability of biofuels depend on productivity. Yield is a key component in biofuel production and in LUC discussions. The improvement in yield depends on breeding and agronomic practices. However, most second generation energy crops (e.g., poplar, willow) are essentially unimproved or have been bred only recently for biofuels, whereas conventional crops (e.g., corn, wheat) have undergone a substantial improvement in yield, in pest resistance and in other agronomic traits. A more complete understanding of the biological system coupled to management practices and to biotechnological advances will speed up energy crops with desirable attributes, such as increased yields and usability, optimal growth, better pest resistance, efficient water and nutrient use and greater resistance to stress. The land allocation module in existing models is restricted to only three land classes: forest, grassland and cropland. Other land classes are not included in this module. Further research is thus needed to extend the land class allocation problem to include other land classes, such as abandoned land, marginal land or degraded lands. Such an extension offers a basis for studying many empirical and policy questions like short‐ and long‐term biofuel demand, dynamic changes and allocation issues in forest conservation and deforestation. Future iLUC models should be improved to include sets of parameters and assumptions (e.g., relocation of displaced activities to degraded lands) that could enable models to predict a negative iLUC carbon intensity. Without this development, our understanding of the full impact of LUC of biofuels will be limited and the validity of existing models will be questioned. Also, a standard and transparent approach to depict the iLUC carbon intensity that can be added to an LCA needs to be developed to correctly assess the sustainability of biofuels. To the extent of our knowledge, no such standard methodology exists or is one under development. Ideally, such an approach should be based on LCA and include all drivers of LUC, all biomass sectors, all land classes and coproducts. It should also consider historical LUC and the fate of the cleared carbon, account for the effects of management and agricultural intensification, conversion efficiency, as well as improvements in yield. Finally, lack of data greatly contributes to the uncertainty about the LUC carbon intensity of biofuels. Datasets on biofuel crop production must be collected, synthesized and standardized to common data formats so as to minimize uncertainties in the input data for the iLUC models. Concluding remarks Models to deal with dLUC and iLUC exist, but each of the models gives different results. The variation in the estimate of the LUC carbon intensity of biofuels is due to model structures, to different data sets and to a number of assumptions made to model LUC. Despite substantial variations and uncertainties involved in the quantification of LUC carbon intensities, this study shows that in some cases, the LUC can potentially alter the GHG benefits of biofuels. Consequently, there is a substantial risk that current biofuel policies will lead to an increase in GHG emissions if emissions from LUC are not accounted for in the life cycle of biofuels. There is currently no way to determine which of the many models yields the most reliable overall LUC carbon intensity. Deciding how to estimate the LUC carbon intensity of biofuels will therefore have major implications for biofuel policies. Finally, key gaps not included in the LUC modelling at present need to be addressed to improve our understanding of biofuel LUC impacts. These include the potential use of coproducts to decrease the impacts of LUC, management practices and the inclusion of other land use classes, such as peatland and fallow land in existing modelling tools. Acknowledgements The research leading to these results has received funding from the European Research Council under the European Community's Seventh Framework Programme (FP7/2007–2013), ERC grant agreement nr. 233366 (POPFULL). We thank Florian Gahbauer for language corrections, four anonymous reviewers for their helpful comments and several authors for providing more detailed information on their published results.

Journal

GCB BioenergyWiley

Published: Jul 1, 2012

There are no references for this article.