178
Planning for persistence in a changing world
simply collations. They are also heavily dependent on
data availability, which is often most problematic in
areas with highest species richness, tropical and ecotonal areas. For quantitative predictions about future
change, we need formal models of species ’ distributions that can be projected to future changes in environmental conditions.
The simplest and most popular approach to building
distribution models with predictive ability is the correlative species distribution model. Here, one correlates a
species ’ distribution with external variables such as
physical environmental parameters or the distribution
of other species or habitat types. Perhaps the simplest
way to correlate a species ’ current distribution with
environmental variables is to use information about
habitat preferences to map the distribution of suitable
habitat, making the assumption that the species of
interest will occur wherever there is suitable habitat. If
the occurrence of suitable habitat is imperfectly
known, its distribution may be modelled statistically
using other environmental variables (Early et al. ,
2008 ).
Habitat models have been widely used in conservation science to predict the response of species to habitat
loss, essentially by using habitat extent as a surrogate
for population size (IUCN, 2001 ; such a process is actually a modelling exercise, even though it is often not
cast in such terms). To predict future change, one still
needs to model how habitats will change over time, but
this approach has been used successfully to build predictions of future changes based on change in land use
patterns.
For example, Lee and Jetz (2008) assessed the infl uence of past and future land - cover transformations on
the global reserve network across biogeographical and
geo - political regions and 5 ° latitudinal bands worldwide. They concluded that past changes in land use
poorly predicted future change and were therefore a
poor basis for making decisions about future conservation prioritization.
Clearly, models of this nature are still crude, but they
do have the advantage of ‘ capturing ’ a wide range of
ecologically important variables. If, for example, we
wanted to model the future range of forest birds in a
region, a map of future forest cover (albeit estimated)
will provide a sensible start point.
Such approaches work less well in situations where
species ’ distributions are limited by non - habitat related
factors, such as competition, predation, food availability or dispersal limitation. Even without such
identify the outer limits to its distribution. Interpolation
along the boundary can be used to trace the boundary
on a map, thus forming a representation of the species ’
distribution. This model usually takes the form of an
irregular, more or less contiguous, surface such as the
range maps printed in many modern fi eld guides.
Isolated populations might be given their own polygon,
but typically there is no formal method to decide
whether the range map should ‘ bud off ’ at a particular
location.
The use of such maps in ecological analysis has been
criticized on the basis that they ignore or grossly simplify the pattern of occupancy within species ’ ranges
(Hurlbert & Jetz, 2007 ), and it is true that maps generated from marginal occurrences tend to exhibit greater
errors of commission (false occurrences) than errors
of omission (false absences; Gaston, 1991 ; Graham &
Hijmans, 2006 ). From a conservation viewpoint, such
representations of species ’ distributions are not particularly useful, because they do not effectively discriminate occupied and unoccupied areas within a species ’
range; they are often at a scale too coarse to be practically useful; and, because they are purely pattern -
based, they cannot be used to make predictions about
the future (Rondinini et al. , 2006 ).
Where the density of records is suffi ciently high,
simply mapping all available records might suffi ce.
However, frequently records can be generalized by
mapping them as occupied/unoccupied cells of an
equal - area grid. Omission errors often outweigh commission errors in distribution maps built in this way,
because sampling is usually poor in some parts of a
species ’ range. This phenomenon can render such distribution models inadequate at any resolution that is
useful for conservation purposes (Graham & Hijmans,
2006 ; Rondinini et al. , 2006 ), although some methods
of interpolating such maps to improve their resolution,
e.g. those based on patterns of sampling effort
(H ö gmander & M ø ller, 1995 ) have been developed.
Despite the higher resolution of this kind of distribution models, they are still inherently pattern - based and
cannot be used to make predictions about future
change.
While such methods of plotting species ’ distributions are useful in representing the distributions of
species, opening the door to spatial prioritization of
conservation activity, they provide little in the way
of predictive power because there is no basis on which
to model response to future environmental change. In
other words, no model underlays these maps – they are
Planning for persistence in a changing world
simply collations. They are also heavily dependent on
data availability, which is often most problematic in
areas with highest species richness, tropical and ecotonal areas. For quantitative predictions about future
change, we need formal models of species ’ distributions that can be projected to future changes in environmental conditions.
The simplest and most popular approach to building
distribution models with predictive ability is the correlative species distribution model. Here, one correlates a
species ’ distribution with external variables such as
physical environmental parameters or the distribution
of other species or habitat types. Perhaps the simplest
way to correlate a species ’ current distribution with
environmental variables is to use information about
habitat preferences to map the distribution of suitable
habitat, making the assumption that the species of
interest will occur wherever there is suitable habitat. If
the occurrence of suitable habitat is imperfectly
known, its distribution may be modelled statistically
using other environmental variables (Early et al. ,
2008 ).
Habitat models have been widely used in conservation science to predict the response of species to habitat
loss, essentially by using habitat extent as a surrogate
for population size (IUCN, 2001 ; such a process is actually a modelling exercise, even though it is often not
cast in such terms). To predict future change, one still
needs to model how habitats will change over time, but
this approach has been used successfully to build predictions of future changes based on change in land use
patterns.
For example, Lee and Jetz (2008) assessed the infl uence of past and future land - cover transformations on
the global reserve network across biogeographical and
geo - political regions and 5 ° latitudinal bands worldwide. They concluded that past changes in land use
poorly predicted future change and were therefore a
poor basis for making decisions about future conservation prioritization.
Clearly, models of this nature are still crude, but they
do have the advantage of ‘ capturing ’ a wide range of
ecologically important variables. If, for example, we
wanted to model the future range of forest birds in a
region, a map of future forest cover (albeit estimated)
will provide a sensible start point.
Such approaches work less well in situations where
species ’ distributions are limited by non - habitat related
factors, such as competition, predation, food availability or dispersal limitation. Even without such
identify the outer limits to its distribution. Interpolation
along the boundary can be used to trace the boundary
on a map, thus forming a representation of the species ’
distribution. This model usually takes the form of an
irregular, more or less contiguous, surface such as the
range maps printed in many modern fi eld guides.
Isolated populations might be given their own polygon,
but typically there is no formal method to decide
whether the range map should ‘ bud off ’ at a particular
location.
The use of such maps in ecological analysis has been
criticized on the basis that they ignore or grossly simplify the pattern of occupancy within species ’ ranges
(Hurlbert & Jetz, 2007 ), and it is true that maps generated from marginal occurrences tend to exhibit greater
errors of commission (false occurrences) than errors
of omission (false absences; Gaston, 1991 ; Graham &
Hijmans, 2006 ). From a conservation viewpoint, such
representations of species ’ distributions are not particularly useful, because they do not effectively discriminate occupied and unoccupied areas within a species ’
range; they are often at a scale too coarse to be practically useful; and, because they are purely pattern -
based, they cannot be used to make predictions about
the future (Rondinini et al. , 2006 ).
Where the density of records is suffi ciently high,
simply mapping all available records might suffi ce.
However, frequently records can be generalized by
mapping them as occupied/unoccupied cells of an
equal - area grid. Omission errors often outweigh commission errors in distribution maps built in this way,
because sampling is usually poor in some parts of a
species ’ range. This phenomenon can render such distribution models inadequate at any resolution that is
useful for conservation purposes (Graham & Hijmans,
2006 ; Rondinini et al. , 2006 ), although some methods
of interpolating such maps to improve their resolution,
e.g. those based on patterns of sampling effort
(H ö gmander & M ø ller, 1995 ) have been developed.
Despite the higher resolution of this kind of distribution models, they are still inherently pattern - based and
cannot be used to make predictions about future
change.
While such methods of plotting species ’ distributions are useful in representing the distributions of
species, opening the door to spatial prioritization of
conservation activity, they provide little in the way
of predictive power because there is no basis on which
to model response to future environmental change. In
other words, no model underlays these maps – they are