Page 287 - Materials Chemistry, Second Edition

11 Uncertainty Management and Sensitivity Analysis 273

uncertainty is a good part of the problem’s root in both cases. Uncertainty is indeed
frequently perceived as potentially discrediting LCA and its results as being too
uncertain, unreliable, and insufficiently capable of distinguishing the compared
options. The often considerable resources required for quantifying and managing
uncertainty in an LCA study is an important barrier for their adequate consideration.
Nevertheless, the presence of uncertainties of different types and from numerous
sources in LCA results is a fact and ignoring them may be more detrimental than
managing them in an integrated manner which allows their meaningful use to
quantify and improve the precision of a study and the robustness of its conclusions.
LCA practice sometimes suffers from an imbalanced perception of uncertainties
and their use in justifying modelling choices and omissions (e.g. excluding impact
categories due to their perceived uncertainty). Identifying prevalent misconcep-
tions, in some cases “myths”, around uncertainties is another central goal of this
chapter. The ambition is to help balancing the discussions around uncertainty in
LCA and establish a positive discourse that focuses on the advantages of uncer-
tainty management. Proper uncertainty management allows for more robust results
and conclusions in support of science-based decision-making, grounded on the
(accurate) recognition and discussion of inevitable and ubiquitous uncertainties.
Consider the following conceptual and simplified example to illustrate how
fundamentally useful uncertainty assessment and management are in LCA.
Figure 11.1 shows the results of an LCA study, performing a comparison of two
alternative options A and B, for a given impact category like water use for example.
The point estimate (i.e. reproducible, single value output from the LCA model
without considering variations in inputs) impact score is 4 for option A and 6 for
option B, which may suggest that option A is preferable, i.e. less environmentally
impacting, over option B by a factor of 1.5. However, considering the uncertainties
(including correlations between both options), the impact scores can be shown as
superposed distributions as demonstrated in Fig. 11.1 (even though this may not be
the best way to compare scenarios as discussed later in this chapter). Where the
distributions are overlapping, option B has certain chances to be preferable over
option A, the opposite of the conclusion drawn above from only looking at the point
estimates. The more the distributions overlap, the higher the chances that option
A may not be preferable to option B. In the left plot, there is a relatively small
overlap of both distributions, and hence a relatively low chance to take the wrong
decision when preferring option A over option B. In the centre plot, it is essentially
impossible to discern the impact scores of both options and the chances to make the
wrong conclusion would be high, no matter which option is chosen. In the right
plot, the dispersions of both options are different (which will usually be the case in
practice) and need to be evaluated in order to derive more reliable results. How to
deal with such cases is discussed further-on in the chapter. That means that if the
uncertainty cannot be further reduced (e.g. by using more certain data or models),
both options are basically equal in terms of their potential environmental impact on
water use.
The consideration and communication of uncertainties related to results obtained
via modelling and/or measurements is vital for their correct interpretation. This is