A Python Package for the Simultaneous Global Sensitivity Analysis of Foreground and Background Parameters in Life Cycle Assessment

Highlights/Discussion Points

• Existing Python packages can perform global sensitivity analysis (GSA) of foreground or background parameters, but not both at once.
• We propose a package based on Python and Brightway that simultaneously performs GSA on foreground parameters (flow amounts, scenarios) and background parameters (impact scores of background processes, fit between foreground and background).
• The package uses an estimator for the Sobol indices, which works from a simple Monte Carlo simulation, thus significantly reducing the compute time required to perform GSA with a lot of parameters
• We show how this package is used to iteratively model the comparative LCA of an environmental sensor.

Concise Description

Context

The reliability of a life cycle assessment depends on the quality of the data used to model the system under study. However, collecting high-quality data requires a lot of time and resources. Furthermore, not all processes and parameters are equally important with respect to the LCA results. Global sensitivity analysis (GSA) techniques such as the variance-based Sobol indices indicate the contribution of input parameters to the overall output variance. There exist Python packages such as lca_algebraic, gsa_framework, and SALib that compute the Sobol indices and require a specific sampling scheme. These methods require a number of samples that scales linearly with the number of parameters under study, which is prohibitive for LCAs that can contain up to hundreds of parameters.

Furthermore, the existing packages do not allow simultaneous evaluation of foreground and background contributions. It is important for the LCA practitioner to consider the contribution from both categories, because their associated uncertainty can be addressed with different actions.

Approach

We have developed the lca_toolbox (temporary name) Python package to perform global sensitivity analysis of foreground and background contributions concurrently. It operates as a layer on top of Brightway 2.5. It estimates Sobol indices based on a Monte Carlo simulation, which can also be used for uncertainty analysis. This reduces the computation load, especially when the number of parameters is high. In this presentation, we will demonstrate the use of our package for the iterative modelling of the comparative LCA of an environmental sensor.

BrightCon Spirit

The presentation is structured as a live demonstration using a Jupyter Notebook. Our package and the demonstration notebook are available on GitHub. We wish to use this opportunity to gauge the interest of the community in our work and discuss further collaborations as well as applications beyond LCA.