model parameters from a distribution of values instead of choosing a single
value for a parameter. With this strategy, a relatively simple model is run
numerous times (hundreds or thousands), producing a distribution of
possible model outputs. The modeler acknowledges uncertainty because the
multiple model outputs are generally presented in a statistical form (e.g.,
20% of the possible outcomes result in a 10% decrease in population size).
In this context, the modeler presents the results in terms of the risk of a
certain event occurring. Unfortunately, probabilistic formulations of model
outputs may be confusing for decision makers because clear, unequivocal
answers are not provided.
An alternative to model-based risk assessment is the use of large,
comprehensive models that attempt to duplicate critical natural processes.
These models typically have lengthy run times, so that running them
hundreds or thousands of times is not feasible. Additionally, these models
are typically used for regulatory purposes, where relative answers may be
insufficient. These models typically use engineering methods to optimize
model parameters and to confirm the performance of the simulation. While
it is not possible to remove all sources of error and uncertainty from these
models, efforts are generally made to optimize model performance, to identify model sensitivity to key parameters through Monte Carlo simulation
(in which certain model parameters are randomly changed), and to describe
the error structure of the model by comparing model predictions to
observed data. Error analysis helps the modeler identify weaknesses of
the model or biases (particular scenarios in which certain state variables
may be systematically underestimated or overestimated). This explicit
representation of uncertainties tends to enhance communication only for
modelers who are comfortable with large, comprehensive models (and not
necessarily for decision makers).
8.4.1.4 Model Standards
Effective communication of model results depends upon adherence to
certain standards in model development. Ecological models are used in at
least two ways, conceptual exploration (research) and projection (decision
making). Exploratory models are used to better understand complex
natural processes so that the driving variables and relationships between
variables can be studied. Exploratory models are often highly specialized,
and their accuracy is evaluated in terms of the statistical variation explained
by a model. Alternatively, models used in a regulatory context to support
decisions and determine policies are often developed and applied by the
engineering profession. Development of engineering models is usually
founded on a mathematical description of conservation of mass and
momentum principles. Model documentation and confirmation are critical
elements in establishing the credibility of a model and its application. It is
important that models, particularly those used in a regulatory context, be
152
Eric Gustafson et al.
value for a parameter. With this strategy, a relatively simple model is run
numerous times (hundreds or thousands), producing a distribution of
possible model outputs. The modeler acknowledges uncertainty because the
multiple model outputs are generally presented in a statistical form (e.g.,
20% of the possible outcomes result in a 10% decrease in population size).
In this context, the modeler presents the results in terms of the risk of a
certain event occurring. Unfortunately, probabilistic formulations of model
outputs may be confusing for decision makers because clear, unequivocal
answers are not provided.
An alternative to model-based risk assessment is the use of large,
comprehensive models that attempt to duplicate critical natural processes.
These models typically have lengthy run times, so that running them
hundreds or thousands of times is not feasible. Additionally, these models
are typically used for regulatory purposes, where relative answers may be
insufficient. These models typically use engineering methods to optimize
model parameters and to confirm the performance of the simulation. While
it is not possible to remove all sources of error and uncertainty from these
models, efforts are generally made to optimize model performance, to identify model sensitivity to key parameters through Monte Carlo simulation
(in which certain model parameters are randomly changed), and to describe
the error structure of the model by comparing model predictions to
observed data. Error analysis helps the modeler identify weaknesses of
the model or biases (particular scenarios in which certain state variables
may be systematically underestimated or overestimated). This explicit
representation of uncertainties tends to enhance communication only for
modelers who are comfortable with large, comprehensive models (and not
necessarily for decision makers).
8.4.1.4 Model Standards
Effective communication of model results depends upon adherence to
certain standards in model development. Ecological models are used in at
least two ways, conceptual exploration (research) and projection (decision
making). Exploratory models are used to better understand complex
natural processes so that the driving variables and relationships between
variables can be studied. Exploratory models are often highly specialized,
and their accuracy is evaluated in terms of the statistical variation explained
by a model. Alternatively, models used in a regulatory context to support
decisions and determine policies are often developed and applied by the
engineering profession. Development of engineering models is usually
founded on a mathematical description of conservation of mass and
momentum principles. Model documentation and confirmation are critical
elements in establishing the credibility of a model and its application. It is
important that models, particularly those used in a regulatory context, be
152
Eric Gustafson et al.