Page 301 - Materials Chemistry, Second Edition

11 Uncertainty Management and Sensitivity Analysis 287

Björklund (2002) and by the authors of the present chapter and which is shown in
Table 11.1. A classification of uncertainty types widely used in many fields of
application distinguishes only three different types: parameter, model, and scenario
uncertainty. Most of the nine uncertainty types listed above are essentially
sub-classes of these three types as indicated in Table 11.1. Parameter uncertainty
comprises variability and uncertainty in model input parameters. Model uncertainty
indicates the uncertainty of the model itself via setup, initial and boundary condi-
tions defined, variables/indicators taken into account, and equations used. Scenario
uncertainty can be interpreted as uncertainty in the application and use of the model
and its results under predefined conditions and assumptions. Whereas parameter
and model uncertainty only contribute to the uncertainty of the numerical model
results, scenario uncertainty may also contribute to uncertainty in the interpretation
of the model results and, hence, that of a consequent decision as illustrated in
Fig. 11.9.
For a number of reasons, parameter uncertainty and variability is the uncertainty
type that is best considered in current LCA practice and it is what most people refer
to when discussing uncertainty in LCA. With occasional, rare exceptions, the few
published LCA studies that include uncertainty, essentially consider parameter
uncertainty and variability. This kind of uncertainty is estimated in LCI databases
such as ecoinvent and in some LCIA methods such as Impact World+ or LC-Impact,
and LCA software allows to include the respective calculations in an LCA study. It is
also a source of uncertainty that practitioners can address by improving data quality
and representativeness, e.g. using primary data for foreground processes, or via
spatialised LCA. This can be illustrated using three axes of data representativeness
as discussed by Weidema et al. (2003), which constitute a three-dimensional space
as shown in Fig. 11.8. LCI data may thus be too detailed, too un-specific, or too
non-representative along one, two, or all three axes. Their distance on each axis to
the range of data needed thereby represents their uncertainty.
It is important to keep in mind that most types of uncertainty and variability
listed in Table 11.1 will contribute, to varying degrees, to the overall uncertainty of
a quantitative LCA result (i.e. impact score). Just because parameter uncertainty is
essentially the most accessible one and therefore the most frequently assessed or
discussed type of uncertainty, it does not mean that it is always the most important
(i.e. most contributing) one. The ninth type in the list above (uncertainty related to
environmental relevance, accuracy or representativeness) refers to how completely
all relevant processes are included in a model, notably to how completely an
environmental mechanism is represented in a given characterisation model for a
given category midpoint or endpoint (as illustrated in Fig. 11.4). Note that com-
pleteness and representativeness relate directly to the goal and scope of an LCA,
e.g. the GWP model may be perfectly representative and complete if the goal of a
study is to calculate a carbon footprint, while it may be incomplete and of low
(environmental) relevance if the goal is to quantify the contribution of an activity to
climate change-related human health impacts. For this reason, uncertainty related to
environmental relevance or representativeness (i.e. termed here as relevance
uncertainty in line with Paparella et al. 2013) cannot be part of the model