A pseudo-statistical approach to treat choice uncertainty: the example of partitioning allocation methods

Purpose: Despite efforts to treat uncertainty due to methodological choices in life cycle assessment (LCA) such as standardization, one-at-a-time (OAT) sensitivity analysis, and analytical and statistical methods, no method exists that propagate this source of uncertainty for all relevant processes simultaneously with data uncertainty through LCA. This study aims to develop, implement, and test such a method, for the particular example of the choice of partitioning methods for allocation in LCA, to be used in LCA calculations and software. Methods: Monte Carlo simulations were used jointly with the CMLCA software for propagating into distributions of LCA results, uncertainty due to the choice of allocation method together with uncertainty of unit process data. In this study, a methodological preference is assigned to each partitioning method, applicable to multi-functional processes in the system. The allocation methods are sampled per process according to these preferences. A case study on rapeseed oil focusing on three greenhouse gas (GHG) emissions and their global warming impacts is presented to illustrate the method developed. The results of the developed method are compared with those for the same case similarly quantifying uncertainty of unit process data but accompanied by separate scenarios for the different partitioning choices. Results and discussion: The median of the inventory flows (emissions) for separate scenarios varies due to the partitioning choices and unit process data uncertainties. Inventory variations are reflected in the global warming results. Results for the approach of this study vary with the methodological preference assigned to the different allocation methods per multi-functional process and with the continuous distribution of unit process data. The method proved feasible and implementable. However, absolute uncertainties only further increased. Therefore, it should be further researched to reflect relative uncertainties, more relevant for comparative LCAs. Conclusions: Propagation of uncertainties due to the choice of partitioning methods and to unit process data into LCA results is enabled by the proposed method, while capturing variability due to both sources. It is a practical proposal to tackle unresolved debates about partitioning choices increasing robustness and transparency of LCA results. Assigning a methodological preference to each allocation method of multi-functional processes in the system enables pseudo-statistical propagation of uncertainty due to allocation. Involving stakeholders in determining these methodological preferences allows for participatory approaches. Eventually, this method could be expanded to also cover other ways of dealing with allocation and to other methodological choices in LCA