Year: 2013
Type: article
Abstract: The amount of time, effort and data required to conduct a thorough and accurate life cycle assessment presents a formidable obstacle to LCA practitioners. Streamlined LCA methods have been developed that make LCAs faster and simpler, generally by reducing the amount of data required. These streamlined LCAs are frequently used when it is impractical or not economical to perform a full-scale LCA, for instance when new products must be screened before purchase or during product development. For a streamlining method to be practical, it must provide results that are accurate enough (compared to full-scale LCA results) to be used in such applications. Linear regression can be used to address both the subjectivity and the lack of error information inherent in streamlined LCA. Regression can predict life cycle results from one or more predictor impacts, and cross-validation can generate estimates of the error involved in the prediction and approximate the distributions of the predicted results. This type of streamlining has a high data requirement, as full-scale LCA data is needed to calculate regression coefficients and prediction errors for each model. However, once the regression coefficients have been calculated, only predictor impact data is required to generate streamlined results. Past work on regression-based streamlining used data from the Ecoinvent database to show that cumulative energy demand (CED) is a good proxy indicator, and therefore a good predictor impact, of many other life cycle environmental impacts. This claim is based on linear regression on log transformed data, which results in a high R2 value. The interpretation of a high R is that a high proportion of variance in the log of the impact data can be explained by regression on the log of CED. We will demonstrate the flaws in this approach by showing that, in the original domain (where predictions and error estimates are needed), the ability of CED to predict other life cycle impact variables is greatly reduced. The regression-based streamlining method is demonstrated with two case studies. The first case study looked at a highly homogeneous set of 26 packaging products, and resulted in models of reasonable accuracy. The second case study consisted of a set of generic construction materials, including wood, glass and metal. This data set resulted in extremely inaccurate models, which demonstrates the importance of applying regression-based streamlining only to sets of similar products.
Source: Figshare
Authors: ISSST, Proc., Hanes, Rebecca, Bakshi, Bhavik R., Goel, Prem K.
Language: English
Subfield: Environmental Engineering
Field: Environmental Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W6921089957