136
Danny C. Lee
to 0.80 (mean = 0.46). A fourth parameter, R, was used to scale ˜
N relative to N 0
(i.e., ˜
N = R∗N 0 ). A value of R greater than 1 will lead to an expected increase in
population size; a value less than 1 will lead to a decline.
Two hundred thousand combinations of N 0 , γ, σ, and R were chosen randomly
from the defined parameter space and used in 100-year simulations of the population. In each replication, the mean annual redd count (run), minimum redd count,
and year of extinction (if applicable) were retained as output variables of interest.
Although average run size provided an indication of the central tendency of the
population, the cumulative frequency distribution of the minimum value recorded
for each iteration for a given set of parameters determined the likelihood of
dropping below a given value and is thus a measure of quasi-extinction risk. Input
parameters and output variables were grouped into discrete levels to build contingency tables that would serve as link matrices between model parameters and
outputs in the completed BBN. On average, 500 replications were used for each
combination of discrete parameter ranges to estimate conditional probabilities.
Linking Land-Use Activities to Habitat
To predict population trends based on habitat, one must know something of the
recent or current habitat conditions and have some expectation of the future
conditions. In this application, I assumed that recent habitat conditions are determined solely by watershed history, in which watershed history is characterized by
the magnitude and timing of disturbance (i.e., undisturbed, minor recent, major
recent, or major old). Because recent habitat conditions are likely to have affected
recent population trends, recent trend was included as an evidence node. If information on the recent population trend is available (decreasing, stable, or increasing), then this information together with watershed history determines my belief
about recent habitat conditions using the dual flows of information discussed
above. Habitat condition, both recent and future, is divided into four levels:
submarginal, marginal, favorable, or superior—depending on the suitability of the
habitat to support a self-sustaining population of Bull Trout. Future habitat condition results from a combination of recent habitat condition and proposed activity,
in which future activity is classified according to its effect on the habitat (high or
low positive, no effect, high or low negative).
The descriptive levels for watershed history and proposed activity are meant to
be relative indicators of the intensity and magnitude of past anthropogenic or
natural disturbance and proposed future management activities. Admittedly, these
nodes may seem overly simplistic and imprecise. In part, this is unavoidable if one
wishes to develop a model that may be generalized across a landscape covering
500,000 km
2 or more. This model is targeted at watersheds on the order of 10,000
ha, and there are thousands of such watersheds within the range of Bull Trout.
Each watershed has its own unique combination of physiographic setting and
disturbance history, including both anthropogenic and natural disturbances. The
level of understanding of landscape impacts on fish is far too crude to develop
tailormade specifications for each watershed. As understanding improves, we can
Danny C. Lee
to 0.80 (mean = 0.46). A fourth parameter, R, was used to scale ˜
N relative to N 0
(i.e., ˜
N = R∗N 0 ). A value of R greater than 1 will lead to an expected increase in
population size; a value less than 1 will lead to a decline.
Two hundred thousand combinations of N 0 , γ, σ, and R were chosen randomly
from the defined parameter space and used in 100-year simulations of the population. In each replication, the mean annual redd count (run), minimum redd count,
and year of extinction (if applicable) were retained as output variables of interest.
Although average run size provided an indication of the central tendency of the
population, the cumulative frequency distribution of the minimum value recorded
for each iteration for a given set of parameters determined the likelihood of
dropping below a given value and is thus a measure of quasi-extinction risk. Input
parameters and output variables were grouped into discrete levels to build contingency tables that would serve as link matrices between model parameters and
outputs in the completed BBN. On average, 500 replications were used for each
combination of discrete parameter ranges to estimate conditional probabilities.
Linking Land-Use Activities to Habitat
To predict population trends based on habitat, one must know something of the
recent or current habitat conditions and have some expectation of the future
conditions. In this application, I assumed that recent habitat conditions are determined solely by watershed history, in which watershed history is characterized by
the magnitude and timing of disturbance (i.e., undisturbed, minor recent, major
recent, or major old). Because recent habitat conditions are likely to have affected
recent population trends, recent trend was included as an evidence node. If information on the recent population trend is available (decreasing, stable, or increasing), then this information together with watershed history determines my belief
about recent habitat conditions using the dual flows of information discussed
above. Habitat condition, both recent and future, is divided into four levels:
submarginal, marginal, favorable, or superior—depending on the suitability of the
habitat to support a self-sustaining population of Bull Trout. Future habitat condition results from a combination of recent habitat condition and proposed activity,
in which future activity is classified according to its effect on the habitat (high or
low positive, no effect, high or low negative).
The descriptive levels for watershed history and proposed activity are meant to
be relative indicators of the intensity and magnitude of past anthropogenic or
natural disturbance and proposed future management activities. Admittedly, these
nodes may seem overly simplistic and imprecise. In part, this is unavoidable if one
wishes to develop a model that may be generalized across a landscape covering
500,000 km
2 or more. This model is targeted at watersheds on the order of 10,000
ha, and there are thousands of such watersheds within the range of Bull Trout.
Each watershed has its own unique combination of physiographic setting and
disturbance history, including both anthropogenic and natural disturbances. The
level of understanding of landscape impacts on fish is far too crude to develop
tailormade specifications for each watershed. As understanding improves, we can