|
Input/Output analysis - Shortcuts to life cycle data? 9. Empirically-derived distributions of life cycle emissions
Gregory Norris, Harvard University and University of New Hampshire9.1 Introduction and MotivationLCA researchers are developing and refining methods for impact assessment, which integrate a variety of approaches to fate and exposure modeling as well as (in some instances) effects modeling. This is true for a variety of impact categories. Researchers also continue to develop and refine methods to treat uncertainty quantitatively in life cycle inventory analysis. For both avenues of research, there is a need to assess the information differences (uncertainty reductions), which are achieved in LCA results by different approaches along spectra such as from generic to site-specific. Evaluation of the proposed methodological advances, and insight to guide their continued development, requires quantitative information about what life cycle inventories really look like, in general. In particular, the field needs quantitative answers to the following questions: from how many sites, in what percentages, with what geographic and temporal distributions, do life cycle emissions originate? How do these results depend on the class of pollutant? How do they depend on the class of product or process whose life cycle is being considered? 9.2 ApproachWe have undertaken empirical investigations into this subject using input/output LCA (IO-LCA) models of the US economy. The research has been undertaken with a 500-sector IO-LCA model for the US that has been constructed from databases published by the US government. Databases from the US Department of Commerce describe the flows of goods and services among the sectors in monetary terms. These can be used to estimate tier-by-tier the economic activity in the supply chain for each of 500 commodity groups. Together these 500 commodities span the entire spectrum of commodities bought and sold in the US. Separate databases from the US EPA report annual pollution releases from each sector. These data are divided by the annual economic output from each sector to derive annual average pollution coefficients for each sector. These coefficients are used with the supply chain computations to estimate supply chain pollution upstream of each commodity. The concept of supply tier is straightforward. The set of all suppliers of the direct inputs to a given using sector are termed that using sector’s "first tier suppliers." The set of all the direct suppliers to these first-tier suppliers comprise the second supply tier of the original using sector, and so on. Tier "zero" is the final sector manufacturing the commodity itself. 9.3 ResultsFirst, we computed "percentiles" for the cumulative upstream pollution by tier for the full set of US commodities, for each of the US EPA’s "criteria air pollutants" (NOx, VOCs, particulates, CO, and SO2), for CO2, and for toxic releases to air, water, land, and underground as reported in the US Environmental Protection Agency’s Toxic Release Inventory (TRI). The tier-wise cumulative percentiles indicate the share of total commodities for which cumulative emissions up to and including that tier in their upstream life cycle account for less than that percentage of the commodity’s total upstream emissions for that pollutant. Thus, the percentiles can be used to judge the probability that for some randomly chosen commodity, modeling a given number of tiers will capture a given fraction of total upstream emissions for a given pollutant. As an example, Figure 9.1 presents the cumulative percentiles for emissions of sulfur dioxide, SO2. The curve for the 5th percentile in Figure 9.1 indicates for only 95 percent (100% – 5%) of commodities (goods and services) produced in the US economy, upstream models that include the 0th through 3rd tiers will capture at least 75% of the total upstream emissions of SO2. The figure also indicates, for example, that for 75% of the commodities, models that span tiers 0 through 4 will capture at least 90% of the total upstream emissions of SO2. Figure 9.2 presents similar results for releases of volatile organic compounds (VOCs). The differences between Figure 9.1 illustrate that the speed of convergence – that is, the probability or percent of commodities for which a given number of tiers captures the bulk of the upstream emissions – this speed of convergence varies from pollutant to pollutant. These differences are captured and summarised in Figure 9.3 which presents the 25th percentiles for all of the pollutants or pollutant categories in the IO-LCA modeling system. These curves indicate that toxic releases to water and to land are the slowest pollutants to converge, with 25% of all commodities still missing, approximately 40% of their total upstream emissions after tiers 0 through 3 have been modelled. At the other end of the spectrum are CO2, toxic releases to air, and a measure of total manufacturers’ waste treatment and disposal costs ("Waste T&D") all converge more rapidly, with tiers 0 through 3 accounting for over 80% of the upstream total for at least 75% of the commodities. Next, we have used nested regional economic input/output models for the US for a geographically diverse selection of 6 US states in order to characterise the degree and speed of geographic dispersion in the supply chains of the 500 commodities. With this subset of the analysis we estimate what share of the total upstream pollution is occurring in each of a set of nested regions around the originating activity. We also estimate the nested region shares of economic activity and pollution on a tier-by-tier basis. As an example of these results, we have found that for building and construction taking place in the relatively large US state of Texas, generally less than 15% of the total upstream pollution is generated by economic activities located within the state of Texas; fully 85% or more of the upstream environmental burden occurs outside the state. The out-of-state share will be larger for most sectors, since building and construction require a relatively high share of very massive (and thus locally-sourced) inputs. Third, we look within the tiers to estimate percentiles for the numbers of individual sites contributing the bulk of the emissions for each pollutant for each tier in the supply chains of each originating commodity. Figure 9.4 presents the results for a specific pollutant (CO2), while Figure 9.5 presents the results for the 10th percentile, or 90 percent of commodities. Figure 9.5 indicates that depending upon which pollutant is selected, capturing 90% of the first tier’s emissions for 90% of the commodities requires modeling only the top 5 polluters in the tier for SO2, while it requires modeling the top 23 emission sources in the tier for toxic releases to air. Finally, we have investigated the implications of these findings for the expected uncertainty reductions in Life Cycle Impact Assessment results, which are achieved using regionalised characterisation factors. We do this for three LCIA impact categories for which regionalised (state level) characterisation factors have been recently developed for the US EPA: acidification, eutrophication, and smog. The results confirm that appreciable uncertainty reductions are achieved by using region-specific characterisation factors in place of national average characterisation factors for any single site. They go on to demonstrate, however, that for the bulk of US commodities, the expected uncertainty reductions achieved by applying these regionalised factors beyond the first supply tier are much smaller due to the number of participating sites. Figure 9.1 Look here! Figure 9.2 Look here! Figure 9.3 Look here! Figure 9.4 Look here! Figure 9.5 Look here! |