Geoscience Reference
In-Depth Information
Understanding of the sensitivity structure of the model helps one to better
anticipate the driving variables that most influence the behavior of the model;
thus Grant (1986) encourages sensitivity analysis as a fundamental step in a
validation of the model. This view is echoed by Dixon et al. (1997:472), who
suggest that “particular effort should be made to ensure that parameters with
high sensitivity are accurately known and that critical parts of the model struc-
ture are appropriately specified.”
Two considerations limit the utility of sensitivity analysis, however. First,
some parameters may not vary much in nature; even if they have high sensi-
tivities, they may not contribute much to the dynamics of the system. For
example, adult survival is invariably a highly sensitive parameter in age-struc-
tured models for long-lived species (Meyer and Boyce 1994). Yet adult survival
often varies much less than juvenile or subadult survival, so variation in adult
survival may not account for much of the variation in population growth rate.
A solution to this problem can be to decompose the variance in population
growth rate into that attributable to variation in selected parameters. Numer-
ically this is approximately equivalent to multiplying the variance times the
sensitivity squared, adjusted for covariance (Brault and Caswell 1993).
My biggest concern about the suggested utility of sensitivity analysis is the
conclusion by Dixon et al. (1997:510) that “sensitivities indicate where man-
agement actions to change vital rates have large effects on the population
growth rate.” In practice, I do not believe that sensitivities will provide much
insight into how well the system can respond to management actions. Yet usu-
ally our objective in applied population modeling is being able to anticipate
how much the system will respond to management actions. Even though pop-
ulation growth rate may have a high sensitivity to adult survival for long-lived
species, management might be at a loss to manipulate adult survival. On the
other hand, protection of breeding habitats might accomplish a dramatic
increase of reproductive output. What we really need to know is how effectively
management actions can achieve a response in a system parameter of interest.
Simple demographic sensitivities or elasticities may not provide this insight.
•
Quantitative assessment of accuracy and precision of model's outputs and
behavior.
Ultimately we want to know how well the model actually compares
with the ecological system. This can be evaluated based on how well the model
describes the data that were used to build the model, but ideally the model is
validated when predictions are compared with future behavior of the system or
when the model is applied to data from another population that was not used
for building the model.
