Dean Amy S. Fleischer and a team of researchers from Villanova University published their paper “Life Cycle Inventory Regionalization and Uncertainty Characterization: a Multilevel Modeling Approach,” in the Journal of Cleaner Production.
Authors: Tao Dai; Amy S. Fleischer; Ross Lee and Aaron P. Wemhoff
Highlights
- Life cycle inventory could be regionalized using multilevel modeling
- Approach in this study yields higher accuracy than current averaging methods•
- Multilevel modeling enables objective uncertainty characterization
Abstract
Life cycle inventory (LCI) databases are essential to evaluate the environmental burden of a product or service using life cycle assessment (LCA). The demand for improved accuracy and credibility of LCA results has increased the need for LCI containing higher data granularity and quantified uncertainty. In this study, a novel data processing method is proposed to facilitate regionalized LCI database compilation and uncertainty characterization. The key step in the proposed method is to retain variations within the datasets by building a relationship between the geographical and temporal quantified characteristics and material/energy input amounts using multilevel modeling (MLM) rather than losing the variations via a traditional aggregation method. The uncertainty is objectively characterized by the model prediction interval. An example case study is conducted on nitrogen fertilizer application for corn production with data collected in 19 U.S. states from 1990 to 2016. The results show that estimates by the proposed approach are more accurate than those by current aggregation methods. For the unit-yield-based input and unit-area-based input, respectively, around 60% and 80% of variations within the datasets are explained by MLM, and 86.1% and 96.8% of the unrepresentativeness factors for the MLM estimates are less than 0.2, compared to 62.3% and 70.5% when traditional three-year aggregation is used. The 95% prediction intervals cover more than 95% of measurement data in both input cases. It is suggested to apply an MLM-based data processing framework for future database compilation or for reinforcing a geographical information system (GIS) based regionalized LCI database.
