were collected and the temporal frequency of the data collection efforts.
Ideally, both the extent and resolution of the existing data should match the
extent and resolution of the model. Increasingly, modelers are taking into
account spatial variation in ecological models. However, the variation or
heterogeneity of data collected in relatively pristine areas may be different
from that of more impacted systems (Stewart and Loar 1994). Further, the
heterogeneity of the area of data collection will influence the relative
importance of extent or resolution as the determining factor of appropriateness. In areas with broad-scale homogeneity or large patches of similar
habitat, the extent of the existing data is more important than resolution to
capture the existing variability. However, when fine-scale heterogeneity or
small, interspersed patches pervade the area of data collection, the resolution of the existing data is more important than extent for characterizing
the variability.
10.3.1.3 Degree of Manipulation
It is important to be aware of the additional assumptions that are added to
the ecological model when data are obtained from an existing source. If
raw field data are available, the additional assumptions added to the model
are related to the sampling design and protocol for the data collection. In
many situations, raw field data are not available from existing sources, and
the available data have been summarized or manipulated from their raw
form. Outliers may have been eliminated; data may have been smoothed,
averaged, or normalized; or only ranges of particular interest may have
been presented. Each of these processes would have been carried out on
the basis of some assumptions, routines, or preferences. The available data
may also be the output result of a previous ecological model. These characteristics do not inherently make the existing data a poor choice for a
model, but it is necessary to know how the data were summarized or
manipulated because the effects of these processes may affect the model
and its results.
10.3.1.4 Measurement Error
Measurement error occurs when the measured sample does not accurately
represent the true value in the population. This error can result from either
lack of precision or bias in the data collection. With sufficient metadata, an
estimate of the measurement error distribution should be included or able
to be estimated. If the measurement error is known, it can be accounted for
in the model, or the sensitivity of the model to the measurement error can
be determined.
Overall, the decision to use existing data sources when constructing an
ecological model rests on an evaluation of sufficient and accurate metadata
associated with the existing data. Without metadata, it is not possible to
evaluate the tradeoff between using the existing data versus investing the
10. Effective Ecological Modeling: Data Issues
189
Ideally, both the extent and resolution of the existing data should match the
extent and resolution of the model. Increasingly, modelers are taking into
account spatial variation in ecological models. However, the variation or
heterogeneity of data collected in relatively pristine areas may be different
from that of more impacted systems (Stewart and Loar 1994). Further, the
heterogeneity of the area of data collection will influence the relative
importance of extent or resolution as the determining factor of appropriateness. In areas with broad-scale homogeneity or large patches of similar
habitat, the extent of the existing data is more important than resolution to
capture the existing variability. However, when fine-scale heterogeneity or
small, interspersed patches pervade the area of data collection, the resolution of the existing data is more important than extent for characterizing
the variability.
10.3.1.3 Degree of Manipulation
It is important to be aware of the additional assumptions that are added to
the ecological model when data are obtained from an existing source. If
raw field data are available, the additional assumptions added to the model
are related to the sampling design and protocol for the data collection. In
many situations, raw field data are not available from existing sources, and
the available data have been summarized or manipulated from their raw
form. Outliers may have been eliminated; data may have been smoothed,
averaged, or normalized; or only ranges of particular interest may have
been presented. Each of these processes would have been carried out on
the basis of some assumptions, routines, or preferences. The available data
may also be the output result of a previous ecological model. These characteristics do not inherently make the existing data a poor choice for a
model, but it is necessary to know how the data were summarized or
manipulated because the effects of these processes may affect the model
and its results.
10.3.1.4 Measurement Error
Measurement error occurs when the measured sample does not accurately
represent the true value in the population. This error can result from either
lack of precision or bias in the data collection. With sufficient metadata, an
estimate of the measurement error distribution should be included or able
to be estimated. If the measurement error is known, it can be accounted for
in the model, or the sensitivity of the model to the measurement error can
be determined.
Overall, the decision to use existing data sources when constructing an
ecological model rests on an evaluation of sufficient and accurate metadata
associated with the existing data. Without metadata, it is not possible to
evaluate the tradeoff between using the existing data versus investing the
10. Effective Ecological Modeling: Data Issues
189