54
Jane Elith
Harrell et al. 1996). Investigation of techniques for reducing the variable set for
methods such as DOMAIN and GARP would be useful. It is likely that, for many
species, expert knowledge could be used to guide the selection of a potential set of
variables for a species. Most of the variables used in this study indirectly influence
a species’ distribution. Most are commonly available in digital form. Austin and
Meyers (1995) demonstrated that the use of variables with a direct or causal effect
on distribution improves modeling success. In any study, it is worth considering
whether effort could be best spent developing variables appropriate to the modeling problem at hand, rather than (perhaps unsuccessfully) using variables that are
conveniently available but only indirectly relevant.
The primary concern is predicting potential habitat for the species, because the
objective is to provide advice on how best to protect the species in the long term. I
assume implicitly that the presence of a species reflects suitable habitat. This is
not necessarily so. For example, species may currently grow in places into which
no further recruitment is possible because of changes in climate, the development
of structural attributes in associated vegetation, or the presence of disease, predators, or competitors (including introduced species) that eliminate young plants. In
this study, analyses based on presence records and on sites considered to be
suitable habitat, regardless of absence, produce qualitatively the same results.
Analyses of habitats that were apparently unsuitable even if the species was
present were not explored.
The results suggest that the kinds of data that are available, their resolution and
accuracy, and the spatial scale at which decisions are to be used will combine to
determine the most appropriate method in any circumstance. In addition, it is clear
that validation of models needs to be done in such a way that it tests predictions in
the context in which they are to be applied.
Acknowledgments. Many people have contributed to the data, analyses, and ideas
presented here. Among them are Jennie Pearce, Guy Carpenter, and David Stockwell (who helped with modeling), Graeme Watson, Simon Ferrier, and Elizabeth
Atkinson (who allowed me to use their S-Plus routines), Paul Yates, Michele
Arundelle, Fiona Young, Adrian Moorrees, Neville Walsh, Fons Vandenberg,
Doug Frood, and Dale Tonkinson (who provided and interpreted data), and Andrew Taplin, David Barratt, Simon Ferrier, Brendan Wintle, Resit Ak¸ cakaya, and
Michael McCarthy (who contributed ideas and comments). Mark Burgman provided invaluable ideas and support. I am very grateful to all of them for their
contributions. The research was supported by project FB-NP22 of Environment
Australia.
Literature Cited
Agresti A (1996) An introduction to categorical data analysis. John Wiley and Sons, New
York
Ak¸ cakaya HR, Atwood JL (1997) A habitat-based metapopulation model of the California
Gnatcatcher. Conservation Biology 11:422– 434
Jane Elith
Harrell et al. 1996). Investigation of techniques for reducing the variable set for
methods such as DOMAIN and GARP would be useful. It is likely that, for many
species, expert knowledge could be used to guide the selection of a potential set of
variables for a species. Most of the variables used in this study indirectly influence
a species’ distribution. Most are commonly available in digital form. Austin and
Meyers (1995) demonstrated that the use of variables with a direct or causal effect
on distribution improves modeling success. In any study, it is worth considering
whether effort could be best spent developing variables appropriate to the modeling problem at hand, rather than (perhaps unsuccessfully) using variables that are
conveniently available but only indirectly relevant.
The primary concern is predicting potential habitat for the species, because the
objective is to provide advice on how best to protect the species in the long term. I
assume implicitly that the presence of a species reflects suitable habitat. This is
not necessarily so. For example, species may currently grow in places into which
no further recruitment is possible because of changes in climate, the development
of structural attributes in associated vegetation, or the presence of disease, predators, or competitors (including introduced species) that eliminate young plants. In
this study, analyses based on presence records and on sites considered to be
suitable habitat, regardless of absence, produce qualitatively the same results.
Analyses of habitats that were apparently unsuitable even if the species was
present were not explored.
The results suggest that the kinds of data that are available, their resolution and
accuracy, and the spatial scale at which decisions are to be used will combine to
determine the most appropriate method in any circumstance. In addition, it is clear
that validation of models needs to be done in such a way that it tests predictions in
the context in which they are to be applied.
Acknowledgments. Many people have contributed to the data, analyses, and ideas
presented here. Among them are Jennie Pearce, Guy Carpenter, and David Stockwell (who helped with modeling), Graeme Watson, Simon Ferrier, and Elizabeth
Atkinson (who allowed me to use their S-Plus routines), Paul Yates, Michele
Arundelle, Fiona Young, Adrian Moorrees, Neville Walsh, Fons Vandenberg,
Doug Frood, and Dale Tonkinson (who provided and interpreted data), and Andrew Taplin, David Barratt, Simon Ferrier, Brendan Wintle, Resit Ak¸ cakaya, and
Michael McCarthy (who contributed ideas and comments). Mark Burgman provided invaluable ideas and support. I am very grateful to all of them for their
contributions. The research was supported by project FB-NP22 of Environment
Australia.
Literature Cited
Agresti A (1996) An introduction to categorical data analysis. John Wiley and Sons, New
York
Ak¸ cakaya HR, Atwood JL (1997) A habitat-based metapopulation model of the California
Gnatcatcher. Conservation Biology 11:422– 434