Environ. Sci. Technol. 2009, 43, 1718–1723
Environmental Decision-Making Using Life Cycle Impact Assessment and Stochastic Multiattribute Decision Analysis: A Case Study on Alternative Transportation Fuels K R I S T I N R O G E R S * ,† A N D THOMAS P. SEAGER‡ Ecological Science and Engineering, Purdue University, 550 Stadium Mall Drive, West Lafayette, Indiana 47907, and Golisano Institute of Sustainability, Rochester Institute of Technology, 111 Lomb Memorial Drive, Rochester, New York 14623
Received April 23, 2008. Revised manuscript received December 3, 2008. Accepted January 13, 2009.
Life cycle impact assessment (LCIA) involves weighing tradeoffs between multiple and incommensurate criteria. Current stateof-the-art LCIA tools typically compute an overall environmental score using a linear-weighted aggregation of characterized inventory data that has been normalized relative to total industry, regional, or national emissions. However, current normalization practices risk masking impacts that may be significant within the context of the decision, albeit small relative to the reference data (e.g., total U.S. emissions). Additionally, uncertainty associated with quantification of weights is generally very high. Partly for these reasons, many LCA studies truncate impact assessment at the inventory characterization step, rather than completing normalization and weighting steps. This paper describes a novel approach called stochastic multiattribute life cycle impact assessment (SMA-LCIA) that combines an outranking approach to normalization with stochastic exploration of weight spacessavoiding some of the drawbacks of current LCIA methods. To illustrate the new approach, SMA-LCIA is compared with a typical LCIA method for crop-based, fossil-based, and electric fuels using the Greenhouse gas Regulated Emissions and Energy Use in Transportation (GREET) model for inventory data and the Tool for the Reduction and Assessment of Chemical and other Environmental Impacts (TRACI) model for data characterization. In contrast to the typical LCIA case, in which results are dominated by fossil fuel depletion and global warming considerations regardless of criteria weights, the SMA-LCIA approachresultsinarankorderingthatismoresensitivetodecisionmaker preferences. The principal advantage of the SMA-LCIA method is the ability to facilitate exploration and construction of context-specific criteria preferences by simultaneously representing multiple weights spaces and the sensitivity of the rank ordering to uncertain stakeholder values.
assessment (IA) phase of LCA brings the total environmental effects into a consistent framework for evaluation by characterizing the inventory data into categories such as global warming, photochemical ozone formation, and acidification (1). Overall environmental performance is typically determined by normalizing the characterized inventory as a fraction of total emissions from an industry, region, nation, or group of nations and applying relative importance weights to aggregate the normalized values in a linear-weighted sum (2). Nevertheless, due to the high risk of subjectivity, ISO14042 (the international standard for LCIA) considers normalization and scoring steps optional (3). As a result, many LCAs are left as a set of characterized data, causing managers and decisionor policymakers to confront multicriteria, multistakeholder problems unaided. This leaves them vulnerable to biases, such as anchoring on first impressions, placing undue emphasis on narrow, albeit salient, aspects of a choice or judging on the basis of stigma or affect. Ultimately, the results of unaided decision processes may differ from those that might, upon further reflection, be preferred (4). To address this issue, some LCA practitioners have borrowed tools from multicriteria decision analysis (MCDA) to facilitate understanding of trade-offs and multiple perspectives in the final weighting phase of LCIA (5, 6). This paper reviews the problematic issues with current LCIA normalization and weighting and compares them with a novel stochastic multiattribute life cycle impact assessment method (SMA-LCIA) using a case study of transportation fuel alternatives. SMA-LCIA applies an outranking normalization method that uses data internal to the specific case study and a stochastic weight set that explores the sensitivity of alternative ranks to the entire feasible weight space. The case study presented in this article was chosen to demonstrate the utility of SMA-LCIA compared with the traditional approach to LCIA. The results are meant to illustrate the method and should not be interpreted as an endorsement of a specific alternative. Normalization. Normalization converts characterized data into dimensionless or identical units by relating the data to an external database or benchmark (external normalization) or relating data from different alternatives to each other (internal normalization) (7). Current LCIA methods use external normalization relative to a reference area (e.g., the total emissions in the United States). Normalized data can be analyzed as-is (e.g., for improvement assessment) or as preparation for further aggregation into an overall score (7). External normalization typically allows full compensation between impact categories under the premise that impact categories are completely substitutable, and poor performance in one category can be compensated for by good performance in another (8). A major disadvantage to external normalization is the risk of over- or underemphasizing
Introduction Life cycle assessment (LCA) is a valuable tool for analyzing the environmental impact of a product or process. The impact * Corresponding author e-mail:
[email protected]. † Purdue University. ‡ Rochester Institute of Technology. 1718
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 6, 2009
FIGURE 1. Outranking normalization. 10.1021/es801123h CCC: $40.75
2009 American Chemical Society
Published on Web 02/06/2009
TABLE 1. Characterization and Normalization Factors, Weight Sets, and Pseudocriteria Characterization Factorsa FF MJ/BTU FF CO2 CH4 N2O VOC CO Nox PM10 PM2.5 Sox
GW CO2-eq/kg
EUT N-eq/kg
SMOG NOx-eq/kg
ACID H+eq/kg
HHCR milli-DALYS/kg
40.04
0.002213 0.083448 0.13908
0.001055 1 21 310
0.002964 0.780645 0.013387 1
0.00429
50.79 Normalization Factorsb MJ/yr
CO2eq/yr
8.53E+13
6.85E+12
N-eq/yr
NOXeq/yr
H+eq/yr
milli-DALYS/yr
5.02E+09
3.38E+10
2.08E+12
1.71E+11
Stakeholder Weights (%)c Short-Term producer user LCA expert Long-Term producer user LCA expert
27 23 19
10 17 9
16 17 5
12 11 5
12 11 1
23 21 61
21 2 6
61 92 87
8 3 3
2 0 1.3
4 3 1.3
4 0 1.3
Pseudocriteriad threshold
MJ/mi.
CO2eq/mi.
N-eq/mi.
NOXeq/mi.
H+eq/mi.
milli-DALYS/mi.
preference indifference
-0.297 -0.148
-0.023 -0.012
-2.84E-07 -1.42E-07
-8.29E-05 -4.15E-05
-0.0058 -0.0029
-4.57E-06 -2.28E-06
a These particular categories were chosen because the pollutants provided in the GREET model do not have impacts in other categories from ref 1. b Total U.S. emissions, characterized to correspond to TRACI impact factors (10). c Stakeholder Weights from (22). d Pseudocriteria are negative because a favorable preference would be a lower criteria value, rather than a higher value.
FIGURE 2. TRACI-characterized inventory, important criteria based on how the reference area is currently performing (9). For example, if the reference area has high smog emissions, normalized smog emissions attributed to the specific decision may be masked. Other drawbacks are related to the quality and completeness of the
external data, as national inventories tend to be highly aggregated industry reports that vary spatially and temporally (7, 10). While completely disallowing compensation between criteria is not feasible, SMA-LCIA uses a partially compensaVOL. 43, NO. 6, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1719
FIGURE 3. LCIA results: normalized inventory. tory, outranking method called PROMETHEE (11) to normalize characterized inventory on a scale of 0 to 1 using a pairwise comparison of alternatives on each impact category. Pseudocriteria called preference and indifference thresholds are used to determine the level of preference between two alternative for each criterion. When the difference d between alternatives a and b is less than the indifference threshold, preference equals 0. When d is greater than the preference threshold, alternative a is strongly preferred over b and preference equals 1. When d is between the preference and indifference thresholds, a is weakly preferred over b, and the preference value is between 0 and 1. Figure 1 is a graphical depiction of the preference and indifference thresholds with a linear function for incomplete preferences. Other preference functions such as stepwise or Gaussian are possible, and it is left to the decision maker to determine their appropriateness for a specific decision (12). Outranking methods are partially compensatory because preference relations (the degree of outranking between alternatives) prevent over performance once the preference threshold is exceeded (13). Several environmental decision making problems have applied outranking methods (14-16), including two in the context of LCIA (17, 18). Although outranking methods tend to be mathematically complex compared to the external normalization currently applied to LCIA, limiting compensation is generally more satisfying for environmental decision makers (8). Weighting. A major consideration that environmental DMs must address is how to incorporate stakeholder values in the analysis. Typically, stakeholder values are manifested as weights to be determined by direct, indirect, or combined value elicitation methods (19). The advantage of a formalized value-elicitation process is that judgments are made explicit allowing value information to clarify the decision making process (20). However, many methods of value elicitation are expensive and time-consuming and run the risk of bias resulting in incompatible or uncertain weights (21). Moreover, the interview process can be confusing, and participants may be hesitant to reveal their values for a variety of reasons including uncertainty and inexperience (20). Therefore, uncertainty and variability in value elicitation is typically very high. For example, in a recent study of preferences for environmental purchasing decisions, “LCA Experts” attached a preference weighting to global warming that ranged from 7% to nearly 70%, depending upon the time horizon of interest (22). Stochastic multiattribute acceptability analysis (SMAA) is a family of MCDA methods that couple traditional MCDA (including both compensatory and partially compensatory normalization) with stochastic weights to explore the entire feasible weight space using Monte Carlo (MC) simulation (23, 24). The results of SMAA for each alternative are a rank acceptability index (RAI) representing the probability that 1720
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 6, 2009
an alternative will be ranked first, second, third (and so on) and a central weights vector characterizing the weight set that results in an alternative being ranked first (24). A stochastic approach to weight assignment allows explicit exploration of the sensitivity of rank ordering results to uncertainty. Consequently, SMAA can be used to eliminate alternatives that rank poorly, prioritize research resources for reducing uncertainty, or allow DMs to explore the sensitivity of results to different weighting representationss thereby facilitating construction of preferences in the context of the specific decision. Case Study: Transportation Fuels. Alternative transportation fuels, especially crop-based ethanol and biodiesel, exemplify the challenges faced by environmental managers and decision- and policy makers by involving a comparison of very different feedstocks, processing methods, and a variety of stakeholder groups with potentially conflicting or changing views on the relative importance of different environmental impacts. A number of state and federal policy initiatives have resulted in rapid expansion of biofuels production capacity (25). For example, the Energy Policy Acts of 1992 and 2005 and Executive Order 13423 require state and federal fleets to have vehicles capable of using alternative fuelssalthough there are few guidelines for assessing the systemic environmental impact of alternative fuel technologies (26, 27). While many of the current studies on alternative fuels emphasize only a few environmental aspects such as global warming or net energy, attention has been turned toward additional impacts in the feedstock stagesparticularly in biofuels (28, 29). As more impacts are added to the assessment, stakeholders and DMs have more difficulty articulating preferences and interpreting the additional datasa situation for which SMAA is particularly suitable.
Methods The data inventory for both the traditional LCIA approach and SMA-LCIA was obtained from the Greenhouse gas Regulated Emissions and Energy in transportation (GREET) model developed at Argonne National Laboratory (30). However, LCIA uses external normalization and a linearweighted sum to determine the environmental performance of individual alternatives, whereas SMA-LCIA uses an outranking normalization method with stochastic weights to determine the likelihood of an alternative being preferred with respect to others. This case study is limited to life cycle air emissions for the fuel alternatives: gasoline (GAS), lowsulfur diesel (LSD), 100% soy-biodiesel (BD100), electric vehicle (EV), and 85% corn-based ethanol (EtOH). The main purpose is to illustrate the SMA-LCIA method, not to provide definitive conclusions about alternative fuels. Inventory. GREET models well-to-wheels (extraction to use phase) fuel life-cycle energy use and emissions of nine criteria air pollutants on a per mile basis, listed in Table 1. GREETv1.8, which provides stochastic data for MC simulations, was used to compile the inventory (31). GREET does not provide characterization or normalization factors. Characterization. Inventories were characterized into six midpoint impact categories using the U.S. Environmental Protection Agency’s Tool for the Reduction and Assessment of Chemical and other environmental Impacts (TRACI) characterization factors: fossil fuel depletion (FF), global warming potential (GW), eutrophication potential (EUT), photochemical ozone formation potential (SMOG), acidification potential (ACID), and human health criteria air pollutants (HHCR), shown in Table 1. These particular categories were chosen because the pollutants provided in the GREET model do not have impacts in other TRACI categories. Crystal Ball 5.5 (32) was used to run 10,000 trial Monte Carlo simulations with GREETv1.8 and SMA-LCIA.
FIGURE 4. Overall environmental scores. Stacked bars represent the partial contribution of respective weighted, normalized impact categories to the overall score.
FIGURE 5. SMA-LCIA results - rank acceptability index. LCIA. The TRACI normalization database, also in Table 1, was used for external normalization in the LCIA case. Characterized values were divided by normalization factors which represent the total U.S. emissions in each impact category. The result is a dimensionless fraction of U.S. emissions attributed to each impact category by the alternative under consideration. Normalized values were aggregated in several different scenarios using stakeholder weights derived from Gloria et al. (22), who queried users of a LCIA software program called Building for Environmental and Economic Sustainability (BEES) developed for environmental and economic assessment of materials for the construction
industry (33). Respondents are classified as either Producers, who sell building materials and services, Users, who buy materials and services, or LCA Experts. Weights related to short-term (10 years) time horizons emphasized the importance of FF and HHCR, while long-term (greater than 100 years) results assigned the greatest weight to GW. SMA-LCIA. The SMA-LCIA case uses an approach based on PROMETHEE outranking (11) to normalize the characterized inventory. For demonstrative purposes, a linear preference function was used for this study with the preference threshold of each impact category equal to the standard deviation of the characterized inventory data in that category and the indifference threshold equal to half of the preference threshold. Pseudocriteria are listed in Table 1. The following algorithm for generating n stochastic weights is derived from ref 24. 1. n-1 uniformly distributed random numbers between 0 and 100 are generated using MC simulation (q1, q2. ..qn-1) 2. the numbers are sorted in ascending order with q0)0 and qn)100 3. the weights are calculated as w1)q1-q0, w2)q2q1...wn)qn-qn-1 so that Σw)100 VOL. 43, NO. 6, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1721
FIGURE 6. Central weights box and whisker plots. The solid line indicates the first quartile, representing the lowest 25% of the data, the gray box represents the second quartile, the white box represents the third quartile, and the dotted line indicates the fourth quartile or the highest 25% of the data set. The diamonds represent the central weights for each alternative. EtOH did not have a RAI for first rank and does not have a central weights vector. This procedure ensures that all weight sets add to 100% and that the generated weights are a uniform sampling of the available weight space. If partial weight information is available, the weights can be constrained by rejecting weight sets that fall outside the constraints. For example, if stakeholders agree that GW will never be valued at less than 25%, all trials where the GW weight is below 25% are removed from the analysis. The unconstrained weight space was used in this study. The results of SMA-LCIA are the RAI and central weights vectors (23). For detailed information on outranking methods and stochastic weights see the Supporting Information.
Results and Discussion Characterized Inventory. The characterized inventory data are presented in Figure 2. Trade-offs are apparent in that no alternative outperforms all other alternatives in all impact categories. Consequently, a DM must make a compromise between impacts regardless of the alternative chosen. LCIA Results. The unweighted (i.e., weighted equally), TRACI-normalized LCIA values are shown in Figure 3. The normalized values are heavily influenced by GW and FF, which dwarf other impact categories such as EUT and HHCR. The advantage of normalization in LCIA is that the characterized inventory is converted into the same units (fraction of U.S. emissions). However, because they are based on physical midpoint parameters (kg-equivalents) rather than specific end-point damages (e.g., mortality) normalized values are still qualitatively different and not directly comparable. Relative importance weights act as a conversion factor between normalized impacts. The normalized values are multiplied by the corresponding weights, and these weighted values are added together to form an overall unitless score that is used to compare alternatives within the analysis. Figure 4 shows the overall environmental scores as stacked, weighted TRACI-normalized values for the six weight sets listed in Table 1. The most interesting finding is that the relative rankings 1722
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 6, 2009
of the alternatives do not change regardless of the weight set applied. Although the magnitude of the overall scores changes between short- and long-term weights, the preference ordering of alternatives is practically identical. In the TRACInormalized case, the overall scores are dominated by FF and GW in the short term and almost exclusively by GW in the long term, masking all other impact categories, despite the variation in weight sets. In this case the lack of sensitivity of results to changing decision maker preferences is more suggestive of bias in normalization than dominance of any one technological alternative. SMA-LCIA Results. The RAI results are shown in Figure 5. Alternatives are arranged according to their highest acceptability. The RAI indicates that LSD has the highest probability of being ranked first, thus it is likely to be the best environmental choice within the boundaries of this study. However, if other factors (e.g., economics, politics) prevent LSD from being feasible, EV and BD100 also have high rank acceptabilities. When a large number of alternatives (>10) are being considered, the RAI may be used as to eliminate alternatives that will never be favorable under any value set. For example, EtOH performs poorly against the other alternatives in the pairwise comparison stage of SMAA. Regardless of weight sets applied, EtOH is dominated by other alternatives and may be eliminated from further consideration based on its low RAI. Central weights vectors are presented in Figure 6 as a box and whisker plot depicting the size of each quartile in the data distribution. Central weights are calculated by taking the average of the range of weights in each impact category that result in the most favorable RAI. The central weights vectors reveal value sets that lead to outcomes that can be used in constructing stakeholder preferences by matching them with descriptions of stakeholder priorities. For example, if stakeholders indicate the highest concern for FF and GW, BD100 would likely be the preferred alternative because it had the highest central weights for both FF and GW. Fossilbased fuels are favored when GW and FF weights are low
and other impact categories (such as eutrophication) are weighted relatively high. EVs are preferred when HHCR and ACID are considered less important, and the remaining impacts are considered approximately equally important. In the SMA-LCIA approach, stakeholders may explore the sensitivity of the final preference ordering of alternatives to different weight expressions. The combination of rank acceptability indices and central weight vectors allows stakeholders to iterate between viewing the rankings that result from a specific expression of weights and the weights that would result in a specific ranking. In this case, approaching the problem both backward and forward may facilitate construction of weighting preferences within the specific decision context, potentially increasing decision maker confidence in the final result. By contrast, the traditional LCIA rank ordering is unchanged regardless of the weight set applied. While it appears from the characterized data set that trade-offs are present, the process of external normalization effectively masks expression of potentially different stakeholder priorities to the point where the weights for FF and GW would have to be so low (i.e., less than 1%) to result in a change in rank ordering that such a weight set would not realistically be acceptable to stakeholder groups. The outranking approach in SMA-LCIA provides a more balanced evaluation of environmental performance focused on a specific decision, and as a result, stakeholder and DM values play a more prominent role in alternative selection. Consequently, the SMA-LCIA approach may be advantageous in revealing potential conflicts between or opportunities for compromise among different stakeholder groups engaged in a deliberative decision process.
Acknowledgments This research was supported by the USDA Food and Agricultural Sciences National Needs Graduate and Postdoctoral Fellowship Grants Program (GRANT # 2005-3842015803), the Ross Fellowship and the Elmer Ballotti Memorial Fellowship through the Civil Engineering Department at Purdue University, and the Golisano Institute for Sustainability at Rochester Institute of Technology.
Supporting Information Available Detailed descriptions of outranking normalization, stochastic weight space exploration techniques, and GREET assumptions, inventory, and characterized inventory statistics. This material is available free of charge via the Internet at http:// pubs.acs.org.
Literature Cited (1) Bare, J. C.; Norris, G. A.; Pennington, D. W.; McKone, T. TRACI: The Tool for the Reduction and Assessment of Chemical and Other Environmental Impacts. J. Ind. Ecol. 2003, 6 (3-4), 49– 78. (2) Bare, J. C.; Gloria, T. P. Critical Analysis of the Mathematical Relationships and Comprehensiveness of Life Cycle Impact Assessment Approaches. Environ. Sci. Technol. 2006, 40 (4), 1104–1113. (3) Ryding, S.-O. ISO14042. Int. J. LCA 1999, 4 (6), 307. (4) Gregory, R.; McDaniels, T. Improving Environmental Decision Processes. In Decision Making for the Environment: Social and Behavioral Science Research Priorities; Brewer, G. D., Stern, P. C., Eds.; National Academies Press: Washington DC, 2005; pp 175196. (5) Miettinen, P.; Hamalainen, R. P. How to Benefit from Decision Analysis in Environmental Life Cycle Assessment (LCA). Eur. J. Oper. Res. 1997, 102, 279–294. (6) Seppala, J.; Basson, L.; Norris, G. Decision Analysis Frameworks for Life-Cycle Impact Assessment. J. Ind. Ecol. 2002, 5 (4), 45– 68. (7) Heijungs, R.; Guinee, J. B.; Kleijn, R.; Rovers, V. Bias in Normalization: Causes, Consequences, Detection and Remedies. Int. J. LCA 2007, 12 (4), 211–216.
(8) Guitouni, A.; Martel, J.-M. Tentative Guidelines to Help Choosing an Appropriate MCDA Method. Eur. J. Oper. Res. 1998, 109, 501–521. (9) Seppala, J.; Hamalainen, R. P. On the Meaning of Distanceto-Target Weighting Method and Normalization in Life Cycle Impact Assessment. Int. J. LCA 2001, 6 (4), 211–218. (10) Bare, J. C.; Gloria, T.; Norris, G. A. Development of the Method and U.S. Normalization Database for Life Cycle Impact Assessment and Sustainability Metrics. Environ. Sci. Technol. 2006, 40, 5108–5115. (11) Brans, J.-P.; Mareschal, B. PROMETHEE Methods. In Multiple Criteria Decision Analysis, Figueira, J., Greco, S., Ehrogott, M., Eds.; Springer: New York, 2005. (12) Brans, J.-P.; Vincke, P.; Mareschal, B. How to select and how to rank projects: The PROMETHEE method. Eur. J. Oper. Res. 1986, 24, 228–238. (13) Roy, B. Paradigms and challenges. In Multiple Criteria Decision Analysis: State of the art surveys; Figueira, J., Greco, S., Ehrogott, M., Eds.; Springer: New York, 2005; pp 3-24. (14) Salminen, P.; Hokkanen, J.; Lahdelma, R. Comparing Multicriteria Methods in the Context of Environmental Problems. Eur. J. Oper. Res. 1998, 104, 485–496. (15) Norese, M. F.; Toso, F. Group Decision and Distributed Technical Support. Int. Trans. Operational Res. 2004, 11, 395–417. (16) Hermans, C.; Erickson, J.; Noordewier; Sheldon, A.; Kline, M. Collaborative environmental planning in river management: An application of multicriteria decision analysis in the White River Watershed in Vermont. J. Environ. Manage. 2007, 84, 534– 546. (17) Le Teno, J. F.; Mareschal, B. An Interval Version Of PROMETHEE For The Comparison Of Building Products’ Design With IllDefined Data On Environmental quality. Eur. J. Oper. Res. 1998, 109, 522–529. (18) Geldermann, J.; Rentz, O. Multi-criteria analysis for technique assessment: Case study from industrial coating. J. Ind. Ecol. 2005, 9 (3), 127–142. (19) Finnveden, G. A Critical Review Of Operational Valuation/ Weighting Methods For Life Cycle Assessment; Swedish Environmental Protection Agency: Stockholm, 1999; pp 1-49. (20) Brewer, G. D.; Stern, P. C. Decision making for the environment: Social and behavioral science research priorities; National Academies Press: Washington DC, 2005; p 33. (21) Hofstetter, P.; Mettier, T. M. What Users Want and May Need: Insights from a Survey of Users of a Life-Cycle Tool. J. Ind. Ecol. 2003, 7 (2), 79–101. (22) Gloria, T. P.; Lippiatt, B.; Cooper, J. Life cycle impact assessment weights to support environmentally preferable purchasing in the United States. Environ. Sci. Technol. 2007, 41, 7551–7557. (23) Lahdelma, R.; Salminen, P. SMAA-2: Stochastic multicriteria acceptability analysis for group decision making. Oper. Res. 2001, 49 (3), 444–454. (24) Tervonen, T.; Lahdelma, R. Implementing Stochastic Multicriteria Acceptability Analysis. Eur. J. Oper. Res. 2007, 178, 500– 513. (25) Schumacher, J. Oilseed, Biodiesel and Ethanol Subsidies and Renewable Energy Mandates: US Federal and Selected State Initiatives; Montana State University: Bozeman, Montana, 2006. (26) Bush, G. W. Executive Order 13423sStrengthening Federal Environmental, Energy, and Transportation Management. In Federal Register; United States, 2007; Vol. 72, pp 3919-3923. (27) Rahm, D.; Coggburn, J. D. Environmentally preferable procurement: Greening U.S. state government fleets. Public Works Manage. Policy 2007, 12 (2), 400–415. (28) von Blottnitz, H.; Curran, M. A. A Review Of Assessments Conducted On Bio-Ethanol As A Transportation Fuel From A Net Energy, Greenhouse Gas, And Environmental Life-Cycle Perspective. J. Clean. Prod. 2007, 17 (7), 607–619. (29) Landis, A. E. Environmental and Economic Impacts of Biobased Production; Ph.D. Dissertation, Civil and Materials Engineering, University of Illinois - Chicago: Chicago, 2007. (30) Wang, M. Q. Development and Use of GREET 1.6 Fuel-Cycle Model for Transportation Fuels and Vehicle Technologies; United States Department of Energy: 2001; p 11-13. (31) Wang, M. GREET-Greenhouse Gases, Regulated Emissions, and Energy Use in Transportation, 1.8; Argonne National Laboratory: Argonne, IL, 2007. (32) Decisioneering Inc. Crystal Ball 5.5; Littleton, CO, 2005. (33) Lippiatt, B.; Boyles, A. S. Using BEES to select cost-effective green products. Int. J. LCA 2001, 6 (2), 76–80.
ES801123H VOL. 43, NO. 6, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1723
