Geoscience Reference
In-Depth Information
of environmental systems. In practice, however, deeper investigation often reveals
greater complexity and non-linearity in social and environmental systems than
expected. This is particularly apparent when addressing large, interdisciplinary,
problems, such as climate change, where the range of responses, and capacity of
the system to adapt (e.g., through structural changes in the land-surface and ocean
currents), will continue to generate large uncertainties.
The concepts of 'complexity', 'scale' and 'variability' are referred to as situational
because they allude to external structures, like the environment or society. However,
their meaning is derived through our
representations
of these structures. Thus,
people may be uncertain about the environment because it appears more complex
than our abstraction and simplifi cations imply, because it is too variable for us to
capture, too large to observe everything at once or too small to observe in suffi cient
detail (Brown, 2004). This is evidenced by the close relationship between environ-
mental scales and scales of measurement, modelling and presentation (Van Asselt
and Rotmans, 1996). There are also close connections between environmental vari-
ability and the variance (un)explained by statistical modelling, as variable processes
are more diffi cult to model than stationary ones (Wainwright and Mulligan, 2004).
These factors have been widely examined in geographical research. For example,
place (time, space and location) is recognised as an important control on the opera-
tion and observed outcomes of geographical processes (Richards et al., 1997).
Assessing and Managing Uncertainty and Risk
In many respects, the literatures on assessing and managing uncertainty have fol-
lowed a similar trajectory to those on risk. Both are dominated by attempts to
quantify, minimise, and control uncertainty. In recent years, there has been a pro-
liferation of studies in which theories of uncertainty have been devised and applied
to geographical problems. Most of these studies have focused on the quantifi cation
of uncertainties in geographic data (Cressie, 1993) and the propagation of uncer-
tainties through geographic models (Heuvelink, 1998). In terms of the latter, prob-
ability distributions may be developed for the uncertain inputs and parameters of
a model, sampled randomly to create different input and parameter combinations
and then propagated through the model by repeat simulation (Hammersley and
Handscomb, 1979). The propagated uncertainties can then feed directly into quan-
titative studies of risk, where probabilities of outcomes are combined with their
expected costs and benefi ts (Ayyub, 2001). Nevertheless, a distinct spectrum of
methods, not all statistical, has emerged for characterising uncertainty. For example,
the Numeral, Unit, Spread, Assessment and Pedigree (NUSAP) scheme of Funtowicz
and Ravetz (1990) employs a combination of numerical scoring and qualitative
assessment to address a range of uncertainties in scientifi c information.
An uncertainty analysis is typically limited to a few sources of uncertainty, which
may be selected by expert judgement or sensitivity testing (Saltelli et al., 2004). In
practice, the uncertainties associated with model inputs and parameters have received
much greater attention than those associated with model structure (Refsgaard et al.,
2006). The latter refers to uncertainty about social and environmental processes
(e.g., what are the dominant process controls?) and how they are manifest in obser-
vations (e.g., is a linear regression appropriate?). Typically, structural uncertainty
will lead to several methods providing reasonable accounts of the observed data or
plausible explanations of system behaviour. For example, in a study of groundwater
