Environmental Engineering Reference
In-Depth Information
The strength of the BN approach is that probabilities can be propagated in
both directions through the use of Bayes' theorem. This makes it a very powerful
tool for inference. For example, if we wished to know what the effect of the
government managing a fishery was, we could set the 'management type' node
to 'government', and see how the causation propagated forward through the
network (Figure 5.8): The probability of having a declining CPUE has now risen
substantially, to 72.6%.
We could alternatively ask what would be required for sustainability, i.e. for all
three of the output nodes to have positive values? This again gives some clear and
interesting answers about what way of structuring the fishery management works
best—for example although clear access rights and fisher representation on man-
agement boards are essential, it makes little difference how many fishers there are
fishing in the area (Figure 5.9).
Source
: Halls
et al
. (2002).
●
The model may look attractive, but end-users will not be able actually to mod-
ify the model without
expert help
, so it would not be a useful legacy from a
project simply to leave the BN model behind.
●
Bayesian networks are also surprisingly
deterministic
compared to Bayesian
statistical models. The conditional probability tables give one value for the
probability of outcome A given input B, but in a standard Bayesian model this
probability would be represented as a distribution. This is the major drawback
of the approach in terms of its ability truly to represent uncertainty.
In the end, Bayesian Networks are another potential modelling approach,
but one that is relatively limited in its applications. A particularly interesting
use of BNs was suggested by Hammond and O'Brien (2001), who demonstrated
their use as the basis for consensus-building in a hypothetical dispute about
whether haddock fishers should be allowed to cull seals to stop them eating
their fish. The idea is that because their graphical outputs are relatively easy
to understand for non-specialists, they can be used in real-time negotiations,
providing a transparent illustration of the consequences of each party's under-
standing of the system. Cain
et al
. (2003) also used BNs in a workshop setting,
this time for deciding about land management and water use in a river basin in
Sri Lanka.
Predictive models are an incredibly useful and powerful tool for the exploration of
conservation options, particularly when empirical data are scarce and hard to come
