develop, parameterize, calibrate, and validate the model. Models are
simplifications based on sets of information, some of which are specific to
the ecosystem, ecological component, or stressor at hand. Some data are
derived from information on other systems, at other locations, or affected
by other stressors. Thus, there is a continual issue of how appropriate the
information used to develop the model is and how well it reflects the actual
response characteristics of the ecosystem and stressor of concern. If the data
sources and applicability are clear, documented, and relevant, then model
confidence is greatly enhanced. Data are often not site-specific to the
ecosystem or species of concern. For example, toxicological data on dose
responses often are based on one or two species of fish that can be tested
in the laboratory as a surrogate for toxicity to a fish species of concern in
the ecosystem but that cannot be tested experimentally. In that situation, a
case needs to be made on how well the tests can be expected to fit the
species of concern, based on, for example, physiological, taxonomic, or ecological similarities. Likewise, data may be derived from another ecosystem,
such as taken at one lake but applied to another lake. Moreover, many times
data relate to one stressor, but another is being assessed, or there are
multiple stressors involved. Again, the case has to be examined as to how
reliably the extrapolation can be made (in the context of resulting uncertainties and the significance of those uncertainties). Confidence in results is
built when the case can be made that the important relationships are
broadly based to cover the specific assessment, such as through reliance on
first principles or through demonstration in other similar cases where the
information has been appropriate. A significant model barrier that continues to require considerable attention, but is beyond the scope of this
paper, is that of cumulative effects from multiple stressors [see Gentile and
Harwell (2001)].
5.2.9 Model Development
What are the model development costs in money and time; how long will
model development take; when will a reliable model be sufficiently ready
for decision support; and what is the value added by having the model?
Even when it is clear that a model would be useful for a decision-making
process, the question arises of costs and time delays in producing the model
and, thus, the decision. Some decisions cannot be delayed until adequate
model development occurs; in that case, other than relying on another
already developed model, there is little to be done for the initial decision
(although the case might be made to proceed with model development
anyway in order to have the tool available for future decisions or to refine
the initial decision.) In other cases, the utility of the model may be very
high, in which case the decision maker has to weigh the pros and cons of
delaying a decision. The value-added assessment basically relates to the
judgment that the model will substantively increase the likelihood of
100
Mark A. Harwell and John H. Gentile
simplifications based on sets of information, some of which are specific to
the ecosystem, ecological component, or stressor at hand. Some data are
derived from information on other systems, at other locations, or affected
by other stressors. Thus, there is a continual issue of how appropriate the
information used to develop the model is and how well it reflects the actual
response characteristics of the ecosystem and stressor of concern. If the data
sources and applicability are clear, documented, and relevant, then model
confidence is greatly enhanced. Data are often not site-specific to the
ecosystem or species of concern. For example, toxicological data on dose
responses often are based on one or two species of fish that can be tested
in the laboratory as a surrogate for toxicity to a fish species of concern in
the ecosystem but that cannot be tested experimentally. In that situation, a
case needs to be made on how well the tests can be expected to fit the
species of concern, based on, for example, physiological, taxonomic, or ecological similarities. Likewise, data may be derived from another ecosystem,
such as taken at one lake but applied to another lake. Moreover, many times
data relate to one stressor, but another is being assessed, or there are
multiple stressors involved. Again, the case has to be examined as to how
reliably the extrapolation can be made (in the context of resulting uncertainties and the significance of those uncertainties). Confidence in results is
built when the case can be made that the important relationships are
broadly based to cover the specific assessment, such as through reliance on
first principles or through demonstration in other similar cases where the
information has been appropriate. A significant model barrier that continues to require considerable attention, but is beyond the scope of this
paper, is that of cumulative effects from multiple stressors [see Gentile and
Harwell (2001)].
5.2.9 Model Development
What are the model development costs in money and time; how long will
model development take; when will a reliable model be sufficiently ready
for decision support; and what is the value added by having the model?
Even when it is clear that a model would be useful for a decision-making
process, the question arises of costs and time delays in producing the model
and, thus, the decision. Some decisions cannot be delayed until adequate
model development occurs; in that case, other than relying on another
already developed model, there is little to be done for the initial decision
(although the case might be made to proceed with model development
anyway in order to have the tool available for future decisions or to refine
the initial decision.) In other cases, the utility of the model may be very
high, in which case the decision maker has to weigh the pros and cons of
delaying a decision. The value-added assessment basically relates to the
judgment that the model will substantively increase the likelihood of
100
Mark A. Harwell and John H. Gentile