Context

 LCA started as a "deterministic" business. Only gradually, the importance of including uncertainty became a serious topic. The result is that LCA is still typically considered as a deterministic model, with add-ons to account for uncertainty. This is, for instance, reflected in the various formats and softwares for LCA, where an amount is specified, with an optional uncertainty parameter.

When you think of LCA as an inherently stochastic model, this makes little sense. Just take the example of a fair die, with equal probabilities of obtaining 1, 2, 3, 4, 5 or 6. Could you specify this as a deterministic outcome (3.5?) with an uncertainty range (2.5?)? No, you would specify the outcome (let's call it X) as X~U(1, 6), with U(a, b) denoting the discrete uniform distribution with parameters a (the minimum) and b (the maximum).

Forcing stochastic LCA data into a "determinstic+uncertainty" structure, various software programs and data formats have made confusing choices. I have been working on unravelling this in a number of publications.

My contributions

In 2005, I wrote with Rolf Frischknecht Representing statistical distributions for uncertain parameters in LCA. Relationships between mathematical forms, their representation in EcoSpold, and their representation in CMLCA (https://doi.org/10.1065/lca2004.09.177). It discusses four standard probability distributions (uniform, normal, triangular and lognormal). The paper refers to software with two links (ecoinvent and Leiden University) which are no longer valid. The software can still be downloaded from this site.

Around 2013, the ecoinvent database moved to a new format, EcoSpold2. Unfortunately, the new format did not improve the situation; actually its documentation contains unclear and conflicting descriptions.

In 2024, I published two further analyses. My book Probability, statistics and life cycle assessment. Guidance for dealing with uncertainty and sensitivity (https://doi.org/10.1007/978-3-031-49317-1; see also here) discusses several formats and computer documentations, for a variety of probability distributions. See in particular Sections 6.6.4-6.6.6 and 11.1.7. This latter section is only discussing the lognormal distribution, but is due to its default status in ecoinvent of critical importance.

The second source is my article Analysis and remediation of the confusing specification of the lognormal distribution (https://doi.org/10.1007/s11367-023-02249-8), which is a dedicated analysis of the lognormal case.

In response to the (highly-welcomed) conversion software, I further discussed the case of the lognormal distribution in the two most popular formats, EcoSpold2 and ILCD. See my letter Interoperability of uncertainty information in LCI databases (https://doi.org/10.1007/s11367-026-02755-5).

Colophon

This page was created in 2026 by Reinout Heijungs, based on my personal knowledge and some random internet searches. See https://sites.google.com/vu.nl/reinoutheijungs/contact for more information, contact details, and other projects.