An exploration of materials taxonomies to support streamlined life cycle assessment

Abstract

As life cycle assessment (LCA) gains prominence as a reliable method of environmental evaluation, concerns about data availability and quality have become more important. LCA is a resource intensive methodology, and thus data gaps pose a frequent challenge, motivating the need for robust streamlining approaches. Existing methods for filling data gaps often ignore the effects of the uncertainty inherent in estimated data. Under-specification, or using structured data to provide less information in product characterization, is one option to incorporate uncertainty, and has been shown to be a viable method both for streamlining and decision-making under uncertainty. However, previous work did not emphasize developing robust data structures intended to balance trade-offs between effectiveness and efficiency in streamlining methods. Furthermore, there was little consideration given to analyzing the environmental profile (multiple impacts) of a process, rather than a single impact. This thesis explores how data mining techniques can be used to quantitatively develop data structures to enable streamlined assessment. The use of clustering and principal component analysis is explored in an effort to identify potential material classifications, and other statistical methods further assess the classifications. These insights are used to create hierarchical taxonomies that are evaluated in terms of effectiveness and efficiency. The method is applied to life cycle inventory process datasets for three material types (metals, polymers, and precious metals). Four environmental midpoints from the TRACI 2.0 impact assessment method are used to illustrate the uncertainty reduction enabled by classification. It was found that the most useful classification method for both metals and polymers was based on price, and for precious metals, material type and recycled content. In general, the method was able to select efficient groupings that accounted for a large percentage of the overall variation in the data. With one additional level in the taxonomy, the overall median percent error rates were approximately 30- 40% for all impacts except non carcinogenicity, which was 65-80%. This is compared to initial error rates that were on average twice as high for the metals and precious metals datasets. Case studies demonstrated how the analysis and structure provided by this methodology can be useful in comparative decision-making, to reduce the number of elements prioritized for detailed data collection in triage methods, and for developing models to predict materials' impacts. This work serves as a framework for structuring data to enable streamlined LCA as well as provides guidance for predictive model development. By showing the feasibility of developing effective and efficient taxonomies, the work demonstrates a method to reduce the amount of information required to characterize a product while achieving relatively low uncertainty in the final product impact.