Geoscience Reference
In-Depth Information
(Clark et al. 1993; Knick and Dyer 1997). It has many interesting properties
as compared to other measures of similarity and dissimilarity, the most appeal-
ing of which is that it takes into account not only the mean values of the envi-
ronmental variables measured at observation sites, but also their variance and
covariance. Thus the Mahalanobis distance reflects the fact that variables with
identical means may have a different range of acceptability and eliminates the
problem that the use of correlated variables can have in the analysis.
Along with the identification of presence-absence data sets, each statistical
method has some specific assumption that must be satisfied for correct appli-
cation of the technique. For example, nonparametric statistical tests may
assume that a distribution is symmetric, whereas a parametric test may assume
that the test data are normally distributed. We will not discuss further the
assumptions of the different statistical methods because they are beyond the
scope of this chapter; we refer the reader to more specific topics and journal
articles on statistical methods.
SPATIAL AND TEMPORAL SCALE
Scale is a central concept in developing species distribution models with
GIS
.As
mentioned earlier in this chapter, this concept is common to both geography
and ecology, the two main disciplines involved in the development of
GIS
species distribution models. The concept of scale evolves from the representa-
tion of the earth surface on maps and is the ratio of map distance to ground
distance. Scale determines the following characteristics of a map (Butler et al.
1986): the amount of data or detail that can be shown, the extent of the infor-
mation shown, and the degree and nature of the generalization carried out.
This group of characteristics determines the quality of the layers derived,
that is, the quality of the environmental variables stored in the
GIS
database and
the type of species-environment relationship that can be investigated (Bailey
1988; Levin 1992; Gaston 1994) using the capabilities of the
GIS
.
The scale of the analysis influences the type of assumptions that need to
hold true for sound modeling. To clarify this concept, we need to consider that
species distribution is the result of both deterministic and stochastic events.
The former tend to be described in terms of the coexistence of a series of envi-
ronmental factors related to the biological requirements of the species, whereas
stochastic processes are regarded as disturbances caused by unpredictable or
unaccountable events (Stoms et al. 1992). Generally distribution models are
built on deterministic events and are averaged over wide spatial and temporal
ranges to minimize the error related to the unaccounted stochasticity.
