Data-Driven Environmental System Analysis: Addressing Data Gaps in Life Cycle Assessment

Abstract

Life Cycle Assessment (LCA) is a widely used analytical tool in environmental system analysis. LCA examines the environmental impacts of a product along its whole life cycle, including raw materials extraction, manufacturing, transport, use, and disposal. LCA studies are data-intensive, requiring two types of data. Unit process data are first used to calculate life cycle consumptions and emissions of a product system. And then characterization factors are used to convert the consumptions and emissions to their potential damage on the ecosystem and human health. Traditional ways to collect the two types of data involve on-site investigation of manufacturing processes and laboratory tests, which are time-consuming and expensive. Therefore, many data in LCA are missing, which generate data gaps and make LCA unable to support decision making effectively. In this research, taking advantage of existing already collected empirical data, I propose three data-driven frameworks to estimate the missing data in LCA. For the unit process data, I develop a link prediction method based on the ecoinvent database. The results show that on average missing data can be accurately estimated when less than 5% data are missing in one process. For the characterization factors, I first develop neural network models based on existing data in USEtox. The results show that the neural network models outperform a traditional quantitative structure-activity relationship (QSAR) model and linear regression models. Also based on USEtox data, I develop random forest models. The results show random forest models outperform neural network models both in prediction accuracy and computational time. Using the validated random forest model, I provide estimated missing ecotoxicity characterization factors for LCA practitioners to use. In summary, I use data-driven approaches to explore the underlying patterns of LCA data and reveal the interrelationship between manufacturing processes and the environment and between properties of contaminants and their hazard impacts. Correctly extracting the patterns behind LCA data helps estimate the missing data without relying on the time-consuming, expensive empirical data collection. The developed data-driven computational approaches will significantly reduce the cost of and save time for LCA studies, therefore help broaden the applications of LCA for sustainability decision making.