Impact of Policy on Greenhouse Gas Emissions and Economics of

May 14, 2014 - However, studies consistently report a high degree of uncertainty in these emissions estimates, raising questions concerning the carbon...
0 downloads 0 Views 2MB Size
Download PDF
Article pubs.acs.org/est

Impact of Policy on Greenhouse Gas Emissions and Economics of Biodiesel Production Elsa Olivetti,*,† Ece Gülşen,‡ Joaõ Malça,§ Érica Castanheira,∥ Fausto Freire,∥ Luis Dias,¶ and Randolph Kirchain‡ †

Department of Materials Science & Engineering, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, Massachusetts 02139, United States ‡ Engineering Systems Division, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, Massachusetts 02139, United States § Department of Mechanical Engineering, ISEC, Coimbra Polytechnic Institute, 3030-199 Coimbra, Portugal ∥ ADAI-LAETA, Department of Mechanical Engineering, University of Coimbra, Pólo II Campus, Rua Luís Reis Santos, 3030-788 Coimbra, Portugal ¶ INESC Coimbra, Faculty of Economics. University of Coimbra, Rua Antero de Quental 199, 3000-033 Coimbra, Portugal S Supporting Information *

ABSTRACT: As an alternative transportation fuel to petrodiesel, biodiesel has been promoted within national energy portfolio targets across the world. Early estimations of low lifecycle greenhouse gas (GHG) emissions of biodiesel were a driver behind extensive government support in the form of financial incentives for the industry. However, studies consistently report a high degree of uncertainty in these emissions estimates, raising questions concerning the carbon benefits of biodiesel. Furthermore, the implications of feedstock blending on GHG emissions uncertainty have not been explicitly addressed despite broad practice by the industry to meet fuel quality standards and to control costs. This work investigated the impact of feedstock blending on the characteristics of biodiesel by using a chance-constrained (CC) blend optimization method. The objective of the optimization is minimization of feedstock costs subject to fuel standards and emissions constraints. Results indicate that blending can be used to manage GHG emissions uncertainty characteristics of biodiesel, and to achieve cost reductions through feedstock diversification. Simulations suggest that emissions control policies that restrict the use of certain feedstocks based on their GHG estimates overlook blending practices and benefits, increasing the cost of biodiesel. In contrast, emissions control policies which recognize the multifeedstock nature of biodiesel provide producers with feedstock selection flexibility, enabling them to manage their blend portfolios cost effectively, potentially without compromising fuel quality or emissions reductions.

■

discussed among researchers16−20 and several popular press articles have been published highlighting the uncertainty of the GHG benefits of biofuels including biodiesel.21−26 LCA studies have demonstrated that the cultivation of feedstocks, and particularly the land use change (LUC) associated with the conversion from reference land to cultivated land, dominate biodiesel lifecycle emissions and the variation in emissions.27,28 Additionally, emissions estimates tend to be quite sensitive to variable agricultural and production inputs, particularly for cultivation, and are subject to inherent uncertainty surrounding emissions processes that are not well-understood, such as emissions due to the changes in soil carbon stock.12 A large and growing number of potential feedstocks with different

INTRODUCTION Promotion of biodiesel as a transport fuel has been supported by governments as part of efforts to reduce dependence on finite petroleum fuels, achieve national energy independence, and cut down tailpipe emissions. Given that transport accounts for about 23% of energy-related carbon dioxide emissions and is projected to rise to more than 40% by 2050,1 biodiesel also offers a lowemissions solution for future transport challenges in combating global warming. Life cycle assessment (LCA) studies, however, have reported an extremely broad range of greenhouse gas (GHG) emissions from biodiesel production. In fact, GHG emissions estimates as high as 700% that of petrodiesel and as low as −150% (sequestration) have been reported under different lifecycle scenarios.2−15 Whatever the cause, these results make it clear that lifecycle biodiesel emissions are highly variable and uncertain. Because of this uncertainty, whether or not biodiesel has GHG emission benefits with respect to petrodiesel is controversial. This controversy has been heavily © 2014 American Chemical Society

Received: Revised: Accepted: Published: 7642

December 4, 2013 May 12, 2014 May 14, 2014 May 14, 2014 dx.doi.org/10.1021/es405410u | Environ. Sci. Technol. 2014, 48, 7642−7650

Environmental Science & Technology

Article

based on the EU directives has been developed. Through the development of optimization models for biofuel production decisions, modeling of GHG emissions of feedstocks, and comparison of different policy frameworks around allowable emissions, this study investigates whether feedstock blending can be used as a tool to directly manage uncertain emissions characteristics of the final fuel without compromising environmental performance. In other words, we explore whether explicit consideration of uncertainty (in GHG emissions as well as technical properties of feedstocks) in blending decisions provides economic benefit to biofuels producers. Through case-based examples, we assess potential economic and GHG implications across different biofuel policy frameworks to understand whether certain policies can translate to less feedstock diversification, which can then translate to economic inefficiencies.

characteristics further contributes to the uncertainty challenge for LCA of biofuels, requiring the collection of vast amounts of information from multiple decision makers along several supply chains.29 Some of the observed variation in GHG emissions derives from explainable (and potentially controllable) variability in feedstock cultivation. The balance derives from inherent uncertainty in knowledge or measurement of emissions. In the interest of brevity, in the balance of this paper, these two sources of variation will be referred to as uncertainty (except where doing so would lead to confusion). More importantly, however, this paper explores how controllable sources of variation can in fact be leveraged to manage the resulting uncertainty in biodiesel emissions. The risk of increasing GHG emissions due to increased biodiesel production has alerted governments to implement emissions control policies (ECPs). The European Renewable Energy directive requires that for biofuels and bioliquids, the GHG emissions savings shall be at least 35%. The requirement will be increased to 50% in 2017, and for biofuels produced in installations where production started on or after 1 January 2017, the savings shall be at least 60% from 1 January 2018.30 In the U.S., GHG thresholds range between 20% and 50% savings depending on the type of feedstock used and the production conditions. As the ECPs are being designed, government-funded research toward more accurate and comprehensive emissions estimations is in progress both in the EU and the U.S.,30−33 and several policy incentives such volume credit trading approaches are available that allow trading of credits generated from biofuels that must also meet emissions reduction thresholds.32,34 The provision of government support is conditional upon meeting certain emissions reduction criteria which are usually established in relation to the emissions of petroleum fuels. More importantly, individual biodiesel pathways (lifecycle stages from feedstock cultivation to fuel production) are evaluated against the established criteria based on their estimated mean emissions. Yet, mean emissions only partly define the emissions characteristics of a biodiesel pathway, and carry the risk of overlooking wide ranges of inherent variation, leading to complications regarding the compliance with the emissions reduction criteria.2,17 Given these complications, emissions performance of (and therefore requirements for) a biodiesel pathway may be better represented with a probability distribution. Probabilistic representation of emissions performances in ECPs has been adopted to a certain extent,33 with individual biodiesel pathways that have 50% and above probability of meeting the criteria considered “compliant”. Despite an increasing recognition of the existence of uncertainty in GHG emissions for individual biodiesel pathways, to date, there has been no particular attention to the prevalent feedstock blending practices in these discussions, and to how blending might change the resulting uncertainty characteristics of the final fuel. However, previous studies in other materials processing systems, such as the recycling or paper industry, have shown that the distributional characteristics of uncertainty in a blend can be substantially different than that of the constituent components.35,36 Moreover, as Mullins et al. has shown for individual biofuel feedstocks, the probability of meeting the emissions policy target is very sensitive to the underlying distributional characteristics.17 Recognizing this gap, this paper investigates the impact of feedstock blending on the lifecycle GHG emissions characteristics and on the cost of biodiesel by implementing a chance-constrained (CC) blend optimization methodology that explicitly considers uncertainty and minimizes feedstock costs at a given time subject to technical and emissions constraints. To investigate GHG emissions constraints, a model that estimates LUC scenario-specific lifecycle GHG emissions

■

MATERIALS AND METHODS Chance-Constrained Optimization Model. To address the complexities associated with feedstock blending decision making, particularly in a GHG-constrained policy landscape, a chance-constrained (CC) optimization model was developed. Conventional approaches to incorporating uncertainty information into an optimal blending problem are known to result in overestimation of uncertainty, and therefore can lead to lower profitability.35,37,38 Optimization under uncertainty has been extensively studied over recent years, and the methods developed constitute promising solutions to increase performance of decisions under uncertainty.39,40 CC optimization, first formulated by Charnes and Cooper,41 is one such method that has been implemented in various optimization problems governed by compositional uncertainty. Owing to its capability to explicitly consider, propagate, and control uncertainty in a mix of uncertain constituents, CC optimization offers a great potential for multifeedstock biodiesel production. In addition, one key characteristic of the CC method is the increased diversification of the solution portfolio compared to conventional ways of formulating constraints, which can benefit the producers by diversifying supply chains and reducing risk to price volatility. Model parameters and the constraints of the CC optimization were chosen such that the resulting optimal blend complies with the currently deployed technical and emissions standards. Similarly, historically observed prices were used to model the feedstock prices. For the analysis presented here, the objective was to minimize the total feedstock cost, P(x), through cost optimal allocation of feedstock volumes, Ai (eq 1), which were the decision variables. This optimization was subject to constraints around normalized demand, DN = 1, as shown in eq 2. Constraints on GHG emissions were added to this formulation (an addition from the previous work by the same authors) and are described generally in eq 3; details on the GHG constraints are provided below. For each technical feedstock property (IV, CN, CFPP, OS) the composition of the final fuel must not violate the technical specifications (eqs 4 and 5). min : P(x) =

∑ PA i i i

subject to:

∑ Ai = D N = 1 i

(2)

μGHG + Χ(α)σGHGB ≤ GHGcons

(3)

Q̅ B + Χ(α)σQ B ≤ Q cons

(4)

B

7643

(1)

dx.doi.org/10.1021/es405410u | Environ. Sci. Technol. 2014, 48, 7642−7650

Environmental Science & Technology

Article

R̅B − Χ(α)σRB ≥ R cons

(5)

Ai ≥ 0

(6)

Because feedstock cost is estimated to be a major part of the production cost (for biodiesel produced by trans esterification, feedstock costs represent between 80% and 85% of the total production cost),43−48 decisions on blending recipe and cost reduction opportunities are strongly dependent on the feedstock prices. These prices not only differ from each other across feedstocks, but also fluctuate to a significant extent over time, thereby changing also with respect to each other. A plot demonstrating these fluctuations is shown in the Supporting Information, Section C. For the four feedstocks considered in this study, the relative price ranking is typically canola > soybean > palm with sunflower varying in rank relative to the others over time. When the relative prices among feedstocks shift based on the market conditions, a producer might need to modify the feedstock proportions used in the batch to remain profitable. The correlations among deflated feedstock prices are either fairly weak and positive, or relatively strong and negative. Given this lack of strong positive correlations, the price behavior suggests that maintaining a diversified blend portfolio could be helpful to hedge against unexpected price changes in the market.49 The ability to quickly adjust the blend portfolio in response to availability and price dynamics as well as comprehend the GHG impacts could bring substantial value to biodiesel producers. Currently, the biodiesel market does not capture the GHG reduction value of different feedstocks. In addition, because feedstock properties are multidimensional and the value of specific properties varies due to the operational strategy of an individual plant, there is no apparent correlation between feedstock properties and their estimated GHG emissions. For all of these reasons, an optimization model provides not only value to an individual producer but also insight around the potential implication of GHG policy on the individual producer. Lifecycle GHG Emissions Prediction Model. A Note on Variation and Uncertainty. Variation in feedstock performance, whether technical or environmental performance, derives from both variability (e.g., differences in feedstock characteristics or cultivation practices) and inherent uncertainty (e.g., limitations in measurement technology, underdeveloped scientific knowledge, and yet unrealized future activities). Conceivably, sources of variability are knowable and can therefore be controlled by the decision-maker while uncertainty cannot be directly controlled. For any real-world decision, the partition between variability and uncertainty is inherently contextual depending upon available information and the range of possible decisions. For example, while feedstock fatty-acid content could be measured, none of the biodiesel producing companies interviewed for this study indicated they intended to do so. As a consequence, for the context of these relevant decision-makers, feedstock fatty-acid content is uncertain. Because the data collection infrastructure around agricultural carbon burden is still evolving, the authors had to select which technically knowable attributes of a feedstock would be assumed to be known to the decision-maker (and therefore controllable) and which would be unknown and uncertain. In doing so, the authors are implicitly assuming that knowable drivers of variability are effectively random. In a market where lower carbon emissions are rewarded we would expect that certain such drivers (e.g., lower carbon tillage and management practices) will become more prevalent. Based on current practices and pending requirements within the European Renewable Energy directive, we selected crop, cultivation location (i.e., soil type and climate region), and reference land use as knowable information, while cultivation practices were assumed to be unknown and uncertain. The mapping

where Pi: feedstock i, price, Ai: feedstock i, volume proportion, D: normalized total demand, X(α): normal distribution test coefficient, one-tailed, μGHGB: GHG emissions in blend, mean value, Q̅ B(or R̅ B): property Q (or property R), in blend mean value, GHGcons: GHG constraint level, Qcons, (Rcons): property Q (or property R), constraint level, σGHGB: standard deviation for GHG in blend, σQB, σRB: property Q (or property R) standard deviation in blend. Standard deviations on the properties of the blend were derived based on the generalized equation as shown in eq 7: σϕB =

∑ ∑ ρij σϕ σϕ AiAj i

i

j

j

(7)

where ϕ represents a specification, ρij is the correlation coefficient between feedstock i and j. By definition ρij = 1 when i = j. Because we assumed no correlation between feedstock properties, ρij = 0 when i ≠ j. In other words, all feedstock scenarios considered in the model were regarded to be statistically independent from each other, whether or not they belong to the same crop species. The volume proportions of each feedstock to be blended, namely Ai, constituted the decision variables of the optimization problem. Choosing a set of Ai values defined the blend recipe and, therefore, determined the chemical composition of biodiesel. Using the previously developed prediction model that related the chemical characteristics with technical properties of the composition, IV, CN, OS, and CFPP could be predicted. The derivation of these properties and the details of standard deviation estimations can be found in the previous work.42 Finally, X(α) was a user input that determines the maximum accepted noncompliance rate based on the value of α. For all the technical properties considered in the model, α was chosen as 95%, allowing a maximum 5% probability of noncompliance rate for each property. On the other hand, we relaxed α down to 80% confidence level for the GHG emissions constraint where applicable. This implies that there is 80% confidence level on the batch emissions or feedstock emissions depending on the ECP framework in consideration. Physical Property Prediction Model. Because the chemical content of each feedstock is slightly different, the quality of biodiesel might vary depending on its feedstock composition. While the producer needs to ensure that the final blend is qualified with respect to each technical specification, a conservatively compliant batch might lead to higher costs with no advantage in market value. A physical prediction model based on feedstock composition may help the producer plan compliant batches with cost reduction opportunities. Therefore, in investigating the impact of blending on the GHG distribution of biodiesel, a model previously developed by the authors was used to predict the key performance attributes of a fuel produced from a blend of raw materials based on the final iodine value (IV), oxidation stability (OS), cold filter plugging point (CFPP), and cetane number (CN). These properties were chosen through a series of interviews with six biodiesel companies, based on what they indicated to be most challenging properties to meet. The details of this model can be found in previous work by the authors.42 7644

dx.doi.org/10.1021/es405410u | Environ. Sci. Technol. 2014, 48, 7642−7650

Environmental Science & Technology

Article

statistical characteristics were estimated for a range of cultivation location (i.e., soil-climate combinations) and reference land use scenarios. Other factors exerting influence on the LUC emissions for a biodiesel feedstock are crop yield, oil extraction yield, and coproduct allocation factor. The RED provides characteristic values of these factors for each crop. The values for creating these distributions were taken from the procedures listed in the Commission Decision.30 The details of the assumptions used in this paper for calculating LUC emissions can be found in the Supporting Information, Section A. The above approach assumes that land conversion is driven by the production of biodiesel feedstocks as opposed to other drivers, and that past land conversion patterns will remain constant in the future and should be allocated to biodiesel feedstock production. Case Development. Identifying Relevant Feedstock Scenarios. As part of the research that led to this paper, the authors conducted interviews with six firms producing biodiesel in Portugal (and in some cases in other parts of Europe). All of these producers made use of multiple oils including soy, canola, palm, and sunflower. It should be noted that two (out of six) of the Portuguese companies do extraction of oil from canola, soy (and some minor amount of sunflower) on site, while palm oil is bought from the global market. Palm oil extraction occurs in close proximity to the plantation. To understand the implication of feedstock blending choices, the authors created a diverse set of feedstock scenarios that attempt to represent plausible cultivation alternatives for these crops. Geographical Mapping for Soil-Climate Combinations of Major Biodiesel Feedstocks. To determine the most likely cultivation locations in terms of soil-climate combinations for each feedstock, and thus limit the number of scenarios considered within the optimization model, a geographical mapping methodology was developed that adheres to EC guidelines. First, the Statistics Division of the Food and Agricultural Organization of the United Nations global crop production database, FAOSTAT,50 was used to determine the production quantity at the country level for various crops including canola, soybean, sunflower and palm. For each crop, the top producer countries were identified such that, when combined, they contribute to at least 80% of the annual global production during the period between 2005 and 2009. Next, global soil and climate layers were separately obtained from the EC’s online transparency platform.51,52 The climate regions were defined by a set of rules based on factors such as annual mean daily temperature, total annual precipitation, total annual potential evapo-transpiration and elevation, and are classified based on the IPCC guidelines. Using spatial analysis techniques, the dominant soil type and climate region combinations of the top producer countries were determined. This, in turn, allowed the determination of the most likely soil and climate conditions for each feedstock under consideration. More details can be found in the Supporting Information, Section B. Identifying Applicable Land Cover Types. To develop out the scenario set, we identified and considered the most likely land conversion scenarios for each feedstock. The EC considers land cover types under five main categories: cropland, perennial vegetation land, perennial cropland, grassland, and forestland. Therefore, when arable land is used for crop cultivation, a number of potential land cover conversions might occur: grassland-tocropland, forestland-to-cropland, cropland-to-cropland, etc. It was assumed that changes resulting from crop rotation or other

exercises described below identified 75 crop-location-reference land use scenarios (each with uncertain emissions characteristics) that were most likely for the four crops considered here. Emissions without Land Use Change. Over the last two decades dozens of studies have been published which quantify GHG emissions associated with various aspects of the cultivation and production of biofuels. These studies have made it clear that many factors influence the intensity of emissions and that there is still much to learn. This paper does not aim to advance the methods of quantifying GHG emissions from a single biodiesel cultivation-production pathway. Nevertheless, to explore the implications of blending practices it was necessary to develop estimates of GHG emissions for a large set of potential cultivation-production pathways, which included estimates of uncertainty, and which were developed using a consistent framework for evaluation (and were therefore comparable). The last requirement precluded assembling estimates directly from a broad collection of published studies. Instead, this study developed an approach for identifying plausible cultivationproduction pathways from across the globe and then applied published guidelines described below to estimate the emissions and uncertainty in emissions for those pathways. GHG emissions occur at various stages of the path from crop to biodiesel. At a high level, these stages can be categorized as cultivation, LUC and indirect LUC (iLUC), processing, transport and distribution. This study used the European Commission’s (EC) guidelines to estimate the processing, transport, and distribution emissions for feedstocks as described in several references.15,17,27 However, in order to represent the inherent uncertainty as discussed by several papers in the literature,2−15 we added a coefficient of variation (COV) of 20% to these deterministic, no-LUC emissions estimates. This value was derived from analogous measures of uncertainty;30 furthermore, this uncertainty relative to the LUC emissions was small. This analysis does not explicitly account for iLUC, but the case analysis incorporates sensitivities including and excluding several of the LUC scenarios to account for the challenges in fully separating LUC from iLUC. These will be described further in the Results section. Emissions from Land Use Change. Conversion of land for cultivation of raw biofuel materials might lead to release of stored carbon in the reference land over time. LUC emissions stem from this carbon release and the EC’s Renewable Energy Directive (RED) clearly states that the full carbon effects of biofuels should be calculated by taking the LUC emissions into consideration. There are many factors that impact the LUC emissions of cultivation. In an attempt to describe the statistical characteristics of LUC emissions, this work developed a methodology that follows the RED. According to the RED, LUC emissions of feedstock cultivation can be determined if the following are known: 1 soil type and climate region of cultivated land 2 reference and actual land cover types 3 cultivation practices (i.e., tilling practice) and inputs (i.e., fertilizer and manure) Specifically, the RED provides estimates (including uncertainty in the estimate) for LUC emissions associated with each of these factors. The results presented here assume that the biodiesel producer can know the crop, cultivation location (i.e., soil type and climate region), and reference land use as knowable information, while cultivation practices and inputs were assumed to be unknown and uncertain. As such, using the RED approach 7645

dx.doi.org/10.1021/es405410u | Environ. Sci. Technol. 2014, 48, 7642−7650

Environmental Science & Technology

Article

Figure 1. Summary of 75 biodiesel lifecycle GHG emissions scenarios. Mean emission estimates are shown relative to petrodiesel (emissions at 1). Black bars represent the lifecycle scenarios without LUC. Dotted red line at 65% of petrodiesel.

cropland-to-cropland conversions do not lead to LUC emissions (no-LUC scenario). The likelihood of the remaining land cover conversions or LUC scenarios was evaluated given the global production and geographic data collected for each major feedstock. First, the land conversion statistics for 2001−2009 from the top producers, determined from satellite imagery, were used as a proxy to eliminate unlikely land conversion scenarios.53 We used information for the percentage of land conversions from forest, grassland, savanna and shrub to cropland by country and weighted by production volume to determine the most likely land conversion scenarios for each major feedstock. This approach assumed that land area used for cultivation is proportional to the production volume of that country (neglecting differences in agricultural yields across countries). This led to six total land conversion scenarios for implementation within the optimization model (grassland-to-cropland, perennial vegetation land-tocropland and forestland-to-cropland for canola, soybean, and sunflower; perennial vegetation land-to-perennial cropland, grassland-to-perennial cropland, and forestland-to-perennial cropland for palm). Figure 1 summarizes the mean emissions estimates of the 75 scenarios, normalized to petrodiesel emissions. The four black bars refer to the no-LUC emission estimates. Out of the remaining 71 LUC scenarios, 30 belong to canola, 10 to soybean, 27 to sunflower, and 4 to palm oil. Within a feedstock type, each scenario represents a type of soil and climate region of cultivated land, a land cover conversion and a particular cultivation practice. As will be described subsequently, only 13 out of 75 scenarios would qualify under ECP in the EU (ECPEU), which only allows feedstocks with mean emissions less than 65% of petrodiesel emissions. Each scenario has a corresponding estimated standard deviation which ranges from 23 to 65% of the scenario mean.

Figure 2. Optimal blend portfolios and costs, April 2007. The full list of feedstocks for each scenario is provided in the Supporting Information, Section D.

■

When 20 feedstocks are available, the CC optimization results blended a small portion of each feedstock to arrive at a lower cost than when 3 feedstocks are available (the model is infeasible for 2 and 1 feedstocks). The optimal blend is highly diversified when the number of available feedstocks is high and the cost nonlinearly increases from $479.8/ton to $486.3/ton (1.4% increase) in deflated dollars as the limit on the number of feedstocks tightens, thereby demonstrating the resulting relationship between diversification and cost reduction. The dynamic character of the model can be better understood by observing the use of an example feedstock from a particular LUC scenario, sunflower (shown in purple, sunflower cultivated

RESULTS Given the methods described above to model GHG emissions and technical properties, the CC optimization formulation was used to evaluate a series of ECP cases. Before these cases are shown, Figure 2 demonstrates an initial result of the CC optimization, blending up to 20 of the (LUC) scenarios for canola, soybean, sunflower and palm feedstocks, using the feedstock prices in April 2007. Each column in Figure 2 shows the percentage of the total blend made up by each feedstock as a function of the number of feedstocks made available, and the resulting cost of the blend on the secondary y-axis (black line). 7646

dx.doi.org/10.1021/es405410u | Environ. Sci. Technol. 2014, 48, 7642−7650

Environmental Science & Technology

Article

on a cropland which was converted from grassland, high activity clay soils, and a cold, temperate, dry climate) in different blends. When the number of feedstocks is high, it comprises around 5% of the total blend. As the feedstock number is lowered to 6, this scenario is included at 14%, and increases to 51% when there are only 3 feedstocks. This example illustrates that restricting the feedstock scenario options for the producers might result in an increase in the cost of biodiesel, particularly as the feedstock prices fluctuate over time. Case: ECP Frameworks for Biodiesel Lifecycle GHG Emissions. Two potential ECP frameworks were initially considered; a third will be shown in the discussion below. The first framework, ECPEU, represented the European Union policy where specific feedstocks with estimated emissions higher than a regulated threshold value are excluded from the biodiesel market supported by governmental fiscal programs. Therefore, the mean emissions of individual feedstocks in the blend could be at most 65% of petrodiesel, eliminating all but 13 of the feedstock scenarios derived above (Figure 1). In this framework there was no constraint on the emissions of the blend (but the mean emissions of the blend can be determined). The second ECP framework, ECPCC, relied on the CC model to control for emissions of the biodiesel blend, but did not have constraints on emissions on the individual constituent feedstocks. ECPCC required that a blended biodiesel batch meets a regulated emissions distribution threshold (65% that of petrodiesel) at a certain confidence level (80%). ECPCC offered more flexibility for the producer while still allowing regulatory criteria to be met. It should be noted that ECPCC is very similar to the framework currently implemented in California under the Low Carbon Fuel Standard (LCFS). LCFS sets an overall emissions (intensity) reduction target to be met each year.54

Based on the descriptions given above, GHG emissions constraints under different policy frameworks can be formulated as in eqs 8 and 9:

ECPEU |μi ≤ ε

(8)

ECPCC|μ + Χ(α)σGHGB ≤ E

(9)

where μi: feedstock, i, mean emissions, μ: mean emissions of blend, ε: feedstock emissions constraint, E: blend emissions constraint, σGHGB: standard deviation of emissions of blend. There was a final set of conditions that were of interest. Because the ECPEU was restricting the individual feedstocks to 65% of petrodiesel, the resulting blend emissions under ECPEU would not be higher than and were generally lower than 65% (the blend emissions criteria for ECPCC). Therefore, in the Discussions section below we propose a third framework to consider. This third framework sets the blended emissions, E, at or below the blend emissions of ECPEU. Single Period Price Data Analysis. First, consider single period price scenarios to observe the differences in optimal blend portfolios under ECPEU and ECPCC. Figure 3a and b show the resulting feedstock blends for each ECP using the deflated feedstock prices observed in January 2005 (two left-hand bars) and in April 2007 (two right-hand bars); the results of prices from 2003 to 2011 will be presented in the next section. For January 2005, these prices were $733, $428, $1072, and $341 per ton of canola, soybean, sunflower and palm oil, respectively. For April 2007, these prices were $567, $485, $457 and $438 per ton of canola, soybean, sunflower and palm oil, respectively. ECPCC leads to 4% and 10% lower feedstock cost for January 2005 and April 2007, respectively. In addition, ECPCC provides more diversified portfolios (as more feedstocks are available given the

Figure 3. Optimal blend portfolios and feedstock costs of ECPEU and ECPCC for (a) January 2005, (b) April 2007. (c) Emissions distribution for ECPEU’s and (d) ECPCC’s optimal blend portfolios in April 2007. 7647

dx.doi.org/10.1021/es405410u | Environ. Sci. Technol. 2014, 48, 7642−7650

Environmental Science & Technology

Article

represented by black dots, show that implementing ECPEU instead of ECPCC results in an average $35/ton higher feedstock costs over time, or a 6% increase. Given the high volumes of biodiesel production, these differences add up to significant expenses for producers. While there is significant overlap in the ranges of costs of these two solutions, the net benefit of ECPcc becomes clearly apparent when the solutions are compared day by day and then the benefits aggregated. Specifically, the cumulative cost savings of ECPCC compared to ECPEU over the time period between 2003 and 2011 is approximately $30 million dollars for a biodiesel producer producing 100,000 tons/year. Also notice that imposing GHG limits on the feedstock level under ECPEU results in more variable feedstock costs over time. When ε = 65%, the producer is only allowed to use canola and sunflower, and this makes feedstock costs more exposed to price fluctuations in the market. This is a significant result that illustrates the impact of emissions control policies on the potential cost of biodiesel in the market.

absence of an individual feedstock constraint; 13 feedstocks for ECPEU as opposed to 24 for ECPCC) as is shown in Figure 2. Figure 3c and d show the emissions distributions of the optimal blends under ECPEU, and under ECPCC depicting the performance of the blends with respect to the 65% petrodiesel emissions threshold for April 2007. The distribution for ECPEU shows that this constraint was met with only 69% probability, whereas for ECPCC the constraint was met with 80% probability. The mean emissions for ECPEU are lower than ECPCC, 60% versus 63%. As opposed to the technical biodiesel constraints that need to be met by each batch, GHG emissions matter on average and the mean emissions is a relevant metric for assessing performance over time. This analysis does not directly account for indirect land use change, but this approach assumes that there is no leakage or shifting by producers. In reality, however, there is leakage and shifting of land use by producers, making this separation between LUC and iLUC challenging. Therefore, three sensitivities were performed in Figure 3. For the first of these, the lifecycle scenarios without LUC were removed (cf. Figure 1) for each feedstock and for the second and third, an average scenario for the most typical land use change was chosen. In other words, for this latter case, the LUC emissions were allocated to a type of feedstock regardless of its origin using the average as a proxy (for palm: forest to perennial cropland, for soy: perennial cropland to cropland, and for canola: grassland to cropland was chosen).3,16,55 For the first sensitivity, the overall change in cost was less than 0.05% and there was a slight increase in the amount of canola in the final blend. This resulted in a cost increase of 1.4% from the ECPCC result. Multiple Period Price Data Analysis. A total of 102 monthly price sets were used in the model, covering the range between January 2003 and July 2011. Figure 4a shows the average percent use of each feedstock for the optimal blends at each price point from 2003 to 2011. Note that multiple LUC scenarios are included within each feedstock. ECPCC has a more diverse portfolio, indicating that each single crop might be economically valuable for the producer over the period of multiple years. Policy frameworks could potentially restrict this flexibility at a potential cost to producers. This suggests that, in the absence of specific feedstock constraints, producers should be prepared to include chemically different crops into their portfolios at any point in time in order to take advantage of fluctuating feedstock prices in the market. Figure 4b summarizes the costs of the resulting multi period cost blends under ECPEU, and ECPCC. Minimum and maximum points refer to the fifth and 95th percentiles, and the rectangles represent the first and third quartiles. The average costs,

■

DISCUSSION The higher blended mean emissions of ECPCC compared to ECPEU suggests an additional framework that leads to more comparable blended emissions with ECPEU. For this new framework, ECPCC‑EU, there was no restriction on the individual feedstocks; however, the mean GHG emissions of the blend should now be equal to or less than the mean emissions of the optimal blend found under ECPEU. All 75 LUC scenarios are made available, but the blend GHG constraint was modified such that the mean emissions of the resulting portfolio is 60% (identical to the mean emissions observed for ECPEU), modifying the constraint found in eqs 3 and 8 to be μ ≤ μECPEU. In order to obtain a blend portfolio with 60% mean emissions, we reduced the constraint on E from 65% to 63%. The composition of feedstocks in the blend included all four feedstocks with 19% palm, 51% sunflower, 4% soybean, and 26% canola. Although ECPCC‑EU has the same mean emissions as ECPEU, it demonstrates a tighter distribution. In addition, the confidence level with respect to the 65% emissions threshold is observed at 97%, which is an improvement over 80%. The resulting blend for April 2007 price data consists of 17 feedstocks and costs $480.3/ton, only slightly higher than ECPCC. Although certainly not conclusive, these results suggest that a blendbased policy could allow for lower production cost with comparable mean emissions and superior emissions risk characteristics. These results demonstrate that GHG emissions thresholds have significant implications on the cost of biodiesel depending on the relative prices observed at a given point in time. As expected, relaxing the GHG emissions constraint leads to more diversified blends because of the increased diversity of scenarios

Figure 4. Multi period price analysis for the two frameworks (a) blend composition and (b) price from 2003 to 2011. 7648

dx.doi.org/10.1021/es405410u | Environ. Sci. Technol. 2014, 48, 7642−7650

Environmental Science & Technology

Article

(7) Malça, J.; Freire, F. Life-cycle studies of biodiesel in Europe: A review addressing the variability of results and modeling issues. Renewable Sustainable Energy Rev. 2011, 15 (1), 338−351. (8) Plevin, R. J.; O’Hare, M.; Jones, A. D.; Torn, M. S.; Gibbs, H. K. Greenhouse gas emissions from biofuels’ indirect land use change are uncertain but may be much greater than previously estimated. Environ. Sci. Technol. 2010, 44 (21), 8015−8021. (9) Pradhan, A.; Shrestha, D.; McAloon, A.; Yee, W.; Haas, M.; Duffield, J.; Shapouri, H., Energy life-cycle assessment of soybean biodiesel. Agric. Econ. Rep. 2009, 845. (10) Reijnders, L.; Huijbregts, M. Biogenic greenhouse gas emissions linked to the life cycles of biodiesel derived from european rapeseed and Brazilian soybeans. J. Clean. Prod. 2008, 16 (18), 1943−1948. (11) Schade, B.; Wiesenthal, T.; Gay, S. H.; Leduc, G. Potential of biofuels to reduce greenhouse gas emissions of the European transport sector. Transp. Moving Clim. Intell. 2011, 243−269. (12) Soimakallio, S.; Mäkinen, T.; Ekholm, T.; Pahkala, K.; Mikkola, H.; Paappanen, T. Greenhouse gas balances of transportation biofuels, electricity and heat generation in finland–dealing with the uncertainties. Energy Policy 2009, 37 (1), 80−90. (13) Thamsiriroj, T.; Murphy, J. Is it better to import palm oil from Thailand to produce biodiesel in Ireland than to produce biodiesel from indigenous Irish rape seed? Appl. Energy 2009, 86 (5), 595−604. (14) Thamsiriroj, T.; Murphy, J. Can rape seed biodiesel meet the European Union sustainability criteria for biofuels? Energy Fuels 2010, 24 (3), 1720−1730. (15) Lange, M. The GHG balance of biofuels taking into account land use change. Energy Policy 2011, 39 (5), 2373. (16) Searchinger, T.; Heimlich, R.; Houghton, R.; Dong, F.; Elobeid, A.; Fabiosa, J.; Tokgoz, S.; Hayes, D.; Yu, T. Use of US croplands for biofuels increases greenhouse gases through emissions from land-use change. Science 2008, 319 (5867), 1238. (17) Mullins, K. A.; Griffin, W. M.; Matthews, H. S. Policy implications of uncertainty in modeled life-cycle greenhouse gas emissions of biofuels. Environ. Sci. Technol. 2011, 45 (1), 132. (18) Actively Engaged in Investigating the Role of Biofuels in Greenhouse Gas Emissions from Indirect Land Use Change; U.S. Department of Energy, 2008. (19) Searchinger, T. D. Response to New Fuels Alliance and DOE Analysts Criticisms of Science Studies of Greenhouse Gases and Biofuels. Princeton University, 2008. (20) Righelato, R.; Spracklen, D. V. Carbon mitigation by biofuels or by saving and restoring forests? Sci. Mag. 2007, 317 (5840), 902. (21) Kanter, J. Questioning Europe’s Math on Biofuels; The New York Times, September 25, 2011. (22) Kanter, J. ’Serious’ Error Found in Carbon Savings for Biofuels; The New York Times, September 14, 2011. (23) Biello, D. For Greening Aviation, Are Biofuels the Right Stuff?; Yale School of Forestry & Environment Studies, 2009. (24) Beer, T., G, T., Campbell, P. K. Biodiesel Could Reduce Greenhouse Gas Emissions, KS54C/1/F2.27; CSIRO: Australia, 2007. (25) Cimitile, M., Corn ethanol will not cut greenhouse gas emissions. Sci. Am. April 20, 2009. (26) Giles, J., Biofuels Emissions may be ’Worse Than Petrol’. New Sci. February 7, 2008. (27) European Commission. Directive 2009/28/EC of the European Parliament and of the Council of 23 April 2009 on the Promotion of the Use of Energy from Renewable Sources and Amending and Subsequently Repealing Directives 2001/77/EC and 2003/30/EC. Off. J. Eur. Union 2009, L140/16. (28) Well-to-Tank and Well-to-Wheels Report; EC Joint Research Centre: 2007. (29) McKone, T. E.; Nazaroff, W. W.; Berck, P.; Auffhammer, M.; Lipman, T.; Torn, M. S.; Masanet, E.; Lobscheid, A.; Santero, N.; Mishra, U.; Barrett, A.; Bomberg, M.; Fingerman, K.; Scown, C.; Strogen, B.; Horvath, A. Grand challenges for life-cycle assessment of biofuels. Environ. Sci. Technol. 2011, 45 (5), 1751−1756.

that can be accessed. Given the interdependence of a multiple feedstock system subject to multiple constraints, including (or excluding) a feedstock into (or from) an optimal portfolio might require a set of other include/exclude decisions, either to keep the blend feasible or to reach to the optimal point with a new set of feedstocks. Therefore, optimal solutions obtained by the chance-constrained model developed here are extremely challenging to achieve by intuitive decisions when all the complexities are taken into account. In addition, given the changing dynamics of price, emissions constraints and new feedstock development, a producer’s experience may not provide the most beneficial blend. Helping operators make decisions about diversification requires a tool that is capable of designing multifeedstock blends as well as predicting the final fuel properties and GHG emissions prior to blending. This capability can enable producers to modify the batch composition over time as prices fluctuate and thereby obtain cost-effective and technically compliant biodiesel capable of competing with petrodiesel.

■

ASSOCIATED CONTENT

S Supporting Information *

Additional information as noted in the text. This material is available free of charge via the Internet at http://pubs.acs.org.

■

AUTHOR INFORMATION

Corresponding Author

*Phone: 617-253-0877; fax: 617-258-7471; e-mail: [email protected]. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS We acknowledge support from the Portuguese Science and Technology Foundation (FCT), in the context of the following 2 projects: MIT/SET/0014/2009 (Biofuel capturing uncertainty in biofuels for transportation: resolving environmental performance and enabling improved use), funded in the scope of the MIT-Portugal program; and PTDC/SEN-TRA/117251/2010 (Extended “well-to-wheels” assessment of biodiesel for heavy transport vehicles). We also acknowledge funding from the National Science Foundation Award No. 1133422, Environmental Sustainability Program (ENG-CBET) that made this work possible. We extend our utmost thanks to Suzanne Greene for her important graphic and organizational contributions.

■

REFERENCES

(1) IEA; OECD. Transport, Energy and CO2: Moving Toward Sustainability; International Energy Agency, 2009. (2) Acquaye, A. A.; Wiedmann, T.; Feng, K.; Crawford, R. H.; Barrett, J.; Kuylenstierna, J.; Duffy, A. P.; Koh, S. C. L.; McQueen-Mason, S. Identification of ‘carbon hot-spots’ and quantification of GHG intensities in the biodiesel supply chain using hybrid LCA and structural path analysis. Environ. Sci. Technol. 2011, 45 (6), 2471. (3) Castanheira, E. G.; Freire, F. M. Greenhouse gas assessment of soybean production: Implications of land use change and different cultivation systems. J. Clean. Prod. 2013, 54, 49−60. (4) Frondel, M.; Peters, J. Biodiesel: A new oildorado? Energy Policy 2007, 35 (3), 1675−1684. (5) Huo, H.; Wang, M.; Bloyd, C.; Putsche, V. Life-cycle assessment of energy use and greenhouse gas emissions of soybean-derived biodiesel and renewable fuels. Environ. Sci. Technol. 2008, 43 (3), 750−756. (6) Malça, J.; Freire, F. Uncertainty analysis in biofuel systems. J. Ind. Ecol. 2010, 14 (2), 322−334. 7649

dx.doi.org/10.1021/es405410u | Environ. Sci. Technol. 2014, 48, 7642−7650

Environmental Science & Technology

Article

(30) European Commission. Comission Decision on the Guidelines for the Calculation of Land Carbon Stocks for the Purpose of Annex V to Directive 2009/28/EC; European Commission, 2010. (31) European Commission. Directive 2009/30/EC of the European Parliament; EEC, 2009. (32) U.S. EPA. Renewable Fuel Standard Program (RFS2) Regulatory Impact Analysis. In U.S. Environmental Protection Agency: Washington, DC, 2010. (33) Notice of Data Availability Concerning Renewable Fuels Produced From Palm Oil Under the RFS Program. Fed. Regist., 2012. (34) Renewable Fuel Standard (RFS2) Delayed RIN Generation Guidance Document for Canola Oil; Washington, DC, 2010. (35) Gaustad, G.; Li, P.; Kirchain, R. Modeling methods for managing raw material compositional uncertainty in alloy production. Resour., Conserv. Recycl. 2007, 52 (2), 180−207. (36) Persson, E.; Engstrand, P.; Karlsson, L.; Nilsson, F.; Wahlgren, M., Utilization of the natural variation in wood properties−A Comparison between pilot plant and mill scale trials. 2003, 2−5. (37) Olivetti, E. A.; Gaustad, G. G.; Field, F. R.; Kirchain, R. E. Increasing secondary and renewable material use: A chance constrained modeling approach to manage feedstock quality variation. Environ. Sci. Technol. 2011, 45 (9), 4118. (38) Wendt, M.; Li, P.; Wozny, G. Nonlinear chance-constrained process optimization under uncertainty. Ind. Eng. Chem. Res. 2002, 41 (15), 3621−3629. (39) Sahinidis, N. V. Optimization under uncertainty: State-of-the-art and opportunities. Comput. Chem. Eng. 2004, 28 (6−7), 971−983. (40) Bertsimas, D.; Sim, M. The price of robustness. Oper. Res. 2004, 35−53. (41) Charnes, A.; Cooper, W. W. Chance-constrained programming. Management Science 1959, 73−79. (42) Gulsen, E.; Olivetti, E.; Freire, F.; Dias, L.; Kirchain, R. Impact of feedstock diversification on the cost-effectiveness of biodiesel. Appl. Energy 2014, 126, 281−296. (43) Demirbas, A. Political, economic and environmental impacts of biofuels: A review. Appl. Energy 2009, 86, S108−S117. (44) Soybeans and Biodiesel: Key Price Relationships; Centrec Consulting Group, 2009. (45) Retka-Schill, S. Walking a tightrope. Biodiesel Mag. 2008, 5 (3), 64−70. (46) Moser, B. R. Influence of blending canola, palm, soybean, and sunflower oil methyl esters on fuel properties of biodiesel. Energy Fuels 2008, 22 (6), 4301−4306. (47) A Technology Performance and Regulatory Review; American Biofuels Association & Information Resources, Inc.: Jefferson City, MO, 1994; p 118. (48) Haas, M.; McAloon, A.; Yee, W.; Foglia, T. A process model to estimate biodiesel production costs. Bioresour. Technol. 2006, 97 (4), 671−678. (49) Markowitz, H. Portfolio selection. J. Finance 1952, 7 (1), 77−91. (50) Food and Agriculture Organization, Food and Agriculture Commodities Production; United Nations, 2012. (51) European Commission. Joint Research Centre, Climatic Zone, 2010. (52) European Commission. Joint Research Centre, Soil Type Classification, 2010. (53) Regulation of fuels and fuel additives: Changes to renewable fuel standard program. Fed. Regist., 2009. (54) California Air Resources Board. Low carbon fuel standard. In 17, California Code of Regulations 95480−95490; CARB: Sacramento, CA, 2007. (55) Wicke, B.; Sikkema, R.; Dornburg, V.; Faaij, A. Exploring land use changes and the role of palm oil production in Indonesia and Malaysia. Land Use Policy 2011, 28, 193−206.

7650

dx.doi.org/10.1021/es405410u | Environ. Sci. Technol. 2014, 48, 7642−7650