4. Quantitative Methods for Modeling Species Habitat
43
• “chromosome” represents a string of variables, which in the current context are
environmental variables, with an associated set of rules that describe the relationship between the variable and the presence or absence of the species
• a pool of chromosomes is created to initiate the algorithm
• random mutations alter the chromosome to a number of different states
• recombination (crossover) occurs between chromosomes
• resulting chromosomes are evaluated (how well do they predict the sample distribution?); the poorest are made “extinct” and replaced by new chromosomes.
GARP (genetic algorithm for rule set production) is a program developed by
Stockwell and co-workers (ERIN 1995) to predict species distributions from
environmental variables in the context of a geographical information system. The
modeling component of the package uses a genetic algorithm to generate, test, and
modify rules for predicting distribution. It develops the models on test and training sets of data that are resampled sets of the original presence-absence or
presence-only species data. Predictions can be interpreted as relative likelihoods
of the presence of the species and could be understood as an index of habitat
quality for the species.
Comparative Features
There are differences between these methods that may define the most appropriate
technique for a particular situation.
Scale and Definition
ANUCLIM uses only climate variables, which are estimates based on long-term
climatic data. Although the resolution of these data is partly dependent on the
resolution of the altitude values (usually derived from a digital elevation model)
used to sample them, the climate estimates cannot provide surrogate measures of
microhabitat features that may be important in some species modeling. Hence
ANUCLIM can be viewed as a mesoscale modeling approach. Methods such as
ANUCLIM and DOMAIN that only use species presence data will also provide
less definition in their estimation of suitable habitat than those that include absence records.
Species Data
The form of species data required varies: ANUCLIM and DOMAIN only require
records of presence, GARP can deal with either presence data or presenceabsence data, and GLMs and GAMs require records of presence (or abundance)
and absence. The requirement for presence-absence data is a serious limitation to
the use of GLMs and GAMs, and there has been some exploration of their
performance with pseudo-absence data generated from nonpresence sites (see,
e.g., Ferrier and Watson 1996).
43
• “chromosome” represents a string of variables, which in the current context are
environmental variables, with an associated set of rules that describe the relationship between the variable and the presence or absence of the species
• a pool of chromosomes is created to initiate the algorithm
• random mutations alter the chromosome to a number of different states
• recombination (crossover) occurs between chromosomes
• resulting chromosomes are evaluated (how well do they predict the sample distribution?); the poorest are made “extinct” and replaced by new chromosomes.
GARP (genetic algorithm for rule set production) is a program developed by
Stockwell and co-workers (ERIN 1995) to predict species distributions from
environmental variables in the context of a geographical information system. The
modeling component of the package uses a genetic algorithm to generate, test, and
modify rules for predicting distribution. It develops the models on test and training sets of data that are resampled sets of the original presence-absence or
presence-only species data. Predictions can be interpreted as relative likelihoods
of the presence of the species and could be understood as an index of habitat
quality for the species.
Comparative Features
There are differences between these methods that may define the most appropriate
technique for a particular situation.
Scale and Definition
ANUCLIM uses only climate variables, which are estimates based on long-term
climatic data. Although the resolution of these data is partly dependent on the
resolution of the altitude values (usually derived from a digital elevation model)
used to sample them, the climate estimates cannot provide surrogate measures of
microhabitat features that may be important in some species modeling. Hence
ANUCLIM can be viewed as a mesoscale modeling approach. Methods such as
ANUCLIM and DOMAIN that only use species presence data will also provide
less definition in their estimation of suitable habitat than those that include absence records.
Species Data
The form of species data required varies: ANUCLIM and DOMAIN only require
records of presence, GARP can deal with either presence data or presenceabsence data, and GLMs and GAMs require records of presence (or abundance)
and absence. The requirement for presence-absence data is a serious limitation to
the use of GLMs and GAMs, and there has been some exploration of their
performance with pseudo-absence data generated from nonpresence sites (see,
e.g., Ferrier and Watson 1996).