Conservation planning in a changing world
179
complications, many species have highly specifi c
microhabitat requirements that are diffi cult to capture
using broad - scale or remotely sensed habitat data.
Because many species ’ distributions are limited by features other than broad habitat type, estimates of suitable habitat are often masked by intersecting some
broad representation of the known geographical range
of the species of interest, so as not to generate predictions beyond the bounds of known or likely occurrence
(Rondinini et al. , 2005 ; Harris & Pimm, 2008 ). This
makes prediction based on habitat mapping an inexact
science.
Rooted in Hutchinson ’ s (1957) niche theory, a powerful suite of modelling techniques collectively known
as species distribution models, niche modelling or bioclimate envelope models have been developed and
refi ned over the past 25 years (Chapter 4 , Section
4.4.1 ; Pearson & Dawson, 2003 ). These techniques are
based on the idea that, for each environmental variable, there is an optimum value at which conditions are
most suitable for the species of interest – and that suitability declines as the value of the environmental variable increase beyond, or decreases below, the optimum.
Hutchinson (1957) conceptualized two kinds of
niches: the fundamental niche , refl ecting the underlying
physiological tolerances of a species to environmental
conditions; and the realized niche , in which the possible
range of environmental conditions in which a species
can exist is limited by biotic interactions such as predation and competition. These concepts also apply to distributions: a species ’ realized distribution often does
not fully occupy the geographical space of its fundamental (i.e. potential) distribution. This distinction is
important when modelling the distributions of species,
and different methods suit the modelling of realized, as
opposed to fundamental (potential), distributions
(Jim é nez - Valverde et al. , 2008 ), as we next discuss.
Most efforts to model species ’ realized distributions
work by spatially sampling across an environmental
gradient and determining whether the species is
present or absent at a range of sites with different environmental conditions. If one assembles several relevant environmental variables, each geographical
location at which a species has been found has an
equivalent position in environmental space. These
inductive models predict species ’ distributions using a
combination of presence/absence or presence - only
locality records (in their raw form or generalized on a
grid), data on spatial variation in environmental variables, and one or more modelling techniques (e.g.
logistic regression, GLM (generalized linear models),
GARP (genetic algorithm for rule - set prediction), or
Maxent : Guisan & Zimmermann, 2000 ; Stockwell,
2007 ; Phillips & Dud í k, 2008 ).
Many such models have employed small numbers of
original locality records (Hernandez et al. , 2006 ;
Pearson et al. , 2007 ) and some have attempted to use
coarse - grained occurrence data to predict distributions
at fi ner resolutions (Collingham et al. , 2000 ; Ara ú jo et
al. , 2005a ; McPherson et al. , 2006 ). Methods based on
habitat suitabilities and statistical modelling have been
developed to predict the potential distribution of a
species within the limits of its occurrence. They tend to
work well when historical factors, dispersal limitation
and sampling biases are not dominant processes,
although some of these effects can be reduced by
careful evaluation of alternative models and by consideration of how the results will be applied (Loiselle et al. ,
2003 ).
Species distribution models can also be based on the
fundamental, rather than realized, niche. Here the
focus is on predicting the locations of physiologically
suitable conditions in which a species is expected to
persist (Porter et al. , 2002 ; Kearney & Porter, 2004 ;
Morin et al. , 2007 ). Such models use physical principles
and physiological data on tolerances to micro climatic
conditions to predict the occurrence pattern of a
species, and can provide surprisingly good predictions
of distributions when constrained to regions in which
the species are known to occur and when suffi cient
physiological detail is known (Gaston & Fuller, 2009 ).
For example, Kearney and Porter (2004) modelled the
fundamental niche of the nocturnal lizard Heteronotia
binoei across the whole of Australia through combining
physiological measurements (thermal requirements
for egg development, thermal preferences and tolerances, metabolic and evaporative water loss rates) and
high - resolution climatic data (air temperature, cloud
cover, wind speed, humidity and radiation) with biophysical models. This methodology allowed them to
calculate the climatic component of the fundamental
niche of the lizard and map it onto the Australian landscape at high resolution.
In general, species distribution models are particularly useful for conservation in situations where data
on a species ’ distribution are patchy and sparse, exemplifi ed by the analysis of the South African protea data
(Grantham et al. , 2009 ). Most information on the distribution of species under - represents certain areas,
often in places where species richness is highest, such
179
complications, many species have highly specifi c
microhabitat requirements that are diffi cult to capture
using broad - scale or remotely sensed habitat data.
Because many species ’ distributions are limited by features other than broad habitat type, estimates of suitable habitat are often masked by intersecting some
broad representation of the known geographical range
of the species of interest, so as not to generate predictions beyond the bounds of known or likely occurrence
(Rondinini et al. , 2005 ; Harris & Pimm, 2008 ). This
makes prediction based on habitat mapping an inexact
science.
Rooted in Hutchinson ’ s (1957) niche theory, a powerful suite of modelling techniques collectively known
as species distribution models, niche modelling or bioclimate envelope models have been developed and
refi ned over the past 25 years (Chapter 4 , Section
4.4.1 ; Pearson & Dawson, 2003 ). These techniques are
based on the idea that, for each environmental variable, there is an optimum value at which conditions are
most suitable for the species of interest – and that suitability declines as the value of the environmental variable increase beyond, or decreases below, the optimum.
Hutchinson (1957) conceptualized two kinds of
niches: the fundamental niche , refl ecting the underlying
physiological tolerances of a species to environmental
conditions; and the realized niche , in which the possible
range of environmental conditions in which a species
can exist is limited by biotic interactions such as predation and competition. These concepts also apply to distributions: a species ’ realized distribution often does
not fully occupy the geographical space of its fundamental (i.e. potential) distribution. This distinction is
important when modelling the distributions of species,
and different methods suit the modelling of realized, as
opposed to fundamental (potential), distributions
(Jim é nez - Valverde et al. , 2008 ), as we next discuss.
Most efforts to model species ’ realized distributions
work by spatially sampling across an environmental
gradient and determining whether the species is
present or absent at a range of sites with different environmental conditions. If one assembles several relevant environmental variables, each geographical
location at which a species has been found has an
equivalent position in environmental space. These
inductive models predict species ’ distributions using a
combination of presence/absence or presence - only
locality records (in their raw form or generalized on a
grid), data on spatial variation in environmental variables, and one or more modelling techniques (e.g.
logistic regression, GLM (generalized linear models),
GARP (genetic algorithm for rule - set prediction), or
Maxent : Guisan & Zimmermann, 2000 ; Stockwell,
2007 ; Phillips & Dud í k, 2008 ).
Many such models have employed small numbers of
original locality records (Hernandez et al. , 2006 ;
Pearson et al. , 2007 ) and some have attempted to use
coarse - grained occurrence data to predict distributions
at fi ner resolutions (Collingham et al. , 2000 ; Ara ú jo et
al. , 2005a ; McPherson et al. , 2006 ). Methods based on
habitat suitabilities and statistical modelling have been
developed to predict the potential distribution of a
species within the limits of its occurrence. They tend to
work well when historical factors, dispersal limitation
and sampling biases are not dominant processes,
although some of these effects can be reduced by
careful evaluation of alternative models and by consideration of how the results will be applied (Loiselle et al. ,
2003 ).
Species distribution models can also be based on the
fundamental, rather than realized, niche. Here the
focus is on predicting the locations of physiologically
suitable conditions in which a species is expected to
persist (Porter et al. , 2002 ; Kearney & Porter, 2004 ;
Morin et al. , 2007 ). Such models use physical principles
and physiological data on tolerances to micro climatic
conditions to predict the occurrence pattern of a
species, and can provide surprisingly good predictions
of distributions when constrained to regions in which
the species are known to occur and when suffi cient
physiological detail is known (Gaston & Fuller, 2009 ).
For example, Kearney and Porter (2004) modelled the
fundamental niche of the nocturnal lizard Heteronotia
binoei across the whole of Australia through combining
physiological measurements (thermal requirements
for egg development, thermal preferences and tolerances, metabolic and evaporative water loss rates) and
high - resolution climatic data (air temperature, cloud
cover, wind speed, humidity and radiation) with biophysical models. This methodology allowed them to
calculate the climatic component of the fundamental
niche of the lizard and map it onto the Australian landscape at high resolution.
In general, species distribution models are particularly useful for conservation in situations where data
on a species ’ distribution are patchy and sparse, exemplifi ed by the analysis of the South African protea data
(Grantham et al. , 2009 ). Most information on the distribution of species under - represents certain areas,
often in places where species richness is highest, such