Keywords
stochastic results, uncertain lca results, principal component analysis, presentation
Start Date
1-7-2004 12:00 AM
Abstract
A significant challenge of an uncertainty assessment is the presentation of the results, since aquantitative uncertainty analysis dramatically increases the already considerable amount of data that needs tobe communicated in an LCA study. This paper investigates three graphical options to interpret outputsamples from quantitative uncertainty analyses. The output samples are from case studies within the coalfiredpower generation sector, and include an assessment of empirical uncertainty from a stochasticuncertainty assessment and an assessment of uncertainty in decision variables from a parametric sensitivityanalysis. Two commonly used representations of probabilistic samples are demonstrated, namely “box andwhisker” plots and plots of the cumulative probability density function, as well as the multivariate geometrictechnique, principal component analysis (PCA). Cumulative probability plots are useful representations ofuncertainty where a quantitative estimate of the relative uncertainty between options is required, but theybecome extremely tedious (many pair-wise combinations) and difficult to interpret when a large number ofoptions are compared over many criteria. In such cases, PCA can be used to provide a valuable overview ofthe results, where it is able to clearly present any trade-offs that have to be made between selection criteria,and the “spread” of the options under consideration over the decision space. Box and whisker plots are goodat representing the relative importance of empirical parameter uncertainty and the uncertainty arising fromthe choice of decision variables, and show the degree of shifting between the options as well as the full rangeover which the options potentially act. The three representations of uncertainty are found to complement eachother, as each enhances different aspects of the results. The most appropriate graphical presentation methodis found to depend on the particular decision context and the particular stage of the analysis.
Enhanced Presentation and Analysis of Uncertain LCA Results with Principal Component Analysis
A significant challenge of an uncertainty assessment is the presentation of the results, since aquantitative uncertainty analysis dramatically increases the already considerable amount of data that needs tobe communicated in an LCA study. This paper investigates three graphical options to interpret outputsamples from quantitative uncertainty analyses. The output samples are from case studies within the coalfiredpower generation sector, and include an assessment of empirical uncertainty from a stochasticuncertainty assessment and an assessment of uncertainty in decision variables from a parametric sensitivityanalysis. Two commonly used representations of probabilistic samples are demonstrated, namely “box andwhisker” plots and plots of the cumulative probability density function, as well as the multivariate geometrictechnique, principal component analysis (PCA). Cumulative probability plots are useful representations ofuncertainty where a quantitative estimate of the relative uncertainty between options is required, but theybecome extremely tedious (many pair-wise combinations) and difficult to interpret when a large number ofoptions are compared over many criteria. In such cases, PCA can be used to provide a valuable overview ofthe results, where it is able to clearly present any trade-offs that have to be made between selection criteria,and the “spread” of the options under consideration over the decision space. Box and whisker plots are goodat representing the relative importance of empirical parameter uncertainty and the uncertainty arising fromthe choice of decision variables, and show the degree of shifting between the options as well as the full rangeover which the options potentially act. The three representations of uncertainty are found to complement eachother, as each enhances different aspects of the results. The most appropriate graphical presentation methodis found to depend on the particular decision context and the particular stage of the analysis.
