138
Danny C. Lee
conditions deteriorate, the population declines. The primary question concerns the
amount of change. There was a general consensus among the group that rapid
declines or increases are relatively rare and that the chances of decline under
deteriorating or stable conditions are greater than the chances of increase under
improving or stable conditions.
Results and Discussion
In the completed Bull Trout BBN, there are five root nodes (proposed activity,
watershed history, natural variation, initial number, and density dependence) and
four leaf nodes (recent trend, average run, minimum run, and years to extinction).
The longest unidirectional path between a root node and a leaf node is between
watershed history and minimum run via environmental noise, which is four arcs in
length. All remaining paths are of length three or less. This produces a network
where substantive changes in the belief vectors at the root nodes have readily
observable impacts on the terminal leaf nodes. In networks with long paths between the root and leaf nodes, signals generally are attenuated such that major
changes in belief at the root nodes may translate into only minor changes at the
leaves. A general feature of BBNs is that the shorter the path between nodes, the
greater the response in one to a change in the other, all things being equal.
Initialization of the BBN begins with specifying the belief vectors associated
with the root nodes. In the Bull Trout network (Fig. 9.3), the root nodes have been
initialized with uniform vectors, which indicates no prior information on the state
of the system (i.e., if we know nothing about a node, all states are equally likely).
There are numerous ways that one can use the Bull Trout network to explore
various management scenarios; all involve entering findings at the appropriate
nodes and examining the resultant changes in belief vectors of interest. For illustration purposes, two types of analyses are demonstrated that could be used to
support management decisions.
Example 1: Analysis of a Specific Watershed
In the first example, a mix of activities is proposed for a hypothetical watershed
for which there is limited information. The watershed has experienced some lowlevel logging activity in recent decades, and the forest plan calls for a series of
small timber sales over the next 50 years, spread throughout the watershed. In
light of evidence that the watershed supports a local population of Bull Trout, the
forest supervisor asks the technical staff to identify a list of options and the
respective risks and costs associated with each. The technical staff identifies five
options covering the range of possible effects on future Bull Trout habitat conditions from high positive to high negative effects. These options include various
levels of logging, riparian-area and hill-slope protection measures, and specific
actions designed to enhance the habitat for Bull Trout.
Danny C. Lee
conditions deteriorate, the population declines. The primary question concerns the
amount of change. There was a general consensus among the group that rapid
declines or increases are relatively rare and that the chances of decline under
deteriorating or stable conditions are greater than the chances of increase under
improving or stable conditions.
Results and Discussion
In the completed Bull Trout BBN, there are five root nodes (proposed activity,
watershed history, natural variation, initial number, and density dependence) and
four leaf nodes (recent trend, average run, minimum run, and years to extinction).
The longest unidirectional path between a root node and a leaf node is between
watershed history and minimum run via environmental noise, which is four arcs in
length. All remaining paths are of length three or less. This produces a network
where substantive changes in the belief vectors at the root nodes have readily
observable impacts on the terminal leaf nodes. In networks with long paths between the root and leaf nodes, signals generally are attenuated such that major
changes in belief at the root nodes may translate into only minor changes at the
leaves. A general feature of BBNs is that the shorter the path between nodes, the
greater the response in one to a change in the other, all things being equal.
Initialization of the BBN begins with specifying the belief vectors associated
with the root nodes. In the Bull Trout network (Fig. 9.3), the root nodes have been
initialized with uniform vectors, which indicates no prior information on the state
of the system (i.e., if we know nothing about a node, all states are equally likely).
There are numerous ways that one can use the Bull Trout network to explore
various management scenarios; all involve entering findings at the appropriate
nodes and examining the resultant changes in belief vectors of interest. For illustration purposes, two types of analyses are demonstrated that could be used to
support management decisions.
Example 1: Analysis of a Specific Watershed
In the first example, a mix of activities is proposed for a hypothetical watershed
for which there is limited information. The watershed has experienced some lowlevel logging activity in recent decades, and the forest plan calls for a series of
small timber sales over the next 50 years, spread throughout the watershed. In
light of evidence that the watershed supports a local population of Bull Trout, the
forest supervisor asks the technical staff to identify a list of options and the
respective risks and costs associated with each. The technical staff identifies five
options covering the range of possible effects on future Bull Trout habitat conditions from high positive to high negative effects. These options include various
levels of logging, riparian-area and hill-slope protection measures, and specific
actions designed to enhance the habitat for Bull Trout.