Uncertainty analysis in agent-based modelling and consequential life cycle assessment coupled models: a critical review

The evolution of life cycle assessment (LCA) from a merely comparative tool for the assessment of products to a policy analysis tool proceeds by incorporating increasingly complex modelling approaches. In more recent studies of complex systems, such as the agriculture sector or mobility, agent-based modelling (ABM) has been introduced as tool for life cycle inventory modelling. The promises of such ABM/LCA coupled models include the consideration of human behaviour and local variabilities in the studied system as well as scenario modelling for emerging systems. The acceptance of this new approach depends, among other things, on the handling of uncertainty and variability forthcoming from various sources. As the complexity of a methodology increases, it also becomes increasingly challenging to adequately handle uncertainty and variability, and be confident about an inference. In the case of ABM/LCA coupled models, the different nature of both parts (non-linear computational ABM and linear deterministic LCA) poses an additional challenge. The sources of uncertainty and variability and the preferable propagation methods differ for both parts and clear guidance is needed. Yet no study, to our best knowledge, has addressed this issue, although its need has been expressed by several authors. In this paper, to make uncertainty analysis of ABM/LCA coupled models operational, the different uncertainty sources in both models are identified and a systematic classification is proposed. The efforts in both fields to propagate these uncertainty sources are reviewed and discussed against three criteria (applicability, accuracy and computational effort). Using ABM within LCA adds new uncertainty sources to the LCI and limits the number of applicable propagation methods, as the coupled model can no longer be expressed as an explicit formula. The context of uncertainty sources (e.g. nature of uncertainty and available information) determines which propagation method is the most appropriate and promises high accuracy, while the choice might be constraint by the computational effort.