Authors
Jaxa-Rozen, Marc; Pratiwi, Astu Sam; Trutnevyte, Evelina
Abstract
Methods for uncertainty and sensitivity analysis are increasingly applied to investigate the effects of uncertainty in life-cycle assessment (LCA). However, standard “one-at-a-time” sensitivity analysis can be misleading when estimating the influence of uncertain inputs in complex models. Methods for global sensitivity analysis are more reliable, but typical variance-based techniques of global sensitivity analysis are computationally demanding and require meeting specific assumptions on the computed distributions of environmental impacts. We demonstrate three emerging techniques that yield new insights on uncertainty in typical LCA applications with any type of impact distributions, trade-offs between impacts, and interactions between uncertain model inputs. To estimate the influence of model inputs, we apply the PAWN technique for distribution-based global sensitivity analysis. To identify trade-offs between impacts and interactions between inputs, we use spectral clustering and the Patient Rule Induction Method for scenario discovery. These three techniques are applicable with generic Monte Carlo sampling and common LCA software. We demonstrate these three techniques, using a case study on life-cycle environmental impacts of geothermal heating in the State of Geneva, Switzerland. We find that the PAWN technique computationally efficiently estimates the influence of uncertain inputs, requiring a smaller number of model executions than variance-based global sensitivity analysis. Combining spectral clustering with scenario discovery also enables precisely identifying combinations of uncertain input values that lead to different outcomes in environmental impacts. We recommend adding these three techniques to uncertainty and sensitivity analysis in LCA as they offer less methodological constraints and extend the range of uncertainty insights gained at a minimal computational cost.