Geoscience Reference
In-Depth Information
yielding acausal and astructural 'black-boxes' that provide little heuristic insight
(e.g., see Sayer, 1992).
Although empirical models are usually seen as focused strongly, if not solely, on
prediction, they can also be used in an explanatory sense. The general intent of most
empirical modelling is establishing a relationship between some variable
x
and a
suite of predictor variables; establishing this relationship allows
indirect
causal
relationships to be established (Mac Nally, 2000). Furthermore, there is increasing
interest in applying statistical frameworks and tools, such as information-theoretic
model selection and Bayesian statistics, to bridge the gap between exploration and
prediction (Hobbs and Hilborn, 2006). In any case, the users of a prediction may
be concerned solely with the reliability of the prediction. In such cases, a black-box
approach may even be more appropriate than a complicated process-based model
that explains the underlying processes responsible for driving the system being pre-
dicted (Demeritt and Wainwright, 2005). Furthermore, such models may also be
suggestive of mechanism and help to generate new hypotheses.
Irrespective of how predictive modelling is best conducted there is, undoubtedly,
a pressing need for reliable prediction to inform (environmental) public policy and
decision making (Sarewitz et al., 1999; Clark et al., 2001; Pielke, Jr., 2003). Nev-
ertheless, the goal of accurate prediction has, itself, been questioned. Clark et al.
(2001, p. 657) take the pragmatic stance that ' “Forecastable” ecosystem attributes
are ones for which uncertainty can be reduced to the point where a forecast reports
a useful amount of information'. However, Oreskes (2003) comments that the very
factors that often lead us to modelling (limited understanding of/empirical informa-
tion about a complex and/or complicated system) restrict the use of models for
quantitative prediction. She argues that successful prediction in science has been
limited to short duration, repetitive systems of low dimensionality, and that, even
in such cases, successful prediction has often been reliant on trial and error. Con-
versely, socio-ecological systems may play themselves out over long durations, be
non-repetitive, exhibit emergent or path-dependent behaviours, and be of high
dimensionality - all traits that seem to preclude prediction (Batty and Torrens,
2001).
Unpredictability is also the key lesson of chaos theory. In chaotic (non-linearly
deterministic) systems infi nitesimally small differences in initial conditions will, in
the long-term, result in completely different dynamics and system-states. These dif-
ferences in initial conditions are much smaller than could ever be measured, and
so, in a practical sense, chaotic systems do not even possess the quality of predict-
ability (Gleick, 1987). Concerns over the ability to make reliable or meaningful
predictions have, for example, been at the centre of the debate over the siting of
the US high-level nuclear waste repository at Yucca Mountain, Nevada. 'Science',
including, but not limited to, modelling, has played a central role in attempting to
assess the performance of Yucca Mountain as a waste disposal site and billions of
dollars (US) have been spent on this process (Ewing and Macfarlane, 2002). With
a regulatory framework demanding safety assessments spanning tens of thousands
of years (!), 'geoscientists in this project are challenged to make unprecedented
predictions...' in a context where epistemic uncertainty is high and the policy
implications of those predictions even higher (Long and Ewing, 2004, p. 364). In
such situations, where science and politics are intertwined and interdependent, there
are important issues at stake about how the predictions scientists make are best
interpreted and used (Macfarlane, 2003).
