thought of as adjusting the model’s parameters to improve its ability to
predict the verification data set. If sufficient data are available, it is possible and common practice to use a portion of them to parameterize the
model and a second portion reserved to validate the model. This verification is not a formal test of the model, merely an attempt to ensure that
the model output fits the observations of the input used to construct the
model.
10.1.5 Model Testing
The purpose of model testing is to determine how well the model can be
extrapolated to conditions beyond those limited data under which it was
constructed. The data required for model testing are similar to those
required for verification/calibration. Whereas the verification data set may
be small, the purposes of testing are best served by larger data sets or multiple data sets that challenge the model over a broad range of conditions.
Generally, extrapolation is one of the motivations for using models to begin
with. For example, Mann et al. (2000) developed a deductive, GIS-based
model of threatened calcareous ecosystems (sometimes called cedar
barrens or slope glades), which was verified and calibrated against small
geographic data sets at Oak Ridge, Tennessee, and Fort Knox, Kentucky.
The model was then extrapolated to predict the distribution of these rare
communities across all of Fort Knox and across a much larger region
(Missouri and Tennessee). Model testing compared the model’s predicted
distributions against known occurrences in these larger areas. As with most
analytical procedures, it is important to examine the limits of extrapolation
that result from the various components of the model, make appropriate
decisions about which model components give the greatest power, and disclose the limits of extrapolation to users of the model.
10.1.6 Model Limitations
Understanding the fundamental ways that data are used, as described
above, can inform the documentation of how the model may be limited.
In addition, assumptions and best professional judgments are frequently
substituted for field data throughout the modeling process. It is incumbent upon users to understand these assumptions as well as the strengths
and limitations of the underlying data. Clearly, any limitations on the
accuracy or extent of the data will affect the output of the model, so
such limitations must be addressed throughout the modeling process and
should be understood by and disclosed to those who use the models or
their results. This documentation should accompany the model in reports,
metadata, and meetings with managers to explain the modeling process and
results.
10. Effective Ecological Modeling: Data Issues
183
predict the verification data set. If sufficient data are available, it is possible and common practice to use a portion of them to parameterize the
model and a second portion reserved to validate the model. This verification is not a formal test of the model, merely an attempt to ensure that
the model output fits the observations of the input used to construct the
model.
10.1.5 Model Testing
The purpose of model testing is to determine how well the model can be
extrapolated to conditions beyond those limited data under which it was
constructed. The data required for model testing are similar to those
required for verification/calibration. Whereas the verification data set may
be small, the purposes of testing are best served by larger data sets or multiple data sets that challenge the model over a broad range of conditions.
Generally, extrapolation is one of the motivations for using models to begin
with. For example, Mann et al. (2000) developed a deductive, GIS-based
model of threatened calcareous ecosystems (sometimes called cedar
barrens or slope glades), which was verified and calibrated against small
geographic data sets at Oak Ridge, Tennessee, and Fort Knox, Kentucky.
The model was then extrapolated to predict the distribution of these rare
communities across all of Fort Knox and across a much larger region
(Missouri and Tennessee). Model testing compared the model’s predicted
distributions against known occurrences in these larger areas. As with most
analytical procedures, it is important to examine the limits of extrapolation
that result from the various components of the model, make appropriate
decisions about which model components give the greatest power, and disclose the limits of extrapolation to users of the model.
10.1.6 Model Limitations
Understanding the fundamental ways that data are used, as described
above, can inform the documentation of how the model may be limited.
In addition, assumptions and best professional judgments are frequently
substituted for field data throughout the modeling process. It is incumbent upon users to understand these assumptions as well as the strengths
and limitations of the underlying data. Clearly, any limitations on the
accuracy or extent of the data will affect the output of the model, so
such limitations must be addressed throughout the modeling process and
should be understood by and disclosed to those who use the models or
their results. This documentation should accompany the model in reports,
metadata, and meetings with managers to explain the modeling process and
results.
10. Effective Ecological Modeling: Data Issues
183