9. Assessing Land-Use Impacts on Bull Trout Using Bayesian Belief Networks
137
begin to stratify watersheds into various types based on their biophysical setting
and sensitivity to various types of disturbances. Belief networks will evolve to
reflect this increased understanding and complexity. The current example, however, is limited to the type of relative, subjective judgments that are at least
consistent with the everyday experience of biologists and land managers and can
be applied universally.
Calibration of the conditional probability matrices linking habitat components
and linking habitat to model parameters involved combining expert opinion with
empirical evidence. A group of six colleagues from the Intermountain Research
Station (see acknowledgments) with extensive experience in fisheries, habitat
assessment, and watershed disturbance and recovery joined me in estimating the
conditional probabilities needed to complete the network. Each individual independently filled out probability matrices such as the ones shown in Tables 9.1 and
9.2 and presented their estimates, with the rationale behind them, in an open
discussion. Consensus values were developed during the open discussion. Finally,
the consensus values were compared with the limited data available from Idaho
and Montana and the matrix values updated to either be more consistent with the
data or to resolve minor logical inconsistencies.
One consensus view reflected in the link matrices is that degraded habitat tends
to recover or improve over time if left alone. Similarly, favorable or superior
habitat is most likely to be found in undisturbed watersheds in which conditions
remain relatively stable. Thus, when a proposed activity is judged to have no
effect on an undisturbed watershed, the belief vectors for recent and future habitat
conditions should be equivalent. For disturbed watersheds, activities having no
effect lead to slight improvements in habitat conditions.
Linking Habitat to Viability Model Parameters
Two parameters of the viability model, the environmental noise component (σ)
and the scalar determining expected trend (R), are directly linked to habitat.
Future habitat interacts with natural variation to determine the environmental
noise exhibited in the viability model. Natural variation refers to the level of
unexplained variation in the time series of redd counts that might occur naturally,
independent of habitat condition Factors that might contribute to natural variation
include such things as the frequency of extreme hydrologic events, the stability of
the parent geologic material within the watershed, or other interactions between
climate, basin topography, and vegetation. Major disturbances such as standreplacing wildfires or high-intensity floods were not included as part of the natural
variation because they would be expected to have longer-lasting impacts; such
risks are not addressed by this model. The link matrix between natural variation,
future habitat, and environmental noise reflects a general perception that favorable and superior habitats tend to dampen or moderate the impacts of natural
variation on environmental noise (Table 9.1).
The link between recent habitat, future habitat, and expected trend reflects the
logical argument that if habitat conditions improve, the population increases; if
137
begin to stratify watersheds into various types based on their biophysical setting
and sensitivity to various types of disturbances. Belief networks will evolve to
reflect this increased understanding and complexity. The current example, however, is limited to the type of relative, subjective judgments that are at least
consistent with the everyday experience of biologists and land managers and can
be applied universally.
Calibration of the conditional probability matrices linking habitat components
and linking habitat to model parameters involved combining expert opinion with
empirical evidence. A group of six colleagues from the Intermountain Research
Station (see acknowledgments) with extensive experience in fisheries, habitat
assessment, and watershed disturbance and recovery joined me in estimating the
conditional probabilities needed to complete the network. Each individual independently filled out probability matrices such as the ones shown in Tables 9.1 and
9.2 and presented their estimates, with the rationale behind them, in an open
discussion. Consensus values were developed during the open discussion. Finally,
the consensus values were compared with the limited data available from Idaho
and Montana and the matrix values updated to either be more consistent with the
data or to resolve minor logical inconsistencies.
One consensus view reflected in the link matrices is that degraded habitat tends
to recover or improve over time if left alone. Similarly, favorable or superior
habitat is most likely to be found in undisturbed watersheds in which conditions
remain relatively stable. Thus, when a proposed activity is judged to have no
effect on an undisturbed watershed, the belief vectors for recent and future habitat
conditions should be equivalent. For disturbed watersheds, activities having no
effect lead to slight improvements in habitat conditions.
Linking Habitat to Viability Model Parameters
Two parameters of the viability model, the environmental noise component (σ)
and the scalar determining expected trend (R), are directly linked to habitat.
Future habitat interacts with natural variation to determine the environmental
noise exhibited in the viability model. Natural variation refers to the level of
unexplained variation in the time series of redd counts that might occur naturally,
independent of habitat condition Factors that might contribute to natural variation
include such things as the frequency of extreme hydrologic events, the stability of
the parent geologic material within the watershed, or other interactions between
climate, basin topography, and vegetation. Major disturbances such as standreplacing wildfires or high-intensity floods were not included as part of the natural
variation because they would be expected to have longer-lasting impacts; such
risks are not addressed by this model. The link matrix between natural variation,
future habitat, and environmental noise reflects a general perception that favorable and superior habitats tend to dampen or moderate the impacts of natural
variation on environmental noise (Table 9.1).
The link between recent habitat, future habitat, and expected trend reflects the
logical argument that if habitat conditions improve, the population increases; if