40
Jane Elith
Mapped distributions may be influenced by the time of year and the inclusion or
exclusion of nonreproducing individuals and vagrants (Gaston 1994). More useful
measures of the area of occupancy of species reflect habitat suitability and probability of occurrence on a continuous scale.
Different places support different densities of individuals of a species, depending on a range of historical, ecological, and demographic parameters. Potential
habitat does not necessarily coincide with a species current or past distribution,
even if these are known without error. Van Horne (1983) pointed out that areas in
which a species is found relatively frequently do not necessarily represent the
areas in a landscape associated with greatest survival or reproductive success.
Individuals of a species may not congregate in the most suitable locations because
of behavior, intraspecific competitive exclusion, interspecific interactions, or
dispersal dynamics. As a result, population sinks may have high population densities but be of relatively limited value in contributing to the likelihood of persistence of a species.
Some of these procedures are based on qualitative information, subjective
estimates, and expert judgment, e.g., habitat suitability indices (USFWS 1980;
Crance 1987; Rand and Newman 1998). This chapter describes several quantitative modeling methods that can be used to map predicted species distibutions from
various combinations of species and environmental data. Five of them were
applied to data from several plant species in a forested landscape in Australia. The
resulting predictions were assessed in the context of the ability of the models to
provide results that could be useful in the management of land for conservation.
Modeling Methods
A broad range of computer modeling methods has been applied to the problem of
predicting species distributions, and some of the more common approaches are
summarized with key references and examples in Table 4.1. Often, the choice of
method for prediction is related to availability of software and local expertise.
Few studies involve comparison of modeling methods. Exceptions include comparisons of generalized linear models (GLM), generalized additive models
(GAM), decision trees, and genetic algorithms by Austin and Meyers (1995), and
ANUCLIM, GLMs, GAMs, and decision trees by Ferrier and Watson (1996). In
this section, the attributes and advantages of five of the methods are described and
evaluated.
ANUCLIM (also known as BIOCLIM) is an example of a climate-mapping
approach to modeling. It uses presence data as point locations together with
elevation data and climatic surfaces developed from long-term rainfall, temperature, and radiation records to construct a climate profile for a species. With several
climatic parameters, the aggregated profile forms a multidimensional rectangle
known as a “bounding box” or “climatic envelope.” Several sets of bounds can be
defined within these profiles: they can be based on the range of all observed
points, on pairs of percentiles, or on means and standard deviations. The mapping
Jane Elith
Mapped distributions may be influenced by the time of year and the inclusion or
exclusion of nonreproducing individuals and vagrants (Gaston 1994). More useful
measures of the area of occupancy of species reflect habitat suitability and probability of occurrence on a continuous scale.
Different places support different densities of individuals of a species, depending on a range of historical, ecological, and demographic parameters. Potential
habitat does not necessarily coincide with a species current or past distribution,
even if these are known without error. Van Horne (1983) pointed out that areas in
which a species is found relatively frequently do not necessarily represent the
areas in a landscape associated with greatest survival or reproductive success.
Individuals of a species may not congregate in the most suitable locations because
of behavior, intraspecific competitive exclusion, interspecific interactions, or
dispersal dynamics. As a result, population sinks may have high population densities but be of relatively limited value in contributing to the likelihood of persistence of a species.
Some of these procedures are based on qualitative information, subjective
estimates, and expert judgment, e.g., habitat suitability indices (USFWS 1980;
Crance 1987; Rand and Newman 1998). This chapter describes several quantitative modeling methods that can be used to map predicted species distibutions from
various combinations of species and environmental data. Five of them were
applied to data from several plant species in a forested landscape in Australia. The
resulting predictions were assessed in the context of the ability of the models to
provide results that could be useful in the management of land for conservation.
Modeling Methods
A broad range of computer modeling methods has been applied to the problem of
predicting species distributions, and some of the more common approaches are
summarized with key references and examples in Table 4.1. Often, the choice of
method for prediction is related to availability of software and local expertise.
Few studies involve comparison of modeling methods. Exceptions include comparisons of generalized linear models (GLM), generalized additive models
(GAM), decision trees, and genetic algorithms by Austin and Meyers (1995), and
ANUCLIM, GLMs, GAMs, and decision trees by Ferrier and Watson (1996). In
this section, the attributes and advantages of five of the methods are described and
evaluated.
ANUCLIM (also known as BIOCLIM) is an example of a climate-mapping
approach to modeling. It uses presence data as point locations together with
elevation data and climatic surfaces developed from long-term rainfall, temperature, and radiation records to construct a climate profile for a species. With several
climatic parameters, the aggregated profile forms a multidimensional rectangle
known as a “bounding box” or “climatic envelope.” Several sets of bounds can be
defined within these profiles: they can be based on the range of all observed
points, on pairs of percentiles, or on means and standard deviations. The mapping