Page 300 - Materials Chemistry, Second Edition

286 R.K. Rosenbaum et al.

often result in a normal distribution of the impact score. This phenomenon is called
the “central limit theorem” which states that the arithmetic mean of a sufficiently
large number of independent values will be approximately normally distributed,
regardless of the underlying input distributions (Pólya 1920). Although, this the-
orem requires certain conditions to be fulfilled (e.g. independence of the included
parameters, existence of a finite expected value and standard deviation for each
parameter), it is reasonable to assume these conditions to be fulfilled by most unit
processes in LCI. This practical assumption offers several ways to significantly and
parsimoniously simplify uncertainty quantification in the LCA context with a likely
acceptable loss of precision when assuming one or only a few distribution types for
LCA input parameters.

11.3 Addressing Uncertainty in LCA

11.3.1 Types and Sources of Uncertainty and Variability
in LCA

There is no shortage of classifications of uncertainty types in literature, ranging
from only two or three classes up to ten or more different types. A very useful
classification for LCA was published by Huijbregts (1998) and comprises the
following classes:
1. Temporal variability (e.g. seasons),
2. Spatial variability (e.g. population density, climate conditions),
3. Variability between objects (e.g. between different individuals),
4. Parameter uncertainty (e.g. inaccuracy, lack or non-representativeness of input
data and model parameters),
5. Model (structure) uncertainty (e.g. algorithms in process and characterisation
models),
6. Uncertainty due to choices (e.g. definition of functional unit and system
boundaries, selection of LCIA method),
to which Björklund (2002) added:
7. Epistemological uncertainty (e.g. lack of relevant knowledge),
8. Mistakes (e.g. choosing the wrong substance or process due to similar names as
references, unit conversions or unclear units like tons vs. metric tons/tonnes),
and to which we add:

9. Relevance uncertainty (e.g. environmental relevance, accuracy or representa-
tiveness of an indicator towards an area of protection).
Huijbregts (1998) also provided an illustrative list of examples of sources of
uncertainty for each type and per LCA phase, which was slightly modified by