Comparing Two Life Cycle Assessment Approaches: A Process Model- vs. Economic Input-Output-Based Assessment

Comparing Two Life Cycle Assessment Approaches: A Process Model- vs. Economic Input-Output-Based Assessment Chris T. Hendrickson, Arpad Horvath, Satish Joshi, Marhs Klausne?, Lester B. Lave and Francis C. McMichael Green Design Initiative Carnegie Mellon University, Pittsburgh, PA 15213 Abstract -We compare two tools for Life Cycle Assessment (LCA). The software GaBi (Ganzheitliche Bilanzierung - Integrated Assessment) from Germany is based on a process model approach, as recommended by the Society of Environmental Toxicology and Chemistry (SETAC). These results are contrasted to those from the method developed by Carnegie Mellon University s Green Design Initiative, Economic Input-Output Life Cycle Analysis $10-LCA). The EIO-LCA model uses economic input-output matrices, and industry sector level environmental and nonrenewable resource consumption data to assesses the economy-wide environmental impacts of products and processes. The results from the alternative approaches are compared in terms of toxic chemical releases, conventional pollutant emissions, energy use by fuel type, and use of ores. We find that most of the values from the two tools are within the same order of magnitude, despite the fundamental differences in the models. We contrast the two approaches to identify their relative strengths and weaknesses. 1. INTRODUCTION Life Cycle Assessment (LCA) systematically considers and quantifies the consumption of resources and the environmental impacts associated with a product or process. By considering the entire life cycle and the associated environmental burdens, LCA identifies opportunities to improve environmental performance. This study contrasts two different approaches to LCA. 0 A process model approach based primarily on the standard recommendations of the Society of Environmental Toxicology and Chemistry (SETAC) [ 11, implemented in software called GaBi, developed by the Institut her Kunststoffpruefing und Kunststofflrunde (IKP) at the University of Stuttgart, Germany [2]. 0 An approach based on the interindustry monetary transactions, or economic input-output data, and pollution discharges and nonrenewable resource consumption data of all industry sectors called EIO- LCA, developed and implemented in sohare at Carnegie Mellon University s Green Design Initiative [3][4][5]. GaBi is representative of a suite of software tools based largely on the process model, or SETAC approach [6]. We selected GaBi particularly since this software is being used by approximately 65 customers, most of them large companies [7]. Data for GaBi were obtained by investigating various industrial processes in Europe with regular updates. LCA has been criticized for several problems [3] [8] [9]: There is lack of comprehensive data for LCA. Data quality is not uniformly high. Defining problem boundaries for LCA is arbitrary and controversial. LCA is too expensive and slow for application in the design process. There is no single method that is universally L acceptable. Equally credible analyses can produce qualitatively different results. Modeling a new product or process is difficult and expensive. LCA cannot capture the dynamics of changing markets and technologies. LCA results may be inappropriate for use in ecolabeling. Existing LCA software tools based on the process model approach (e.g., SimaPro from Pre Consultants, The Netherlands, TEAM from Ecobalance, Rockville, MD, EcoManager from Franklin Associates, Prairie Village, KS, etc.) cany many of the above limitations. Within a modem economy, each sector contributes to every other sector, directly or indirectly. The SETAC approach sinks under the weight of attempting to specify each of these relationships and to estimate their materials use and environmental discharges. The analyst is forced to draw a boundary around the small part of the problems that can be included in the analysis. Contact person: cth@cmu.edu, phone: (412) 268-2941, fax: (412) 268-7813 Carnegie Mellon University and Robert Bosch GmbH, Germany. 0-7803-3808-1 /97$10.0oO1997 IEEE 176

EIO-LCA is a new approach to LCA. Interindustry relationships are quantified in an input-output matrix. Derived from U.S. Census data, the matrix represents all of the interactions among sectors, even the remote indirect ones. For example, a steel mill requires iron ore, coal and electricity, and is usually included in a SETAC analysis. However, indirect suppliers such as office equipment, paper, food, vehicles, etc., are generally excluded. Each kilogram of steel produced results in environmental discharges in several other industry sectors, which, depending upon the sector, might vary by several orders of magnitude. The process model analysis is expensive and time consuming because inputs and environmental burdens have to be either empirically gathered or obtained from literature (if available). In EIO-LCA, we augment the 519x519 sector inputoutput tables of the U.S. economy (1987) with sector level coefficients for nonrenewable resource use, fuel use and pollutant emissions. The augmented input-output matrices are then used to estimate the economy-wide resource requirements and pollutant discharges [3][4][5]. The 519x519 matrix can be disaggregated by using the detailed wormiles of the published input-output tables, available from the Bureau of Economic Analysis [lo]. These workfiles were used to extract data on the purchase values of various ores like iron ore, copper ore, bauxite, etc., and fuels such as coal, natural gas, motor gasoline, distillate fuel oil, residual he1 oil, etc., by each industry sector. Physical quantities were estimated from these dollar values using the price data from the Bureau of Mines and the Energy Information Administration [Ill [12]. Conventional pollutant emissions from fuel use were derived using the emission factors developed by the U.S. Environmental Protection Agency (EPA) [13]. The emission coefficients for toxic releases are derived from the EPA s Toxics Release Inventory (TRI) data [14]. Comparing the results of the two approaches provides insights. In this study, we focus on the production of steel and aluminum, major input materials to many manufacturing processes. Inputs in LCA are defined as resources used for production of a material (energy and ore use, etc.). Outputs are environmental discharges and byproducts associated with products. We chose to compare GaBi s input and output data on three kinds of steel products (cold-rolled steel, hot-dip galvanized steel sheet, and electrolytically galvanized steel sheet) and one kind of aluminum product (aluminum sheet, import mix) to EIO-LCA s results for steel products (the blast fimaces and steel mills sector - SIC 3312), and aluminum products (the primary aluminum sector - SIC 3334). 11. COMPARISON OF NUTS The inputs reported in GaBi include fossil fuels like brown coal, hard coal, natural gas and crude oil, metallic ores like iron ore, bauxite, copper ore, etc., and other inputs like limestone, hydro power, air, cooling and process water. GaBi provides a detailed breakdown of fossil fuels by the country of origin. However, GaBi does not provide details on the consumption of individual petroleum derivatives. GaBi might trace all the inputs to the ultimate extraction stage, although no details are apparent to the user. The current dataset in EIO-LCA covers fossil fuels like coal and natural gas, and metallic ores like iron ore and copper ore, as well as alumina. EIO-LCA also provides estimates of direct and indirect demand for individual petroleum products like motor gasoline, residual fuel oil, distillate fuel oil, aviation and jet fuel, etc. However, it does not trace the origins of the petroleum products. Inorganic chemical inputs are not currently included in EIO-LCA. The underlying input-output structure of the economy and economy-wide changes in sectoral outputs in response to incremental demand for the output of a sector are apparent to the user. Despite using completely different approaches and data sources, the estimates of main ore inputs, i.e., iron ore for steel and bauxite/alumina for aluminum were similar, within 20%. The estimates of coal and natural gas inputs for steel making were close, within 5%. However, the estimates of natural gas and coal inputs for aluminum manufacture were higher in the case of GaBi, probably on account of greater use of hydro power in the U.S. aluminum industry and dependence on imported alumina. The petroleum product requirements were not comparable since GaBi only reports total crude oil requirements, while EIO-LCA provides details on individual derivatives. 111. COMPARISON OF OUTPUTS Table 1 shows the aggregated output values for steel production, while Table 2 lists the same for aluminum production from the GaBi and EIO-LCA software. We are able to report only aggregated values since the GaBi data are proprietary. We arrived at the values in the tables by aggregating the mass of individual chemical substances discharged to the environment for every kilogram of steel and aluminum produced. Nineteen chemicals for steel and 22 substances for aluminum appeared in both GaBi and EIO-LCA, and these are included in the table. Substances discharged to air and water include heavy metals (e.g., lead, cadmium, 177

nickel, etc.), various inorganic and organic emissions (e.g., ammonia, formaldehyde, etc.), sulfuric acid, hydrochloric acid, hydrogen fluoride, etc. In comparing the values for these chemical emissions from GaBi and EIO-LCA, we were looking for order of magnitude differences, due to uncertainty in the data. For steel, we found that 12 discharge figures were within the same order of magnitude, two were one order of magnitude lower in EIO-LCA (hydrochloric acid and benzene), four were one order of magnitude lower in GaBi (ammonia, formaldehyde, chromium and antimony), and manganese emissions to air and water were two orders of magnitude lower in GaBi. For aluminum, 15 values out of 22 were within the same order of magnitude in GaBi and EIO- LCA, benzene was one order of magnitude lower in EIO- LCA, sulfuric acid and phenol were two orders of magnitude lower in EIO-LCA, copper and formaldehyde were one order of magnitude lower in GaBi, and chlorine and hydrocyanic acid were as much as the orders of magnitude lower in GaBi. The compared compounds form only a subset of the reported outputs from the two tools. In addition, GaBi quantifies radioactive emissions and waste, municipal, industrial and hazardous waste, residual substances, waste waterhewage, exhaudused air, other inorganic emissions to air (5 for steel, 4 for aluminum) and water (9 for steel, 11 for aluminum), other heavy metals into air (14 for steel, 12 for aluminum) and water (4 for steel, 7 for aluminum), stockpiled goods, etc. In contrast, EIO- LCA reports values on 87 additional TRI chemical substances (carcinogens, ozone depleters, etc.) for steel and 100 for aluminum production that constitute emissions to air, water, land and underground injection, and more if we account for chemicals not discharged to the environment directly, but transferred off-site for treatment. For many outputs no comparison is possible. EIO-LCA can qumtiq hazardous waste generated, received, managed and shipped per unit of product from an economic sector (based on the EPA's RCRA data base), but we cannot compare these numbers to the ones from GaBi because the definition of hazardous waste is not stated in GaBi. Similarly, the composition of exhaustbed air is not defined. For ozone depleters, GaBi reports two kinds of CFC emissions (CFC 116 and 14) for steel and aluminum, while EIO-LCA quantifies as many as 7 ozone depleter emissions: CFCs, halons, etc. (the TRI provides information on 22 ozone depleters.) We cannot compare, for example, GaBi's figures for chloride ions (Cl-) (or any other ions) since EIO-LCA's emissions data base (TRI) does not report ions, but those ions are contained in the reported substances (e.g., chlorine). Table 1 and 2 include air and water emissions. EIO-LCA also reports discharges to land and underground injection wells. Reporting of these values is especially crucial information for substances such as nickel or manganese compounds for which 95% of environmental discharges in the EIO-LCA assessment constituted releases to land, or for hydrochloric acid, for which 74% of discharges were injected underground. Assessment of environmental discharges requires a systems view. Considering all media, the number of substances GaBi reports, but EIO- LCA omits, is less. In addition, EIO-LCA also reports transfers of TRI substances to publicly owned treatment works (POTW) and the sum of off-site transfers (for recycling, energy recovery, treatment, disposal or for unspecified activity) as important information for environmental management. Reported values in Table 1 and 2 should be considered in light of the uncertainty of the data. Theoretically, EIO- LCA should yield larger numbers because the model includes not only the direct, but the indirect supplied environmental burdens as well. However, 1. GaBi, for some products, might include indirect effects (see section IV. Sources of Uncertainty and Differences in Results), 2. TFU releases could *J systematically underestimated [15], and 3. some of the direct and indirect suppliers in the EIO-LCA model do not yet report to the TRI. For example, mining industries and electric utilities are important suppliers to steel and aluminum making, but they are exempt from TRI reporting. EPA has proposed to expand reporting requirements to sectors that support manufacturing activities: energy production, materials extraction, materials distribution and waste management [14]. Therefore, EIO-LCA is expected to give a more complete picture of environmental discharges as these data become available in the future. After accounting for uncertainty, there appears to be no significant difference in the values from GaBi and EIO- LCA, shown in Table 1 and 2. In fact, we see no difference between the numbers for GaBi's three kinds of steel products either, but them might be significant differences when comparing individual input and output components. We not only reported the aggregated kilograms of air and water releases (unweighted), but we also weighted the emissions by the relative toxicity of the substances, using the American Conference of Governmental Industrial Hygienists' (ACGIH) Threshold Limit Values (TLV), to arrive at the CMU-ET weighted toxic emissions [16]. The CMU-ET score gives a better way to characterize the relative environmental harm (toxicity) of discharges in the aggregated values, obtained from GaBi and EIO-LCA, respectively. We still find no significant difference in the numbers. EIO-LCA also provided figures for total environmental discharges 178

TABLE 1. COMPARING SOME CHEMICAL SUBSTANCE OUTPUTS OF STEEL MANUFACTURING FROM GaBi AND EIO-LCA [kg of outputs per kg of steel product] GaBi cold-rolled steel hot-dip galvanized electrolytically galvanized steel sheet steel sheet EIO-LCA steel sector (unweighted) Air + water releases (CMU-ET) Total envir. releases (unweighted) Total envir. releases (CMU-ET) Total releases &transfers (unweig hted) 1 E-04 2E-04 2E-04 2E-04 9E-04 3E-03 5E-03 TABLE 2. COMPARING SOME CHEMICAL SUBSTANCE OUTPUTS OF ALUMINUM MANUFACTURING FROM GaBi AND EIO-LCA [kg of outputs per kg of aluminum product] I I GaBi I EIO-LCA I I Aluminum sheet import mix I Primary aluminum sector Outputs (22 chemical substances) Air + water releases I 3E-03 I 2E-03 Junweig h ted) Air + water releases (CMU-ET) 2E-03 1 E-03 Total envir. releases (unweighted) 3E-03 I Total envir. releases I 5E-03 I (unweighted) I I Total releases & transfers I 2E-02 I Notes: Total environmental releases: TRI releases to air, water, land and underground wells. Transfers: off-site transfers of TRI substances. CW-EE emissions weighted by toxicity, using the ACGIH-TLVs [ 161. 179

(air + water + land + underground) and for total environmental releases and transfers (releases + off-site transfers). Unweighted numbers do not significantly change for either steel or aluminum, but the CMU-ET appears to be increasing by an order of magnitude from air + water releases to total environmental releases to total releases and transfers for the steel making process, and also about an order of magnitude from total releases to total releases and transfers for aluminum. The estimated emissions of carbon dioxide, nitrogen oxides and volatile organic compounds per kilogram of steel were close and those for aluminum manufacture were within the same order of magnitude. The estimated CO2 emissions per kilogram of aluminum were higher for GaBi, probably due to differences in fuel mix. w. SOURCES OF UNCERTAINTY AND DIFFERENCES IN RESULTS Even though the basic processes are essentially the same in the U.S. and Europe (with a certain mix of blast fimaces and electric arc haces in steel making), the two LCAs input and output results may differ due to several factors: Comprehensiveness of process inclusion. It is unclear where the boundaries of the GaBi analysis are drawn. If data were available for indirect suppliers, they were inc1udc-f. in the assessment [ 171, but this is not specifically documented for each product or process. EIO-LCA includes all the direct and indirect suppliers of the steel and aluminum making processes, as a feature of the model. For both steel and aluminum manufacturing, EIO-LCA identifies almost all sectors of the economy as direct and indirect suppliers. Differences in the energy mix for each country. Differences in energy efficiency. This results in different amounts of inputs and environmental discharges associated with the considered processes. For example, U.S. industry is less energy efficient on average than that of Germany and many other countries [ 181. Comprehensiveness of impact inclusion. GaBi reports impacts that EIO-LCA does not, and vice versa (as detailed above). GaBi s environmental emissions are to air and water, and in the form of solid or liquid municipal and hazardous waste. EIO- LCA differentiates between discharges to air, water, land and underground wells, as well as transfer of TRI chemicals to POTWs and other off-site treatment facilities. Differences in end-of-pipe treatment due to distinct regulatory requirements. Environmental emissions to various media might be different due to different regulations in each country. One country might have relatively stricter regulations for air or water discharges than the other, affecting the relative proportion of discharges to one media in that country. Therefore, environmental discharges to all media, not only air and water should be quantified and reported. For environmental management decisions, it is also important to report transfer of chemical substances to POTWs and off-site treatment facilities. Temporal differences. A kilogram of steel produced in 1987 might have had different environmental impacts than the same amount made in 1997 due to changes in regulations, efficiency of processes, quality and quantity of inputs and outputs, etc. GaBi claims to have used data from the early 1990s. EIO-LCA in its current version used economic data from 1987 (the newest input-output matrix of the U.S. economy for 1992 should appear in 1997) and TRI emission data from 1993 (the latest year available from the EPA is 1994). Our assumption is that the economic matrix coefficients have not significantly changed between 1987 and 1997. Differences in definitions of outputs. Nineteen chemical substances are common on the GaBi and EIO-LCA output list for steel, and 22 for aluminum making. However, even the commonly listed substances like lead or sulfuric acid pose a problem for direct comparison. Is lead defined as pure lead or as a lead compound with mostly lead and some impurities? What is the concentration of sulfuric acid? The exact definitions are not given either in GaBi or in the TRI (the data base for EIO-LCA toxic chemical emissions). For better information, in our EIO-LCA study, we included both discharges of chemicals (e.g., lead) and chemical compounds (e.g., lead compounds). Comprehensiveness of measurement and error in reporting. GaBi relies on data collected from facilities. EIO-LCA uses data reported to various government agencies (EPA, Department of Commerce and Energy). All these data are bound to have significant measurement and reporting errors. For example, only certain manufacturing fadities have to report to the TRI at present. The TRI has numerous other limitations [16], and a report [15] even suggests that it might be underreporting emissions by as much as a factor of 20. However, it is the most comprehensive toxic chemical release reporting system in the U.S., and it constitutes valuable public information. Level of aggregation. While GaBi assesses specific product types (e.g., galvanized steel), EIO-LCA in 180

this comparison uses entire economic sectors (e.g., steel sector). Current research at Carnegie Mellon University focuses on product-level EIO-LCA to disaggregate economic sectors to make a hybrid of the process model and EIO-LCA approaches [5]. Uncertainty of the data in both LCAs is a significant problem, but we cannot expect that data of morexeliable quality will be available any time soon, whether collected directly from plants or using publicly available data bases. A reasonable approach is to deal with uncertainty by looking for same orders of magnitude in results. LCA on a more refined level may not be meaningfkl at present. V. CONCLUSIONS We have compared results of two different LCA approaches. We found that results on the level of most individual inputs and outputs, as well as on a more aggregate level, were within a factor of 10. We conclude that EIO-LCA leads to comparable results with less effort in data gathering and updating. EIO-LCA s theoretical model allows for assessing all the environmental impacts of product suppliers, both direct and indirect. Both GaBi and EIO-LCA have input and output data that are tool-specific and are not quantified in the other software. Moreover, EIO-LCA can assess more types of environmental impacts to various media, not only air and water. While we contrasted two LCA approaches and software implementations, it is possible to combine process model stages and EIO-LCA by augmenting the input-output matrix [5]. This hybrid approach may be advantageous [ 191. The differences limit the possibility of a direct comparison of the results obtained by the two approaches and software tools. In fact, such differences would limit any comparison of results from different LCA approaches and tools. However, such comparisons are necessary in order to validate the results and benchmark the different models against each other. ACKNOWLEDGMENTS The authors gratehlly acknowledge the support of the Department of Energy - Ofice of Health and Environmental Research, Robert Bosch GmbH (Germany), and the CMU Green Design Initiative. REFERENCES Fava, J. A. et al., A Technical Framework for Life-Qcle Assessment, Society of Environmental Toxicology and Chemistry, Washington, DC, November 1991. Gediga, J., M. Harsch, K. Saur, M. Schuckert and P. Eyerer, Lifecycle Assessment - An Effective Tool for Environmental Management, Surface Mining 1996, Johannesburg, South African Institute of Mining and Metallurgy, pp. 307-311,1996. Lave, L. B., E. Cobas-Flores, C. T. Hendrickson and F. C. McMichael, Using Input-Output Analysis to Estimate Economy-Wide Discharges, Environmental Science & Technology, 29(9), pp. 420A-426A, September 1995. Cobas-Flores, E., C. Hendrickson, L. Lave and F. McMichael, Economic Input-Output Analysis to Aid Life Cycle Assessment of Electronics Products, 1995 IEEE International Symposium on Electronics and the Environment, Orlando, FL, May 1995. Hendrickson, C. T., E. Cobas-Flores, L. B. Lave and F. C. McMichael, Life Cycle Analysis of Batteries Using Economic Input-Output Analysis, 1996 IEEE International Symposium on Electronics and the Environment, Dallas, TX, May 1996. Menke, D. M., G. A. Davis and B. W. Vigon, Evaluation of Life-Cycle Assessment Tools, Unpublished Report, Hazardous Waste Branch, Environment Canada, Ottawa, ON, Canada, August 1996. Personal communication with Harald Florin, IJSP, University of Stuttgart, February 1997. Portney, P. R, The Price is Right: Making Use of Life Cycle Analyses, Issues in Science and Technology, 10(2), pp. 69-75, 1993-1994. Fiksel, J., ed, Design for Environment, McGraw-Hill, New York, 1996, ISBN 0-07-020972-3. U.S. Department of Commerce, Interindustry Economics Division, Input-Output Accounts of the U.S. Economy, 1987 Benchmark, Computer Diskettes, Washington, DC, 1994. U.S. Department of the Interior, Bureau of Mines, Minerals Yearbook 1988, Washington, DC, 1989. Department of Energy, Energy Information Administration, Annual Energy Review 1990, Washington, DC, 1991. U.S. Environmental Protection Agency, Air CHIEF, CD- ROM, Version 4.0, July 1995. U.S. Environmental Protection Agency, OBce of Pollution Prevention and Toxics, 1994 Toxics Release Inventory Public Data Release, Washington, DC, EPA 745-R-96-002, June 1996. Savitz, J. D., C. Campbell, R. Wiles, C. hart ma^, Dishonorable Discharge, Toxic Pollution of America s Waters, Environmental Working Group, Washington, DC, August 1996. Horvath, A., C. T. Hendrickson, L. B. Lave, F. C. McMichael and T-S. Wu, Toxic Emissions Indices for Green Design and Inventories, Environmental Science & Technology, 29(2), pp. 86A-90A, February 1995. Personal communication with Harald Florin, IKP, University of Stuttgart, January 1997. Cairncross, F., Costing the Earth: The Challenge for Governments, the Opportunities for Business, Harvard Business School Press, Boston, MA, 1993. Joshi, S., Comprehensive Product Life Cycle Analysis Using Input-Output Analysis Techniques, Unpublished Ph.D. Thesis Proposal, Heinz School of Public Policy and Mwgament, Carnegie Mellon University, December 1996. 1x1