J. theor. Biol. (2002) 218, 419­433
doi:10.1006/yjtbi.3089, available online at http://www.idealibrary.com on
Long-Term Persistence of Species and the SLOSS Problem
Otso Ovaskainenn
w
wMetapopulation Research Group, Department of Ecology and Systematics, P.O. Box 65,
Viikinkaari 1, FIN-00014 University of Helsinki, Finland
(Received on 15 January 2002, Accepted in revised form on 21 May 2002)
The single large or several small (SLOSS) problem has been addressed in a large number of
empirical and theoretical studies, but no coherent conclusion has yet been reached. Here I
study the SLOSS problem in the context of metapopulation dynamics. I assume that there is
a fixed total amount A0 of habitat available, and I derive formulas for the optimal number n
and area A of habitat patches, where n  A0=A: I consider optimality in two ways. First, I
attempt to maximize the time to metapopulation extinction, which is a relevant measure for
metapopulation viability for rare and threatened species. Second, I attempt to maximize the
metapopulation capacity of the habitat patch network, which corresponds both with
maximizing the distance to the deterministic extinction threshold and with maximizing the
fraction of occupied patches. I show that in the typical case, a small number of large patches
maximizes the metapopulation capacity, while an intermediate number of habitat patches
maximizes the time to extinction. The main conclusion stemming from the analysis is that the
optimal number of patches is largely affected by the relationship between habitat patch area
and rates of immigration, emigration and local extinction. Here this relationship is
summarized by a single factor z; termed the patch area scaling factor.
r 2002 Elsevier Science Ltd. All rights reserved.
Introduction
The acronym SLOSS stands for the phrase
single large or several small, referring to the
question of whether it would be beneficial for the
conservation of species to divide a given total
amount of habitat into one large or several small
habitat patches. The SLOSS problem has been
addressed in a large number of empirical and
theoretical studies. However, the objectives,
methods and interpretations of these studies
have been so distinctive that no coherent
conclusion has yet appeared.
Throughout this paper, I will denote by SL
and SS the ``single large'' and ``several small''
strategies, respectively. More precisely, SL
stands for a strategy where a few (if not just
one) large patches are preferred, and SS for a
strategy in which a large number of small
patches is preferred. I will denote by IN the
intermediate strategy, where a compromise
between the SL and SS ends of the continuum
is optimal.
I will address the SLOSS problem in the
context of single-species metapopulation dynamics,
assuming that the long-term persistence
of the species is based on recolonizations of
empty habitat patches by migrants from extant
populations. I will seek for the optimal number
of habitat patches with two specific objectives in
mind. First, I attempt to maximize the time to
metapopulation extinction, which is an adequate
n
E-mail address: otso.ovaskainen@helsinki.fi
(O. Ovaskainen).
0022-5193/02/$35.00/0 r 2002 Elsevier Science Ltd. All rights reserved.
objective for the conservation of rare species.
Second, I attempt to maximize the metapopulation
capacity of the habitat patch network.
As shown by Hanski & Ovaskainen (2000),
metapopulation capacity sets the deterministic
threshold condition for persistence, and thus
maximizing metapopulation capacity is equivalent
with minimizing the risk of deterministic
extinction. In addition, maximizing metapopulation
capacity corresponds with maximizing the
average fraction of occupied patches (Hanski &
Ovaskainen, 2000), and it relates thus to maximizing
the abundance of the species.
I start by discussing the existing SLOSS
literature according to the objective assigned
for the SLOSS problem. A sample of the
literature is summarized in Table 1.
1. Maximize the number of species that will
eventually survive: From the conservational point
Table 1
Sample of the SLOSS literature according to the objective assigned for the SLOSS problem: (1)
maximize the number of species that will eventually survive, (2) maximize the number of currently
occurring species, (3) maximize time to extinction, (4) maximize metapopulation capacity.
Type of study Objective Conclusion Reference
Biogeographical theory, birds in islands 1 SL Diamond (1975)
Biogeographical theory 1 SL May (1975)
Biogeographical theory 1 SL Wilson & Willis (1975)
Biogeographical theory 1 SL Diamond & May (1976)
Arthropods in experimental islands 2 SS Simberloff & Abele (1976)
Higher plants in quarry reserves 2 SS Higgs & Usher (1980)
Birds in islands 2 IN Gilpin & Diamond (1980)
Literature study 2 SS Simberloff & Abele (1982)
Vascular plants in deciduous woods 2 SS Ja¨ rvinen (1982)
Mammals in islands 2 SL Patterson & Atmar (1986)
Birds in islands 2 SL Patterson (1987)
32 data sets of different taxa 2 SS Quinn & Harrison (1988)
Vascular plants in islands 2 SL Deshaye & Morriset (1989)
Vascular plants in woodlots 2 SS Dzwonko & Loster (1989)
Vascular plants in woodlots 2 SS Zacharias & Brandes (1990)
Birds in woodlots 2 SL Blake (1991)
27 data sets of different taxa 2 SLOSS Wright & Reeves (1992)
Vascular plants in deciduous woods 2 SS Stersdal (1994)
10 data sets of different taxa 2 IN Lomolino (1994)
Butterflies in forest fragments 2 SS Baz & Garcia-Boyero (1996)
Small mammals in mountaintop ``islands'' 2 SS Skaggs & Boecklen (1996)
148 data sets of different taxa 2 SS Boecklen (1997)
Vascular plants in mires 2 SS Virolainen et al. (1998)
Vascular plants in woodlots 2 SS Honney et al. (1999)
Analytical model 3 SL Wright & Hubbell (1983)
Analytical model 3 SL or IN Quinn & Hastings (1987)
Stochastic simulation model 3 SL Burkey (1989)
Experiment with Drosophila flies 3 SL Forney & Gilpin (1989)
Analytical model and five data sets 3 SL Burkey (1995)
Experiment with bacteria and protozoa 3 SL Burkey (1997)
Stochastic simulation model 3 SLOSS Pelletier (2000)
Analytical model 3 SL Etienne & Heesterbeek (2000)
Analytical model 3 IN Present study
Analytical model 4 IN Etienne & Heesterbeek (2000)
Analytical model 4 SL Present study
n
The abbreviations SL and SS stand for the ``single large'' and ``several small'' strategies, respectively, and the
abbreviation IN stands for an intermediate strategy.
O. OVASKAINEN420
of view, the most natural definition of the
SLOSS problem is to ask which kind of network
would ensure the long-term persistence of the
largest number of species. In spite of its
fundamental importance, this question has been
addressed almost solely in the beginning of the
SLOSS debate. In their seminal works, Diamond
(1975), May (1975) and Wilson & Willis (1975)
approached this question through the theory of
island biogeography (MacArthur & Wilson,
1967), the species­area curve and exponentially
distributed relaxation times. The authors gave
support for the SL strategy, reasoning that many
species that would have a good chance of
surviving in a single large reserve would have
their survival chances reduced if the same area
were apportioned among several smaller reserves.
These studies did not account for
metapopulation dynamics, i.e. that a network
of habitat patches may buffer the species against
extinction by allowing for recolonizations of
empty habitat patches by migrants from extant
populations.
Simberloff & Abele (1976) conducted an
experimental study in a mangrove island archipelago,
and found out that an SS network
contained more species than an SL network,
and questioned the earlier results. Terborgh
(1976) noted that the objectives of these two
studies had been different: in the experiment of
Simberloff & Abele (1976), there was a mainland
acting as a continuous source of migrants,
whereas the studies based on the theory of island
biogeography had assumed that all habitat
outside of the nature protection network would
be lost. The debate continued, but the original
question was almost forgotten. Only recently,
the interest has returned to the question of
maximizing the number of species that will
eventually survive (Cabeza & Moilanen, 2001).
2. Maximize the number of currently occurring
species: Like the pioneering study of Simberloff
& Abele (1976), most empirical SLOSS studies
have compared whether an SL or an SS habitat
patch network currently contains more species,
not considering whether the species will survive
or not. The majority of these studies has
concluded that SS is the better strategy, but
there are also exceptions (Table 1). The most
coherent conclusion from these studies is that
the result depends on the level of nestedness
(Simberloff & Mertin, 1991; Bolger et al., 1991;
Wright & Reeves, 1992). A biota is said to
exhibit a nested pattern if species compositions
from species poor samples are largely subsets of
those from species rich samples (Patterson,
1987). If the biota is highly nested, small habitat
patches typically contain species that are found
as a subset of species present in a large habitat
patch, and consequently SL is found to be the
better strategy. However, if the biota is not
highly nested, SS tends to maximize the number
of species.
Though comparing whether an SL or an SS
network contains more species has an interest in
itself, it has been criticized that the numerous
studies on this topic do not really contribute to
the design of nature conservation networks. Cole
(1981) argued that many of the empirical studies
finding the SS strategy to be better were
influenced by a mainland acting as a source. If
the mainland would not be protected, many of
the species in the small habitat patches would be
lost. It has also been questioned whether it is
simply the species number that matters, or
whether one should weigh the species in one
way or another. For example, in a study of
vascular plants in fragments of old-growth
forests, Fukamachi et al. (1996) found that the
maximum number of species was achieved by the
SS strategy, but most of the rarest species were
found solely in the largest forest patches.
3. Maximize time to extinction: Much of the
theoretical SLOSS literature has focused on
maximizing the persistence time of a single
species. The models have been typically based
on some specific type of local dynamics connected
by some specific type of metapopulation
dynamics. Most of these studies have concluded
that the SL strategy is optimal (Table 1). It is,
however, important to note that this conclusion
depends critically on the underlying causes of
local extinctions (see, e.g. Quinn & Hastings,
1987; Etienne & Heesterbeek, 2000). Let me
illustrate by considering two extreme cases.
First, if local extinctions are caused solely by
demographic stochasticity, the lifetime of a local
population scales exponentially with the number
of individuals (Goel & Richter-Dyn, 1974) and
thus (assuming constant density) exponentially
PERSISTANCE OF SPECIES AND THE SLOSS PROBLEM 421
with the size of the habitat patch. Almost
independent of the type of metapopulation
dynamics, this assumption will favour the SL
strategy. The other extreme is to assume that
local extinctions are caused solely by catastrophic
events, which always kill the local
population independent of its size. This assumption
is likely to lead to the SS strategy, as now
the risk of extinction should be spread among a
large set of small populations. Another important
factor is the type of metapopulation
dynamics assumed. If there is no migration
between habitat patches, recolonizations are
not possible, and practically all that matters
are the lifetimes of local populations, favouring
the SL strategy. The only factor that could
favour the SS strategy is that a large number of
patches provides a large number of independent
populations, and the metapopulation is extinct
only after all of the patches have become extinct.
However, this advantage is likely to be negligible
as compared to the advantages of the SL
strategy. In contrast, if migration is included,
the possibility of recolonizations may change the
result completely, now in favour of the SS
strategy.
For example, Quinn & Hastings (1987)
assumed no migration between habitat patches.
They found that if extinctions are caused by
demographic stochasticity, SL is a better strategy
once local populations reach a small threshold
size. If local extinctions were caused by
environmental stochasticity, Quinn & Hastings
(1987) found that the time to extinction is
maximized when the number of habitat patches
equals the square root of the number of
individuals potentially inhabiting them, suggesting
an IN strategy. Burkey (1989) and Wright &
Hubbell (1983) modelled migration between
habitat patches, but assumed solely demographic
stochasticity, concluding that SL is the optimal
strategy.
An exception to the general pattern was given
by Pelletier (2000), who concluded that SS is the
optimal strategy if migration is not allowed for,
even if local population dynamics are driven
solely by demographic stochasticity. Pelletier
(2000) derived his conclusions from a stochastic
simulation model, but as he did not specify the
details of the model, it is not possible to assess
the reason for the unexpected result. Pelletier
(2000) found that SL is the optimal strategy if
migration with high mortality is assumed. This
is to be expected, as in this case the loss of
emigrating individuals is not compensated by
immigration, and a large patch is expected to
have a low per capita emigration rate. Assuming
a lower mortality level, he concluded that neither
of the SL nor the SS strategies is optimal, but the
species persistence is maximized by a self-similar
distribution of patch sizes.
Etienne & Heesterbeek (2000) based their
analysis on the plausible assumption that, for a
fair comparison, SL, IN and SS networks should
have the same density of habitat. As Etienne &
Heesterbeek (2000) restricted their analysis to
nearest-neighbour dispersal, this assumption entails
that larger patches have larger interpatch
distances. Assuming further that dispersal
success decreases exponentially with distance,
Etienne & Heesterbeek (2000) concluded that
close to the SL end of the continuum, the time to
metapopulation extinction is roughly given by
the lifetime of a local population, in which case
the SL strategy is optimal. However, the verbal
argument of Etienne & Heesterbeek (2000)
essentially ignores the possibility that metapopulation
level dynamics may compensate what is
lost at the level of local population dynamics.
There appear to be no field studies that would
have addressed the persistence time of a single
species in the SLOSS context. In experimental
laboratory studies, both Forney & Gilpin (1989)
and Burkey (1997) found that the SL strategy is
optimal for maximizing the time to extinction.
This is consistent with the theoretical studies in
the sense that the effect of environmental
stochasticity is likely to be low in laboratory
studies.
4. Maximize metapopulation capacity: Hanski
& Ovaskainen (2000) considered a deterministic
metapopulation model with variation in patch
areas and connectivities. They concluded that
metapopulation persistence is possible only if
the metapopulation capacity of the landscape is
greater than the ratio of extinction and colonization
rate parameters, that is, if lM4d holds,
where lM is the metapopulation capacity of the
landscape and d  e=c is a species parameter
determined by the extinction and colonization
O. OVASKAINEN422
rate parameters e and c; respectively. Metapopulation
capacity is a landscape index, which
characterizes the ability of a habitat patch
network to support a viable metapopulation of
the focal species. The metapopulation capacity
of a particular landscape depends on the spatial
scale at which the focal species migrates, but is
independent of the extinction and colonization
rate parameters of the species. As a first
approximation, it may be considered to measure
the amount of habitat available, but in addition
it takes into account the actual spatial configuration
of the habitat patch network. Metapopulation
capacity increases with increasing
number of habitat patches, with increasing sizes
of habitat patches, and with decreasing interpatch
distances. A related and widely used
concept in epidemiology is the basic reproductive
ratio R0; which has been called the
``colonization potential'' in metapopulation theory
(Etienne & Heesterbeek, 2000), and is given
by R0  lM=d: R0 gives the expected number of
colonizations that a single occupied habitat
patch will generate during its lifetime in an
initially empty network, the threshold condition
for persistence being given by R041:
Maximizing the metapopulation capacity lM
is equivalent to maximizing the species abundance
(in the sense of a weighted fraction of
occupied patches pn
at equilibrium), as the two
quantities are related as pn
 1  d=lM (Hanski
& Ovaskainen, 2000; Ovaskainen et al., 2002).
As will be shown below, maximizing the
metapopulation capacity lM (or equivalently
R0) differs from maximizing the time to extinction
in the sense that the deterministic approach
excludes the possibility of extinction due to
extinction­colonization stochasticity in the
patch occupancy level. The only study that has
so far addressed the SLOSS problem with the
objective of maximizing metapopulation capacity
concluded that an IN strategy would be
optimal (Etienne & Heesterbeek, 2000). This
conclusion depends strongly on the assumptions
that dispersal is restricted to nearest neighbours,
and that dispersal success decreases exponentially
with distance.
The above classification of the objectives
assigned to the SLOSS problem is by no means
exhaustive. For example, a possible objective
could be to maintain genetic variation, both over
the entire species range, and, if inbreeding
depression is important, within local populations.
However, it should be clear already by the
above discussion that the SLOSS debate is far
from being solved, but has rather become
increasingly confused. Table 1 illustrates that
practically any possible combination of reasoning
and conclusion has been used during the 25
years of the SLOSS debate. It is no wonder that
Soule´ & Simberloff (1986) suggested that the
SLOSS issue should be buried as being not
useful in designing nature reserves. Their idea
was that decisions on reserve area should be
based on the minimum area needed to sustain a
viable population, habitat diversity, and autoecological
data. Worthen (1996) also emphasized
the need for species- and site-specific information,
arguing that ``no single index should be
expected to distill the informational content of
an entire community, let alone predict how it will
react to habitat reduction or fragmentation''.
In spite of the above pessimism, with which I
largely agree, I will use here a single-species
metapopulation model to address the SLOSS
problem. The main reason for this exercise is not
to provide an extra combination of reasoning
and conclusion, but rather to draw attention on
some key factors that are relevant to the SLOSS
framework. My main conclusion is that the
optimal solution to the SLOSS problem is
determined by the patch area scaling factor z;
which integrates the effect of patch area on
metapopulation dynamics. Somewhat less importantly,
my ``solution'' to the SLOSS problem
will be that the IN strategy is optimal for
maximizing the time to extinction, while the SL
strategy is optimal for maximizing the metapopulation
capacity.
In the metapopulation context, the SLOSS
problem is basically an attempt to find the
optimal compromise between the advantages of
the SL and SS ends of the continuum. The
advantages of the SL end are found at the local
scale. The survival chance of a local population,
and the number of individuals emigrating out
and immigrating in are typically expected to
increase with increasing patch size. The two
former effects are mostly due to increasing
population size with patch area. The advantages
PERSISTANCE OF SPECIES AND THE SLOSS PROBLEM 423
of the SS end are due to the fact that ``all eggs
are not in the same basket'', the extinction risk
being spread among a larger set of patches. This
advantage is especially strong if local dynamics
are assumed to be independent of each other, as
in this case the extinction time grows exponentially
with the number of habitat patches
(Andersson & Djehiche, 1998; Ovaskainen,
2001). Although I will follow this assumption,
I note that this is not always the case, as
for instance, regional stochasticity may result
in a spatially correlated extinction risk, which
may greatly increase the risk of metapopulation
extinction (Hanski, 1999; Ovaskainen,
2002).
The present work stems from the analysis of
Hanski (1996), who assumed that the extinction
risk E of a local population depends on patch
area A as E  e=Azex ; where e is a parameter and
zex describes how extinction risk depends on
patch area. Hanski (1996) measured the performance
of a metapopulation by the average
fraction of occupied patches, and used stochastic
simulations to illustrate that the optimal size of
habitat patches increases with increasing scaling
parameter zex: However, it is not only the
extinction risk that scales with patch area, but
also the processes of immigration and emigration
(for empirical examples, see e.g. Moilanen
et al., 1998; Hanski et al., 2000; Ovaskainen,
2002). I will extend the result of Hanski (1996)
by showing that the optimal size of habitat
patches depends on the patch area scaling factor
z  zim  zem  zex; where the three terms indicate
how the key metapopulation processes,
namely immigration, emigration, and extinction
of local populations, scale with patch area.
I will start by formulating a metapopulation
model (the spatially realistic Levins model),
which I will then use to address the SLOSS
problem. Finally, I will attempt to estimate the
realistic range of the patch area scaling factor z
from the available data.
The Model
I assume that there is a fixed total amount A0
of habitat available, and I attempt to derive a
formula for the optimal number n and area A
of habitat patches, where n  A0=A: I consider
``optimality'' in two ways: in the sense of
maximal metapopulation capacity and in the
sense of maximal time to metapopulation
extinction.
Although I concentrate on a spatially homogeneous
case, meaning that all patches are
assumed to be identical and equally connected
to each other, it is instructive to start with a
spatially structured model. The reason for this is
two-fold. First, it is not only the patch sizes that
matter in the SLOSS context, but also the
interpatch distances, and the latter cannot be
addressed without a spatial settings. Second, real
patch networks are not homogeneous, but
typically show great variations in patch size
and connectivity. Although I restrict my analysis
to the homogeneous case, the results may be
generalized to heterogeneous patch networks
(see below).
I will consider the spatially realistic Levins
model (Hanski & Ovaskainen, 2000; Ovaskainen
& Hanski, 2001), where the phrase ``spatially
realistic'' means that the model accounts for
patch areas and locations, but ignores any
further details of the spatial configuration of
the habitat patch network. The model belongs to
the class of patch occupancy models, in which
patches are classified simply as empty or
occupied. Instead of explicitly modelling local
population dynamics, the extinction and colonization
rates are directly connected to the spatial
structure of the habitat patch network. The
extinction rate Ei of an occupied patch i is
assumed to be Ei  e=Azex
i ; where e is a speciesspecific
extinction rate parameter, Ai is the area
of patch i; and zexX0 describes how patch area
affects the risk of extinction. The reasoning
behind this assumption is that the expected
population size is larger in a larger patch, and
that the risk of extinction generally decreases
with population size. The colonization rate Ci of
an empty patch i is given by Ci 
P
jai cijoj;
where cij is the contribution that an occupied
patch j makes to the colonization rate of the
empty patch i; and oj  1 if patch j is occupied
and oj  0 if it is empty. I assume that cij 
cAzim
i Azem
j f dij; where c is a colonization rate
parameter, and zimX0 and zemX0 describe how
patch area affects immigration and emigration,
respectively. The function f describes the
O. OVASKAINEN424
dispersal kernel, i.e. the effect of the interpatch
distance dij on migration success. The reasoning
behind the assumptions made on the colonization
rate are that only occupied patches may
contribute to the colonization of an empty
patch, that large occupied patches are expected
to send more emigrants, and that large empty
patches are expected to attract more immigrants.
Although I assumed above that all the three
scaling factors are positive, it might be possible
that some of them are negative. For example, the
per capita emigration rate may be greatly larger
in small patches, which may lead to zemo0 if the
expected population size grows sufficiently
slowly with patch size. The assumption that the
scaling factors are positive is not needed in the
analysis, and the results to be given thus hold
also for negative scaling factors.
The model may be considered either in a
stochastic or in a deterministic setting. The
stochastic version of the model is defined as a
Markov process determined by the above
extinction and colonization rates, the state space
consisting of all the 2n
possible combinations of
empty and occupied patches. The deterministic
model is defined as the mean-field approximation
of the stochastic model, modelling the rate
of change in the probability for patch i being
occupied as
dpi
dt

X
jai
cijpj
!
1  pi  Eipi: 1
As shown by Ovaskainen & Hanski (2001), the
deterministic model predicts that long-term
persistence is possible only if the condition
lM4d is valid. Here lM is the metapopulation
capacity of the landscape, which measures the
capacity of the habitat patch network to support
the persistence of the species. Mathematically,
lM is given by the leading eigenvalue of the n  n
nonnegative matrix M; the elements of which are
defined by mij  Azexzim
i Azem
j f dij for iaj and
mii  0. Here the factor Azex
i is proportional to
the expected lifetime of the local population in
patch i and Azim
i Azem
j f dij is proportional to the
rate at which immigrants originating from patch
j succeed to colonize patch i: The threshold
condition for persistence is determined by the
species parameter d; which is defined as the ratio
of the extinction and colonization rate parameters,
d  e=c:
In the rest of the paper, I will restrict the
analysis to the homogeneous version of the
model. In this case, the extinction rate of an
occupied patch reduces to E  e=Azex and the
colonization rate of an empty patch reduces to
C  ckAzimzem f d; where k is the number of
occupied patches. In the homogeneous model,
all the n patches are assumed to be at the same
distance d from each other. As this assumption
is impossible already from the geometric point
of view, it is important to make the reasoning
behind the assumption precise. As specified
above, I do not assume that dispersal is
restricted to nearest neighbours only, but I
assume that dispersal success reduces gradually
with increasing distance. With this assumption, d
should be interpreted as the effective dispersal
distance between the habitat patches, characterizing
typical migration distances within the patch
network. Etienne & Heesterbeek (2000) assumed
that the fraction of habitat is independent of the
size of habitat patches, meaning that the n
patches are located within a landscape of fixed
size. Restricting dispersal to nearest neighbours,
they argued that the interpatch distance d should
increase with increasing patch area. Contrary to
the study of Etienne & Heesterbeek (2000), I
make here the assumption that the effective
distance d is independent of the number of
habitat patches. Figure 1 demonstrates that, in
the present case, this assumption is roughly
consistent with the assumption that the n patches
are located into a total area of fixed size,
illustrating an important distinction between
gradually decreasing dispersal success in the
present model and the nearest-neighbour dispersal
in the Etienne & Heesterbeek (2000)
model.
Assuming homogeneity greatly simplifies the
analysis of the model, but also restricts the
applicability of the results. However, as mentioned
above, the results to be given here
generalize to heterogeneous habitat patch networks.
The reason for this is that metapopulation
dynamics in a heterogeneous patch network
can be approximated by metapopulation dynamics
in a homogeneous patch network with
PERSISTANCE OF SPECIES AND THE SLOSS PROBLEM 425
appropriately transformed parameter values
(Ovaskainen & Hanski, 2001, 2002; Ovaskainen,
2002).
SLOSS
For a homogeneous habitat patch network,
the metapopulation capacity is given by
lM  n  1Az
f d; 2
where the patch area scaling factor z  zim 
zem  zex measures how strongly patch area
affects metapopulation dynamics (Ovaskainen
& Hanski, 2001). Approximating n  1 by n and
substituting n  A0=A in eqn (2), I obtain
lMEAz1
A0 f d  n1z
Az
0 f d: 3
Equation (3) suggests that if z41; SL is the
optimal strategy for maximizing lM; whereas
zo1 would lead to the SS strategy. This is to be
expected, as z41 zo1 indicates that the
metapopulation capacity of a patch network
grows with patch area faster (slower) than
linearly. Equation (3) depends on the approximation
n  1=nE1; which is not expected to
hold for very small n; as illustrated by the fact
that n  1 leads to lM  0: Equation (3) thus
actually suggests that for z41 the metapopulation
capacity is not maximized by a single patch,
but rather by a small number of habitat patches.
This is also evident in Fig. 2, where the value of
lM attains its maximum either at n  2 or at
n  3:
It is worth emphasizing here that the patch
area scaling factor z integrates the effect of patch
area on metapopulation dynamics. It is convenient
that metapopulation capacity does not
depend separately on the individual components
zim; zem and zex; so that it suffices just to consider
the integrated patch area scaling factor z:
As the time to extinction is exponentially
distributed, we will lose no generality by
attempting to maximize the expected time to
extinction. The expected time to metapopulation
extinction (using the lifetime of a local population
as a time unit) behaves for lM4d as
(Andersson & Djehiche, 1998; Ovaskainen,
2001)
T 
ffiffiffiffiffiffi
2p
n
r
en #d1log #d

1  #d
1  O1=n; 4
where #d  d=lM is the scaled species parameter.
As the exponential term dominates already with
small n; I will simplify the analysis by attempting
to maximize
gn; #d  n #d  1  log #d: 5
If zo1; #d  1  log #d is maximized as n-N;
and thus the SS strategy is optimal. The
interesting case occurs if z41; as in this case
there is a trade-off between maximizing the
factors n and #d  1  log #d in eqn (5). For a
given species (i.e. a given species parameter d)
and fixed amount of total habitat, n behaves [by
eqn (3)] as #d
1=z1
: Thus, we should maximize
h #d   #d  1  log #d #d
1=z1
: Setting h0
 #d  0; it
turns out that the maximum occurs when  #d 
1z  log #d: This equation cannot be solved
analytically, but #dzE1=z2
gives a reasonable
approximation (Fig. 3). By eqn (3), the optimal
Number of patches
Effectivedistance
10 20 30 40 50 60
0.2
0.4
0.6
0.8
1
1.2
1.4
Fig. 1. Dependence of the effective interpatch distance
on the number of habitat patches. An N Â N patch
network (with N  2; y; 8) of n  N2
patches was located
uniformly in the unit square. The metapopulation capacity
of the habitat patch network was calculated with the
spatially realistic Levins model, where the dispersal kernel
was defined as f d  ead
: The dots depict the effective
interpatch distance, defined as the distance which gives the
same metapopulation capacity for a homogeneous model
with equal distances among patches. The parameter a
attains the values a  1; 2; 4; corresponding to the dots
from top to bottom. The result is independent of the
parameter z:
O. OVASKAINEN426
size and number of habitat patches are given by
AE
d
#dzA0f d
 1=z1
;
nE
Az
0f d #dz
d
" #1=z1
: 6
Based on eqn (6), I conclude that for z41:
1. The optimal size of habitat patches increases
(and thus their number decreases) with
increasing patch area scaling factor z (Fig. 2,
panels a and b). This is intuitive, as z describes
the strength of area dependence in metapopulation
dynamics. However, now the result is not as
extreme as with maximizing the metapopulation
capacity, where z41 led to the SL strategy.
When maximizing the time to metapopulation
extinction, the number of habitat patches should
always be kept reasonably large to gain the
advantages of risk spreading, leading to an IN
strategy. The expected time to metapopulation
extinction is more sensitive to the number of
Number of patches
10 20 30 40 50 60
1
2
3
4
10 20 30 40 50 60
50
100
150
200
10 20 30 40 50 60
1
2
3
4
5
6
10 20 30 40 50 60
5
10
15
20
25
10 20 30 40 50 60
1
2
3
4
10 20 30 40 50 60
2
4
6
8
10
12
14
10 20 30 40 50 60
0.5
1
1.5
2
2.5
3
10 20 30 40 50 60
2.5
5
7.5
10
12.5
15
17.5
TimetoextinctionMetapopulationcapacity
Number of patches
Number of patches
Number of patches
MetapopulationcapacityTimetoextinction
Number of patches
Number of patches
MetapopulationcapacityTimetoextinction
Number of patches
Number of patches
TimetoextinctionMetapopulationcapacity
(a) (b)
(c) (d)
Fig. 2. Dependence of the SLOSS problem on the parameters z; d and A0: Each panel shows how the metapopulation
capacity (lower figure) and the time to extinction (upper figure) depend on the number of habitat patches. Panel a is a
reference case with parameters z  1:5; d  1; A0  5 and f d  1: In each of the panels b­d, one of the parameters is
changed while all the other parameters are held fixed. The changed parameters are z  2:0 (panel b), d  1:5 (panel c) and
A0  4 (panel d). In each panel, the horizontal line corresponds to #d  1=z2
; the vertical line demonstrating that this roughly
coincides with the maximal mean time to extinction.
PERSISTANCE OF SPECIES AND THE SLOSS PROBLEM 427
habitat patches if z is large, as demonstrated by
the more narrow peak in the time to extinction
curve shown in Fig. 2(b) as compared to that of
Fig. 2(a).
2. The optimal size of habitat patches increases
with increasing d (Fig. 2, panels a and c).
This may be interpreted as follows. Consider two
species, labelled A and B; which are otherwise
identical (same f and z), but species A has a
smaller value of d; either because it is a better
colonizer than B (greater c) or because it persists
better as local populations (smaller e) than
species B: Thus in the same environment, the
abundance of species A is greater than that of
species B: With a given amount of habitat
available, it would be beneficial to divide the
habitat to a larger number of habitat patches for
species A than for species B: This makes sense, as
we note from eqn (4) that as soon as the scaled
species parameter #d is relatively small, the
expected time to extinction starts to grow
exponentially with the number of habitat
patches.
3. The optimal size of habitat patches decreases
with increasing A0 (Fig. 2, panels a and
d). This may seem counterintuitive, as one might
expect that with a large amount of habitat
available, it would be optimal to have a reasonable
number of large habitat patches. However,
as noted before, eqn (4) predicts exponential
growth in time to extinction with the number of
habitat patches. Although the metapopulation
capacity lM decreases with decreasing A; lM is
still reasonably large for small A if the total
amount of habitat is very large. This implies that
the scaled species parameter #d is reasonably
small, in which case it is the number of habitat
patches that really matters. On the contrary, if
the total amount of habitat is small, persistence
as a true metapopulation is not likely, in which
case it is better to have a few patches that are as
large as possible in order to enhance the survival
of local populations.
Note that the above analysis is based on the
assumption that the metapopulation is above the
deterministic threshold, as eqn (4) is based on
the assumption lM4d: If this is not the case, the
metapopulation is expected to go deterministically
extinct. As an example, assume that a rapid
episode of habitat loss has deteriorated a landscape
below the extinction threshold. In such a
situation, the time to extinction calculated from
the quasi-stationary distribution is not expected
to be of major importance, as the quasistationary
distribution may not be reached at
all before extinction. As occupied patches will go
extinct, but recolonizations are expected to be
rare, it is the initial patch occupancy pattern that
really matters. In such a case, large occupied
patches are of major importance, as these are
expected to delay the inevitable extinction as
much as possible.
The Patch Area Scaling Factor f
As shown by the above analysis, the optimal
solution to the SLOSS problem depends on
several factors, most importantly on the patch
area scaling factor z: A qualitative change occurs
at the critical value z  1: I will next discuss
factors that affect the value of z; considering
separately the three components zex; zim and zem:
1. Extinction: It is often difficult to state one
clear-cut reason that dominates the extinction
risk, as the risk of extinction is determined
by several mechanisms, such as demographic
stochasticity, environmental stochasticity, catastrophic
events, predator­prey or host­parasite
interactions, inbreeding depression, ephemeral
nature of the habitat, or human influence of
various types. The variation that is observed in
the value of zex may be explained by differences
1.5 2 2.5 3
0.2
0.4
0.6
0.8
1
Patch area scaling factor 
()
Fig. 3. The exact solution #dz to the equation  #d 
1z  log #d (solid line) and its approximation #dzE1=z2
(dashed line).
O. OVASKAINEN428
in the relative importance of the underlying
causes that drive populations to extinction.
Indeed, if the extinction risk is determined solely
by demographic stochasticity, the extinction
probability should decrease exponentially with
population size and thus with patch area, if
population density is assumed to be constant
(e.g. Goel & Richter-Dyn, 1974; Lande, 1993). If
the extinction risk originates solely from environmental
stochasticity or by random catastrophes,
the extinction probability should
decrease as a power of the patch area (Lande,
1993; Foley, 1997). Thus, in sufficiently large
patches, environmental stochasticity and random
catastrophes pose a greater risk of extinction
than demographic stochasticity, suggesting
that the polynomial form e=Azex used here is
adequate. In small populations, where demographic
stochasticity may be of major importance,
the polynomial scaling of the extinction
risk may be considered to approximate the true
exponential scaling if zex is sufficiently high. In
line with the above reasoning, Hanski (1992,
1998) has suggested that the value of zex may be
used as a measure describing how sensitive the
species is to environmental stochasticity, a small
value of zex indicating that environmental
stochasticity has a large effect on the extinction
risk as compared to the effect of demographic
stochasticity.
As discussed by Hanski (1992), Cook &
Hanski (1995) and Hanski (1998), the parameter
zex is among vertebrates often related to the
body size of the species; for instance, smaller
animals often suffer more from environmental
stochasticity because of their sensitivity to
temporal variation in food availability. Hanski
(1992) reported the values zex  0:46; 0:91; 2:30
for three species of shrews, Wahlberg et al.
(1996) the values zex  0:84; 0:95; 0:96; 0:98 for
four species of butterflies, and Moilanen et al.
(1998) the value zex  1:28 for the American
pika. Cook & Hanski (1995) estimated zex for 49
species of birds, the minimum, median, and
maximum values being 0.05, 0.75 and 2.10,
respectively. However, many of these studies
have ignored the effect of patch area on
immigration rate, in which case the estimates
given above may represent estimates for the sum
zex  zim; and may thus overestimate zex: Separating
the effects of zex and zim; Ovaskainen
(2002) estimated the value zex  0:32 for a
species of butterfly.
2. Emigration: The emigration parameter zem
describes how patch area affects the number
of individuals that emigrate from an occupied
habitat patch per unit time. The parameter zem is
influenced by two factors, which are the relation
between population size and patch area, and the
relation between per capita emigration rate and
patch area. Both of these are determined by a
number of factors, but it is reasonable to assume
that the former increases with patch area,
whereas the latter decreases with patch area.
For example, in a study of a butterfly metapopulation,
Hill et al. (1996) concluded that the per
capita emigration rate was significantly higher in
small patches, but that in absolute numbers
more emigrants came from the largest patches
where the source population was the largest. In a
study of a field vole archipelago metapopulation,
Crone et al. (2001) concluded that emigration
from tiny, ephemeral skerry populations was
about as important as emigration from more
persistent populations on large islands. The
pattern resulted from less stable vole densities
on smaller islands, combined with increased
emigration before local extinction. Note that
although the per capita emigration rate may
depend on the density of individuals, I assume
here that population density depends on patch
area, and thus the per capita emigration rate is a
function of patch area.
As a theoretical remark, I note that the value
zem  0:5 would follow by assuming that population
density is independent of patch size, and
that the probability for an individual to emigrate
is proportional to the ratio of boundary length
and patch area. Moilanen et al. (1998) reported
the value zem  0:74 for the American pika, and
Ovaskainen (2002) estimated the value zem 
0:28 for a species of butterfly.
3. Immigration: Much of the theoretical metapopulation
literature assumes that the immigration
rate is independent of patch area,
corresponding to zim  0: However, it is clear
that patch area may have an effect on immigration
rate. Large patches are easier to find, they
are possibly more attractive for migrants, and
the probability of successful colonization per
PERSISTANCE OF SPECIES AND THE SLOSS PROBLEM 429
attempted colonization may be greater. Assuming
that the localization of an empty patch is the
main factor affecting zim; and assuming that
dispersing individuals search for patches along
straight lines at the local scale around a habitat
patch would lead to the scaling zim  0:5 As
species exhibit a wide variety of dispersal
strategies, including the possibility of detecting
habitat patches from a distance, it is clear that
the scaling factor may vary greatly among
species. Hanski et al. (2000) reported the scaling
0.26 for a species of butterfly, but they considered
only whether the individuals were able to
locate the habitat patch, thus neglecting whether
they attempted colonization or if that attempt
was successful. Ovaskainen (2002) restricted his
analysis to succesful colonizations, and estimated
the value zim  0:44 for another species
of butterfly.
Although current empirical knowledge about
the scaling factors is admittedly very limited, the
reasoning above suggests that in a ``typical''
metapopulation, the value of z would be in the
range 1:0ozo2:0: In the context of the SLOSS
problem, I thus conclude that in a ``typical'' case,
SL is the optimal strategy to maximize the
metapopulation capacity of a habitat patch
network, whereas IN maximizes the expected
time to metapopulation extinction.
Discussion
I have addressed the SLOSS problem from the
viewpoint of classical metapopulation dynamics.
Assuming that long-term persistence of the focal
species is based on compensation of local
extinctions by recolonizations, I have shown
that the time to extinction is maximized with the
IN strategy, whereas the metapopulation capacity
of a patch network is typically maximized
with the SL strategy. The location of the
optimum point along the SL to SS axis is largely
determined by the patch area scaling factor z; the
optimum moving towards the SL end with
increasing z: The patch area scaling factor z
integrates the effects of patch area on metapopulation
dynamics, including the processes of
immigration, emigration and local extinction.
Metapopulation capacity lM sets the threshold
for persistence as lM4d; where d is a species
parameter. Below the threshold, the species is
expected to go deterministically extinct, meaning
that the drift affecting the patch occupancy state
is negative. The other quantity considered in this
study, time to metapopulation extinction, is a
stochastic quantity, with the mean value not
depending only on the drift term but also on the
variance around the mean. If the number of
habitat patches is large the variance is small and
thus the time to extinction is smaller than in an
otherwise similar situation with a small number
of habitat patches. Thereby, the time to extinction,
unlike metapopulation capacity, accounts
also for the advantages of risk spreading, leading
to a strategy towards the SS end of the SLOSS
continuum.
I have considered a homogeneous metapopulation
model, which assumes that all habitat
patches are identical and equally connected.
Most real habitat patch networks are not
homogeneous. The justification for restricting
the present study to the homogeneous case is
that the behaviour of an appropriately weighted
average of the occupancy state in a heterogenous
patch network is often well approximated by a
homogeneous model (Ovaskainen & Hanski,
2001, 2002; Ovaskainen, 2002). Hence, the
present results generalize to heterogeneous patch
networks, where the question is to optimize the
average size (and thus number) of habitat
patches. At the optimum, the scaled species
parameter #d satisfies approximately #dzE1=z2
:
Comparing the scaled species parameter (which
can in an equilibrium situation be estimated
from a snapshot of occupancy data; Hanski &
Ovaskainen, 2000) with the patch area scaling
factor gives a rough way of estimating whether
the survival of the species suffers more from the
smallness or from the fewness of the habitat
patches. If #dzo1=z2
; the critical factor is the
number of patches, whereas for #dz41=z2
the
critical factor is the average patch size.
One of the main assumptions of the present
model is that metapopulation processes scale as
a power of patch area. While this is likely to be a
reasonable assumption even for networks with
considerable variation in patch areas (Hanski,
1999), it is clear that the range of patch areas
O. OVASKAINEN430
over which the assumption is valid has a limit.
This is illustrated in Fig. 4, where I show the
probability that a local population of a hypothetical
species would go extinct in unit time.
The extinction probability is shown over a wide
range of patch areas as emphasized by the
logarithmic scale in the x-axis. As zex measures
how sensitive the extinction rate is to patch area,
a deep slope in Fig. 4 corresponds to a high
value of zex: The dashed lines classify the patch
sizes to four different classes. In size class A; the
patches are so small that they are unable to
support a local population. In size class B; the
patches are large enough to support a local
population, but the expected population size
is so small that demographic stochasticity is
expected to be one of the main risks of
extinction. In this size class, the extinction
probability is very sensitive to patch area and
consequently zex is large. In size class C; the
patches are so large that the probability of
extinction due to demographic stochasticity is
negligible, and the main cause of extinction is
due to environmental stochasticity. For example,
several consecutive years with adverse environmental
conditions may be needed to cause a local
extinction. Patch size has an effect on the
extinction risk, but the effect is much milder
than in size class B; and consequently zex is
lower. In size class D; the habitat patches are
large enough to buffer the species against
environmental stochasticity, and the extinction
risk is now largely determined by catastrophes.
For example, human land use practices may
change in such a way that the species cannot
persist anymore. In this size class, the extinction
probability is almost independent of patch size,
corresponding to a very low value of zex: The
metapopulation approach taken in this paper is
most appropriate for networks with patches
from the intermediate size classes (B and C in
Fig. 4), as in such networks local extinctions and
recolonizations are expected to be a regular part
of regional scale dynamics.
There are several reasons why the conclusions
of the present study should be interpreted with
caution even for species which persist as classical
metapopulations at the spatial scale that is
relevant to the problem setting. Most importantly,
the model assumes that the extinction
rates in the different habitat patches are not
temporally or spatially correlated, which is a
critical assumption for the exponential scaling of
the mean time to extinction with respect to n:
Interpatch correlation in extinction or colonization
rates could drastically reduce the effective
number of habitat patches (Ovaskainen, 2002).
In such a case the extinction of the metapopulation
is most likely to occur when multiple
extinctions happen simultaneously (Harrison &
Quinn, 1989).
Most species are not likely to follow classical
metapopulation dynamics, or at least the spatial
scale over which they do so is limited. Nevertheless,
due to habitat loss and fragmentation,
most populations are increasingly patchy at all
spatial scales, and metapopulation dynamics are
likely to play an increasingly important role in
conservation biology (Hanski, 1999). The main
contribution of the present study to the SLOSS
debate is the demonstration that the effects of
patch area on emigration, immigration and
extinction should be considered while assessing
the relative advantages of a large number of
small patches vs. a small number of large
patches.
I thank Marc Belisle, Rampal Etienne, Ilkka
Hanski and Saskya van Nouhuys for valuable
comments. This study was supported by the Academy
of Finland (Grant Number 50165 and the Finnish
Centre of Excellence Programme 2000-2005, Grant
Number 44887).
log Patch area
Extinctionprobability
2 4 6 8 10
0.2
0.4
0.6
0.8
1
A B DC
Fig. 4. A qualitative plot of the local extinction
probability of a hypothetical species over a wide range of
patch areas. The letters A; B; C and D refer to different size
classes, for which the main risks of local extinction are
different.
PERSISTANCE OF SPECIES AND THE SLOSS PROBLEM 431
REFERENCES
Andersson, H. & Djehiche, B. (1998). A threshold limit
theorem for the stochastic logistic epidemic. J. Appl.
Probab. 35, 662­670.
Baz, A. & Garcia-Boyero, A. (1996). The SLOSS
dilemma: a butterfly case study. Biodiversity Conserv. 5,
493­502.
Blake, J. G. (1991). Nested subsets and the distribution of
birds on isolated woodlots. Conserv. Biol. 5, 58­66.
Boecklen, W. J. (1997). Nestedness, biogeographic theory,
and the design of nature reserves. Oecologia 112,
123­142.
Bolger, D. T., Alberts, C. & SouleŁ , M. E. (1991).
Occurrence patterns of bird species in habitat fragments:
sampling, extinction, and nested species subsets. Am. Nat.
137, 155­166.
Burkey, T. V. (1989). Extinction in nature reserves: the
effect of fragmentation and the importance of migration
between reserve fragments. Oikos 55, 75­81.
Burkey, T. V. (1995). Extinction rates in archipelagos:
implications for populations in fragmented habitats.
Conserv. Biol. 9, 527­541.
Burkey, T. V. (1997). Metapopulation extinction in
fragmented landscapes: using bacteria and protozoa
communities as model ecosystems. Am. Nat. 150,
568­591.
Cabeza, M. & Moilanen, A. (2001). Design of reserve
networks and the persistence of biodiversity. Trends Ecol.
Evol. 16, 242­248.
Cole, B. J. (1981). Colonization abilities, island size, and
the number of species on archipelagoes. Am. Nat. 117,
629­638.
Cook, R. R. & Hanski, I. (1995). On expected lifetimes of
small and large species of birds on islands. Am. Nat. 145,
307­315.
Crone, E. E., Doak, D. & Pokki, J. (2001). Ecological
influences on the dynamics of a field vole metapopulation.
Ecology 82, 831­843.
Deshaye, J. & Morriset, P. 1989. Species­area relationships
and the SLOSS effect in a subarctic archipelago.
Biol. Conserv. 48, 265­276.
Diamond, J. M. (1975). The island dilemma: lessons of
modern biogeographic studies for the design of natural
reserves. Biol. Conserv. 7, 129­146.
Diamond, J. M. & May, R. M. (1976). Island biogeography
and the design of natural reserves. In: Theoretical
Ecology: Principles and Applications (May, R. M., ed.),
pp. 163­186. Oxford: Blackwell.
Dzwonko, Z. & Loster, S. (1989). Distribution of
vascular plant species in small woodlands on the western
Carpathian foothills. Oikos 56, 77­86.
Etienne, R. S. & Heesterbeek, J. A. P. (2000). On
optimal size and number of reserves for metapopulation
persistence. J. theor. Biol. 203, 33­50, doi:10.1006/
jtbi.1999.1060.
Foley, P. (1997). Extinction models for local populations.
In: Metapopulation Biology: Ecology, Genetics, and
Evolution (Hanski, I. & Gilpin, M. E., eds), pp. 215­
246. San Diego: Academic Press.
Forney, K. A. & Gilpin, M. E. (1989). Spatial structure
and population extinction: a study with Drosophila flies.
Conserv. Biol. 3, 45­51.
Fukamachi, K., Iida, S. & Nakashizuka, T. (1996).
Landscape patterns and plant species diversity of forest
reserves in the Kanto region, Japan. Vegetatio 124,
107­114.
Gilpin, M. E. & Diamond, J. M. (1980). Subdivision of
nature reserves and the maintenance of species diversity.
Nature 285, 567­568.
Goel, N. S. & Richter-Dyn, N. (1974). Stochastic Models
in Biology. New York: Academic Press.
Hanski, I. (1992). Inferences from ecological incidence
functions. Am. Nat. 139, 657­662.
Hanski, I. (1996). Metapopulation ecology. In: Population
Dynamics in Ecological Space and Time (Rhodes Jr, O. E.,
Chesser, R. K. & Smith, M. H., eds), pp. 12­43. Chicago:
The University of Chicago Press.
Hanski, I. (1998). Connecting the parameters of local
extinction and metapopulation dynamics. Oikos 83,
390­396.
Hanski, I. (1999). Metapopulation Ecology. Oxford:
Oxford University Press.
Hanski, I. & Ovaskainen, O. (2000). The metapopulation
capacity of a fragmented landscape. Nature 404,
755­758.
Hanski, I., Alho, J. & Moilanen, A. (2000). Estimating
the parameters of survival and migration of individuals in
metapopulations. Ecology 81, 239­251.
Harrison, S. & Quinn, J. F. (1989). Correlated environments
and the persistence of metapopulations. Oikos 56,
293­298.
Higgs, A. J. & Usher, M. B. (1980). Should nature
reserves be large or small? Nature 285, 568­569.
Hill, J. K., Thomas, C. D. & Lewis, O. T. (1996). Effects
of habitat patch size and isolation on dispersal by
Hesperia comma butterflies: implications for metapopulation
structure. J. Anim. Ecol. 65, 725­735.
Honnay, O., Hermy, M. & Coppin, P. (1999). Effects of
area, age and diversity of forest patches in Belgium on
plant species richness, and implications for conservation
and reforestation. Biol. Conserv. 87, 73­84.
Ja rvinen, O. (1982). Conservation of endangered plant
populations: single large or several small reserves? Oikos
38, 301­307.
Lande, R. (1993). Risks of population extinction from
demographic and environmental stochasticity and random
catastrophes. Am. Nat. 142, 911­927.
Lomolino, M. V. (1994). An evaluation of alternative
strategies for building networks of nature reserves. Biol.
Conserv. 69, 243­249.
MacArthur, R. H. & Wilson, E. O. (1967). The Theory
of Island Biogeography. Princeton, NJ: Princeton University
Press.
May, R. M. (1975). Island biogeography and the design of
wildlife preserves. Nature 254, 177­178.
Moilanen, A., Smith, A. T. & Hanski, I. (1998). Longterm
dynamics in a metapopulation of the American
pika. Am. Nat. 152, 530­542.
Ovaskainen, O. (2001). The quasi-stationary distribution
of the stochastic logistic model. J. Appl. Probab. 38,
898­907.
Ovaskainen, O. (2002). The effective size of a metapopulation
living in a heterogeneous patch network. Am. Nat.
(in press).
Ovaskainen, O. & Hanski, I. (2001). Spatially structured
metapopulation models: metapopulation capacity and
threshold conditions for persistence. Theor. Popul. Biol.
60, 281­304, doi:10.1006/tpbi.2001.1548.
O. OVASKAINEN432
Ovaskainen, O. & Hanski, I. (2002). Transient dynamics
in metapopulation response to perturbation. Theor.
Popul. Biol. 61, 285­295. doi:10.1006/tpbi.2002.1586.
Ovaskainen, O., Sato, K., Bascompte, J. & Hanski, I.
(2002). Metapopulation models for extinction threshold
in spatially correlated landscapes. J. theor. Biol. 215,
95­108. doi:10.1006/jtbi.2001.2502.
Patterson, B. D. (1987). The principle of nested subsets
and its implications for biological conservation. Conserv.
Biol. 1, 323­334.
Patterson, B. D. & Atmar, U. (1986). Nested subsets and
the structure of insular mammalian faunas and archipelagoes.
Biol. J. Linnean Soc. 28, 65­82.
Pelletier, J. D. (2000). Model assessments of the optimal
design of nature reserves for maximizing species longevity.
J. theor. Biol. 202, 25­32, doi:10.1006/
jtbi.1999.1030.
Quinn, J. F. & Harrison, S. P. (1988). Effects of habitat
fragmentation and isolation on species richness: evidence
from biogeographic patterns. Oecologia 75, 132­140.
Quinn, J. F. & Hastings, A. (1987). Extinction in
subdivided habitats. Conserv. Biol. 1, 198­208.
Saetersdal, M. (1994). Rarity and species/area relationships
of vascular plants in deciduous woods, western
NorwayFapplications to nature reserve selection. Ecography
17, 23­38.
Simberloff, D. & Abele, L. G. (1976). Island biogeography
theory and conservation practice. Science 191,
285­286.
Simberloff, D. & Abele, L. G. 1982. Refuge design and
island biogeographic theory: effects of fragmentation.
Am. Nat. 120, 41­50.
Simberloff, D. & Martin, J.-L. (1991). Nestedness of
insular avifaunas: simple summary statistics masking
complex species patterns. Ornis Fennica 68, 178­192.
Skaggs, R. W. & Boecklen, W. J. (1996). Extinctions of
montane mammals reconsidered: putting a global-warming
scenario on ice. Biodiversity Conserv. 5, 759­778.
SouleŁ , M. E. & Simberloff, D. (1986). What do genetics
and ecology tell us about the design of nature reserves?
Biol. Conserv. 35, 19­40.
Terborgh, J. W. (1976). Island biogeography and
conservation: strategy and limitations. Science 193,
1029­1030.
Wahlberg, N., Moilanen, A. & Hanski, I. (1996).
Predicting the occurrence of endangered species in
fragmented landscapes. Science 273, 1536­1538.
Wilson, E. O. & Willis, E. O. (1975). Applied biogeography.
In: Ecology and Evolution of Communities (Cody,
M. L. & Diamond, J. M. eds), pp. 523­534. Cambridge,
MA: Belknap Press of Harvard University.
Virolainen, K. M., Suomi, T., Suhonen, J. & Kuitunen,
M. (1998). Conservation of vascular plants in single
large and several small mires: species richness, rarity
and taxonomic diversity. J. Appl. Ecol. 35,
700­707.
Worthen, W. B. (1996). Community composition and
nested-subset analyses: basic descriptors for community
ecology. Oikos 76, 417­426.
Wright, S. J. & Hubbell, S. P. (1983). Stochastic
extinction and reserve size: a focal species approach.
Oikos 41, 466­476.
Wright, D. H. & Reeves, J. H. (1992). On the meaning
and measurement of nestedness of species assemblages.
Oecologia 92, 416­428.
Zacharias, D. & Brandes, D. (1990). Species­area
relationships and frequency­floristical data analysis of
44 isolated woods in northwestern Germany. Vegetatio
88, 21­29.
PERSISTANCE OF SPECIES AND THE SLOSS PROBLEM 433