Comparing sources and analysis of uncertainty in consequential and attributional life cycle assessment: review of current practice and recommendations

Life cycle assessment (LCA) is intended as a quantitative decision support tool. However, the large amount of uncertainty characteristic of LCA studies reduces confidence in results. To date, little research has been reported regarding the comparative sources of uncertainty (and their relative importance) and how, or how commonly, they are quantified in attributional and consequential LCA. This paper answers these questions based on a review of recent LCA studies and methods papers, and advances recommendations for improved practice. All relevant LCA methods papers as well as case studies (amounting to 2687 journal articles) published from 2014 to 2018 in the top seven journals publishing LCA studies were reviewed. Common sources and methods for analysis of uncertainty in both attributional and consequential LCA were described, and their frequency of application evaluated. Observed practices were compared to best practice recommendations from methods papers, and additional recommendations were advanced. Less than 20% of LCA studies published in the past five years reported any kind of uncertainty analysis. There are many different sources of uncertainty in LCA, which can be classified as parameter, scenario or model uncertainty. Parameter uncertainty is most often reported, although the other types are considered equally important. There are also sources of uncertainty specific to each kind of LCA—in particular related to the resolution of multi-functionality problems (i.e. allocation in attributional LCA versus the definition of market-mediated substitution scenarios in consequential LCA). However, there are currently no widely applied methods to specifically account for these sources of uncertainty other than sensitivity analysis. Monte Carlo sampling was the most popular method used for propagating uncertainty results, regardless of LCA type. Data quality scores and inherent (i.e. stochastic) uncertainty data are widely available in LCI databases, and researchers should generally be able to define comparable uncertainty information for their primary data. Moreover, uncertainty propagation for parameter uncertainty is supported by LCA modelling software. There are hence no obvious barriers to quantifying parameter uncertainty in LCA studies. More standardized methods based upon context-specific data that strike the right balance between comprehensiveness and usability are, however, necessary in order to better account for both the shared and unique sources of uncertainty in attributional and consequential LCAs. More frequent and comprehensive reporting of uncertainty analysis is strongly recommended for published LCA studies. Improved practices should be encouraged and supported by peer-reviewers, editors, LCI databases and LCA software developers.