Global Sensitivity Analysis in LCA
Life Cycle Assessment (LCA) is a well established tool used for quantification of product impacts that arise throughout the entire value chain. It aims at support- ing environmentally informed decisions in policy-making, product development, and consumer choices. However, LCA results (environmental impact scores) can be highly uncertain due to the large amounts of measured and simulated data that go into LCA models. Therefore, results can only be interpreted confidently if this uncertainty is sufficiently narrow.
The number of uncertain datasets (inputs) in LCA models can reach hun- dreds of thousands. Improving quality of all these datasets is not only infeasible but also unnecessary, because just a small fraction of them drives uncertainty in LCA results. In order to support prioritized data collection and uncertainty reduction, we use Global Sensitivity Analysis (GSA) - the study of how un- certainty in the output of a model can be apportioned to different sources of uncertainty in the model input [1].
A typical LCA model can be written in the matrix-based form as follows
With respect to GSA, model inputs can be entries with uncertainties of one of the matrices or their combinations. Importance of each input is computed with Monte Carlo analysis and various statistical estimators, for example Sobol’ and moment-independent indices [2]. Typically, this class of models allows numerous Monte Carlo simulations as one model run is computationally cheap (< 1s). On the other hand, number of model inputs can reach hundreds of thousands, which makes LCA models unusual and challenging for GSA application.
Closely related to GSA is the field of dimensionality reduction - the process of reducing the number of dimensions in the input space [3]. Here we consider supervised methods, such as decision trees.
References
[1] Andrea Saltelli, Stefano Tarantola, Francesca Campolongo, and Marco Ratto. Sensitivity analysis in practice: a guide to assessing scientific models. John Wiley & Sons, 2004.
[2] Francesca Pianosi, Keith Beven, Jim Freer, Jim W Hall, Jonathan Rougier, David B Stephenson, and Thorsten Wagener. Sensitivity analysis of environ- mental models: A systematic review with practical workflow. Environmental Modelling & Software, 79:214–232, 2016.
[3] Isabelle Guyon and Andr ́e Elisseeff. An introduction to variable and feature selection. Journal of machine learning research, 3(Mar):1157–1182, 2003.