9. Assessing Land-Use Impacts on Bull Trout Using Bayesian Belief Networks
145
models. In my work, I have taken complex population models of trout and salmon
dynamics (e.g., Lee and Hyman 1992) that generally are used only by technical
experts and recast them as BBNs (Lee and Rieman, 1997). These networks were
made accessible in spreadsheet formats and are being used by biologists throughout the Northwest (see Shepard et al. 1997). One can explore stochastic population dynamics far more efficiently with the BBNs than was ever possible by using
the original format.
There is a loss in precision by using BBNs for viability analysis. BBNs are
unlikely to replace more traditional analyses when there are sufficient data available to support them. In many cases, however, traditional approaches that rely
solely on point estimates for parameters exaggerate the true precision of estimates
and vastly understate the uncertainty in a particular course of action. In that
regard, we might be better served by BBNs that highlight inherent uncertainties
and keep us honest in our assessments.
Beyond their usefulness to practicing biologists and managers, BBNs also
benefit researchers who strive to understand underlying mechanisms linking landuse activities to population viability. When individuals participate in developing a
network and assigning values within the link matrices, they are forced to examine
their assumptions about causal mechanisms and the available scientific evidence
to support those assumptions. At times, the behavior of the network may challenge preconceived notions. Unexpected results prompt one to either reexamine
one’s preconception or examine the underlying causal linkages more carefully to
ensure that the overall network behavior conforms to general experience. One of
the more useful outcomes from developing this application was that individuals
were forced to examine their research and experience from a novel perspective.
This led to considerable discussion on the merits of various suppositions. Such
discussions should lead to more focused research efforts that have direct policy
implications.
Acknowledgments. I am grateful to my colleagues at the Rocky Mountain Research Station who assisted in developing the belief network demonstrated here
and provided useful comments on the manuscript: John McIntyre, Bruce Rieman,
Russ Thurow, Kerry Overton, James Clayton, and Gwynne Chandler. The use of
trade or firm names in this publication is for reader information and does not
imply endorsement by the U.S. Department of Agriculture of any product or
service.
Literature Cited
Andersen SK, Olesen KG, Jensen FV, Jensen F (1989) Hugin—a shell for building Bayesian belief universes for expert systems. In: Proceedings of the Eleventh International
Congress on Artificial Intelligence, pp 1080–1085. Reprinted in Shafer G, Pearl J (1990)
Readings in uncertainty. Morgan Kaufman, San Mateo, CA
Bayes T (1763) An essay towards solving a problem in the doctrines of chances. Philosophical Transactions of the Royal Society of London 53:370– 418. Reprinted (1958) in
Biometrika 45:293–315
145
models. In my work, I have taken complex population models of trout and salmon
dynamics (e.g., Lee and Hyman 1992) that generally are used only by technical
experts and recast them as BBNs (Lee and Rieman, 1997). These networks were
made accessible in spreadsheet formats and are being used by biologists throughout the Northwest (see Shepard et al. 1997). One can explore stochastic population dynamics far more efficiently with the BBNs than was ever possible by using
the original format.
There is a loss in precision by using BBNs for viability analysis. BBNs are
unlikely to replace more traditional analyses when there are sufficient data available to support them. In many cases, however, traditional approaches that rely
solely on point estimates for parameters exaggerate the true precision of estimates
and vastly understate the uncertainty in a particular course of action. In that
regard, we might be better served by BBNs that highlight inherent uncertainties
and keep us honest in our assessments.
Beyond their usefulness to practicing biologists and managers, BBNs also
benefit researchers who strive to understand underlying mechanisms linking landuse activities to population viability. When individuals participate in developing a
network and assigning values within the link matrices, they are forced to examine
their assumptions about causal mechanisms and the available scientific evidence
to support those assumptions. At times, the behavior of the network may challenge preconceived notions. Unexpected results prompt one to either reexamine
one’s preconception or examine the underlying causal linkages more carefully to
ensure that the overall network behavior conforms to general experience. One of
the more useful outcomes from developing this application was that individuals
were forced to examine their research and experience from a novel perspective.
This led to considerable discussion on the merits of various suppositions. Such
discussions should lead to more focused research efforts that have direct policy
implications.
Acknowledgments. I am grateful to my colleagues at the Rocky Mountain Research Station who assisted in developing the belief network demonstrated here
and provided useful comments on the manuscript: John McIntyre, Bruce Rieman,
Russ Thurow, Kerry Overton, James Clayton, and Gwynne Chandler. The use of
trade or firm names in this publication is for reader information and does not
imply endorsement by the U.S. Department of Agriculture of any product or
service.
Literature Cited
Andersen SK, Olesen KG, Jensen FV, Jensen F (1989) Hugin—a shell for building Bayesian belief universes for expert systems. In: Proceedings of the Eleventh International
Congress on Artificial Intelligence, pp 1080–1085. Reprinted in Shafer G, Pearl J (1990)
Readings in uncertainty. Morgan Kaufman, San Mateo, CA
Bayes T (1763) An essay towards solving a problem in the doctrines of chances. Philosophical Transactions of the Royal Society of London 53:370– 418. Reprinted (1958) in
Biometrika 45:293–315