144
Danny C. Lee
biologist is able to recommend a conservation strategy using a mix of watersheds
in various stages of disturbance instead of being limited to only pristine areas.
This increased flexibility allows a more robust strategy that incorporates a mosaic
of conditions, which reduces overall risks from sources not directly addressed by
the belief network (Rieman and McIntyre 1993). As expected, fewer options are
available to reduce the risks sufficiently in the most degraded watersheds.
Conclusions
Conservation of Bull Trout and other TES species is a very complex and contentious issue that involves much uncertainty. Difficult land-use choices will have to
be made to ensure population persistence. As this application shows, Bayesian
Belief Networks can assist the decision process by quantifying and displaying the
risks associated with alternative courses of action.
Several important features of BBNs are demonstrated above. First, the reasoning, beliefs, and assumptions that lead to an assessment are open to scrutiny by all
parties affected by a decision. In this example, I relied on the expert knowledge of
colleagues to construct the link matrices that project expected population trends
and environmental noise from habitat conditions and proposed activities. The use
of subjective probabilities from experts is problematic. Morgan and Henrion
(1990) provide an excellent overview of potential biases and error that can arise
with elicited expert judgment. Errors can arise because individuals lack the cognitive skills to assess probabilities accurately or because they might sense a strategic
advantage in skewing the probabilities in one direction or another. Morgan and
Henrion (1990) conclude that “one can only proceed with care, simultaneously
remembering that elicited expert judgments may be seriously flawed, but often are
the only game in town.” The use of expert judgment in management of natural
resources is not only unavoidable, it is desirable given the complexity of the issues
involved. BBNs simply provide a convenient means of capturing expert opinions
in ways that can be easily scrutinized.
Second, the approach recognizes that although the future outcome of an activity
is always uncertain, this uncertainty can be quantified in ways that allow meaningful comparisons. The use of probability theory within the field of risk assessment
and policy analysis is well established. As Ramsey (1926), de Finetti (1974), and
others have shown, the rules of probability apply even when the probability
merely reflects one’s subjective degree of belief (Press 1989; Howson and Urbach
1991). Also, as Morgan and Henrion (1990) note, if we must trust management to
experts, should we not at least know how certain they are of their answers?
Third, complex relationships can be captured in a format that promotes exploration of alternatives. The network described here and the viability model that it
relies on are both relatively simple compared with the complex system models
familiar to many ecological modelers. Yet, I contend that it is far easier to explore
the dynamics of this system by using a BBN than it would be using a more
traditional modeling approach, and the same can be said for more complex
Danny C. Lee
biologist is able to recommend a conservation strategy using a mix of watersheds
in various stages of disturbance instead of being limited to only pristine areas.
This increased flexibility allows a more robust strategy that incorporates a mosaic
of conditions, which reduces overall risks from sources not directly addressed by
the belief network (Rieman and McIntyre 1993). As expected, fewer options are
available to reduce the risks sufficiently in the most degraded watersheds.
Conclusions
Conservation of Bull Trout and other TES species is a very complex and contentious issue that involves much uncertainty. Difficult land-use choices will have to
be made to ensure population persistence. As this application shows, Bayesian
Belief Networks can assist the decision process by quantifying and displaying the
risks associated with alternative courses of action.
Several important features of BBNs are demonstrated above. First, the reasoning, beliefs, and assumptions that lead to an assessment are open to scrutiny by all
parties affected by a decision. In this example, I relied on the expert knowledge of
colleagues to construct the link matrices that project expected population trends
and environmental noise from habitat conditions and proposed activities. The use
of subjective probabilities from experts is problematic. Morgan and Henrion
(1990) provide an excellent overview of potential biases and error that can arise
with elicited expert judgment. Errors can arise because individuals lack the cognitive skills to assess probabilities accurately or because they might sense a strategic
advantage in skewing the probabilities in one direction or another. Morgan and
Henrion (1990) conclude that “one can only proceed with care, simultaneously
remembering that elicited expert judgments may be seriously flawed, but often are
the only game in town.” The use of expert judgment in management of natural
resources is not only unavoidable, it is desirable given the complexity of the issues
involved. BBNs simply provide a convenient means of capturing expert opinions
in ways that can be easily scrutinized.
Second, the approach recognizes that although the future outcome of an activity
is always uncertain, this uncertainty can be quantified in ways that allow meaningful comparisons. The use of probability theory within the field of risk assessment
and policy analysis is well established. As Ramsey (1926), de Finetti (1974), and
others have shown, the rules of probability apply even when the probability
merely reflects one’s subjective degree of belief (Press 1989; Howson and Urbach
1991). Also, as Morgan and Henrion (1990) note, if we must trust management to
experts, should we not at least know how certain they are of their answers?
Third, complex relationships can be captured in a format that promotes exploration of alternatives. The network described here and the viability model that it
relies on are both relatively simple compared with the complex system models
familiar to many ecological modelers. Yet, I contend that it is far easier to explore
the dynamics of this system by using a BBN than it would be using a more
traditional modeling approach, and the same can be said for more complex