Geoscience Reference
In-Depth Information
ist as independent estimates of the “true” species-environment relationship
and evaluate its variability based on these estimates.
Inductive-analytical techniques rely on samples of locations that are ana-
lyzed with some sort of statistical procedure. Different techniques have been
used, including generalized linear models (
GLM
s; McCullagh and Nelder 1988;
for applications see Akçakaya et al. 1995; Bozek and Rahel 1992; Pausas et al.
1995; Pearce et al. 1994; Pereira and Itami 1991; Thomasma et al. 1991; Van
Apeldoorn et al. 1994), Bayes theorem approach (Aspinall 1992; Aspinall and
Matthews 1994; Pereira and Itami 1991; Skidmore 1989a), classification trees
(Walker 1990; Walker and Moore 1988; Skidmore et al. 1996), and multi-
variate statistical methods such as discriminant analysis (Dubuc et al. 1990;
Flather and King 1992; Haworth and Thompson 1990; Livingston et al.
1990; Verbyla and Litvaitis 1989), discriminant barycentric analysis (Genard
and Lescourret, 1992), principal component analysis (
PCA
) (Lehmkuhl and
Raphael 1993; Picozzi et al. 1992; Ross et al. 1993), cluster analysis (Hodgson
et al. 1987), and Mahalanobis distance (Clark et al. 1993; Knick and Dyer
1997; Corsi et al. 1999).
Models that use simple univariate statistics, such as
ANOVA
, Pearson rank
correlation, and Bonferroni, pertain to a different subgroup because these
analyses do not generally allow for definition of the relative importance of the
environmental variables.
Further differences should be outlined for models that rely on the interpo-
lation of density or census estimates to extrapolate distribution patterns.
Although we have included these models in the inductive-analytical group,
the geostatistical approach (Steffens 1992) on which they are generally based
suggests putting them into a slightly different subgroup.
Finally, another means of classifying
GIS
distribution models can be based
on their outputs. Essentially, these can be distinguished as categorical-discrete
models and probabilistic-continuous models. Most often the products of the
first type of models are polygon maps in which each polygon is classified accord-
ing to a presence-absence criterion or a nominal category (e.g., frequent, scarce,
absent). The products of the second type of model are continuous surfaces of
an index that describes species presence in terms of the relative importance of
any given location with respect to all the others. Indices that have been used are
the suitability index (Akçakaya et al. 1995; Pereira and Itami 1991), probabil-
ity of presence (Agee et al. 1989; Skidmore 1989a; Aspinall 1992; Clark et al.
1993; Walker 1990), ecological distances from “optimum” conditions (Corsi et
al. 1999), and species densities (Palmeirin 1988; Steffens 1992). All these
indices can be mapped as a continuous surface throughout the species range.
