The Bias of Complementarity Hotspots toward

Marginal Populations
MIGUEL B. ARAÚJO* AND PAUL H. WILLIAMS†
*Biogeography and Conservation Lab, The Natural History Museum, Cromwell Road SW7 5BD London, U.K.,
email m.araujo@nhm.ac.uk
†Department of Geography, University College, WC1H 0AP London, U.K.

Abstract: It has been suggested that using complementarity to identify networks of important areas for con-
serving biodiversity may preferentially select areas within the margins of species ranges. We tested this idea
by examining the location of complementarity hotspots in relation to two measures of range core-periphery.
The first measures patterns of aggregation among records within each species’ range to identify areas within
the core (i.e., areas with aggregated distributions) and periphery (i.e., areas with scattered distributions) of
the range. The second measures spatial turnover among species to identify areas with a high density of range
edges. For three selected groups of terrestrial vertebrates in Europe—mammals, birds, and herptiles—areas
chosen based on complementarity were located within the margins of species’ ranges more often than ex-
pected by chance. This pattern was consistent for the two measures of core-periphery we used. The bias of
complementarity hotspots toward marginal populations is especially important for species with restricted
range sizes. If extinctions are determined mainly by demographic factors, then selecting areas at the peripher-
ies of species’ ranges might be a poor option. But if extinctions are determined mainly by extrinsic factors,
then peripheral populations might be important to ensure the long-term persistence of species.

El Sesgo de la Complementaridad de Áreas Importantes par la Biodiversidad (Hotspots) Hacia Poblaciones Marginales
Resumen: Se ha sugerido que el uso de la complementaridad para identificar redes de áreas importantes
para la conservación de la biodiversidad puede seleccionar preferentemente a áreas dentro de los márgenes
de los rangos de las especies. Evaluamos esta idea examinando la ubicación de áreas importantes comple-
mentarias en relación con dos medidas de rango nucleo-periferia. La primera midió patrones de agregación
entre los registros dentro de los rangos de cada especie para identificar las áreas dentro del nucleo (por ejem-
plo áreas con distribuciones agregadas) y la periferia (áreas con distribución dispersa). La sequnda mide el
cambio espacial entre especies para identificar áreas con altas densidades de bordes de rango. De tres grupos
seleccionados de vertebrados terrestres dentro de Europa (mamíferos, aves y reptiles) las áreas seleccionadas
en base a la complementaridad estuvieron más frecuentemente localizadas dentro de los márgenes de los
rangos de especies, que lo esperado por el azar. Este patrón fue consistente para las dos medidas de núcleo-
periferia empleadas. Este sesgo hacia poblaciones marginales de áreas importantes complementarias es espe-
cialmente importante para especies con rangos de tamaño restringido. Si las extinciones son determinadas
principalmente por factores demográficos, entonces, la selección de áreas en las periferias de los rangos de las
especies puede ser una opción pobre. Por otro lado, si las extinciones son principalmente determinadas por
factores extrínsecos, entonces las poblaciones periféricas pueden ser importantes para asegurar la persisten-
cia de las especies a largo plazo.

Paper submitted September 27, 1999; revised manuscript accepted February 7, 2001.
1710

Conservation Biology, Pages 1710–1720

Volume 15, No. 6, December 2001
Araújo & Williams
Introduction
Reserve-selection techniques based on complementarity
seek to maximize representation of biodiversity within
the limitations of cost (e.g., Margules & Pressey 2000).
These methods are specifically tailored to problems of
representation and are more efficient than scoring pro-
cedures, such as hotspots of richness or rarity (Pressey
& Nicholls 1989; Williams et al. 1996, 2000; Reid 1998;
Araújo 1999). In the United States, for example, comple-
mentarity-based reserve-selection techniques have been
coupled with gap analysis to prioritize the selection of
additional areas for conservation (Kiester et al. 1996). In
Australia, federal and state governments use complemen-
tarity in the decision process to negotiate long-term agree-
ments for forest reserves (Finkel 1998; Pressey 1998).
Fine-grained demographic and genetic data for different
taxonomic groups are usually unavailable, so complemen-
tarity areas (also called “hotspots” of complementarity by
Dobson et al. [1997]) are often selected without an ex-
plicit treatment of persistence ( but see Kershaw et al.
1995; Turpie 1995; Nicholls 1998; Williams 1998; Cowl-
ing et al. 1999; Araújo & Williams 2000; Williams &
Araújo 2000). All other factors being equal, we might
hope that complementarity hotspots perform at least as
well as areas chosen at random in terms of the expected
persistence of species. Branch et al. (1995) suggest,
however, that this might not always be the case. We in-
vestigated their report that complementarity-based re-
serve-selection algorithms could select areas preferen-
tially for populations in the margins of species’ ranges,
and we discuss the likely consequences of this for spe-
cies persistence.
Branch et al. (1995) studied a small group of 20 spe-
cies of tortoises and terrapins in southern Africa; their
observations were also limited by selection of a small num-
ber of reserves (eight). But if complementarity hotspots
were generally more likely to be selected within the mar-
gins of species’ ranges, then a criticism could be made that
the populations chosen might have relatively poor prog-
noses for persistence. This is because population densi-
ties are often expected to be lower ( Brown 1984; Caugh-
ley et al. 1988; Lawton 1995) and more variable (Curnutt
et al. 1996) near the periphery of the range. As a result,
peripheral populations might be less resilient to sto-
chastic threats than are core populations (Curnutt et al.
1996), increasing the risk of extinction (e.g., Goodman
1987). Consistent with this prediction, Griffith et al.
(1989) showed that attempts to reintroduce birds and
mammals into the core of their former distributions were
more likely to succeed, and Wolf et al. (1996) report sim-
ilar results. Therefore, selecting areas at the margins of
species’ contemporary ranges might compromise the ul-
timate aim of biodiversity conservation, which is the
persistence of these species (Frankel & Soulé 1981).
We tested for a bias toward selecting areas within spe-
Complementary Hotspots and Range Margins 1711
cies’ range margins with data on the distribution of three
groups of terrestrial vertebrates in Europe. We discuss
possible reasons for the observed patterns and their im-
plications for species’ persistence and reserve selection.
Methods
Data
We used atlas records for the occurrence of 149 species
of amphibians and reptiles (34,137 records; median 82
per species), 445 breeding birds (291,390 records; me-
dian 389 per species), and 187 mammals (81,309 records;
median 187 per species) distributed among cells of a
50-km grid across Europe (Gasc et al. 1997; Hagemeijer &
Blair 1997; Mitchell-Jones et al. 1999). Data from these at-
lases were converted to a common grid system based on
the one used for the Atlas Florae Europaeae (Lahti &
Lampinen 1999). This grid follows the 50-km lines of the
UTM grid, except near the border of the 6 compensation
zones and along the coasts. Although all of these atlases
were based on the UTM 50-km grid, conversion was
needed because they used different rules to merge cells
near the border of the 6 compensation zones and we had
to accommodate coasts and islands. Where possible, con-
verting data for the selected groups depended on identify-
ing unique (although sometimes approximate) correspon-
dence between cells in these grids ( Williams et al. 2000).
The mapped area (2434 grid cells) excluded most of the
countries of eastern Europe and the former Soviet Union,
where sampling effort is relatively low. Nonetheless, this
remains one of the world’s most extensive and represen-
tative data sets for terrestrial vertebrates.
Identifying Core and Peripheral Ranges for
Individual Species
Identifying the core and periphery within the range of a
single species is not straightforward; many related mea-
sures give only approximate results (e.g., Blackburn et
al. 1999). Empirical studies have shown that population
density often decreases from the core of a species’ range
toward its margins ( Whittaker 1967; Hengeveld & Haeck
1981; Rapoport 1982; Brown 1984; Caughley et al. 1988).
Spatial patterns of density are often more complex than a
gradual decline, however, with multimodal patterns of den-
sity being common and perhaps even the norm ( Lawton
1995; Maurer 1999). Although, in principle, measuring pat-
terns of density directly may be preferable to measuring
distances to some central point within the range, in prac-
tice, suitable estimates of density are rarely available.
Density is used here in the sense of the number of obser-
vations within a given area ( Lincoln et al. 1998). Direct
counts of individuals are popular measures of density at rel-
atively fine scales. At larger spatial scales, coarser-resolution
Conservation Biology
Volume 15, No. 6, December 2001
1712 Complementary Hotspots and Range Margins
measures, such as frequency of occurrence (e.g., Henge-
veld & Haeck 1981; Gates et al. 1994) or degree of aggrega-
tion among records (e.g., Rapoport 1982; Kunin 1998),
have also been used. Aggregation is based on both the fre-
quency of occurrence and the distance among records,
which can be measured using contagion indices. These are
similar to measures of aggregation for point and lattice data,
such as Kernel estimation and nearest neighbor measures
(e.g., Bailey & Gatrell 1995). With respect to occurrence
records within grid cells, contagion can be measured as
ka
 
 ∑ w ab y b
 b--------------------------
=1
-
contagion =  ka *100, (1)
 
 ∑ w ab 
 b=1  portion of species with aggregated distributions (and/or
 
where contagion is a weighted average of the number of
occupied grid cells among a set of ka neighbors of a cen-
tral grid cell ya. The weight given to the grid cell yb is wab 
1/dab, where dab is the distance between grid cells ya
and yb. We used two orders of neighbors, assigning a
weight of d  1 to the first order and a weight of w  2 to
the second order. Neighbors in the first order were the 8
cells touching the central cell along the edges and at the
corners within a rectangular grid. The second-order neigh-
bors were the 16 grid cells concentric to the first order.
A contagious distribution can be defined as a pattern
in which occurrences are more aggregated or clustered
than in a random distribution, so that one occurrence in-
creases the probability of another occurrence nearby
( Lincoln et al. 1998). Contagion index values range be-
tween 0 and 100% of the maximum aggregation possi-
ble. Here, we expected the maximum aggregation to oc-
cur within the “core” and minimum values to occur
within the “periphery” of a species’ range (Fig. 1).
Areas with high contagion values can be interpreted in
terms of persistence if three important ideas are linked.
First, spatial aggregation of records at large scales is pre-
dictive of density at lower scales (Kunin 1998). Second,
aggregation reflects the response of populations to local
conditions, which in turn reflects the extent to which lo-
cal environments meet species’ ecological requirements
(Hutchinson 1957; Hengeveld & Haeck 1981; Brown 1984;
Lawton 1995). Third, the spatial aggregation of populations
is likely to correlate positively with the ability of individuals
to disperse from one population to another, which may
lead to more-stable metapopulations (e.g., Gonzalez et al.
1998).
Combining Patterns of Peripheral Range for Large Numbers
of Species
Measures that quantify distinct aspects of the core-
periphery pattern for large numbers of species may be
Conservation Biology
Volume 15, No. 6, December 2001
Araújo & Williams
used to explore shared patterns. There are many possi-
ble ways to combine measures of core and peripheral
ranges of individual species into a single composite num-
ber. The principal problem is that the cores and peripher-
ies of different species do not coincide, so any single mea-
sure combining core-periphery values among species will
capture one facet of complex patterns of covariation. Here
we considered two related measures, median contagion
and range edge.
One way in which combined patterns of core and pe-
riphery may be scored is to plot the central tendency in
the species’ contagion scores for each grid cell. High
median contagion scores identify areas with a high pro-
portion of species with aggregated distributions. Low
median contagion scores identify areas with a low pro-
a high proportion of species with scattered distribu-
tions). Because restricted species have generally lower
contagion values than more abundant ones, contagion
scores for all individual species are rescaled between 0
and 100.
Another approach is to identify areas adjacent to many
species’ range edges within the grid. This does not help
identify areas with cores for many species, but areas
with high scores for these measures are expected to co-
incide with the distributional margins of many species.
Counts of the density of range edges have often been
used to detect transition zones in the composition of
faunas and floras, and we used a familiar “absolute-range
edges index” ( Williams et al. 1999). This index is a sim-
ple count of the number of species with presence/ab-
sence differences between each cell and any of its neigh-
boring eight cells within the grid:
absolute range edges = ∑ i :1 . . .s fk:1 . . . 8 ( pi,k, pi, 0 ), (2)
where pi,k is the presence/absence of species i in the
kth of the eight neighboring cells (cells 1–8), pi,0 is the
presence/absence of species i in the central cell of the
neighborhood (cell 0), the simple mismatch function
f( pi,k, pi,0 ) is equal to 1 if pi,k and pi,0 have different val-
ues for any of the values of k (1–8) or is equal to 0 if they
have the same value, and s is the total number if species
in the database.
Area Selection
We selected 211 complementarity hotspots (approxi-
mately 10% of the grid cells) for each of the three groups
of terrestrial vertebrates to maximize species’ represen-
tation. The number of hotspots selected was arbitrary
but followed the recommendation of the World Conser-
vation Union for countries to establish minimum conser-
vation areas up to 10% of their total area (Soulé & Sanjayan
1998). To solve this maximum-coverage problem (Church
et al. 1996), we used a heuristic technique adapted from
Araújo & Williams
the near-minimum-set algorithm of Margules et al.
(1988), but with (1) a test to reject any areas that in
hindsight are redundant and (2) a reordering of areas to
provide an approximate solution for the kind of maxi-
mum-coverage problem we faced (see the Appendix for
a full description of the algorithm).
Randomization Tests
Two tests of randomization were performed. The first
assessed whether, for all species combined, scores for
patterns of range core and periphery in the hotspots
were higher or lower than expected by chance. We im-
plemented this test by selecting 211 areas (i.e., grid
cells) randomly without replacement 1000 times and
then comparing the median value observed from com-
plementarity hotspots with the frequency distribution of
values for each measure (median contagion and range
edges) obtained by simulated random selection. Ob-
served contagion values below the lower 5% tail of the
random distribution indicated that hotspots were lo-
cated at the peripheries of species’ combined range
scores more often than expected by chance (single-
tailed p  0.05). Observed range-edge values above the
upper 95% tail of random indicated that hotspots were
located in areas of transition between biotas more often
than expected by chance (single-tailed p  0.05).
Complementary Hotspots and Range Margins 1713
Figure 1. Map showing patterns of
core and peripheral range distribu-
tion for a single species, alpine cham-
ois ( Rupicapra rupicapra Linnaeus),
based on a contagion measure and
50  50 km atlas grid-cell records
from Mitchell-Jones et al. (1999). Con-
tagion scores are divided into six
gray-scale classes of approximately
equal size by number of grid cells
(equal-frequency), so that maximum
scores (range core) are shown in
black and minimum scores (range
periphery) are shown in light gray.
Scores are shown only within re-
corded distribution.
The second test assessed whether, for each individual
species, the observed central tendency in values of con-
tagion within hotspots was lower than expected by
chance. Within the 211 selected areas, we counted the
number of occurrences n of each species. For each spe-
cies, the respective number of areas n was drawn from
within species’ distributions, and the mean contagion
score among these areas was calculated. This was re-
peated 1000 times to generate an expected frequency
distribution for the mean contagion score for each spe-
cies when n areas were selected from within its range.
An observed mean contagion value within the lower 5%
tail of this distribution indicated that areas were tending
to be selected in the range periphery for this species
(single-tailed p  0.05).
Data-handling procedures, the contagion and range-
edge measures, area selection, and the random simula-
tion were implemented with WORLDMAP software (Wil-
liams 1999).
Results
Combined Patterns of Core-Periphery
For all three groups of vertebrates there was a pattern of
increasing density of species’ range edges (Fig. 2, col-
Conservation Biology
Volume 15, No. 6, December 2001
1714 Complementary Hotspots and Range Margins
umn 2) and decreasing contagion values ( Fig. 2, col-
umns 1 & 2) along the mountain ranges of Spain (e.g.
Cantabrian, Sierra de Gredos, and Sierra Nevada), the
Pyrenees, the Alps, and the Balkan peninsula. As ex-
pected, the range-edge measure was negatively corre-
lated with the contagion measure, (Spearman rank coef-
ficient rs  0.50 for mammals, rs  0.44 for birds, rs 
0.24 for herptiles), supporting the idea that areas with
high spatial turnover among species should coincide
with the range margins of many species. These cells
with low contagion and high density of range edges also
coincided with some of the major European centers of
endemism (Akeroyd & Heywood 1994; Gaston & David
1994; Dennis & Williams 1995; Williams et al. 2000),
which are located predominantly in southern Europe
and in the mountains. In agreement with this observa-
tion, we found a positive correlation between species’
range edges and the density of species of restricted
range size (measured as the sum of the inverse range-
Conservation Biology
Volume 15, No. 6, December 2001
Araújo & Williams
size values for each species) for mammals (Spearman
rank correlation rs  0.43, p  0.001) and amphibians
and reptiles (rs  0.73, p  0.001). This relationship was
not apparent for birds (rs  0.063, p  0.001), although
this might be explained by their strong dispersal abilities
and low endemism within Europe (Long et al. 1996). In-
deed, these data indicate that birds have the largest me-
dian range size (see description of data in methods).
Complementarity Hotspots and Range Margins
The randomization tests for the combined species mea-
sures showed that hotspots of complementarity are lo-
cated in the range margins for many species more often
than expected by chance (Table 1). For both mammals
and birds there was a consistent pattern ( p  0.05) of
complementarity hotspots being selected in areas with
low combined median contagion scores. This pattern
was not found for herptiles, although there was a signifi-
Figure 2. Patterns of core periphery
for many species among 50  50
km atlas grid cells, based on median
contagion and range edges for the
three groups of terrestrial verte-
brates in Europe. Complementarity
hotspots are also shown. Quantita-
tive scores were divided into 33
equal-frequency color classes, with
maximum scores shown in red and
minimum (non-zero) scores in blue.
High scores among the contagion
measure indicate the predominance
of species’ range cores, and mini-
mum scores indicate the predomi-
nance of range peripheries. The
color scale for the range-edges mea-
sure was inverted, so high rates of
spatial turnover (i.e., high density of
range edges) are shown in blue and
low densities of range edges are
shown in red. Svalbard and the
Azores are displaced from their true
relative positions.
Araújo & Williams
Table 1. Results of a randomization test to assess whether the
observed median contagion and range-edge values* obtained for
211 complementarity hotspots are likely to have arisen by chance.
Mammals Birds Herptiles
Test ( p) ( p) ( p)
Median contagion
observed in
hotspots 69.2 (0.05) 86.5 (0.05) 74.9
lower 5% tail
from random 78.5 89.6 74.6
Range edges
observed in
hotspots 39.5 (0.05) 90.7 26.7 (0.05)
upper 5% tail
from random 37.3 92.9 18.8
*Hotspots are considered selected at the margins of species’ ranges if
the observed values fall within the lower 5% tail ( p  0.05) from the
random simulation for median contagion or if they fall within the
upper 5% tail for range edges.
cant association ( p  0.05) between complementarity
hotspots selected for herptiles and areas with high den-
sity of range edges. Hotspots for mammals were also lo-
cated in areas with many range edges ( p  0.05), al-
though this was not true of birds (Appendix).
Randomization tests for individual species show that all
species in complementarity hotspots have contagion val-
ues lower than expected by chance ( p  0.05), supporting
the idea of a bias toward the range periphery (Fig 3). The
cluster points close to the lower left of Fig. 3 show that
hotspots of complementarity tend to sample restricted spe-
cies disproportionately closer to the edge of the range.
Discussion
Complementarity Hotspots and Range Margins
Our results suggest that complementarity hotspots tend
to represent areas near the periphery of species’ ranges
more often than expected by chance. This bias toward
marginal populations was greater among species of more
restricted ranges. To understand this pattern it is impor-
tant to consider some of the properties of complementar-
ity and relate them to the biogeographic factors likely to
contribute to it.
Complementarity is the amount that sets of areas con-
tribute to the representation of otherwise unrepresented
attributes, such as species ( Vane-Wright et al. 1991). This
is influenced by both the relative number of attributes in a
given area (i.e., alpha diversity) and the degree of differ-
ence in attributes between areas (i.e., beta diversity or
spatial turnover). In practice, hotspots of complementar-
ity tend to be located in areas with high local species rich-
ness, because they represent many species in a limited
area, and in areas with many species of restricted range
size, because there are fewer opportunities to represent
Complementary Hotspots and Range Margins 1715
them and they are less likely to co-occur. The degree to
which complementarity-based area selection is driven by
alpha or beta diversity is also related to the relative effi-
ciency of different heuristic algorithms. “Greedy-richness”
algorithms select areas by increments in attribute values
such as complementarity richness and are likely to be par-
ticularly sensitive to local species richness, especially in
the first steps of the iteration. Conversely, “progressive-rar-
ity” algorithms (Appendix) are more sensitive to the loca-
tion of the most restricted species. As expected, our ran-
domization tests (not shown) suggested that, for all groups,
hotspots of complementarity had more restricted species
than expected by chance ( p  0.001). But only comple-
mentarity hotspots for mammals were preferentially lo-
cated in areas with high species richness ( p  0.001). For
birds they were preferentially located in areas with low
species richness ( p  0.001), and for herptiles there was
no significant relationship with richness.
There are many possible mechanisms likely to cause
complementarity hotspots to overlap with species’ range
margins. Some of these mechanisms are spatially coin-
cident, making it difficult to disentangle causes and ef-
fects of the observed bias. For example, complementar-
ity hotspots may coincide with a high density of range
edges because of Odum (1971) “edge effect” of increas-
ing species richness as a result of the spatial overlap of
range margins of species from neighboring assemblages.
At large biogeographical scales, edge effects may be re-
lated to broad patterns of species’ spatial turnover, such
as latitudinal gradients of richness and species replace-
ment (e.g. Williams et al. 1999), or to more local pat-
terns of turnover associated with elevational gradients.
Mountains may act as barriers to dispersal between bio-
geographical units, so they are likely to circumscribe the
edge of many species’ ranges, especially those with poor
dispersal abilities. Simultaneously, species richness in
mountainous regions may be boosted in part by the
edge effect resulting from overlaps in species’ eleva-
tional replacements (e.g., Stevens 1992; Rahbek 1995).
Another reason for coincidence between complemen-
tarity hotspots and a high density of range edges is the
positive correlation between the distribution of restricted
species and concentrations of marginal populations. This
correlation may occur for at least two reasons. First, it
could be an artifact of the way range sizes are measured.
If many species are on the edge of their ranges in Europe,
then the edges of the region will have many species that
appear to have small ranges but are widespread in other
regions. This may have been the case in a study by Branch
et al. (1995 ), in which many marginal populations were
selected on the border between South Africa and Mozam-
bique. Our data show similar patterns on the eastern and
southern borders of Europe, where many complementar-
ity hotspots were selected ( Fig. 2).
Second, areas with many species of restricted range
size (e.g., areas of narrow endemism) might score high
Conservation Biology
Volume 15, No. 6, December 2001
1716 Complementary Hotspots and Range Margins Araújo & Williams

Figure 3. Ratio between the mean

contagion value obtained for each
individual species in the complemen-
tarity hotspots and the lower 5%
threshold value obtained from range
randomization (see methods)
against species range size within the
European atlas region. Each data
point is a species. Values above 1 in-
dicate species whose contagion val-
ues in hotspots are higher than ex-
pected by chance; values below 1 are
species whose contagion values in
hotspots are lower than expected by
chance (preferentially located at the
periphery).

Conservation Biology
Volume 15, No. 6, December 2001
Araújo & Williams
for edge density simply because edges may be close to-
gether for restricted species. In Europe, a disproportion-
ate number of restricted species occurs in mountains.
This may be due to climatic isolation of mountain tops,
which tends to reduce dispersal, or to strong environ-
mental gradients (often from arid to wet and from warm
to cold within short vertical distances), which constrain
ranges according to the environmental tolerances of in-
dividual species. Mountains may also harbor endangered
populations of once widespread species that currently
survive in only a few relatively undisturbed areas (e.g.,
brown bear [Ursus arctos] and wolf [Canis lupus]).
This pattern is particularly important in Europe, where
most of the lowlands have been converted to agricul-
tural and urban uses. But some of the most restricted spe-
cies may not have a measurable pattern of core-periphery
within their ranges at the grid-cell scale of 50 km. Rather
than measuring range periphery for the most restricted
species, the range edge measure may simply be capturing
a pattern of coincidence among endemic species’ ranges.
Implications for Conservation
The bias of complementarity hotspots toward marginal
populations is due partly to increasing richness caused
by edge effects but also to the effects of a high concen-
tration of endemics in the measurement of range periph-
eries. Is this a problem for conservation?
Choosing areas for conservation where many species
are within the margins of their range may not be a bad
thing when it is necessary to represent endemic and vul-
nerable species in what is left of their historical ranges.
In fact, species often persist at the edges of their histori-
cal distributions even when populations at the core are
extirpated ( Lomolino & Channell 1995; Channell & Lo-
molino 2000a, 2000b). This pattern of range collapse
away from the apparently more favorable core might be
expected for at least two reasons. First, if threatening
processes are spatially autocorrelated, then it is likely
that effects spread contagiously, like a disease, across
geographical space (e.g., Burgman et al. 1993; Lomolino
& Channell 1995). Even if a contagious process of ex-
tinction starts at one edge of a species’ range (as pre-
dicted by theory), it is likely that the last population to
be affected will be a distant or isolated population, prob-
ably located at the opposite edge of the range (Channell
& Lomolino 2000b). It follows that the effect of conta-
gion from threatening processes is likely to be higher in
dense and interconnected populations within the core
than in sparse and isolated populations at the range mar-
gins (Burgman et al. 1993). Second, because of their ge-
netic isolation and exposure to different selection pres-
sures, peripheral populations may be more resistant to
the effects of environmental change. For example, Kark
et al. (1999) found that populations of the Chukar Par-
tridge (Alectoris chukar) in Israel reach their highest
Complementary Hotspots and Range Margins 1717
overall genetic diversity (as well as unique alleles) in an
area of transition between ecosystems where a sharp en-
vironmental gradient exists. If this pattern of increasing
genetic diversity in apparently suboptimal conditions holds
for a wider variety taxa and regions, we might expect mar-
ginal populations to be important in maximizing intraspe-
cific variation and therefore increasing the species’ ability
to survive environmental changes (e.g., Lessica & Allendorf
1995).
The observations of Channell and Lomolino do not
contradict the idea that central populations are more
abundant (Brown 1984; Caughley et al. 1988; Lawton
1995) and less variable (Curnutt et al. 1996) than periph-
eral populations, but they do reinforce the idea that the
periphery—rather than being the “land of the living
dead” (e.g., Wolf et al. 1996)—may provide valuable op-
portunities to conserve species (Channell & Lomolino
2000a, 2000b), especially when contagious threats and
environmental change are important influences on cur-
rent and future distributions.
The question remains, however, whether greater pri-
ority should be given to central or peripheral popula-
tions in area selection. Ideally, both should be included
in conservation-area networks to cover both current and
future issues of persistence. Normally, the larger the
overall area devoted to in situ conservation, the more
species are expected to persist (e.g., Soulé & Sanjayan
1998). But resources available for conservation are
scarce, and increasing the representation target for one
set of species often implies a reduction in representation
for others. In moving from single-species conservation
to the broader problem of maximizing representation of
the biodiversity in conservation-area networks, we may
need to choose between selecting for core populations
and selecting for peripheral populations. If the choice is
to favor central populations, a possible solution would
be to constrain selection of complementarity hotspots
to areas near the core of each individual species’ range
(e.g., Nicholls 1998; Williams 1998).
One of the likely outcomes of adding additional con-
straints to area selection is a reduction in area efficiency
(Nicholls & Margules 1993; Nantel et al. 1998; Nicholls
1998; Pressey & Logan 1998; Cowling et al. 1999; Araújo
& Williams 2000). This means that constraining the algo-
rithms to favor selection of central populations will pro-
duce solutions that are either more expensive (if we
accept an increase in budget) or represent fewer species
(if the budget remains the same). The appropriate
trade-off between cost, representation, and persistence
of biodiversity needs to be addressed more explicitly if
conservation strategies are to be effective (e.g., Margules
& Pressey 2000). Approaches are now being investi-
gated to optimize cost and conservation return, mea-
sured in terms of the probability of species persistence
(e.g,. Araújo & Williams 2000; Williams & Araújo 2000).
These goals are targeted more explicitly toward the ulti-
Conservation Biology
Volume 15, No. 6, December 2001
1718 Complementary Hotspots and Range Margins
mate aim of biodiversity conservation, which is the per-
sistence of valued biodiversity (Frankel & Soulé 1981).
Acknowledgments
Distribution data were kindly supplied by J. P. Gasc of
the Societas Europaea Herpetologica, A. J. Michell-Jones
of the Societas Europaea Mammalogica, and W. J. M.
Hagemeijer of the European Bird Census Council. Data
were combined with help from R. Rees, M. Sadka, and S.
Gervás. We thank B. Pressey, C. Harrison, C. Humphries,
C. Margules, R. Vane-Wright, M. Lomolino, R. Channell,
T. Brooks, and two anonymous referees for comments
on the manuscript. M. B. A. was supported by the Portu-
guese Foundation for Science and Technology/ Praxis
XXI studentship ( BD/9761/96).
Literature Cited
Akeroyd, J. R., and V. H. Heywood. 1994. Regional overview: Europe.
Pages 39–54 in S. D. Davis, V. H. Heywood, and A. C. Hamilton, ed-
itors. Centers of Plant Diversity. World Conservation Union, Cam-
bridge, United Kingdom.
Araújo, M. B. 1999. Distribution patterns of biodiversity and the design
of a representative reserve network in Portugal. Diversity and Dis-
tributions 5:151–163.
Araújo, M. B., and P. H. Williams. 2000. Selecting areas for species per-
sistence using occurrence data. Biological Conservation 96:331–345.
Bailey, T. C., and A. C. Gatrell. 1995. Interactive spatial data analysis.
Longman, Essex, United Kingdom.
Blackburn, T. M., K. J. Gaston, R. M. Quinn, and R. D. Gregory. 1999.
Do local abundances of British birds change with proximity to
range edge? Journal of Biogeography 26:493–505.
Branch, W. R., G. A. Benn, and A. T. Lombard. 1995. The tortoises
( Testunidae) and terrapins (Pelomedusidae) of southern Africa:
their diversity, distribution and conservation. South Africa Journal
of Zoology 30:91–102.
Brown, J. H. 1984. On the relationship between abundance and distri-
bution of species. The American Naturalist 124:255–279.
Burgman, M. A., S. Ferson, and H. R. Akçakaya. 1993. Risk assessment
in conservation biology. Population and Community Biology Series
12. Chapman and Hall, London.
Caughley, G., D. Grice, R. Barker, and B. Brown. 1988. The edge of the
range. Journal of Animal Ecology 57:771–785.
Channell, R., and M. Lomolino. 2000a. Dynamic biogeography and
conservation of endangered species. Nature 403:84–86.
Channell, R., and M. Lomolino. 2000b. Trajectories of extinction: spa-
tial dynamics of the contraction of geographical ranges. Journal of
Biogeography 27:169–179.
Church, R. L., D. M. Stoms, and F. W. Davis. 1996. Reserve selection as
a maximal covering location problem. Biological Conservation 76:
105–112.
Cowling, R. M., R. L. Pressey, A. T. Lombard, P. G. Desmet, and G. Ellis.
1999. From representation to persistence: requirements for a sustain-
able system of conservation areas in the species-rich Mediterranean-
climate desert of South Africa. Diversity and Distributions 5:51–71.
Curnutt, J. C., S. L. Pimm, and B. A. Maurer. 1996. Population variabil-
ity of sparrows in space and time. Oikos 76:131–144.
Dennis, R. L. H., and W. R. Williams. 1995. Implications of biogeo-
graphical structures for the conservation of European Butterflies.
Pages. 213–229 in A. S. Pullin, editor. Ecology and conservation of
butterflies. Chapman and Hall, London.
Conservation Biology
Volume 15, No. 6, December 2001
Araújo & Williams
Dobson, A. P., J. P. Rodriguez, W. M. Roberts, and D. S. Wilcove. 1997.
Geographic distribution of endangered species in the United
States. Science 275:550–553.
Finkel, E. 1998. Software helps Australia manage forest debate. Science
281:1789–1790.
Frankel, O. H., and M. E. Soulé. 1981. Conservation and evolution.
Cambridge University Press, Cambridge, United Kingdom.
Gasc, J.-P., A. Cabela, J. Crnobrnja-Isailovic, D. Dolmen, K. Grossen-
bacher, P. Haffner, J. Lescure, H. Martens, J. P. Martínez Rica, H.
Maurin, M. E. Oliveira, T. S. Sofianidou, M. Veith, and A. Zuiderwijk.
1997. Atlas of amphibians and reptiles in Europe. Societas Europaea
Herpetologica and Muséum National d’Histoire Naturelle, Paris.
Gaston, K. J., and R. David. 1994. Hotspots across Europe. Biodiversity
Letters 2:108–116.
Gates, S., D. W. Gibbons, P. C. Lack, and R. J. Fuller. 1994. Declining
farmland bird species: modelling geographical patterns of abun-
dance in Britain. Pages 153–177 in P. J. Edwards, R. M. May, and N.
R. Webb, editors. Large-scale ecology and conservation biology.
Blackwell Science, Oxford, United States.
Gonzalez, A., J. H. Lawton, F. S. Gilbert, and T. M. Blackburn. 1998.
Metapopulation dynamics, abundance, and distribution in a micro-
ecosystem. Nature 281:2045–2047.
Goodman, D. 1987. The demography of chance extinction. Pages 11–
34 in M. E. Soulé, editor. Viable populations for conservation. Cam-
bridge University Press, Cambridge, United Kingdom.
Griffith, B., J. M. Scott, J. W. Carpenter, and C. Reed. 1989. Transloca-
tion as a species conservation tool: status and strategy. Science
245:477–480.
Hagemeijer, W. J. M., and M. J. Blair. 1997. The EBCC atlas of European
breeding birds, their distribution and abundance. European Bird
Census Council, Poyser, London.
Hengeveld, R., and J. Haeck. 1981. The distribution of abundance. II.
Models and implications. Procedings of the Koninklijke Neder-
landse Akademie van Wetenschappen C84:257–284.
Hutchinson, G. E. 1957. Concluding remarks. Cold Spring Harbor Sym-
posia on Quantitative Biology 22:415–427.
Kark, S., P. U. Alkon, U. N. Safriel, and E. Randi. 1999. Conservation
priorities for Chukar partridge in Israel based on genetic diversity
across an ecological gradient. Conservation Biology 13:542–552.
Kershaw, M., G. M. Mace, and P. H. Williams. 1995. Threatened status, rar-
ity and diversity as alternative selection measures for protected areas: a
test using Afrotropical antelopes. Conservation Biology 9:324–334.
Kiester, A. R., J. M. Scott, B. Csuti, R. Noss, B. Butterfield, K. Sahr, and
D. White. 1996. Conservation prioritization using gap data. Conser-
vation Biology 10:1332–1342.
Kunin, W. I. 1998. Extrapolating species abundance across spatial
scales. Science 281:1513–1515.
Lahti, T., and R. Lampinen. 1999. From dot maps to bitmaps: Atlas Flo-
rae Europaeae goes digital. Acta Botanica Fennica 162:5–9.
Lawton, J. H. 1995. Population dynamic principles. Pages 147–163 in
J. H. Lawton and R. M. May, editors. Extinction rates. Oxford Uni-
versity Press, Oxford, United Kingdom.
Lessica, P., and F. W. Allendorf. 1995. When are peripheral popula-
tions valuable for conservation. Conservation Biology 9:753–760.
Lincoln, R., G. Boxshall, and P. Clark. 1998. A dictionary of ecology,
evolution, and systematics. 2nd edition. Cambridge University
Press, Cambridge, United Kingdom.
Lomolino, M. V., and R. Channell. 1995. Splendid isolation: patterns of
range collapse in endangered mammals. Journal of Mammalogy 76:
335–347.
Long, A. J., M. J. Crosby, A. J. Stattersfield, and D. C. Wege. 1996. To-
wards a global map of biodiversity: patterns in the distribution of
restricted-range birds. Global Ecology and Biogeography Letters 5:
281–304.
Margules, C. R., and R. L. Pressey. 2000. Systematic conservation plan-
ning. Nature 405:243–253.
Margules, C. R., A. O. Nicholls, and R. L. Pressey. 1988. Selecting net-
Araújo & Williams
works of reserves to maximize biological diversity. Biological Con-
servation 43:63–76.
Maurer, B. A. 1999. Untangling ecological complexity: the macro-
scopic perspective. University of Chicago Press, Chicago.
Mitchell-Jones, A. J., G. Amori, W. Bogdanowicz, B. Krytufek, P. J. H.
Reijnders, F. Spitzenberger, M. Stubbe, J. B. M. Thissen, V. Voh-
ralík, and J. Zima. 1999. Atlas of European mammals. Academic
Press, London.
Nantel, P., A. Bouchard, L. Brouillet, and S. Hay. 1998. Selection of ar-
eas for protecting rare plants with integration of land use conflicts:
a case study for the West Coast of Newfoundland, Canada. Biologi-
cal Conservation 84:223–234.
Nicholls, A. O. 1998. Integrating population abundance, dynamics and
distribution into broad scale priority setting. Pages 251–272 in G.
Mace, A. Balmford, and J. R. Ginsberg, editors. Conservation in a
changing world: integrating process into priorities for action. Cam-
bridge University Press, Cambridge, United Kingdom.
Nicholls, A. O., and C. R. Margules. 1993. An upgraded reserve selec-
tion algorithm. Biological Conservation 64:165–169.
Odum, E. O. 1971. Fundamentals of ecology. 3rd edition. W. B. Saun-
ders, Philadelphia.
Pressey, R. L. 1998. Algorithms, politics and timber: an example of the
role of science in a public political negotiation process over new
conservation areas in production forests. Pages 73–87 in R. T. Wills
and R. J. Hobbs, editors. Ecology for everyone: communicating
ecology to scientists, the public and the politicians. Surrey Beatty
and Sons, Sydney, Australia.
Pressey, R. L., and V. S. Logan. 1998. Size of selection units for future
reserves and its influence on actual vs. targeted representation fea-
tures: a case study in western New South Wales. Biological Conser-
vation 85:305–319.
Pressey, R. L., and A. O. Nicholls. 1989. Efficiency in conservation eval-
uation: scoring vs. iterative procedures. Biological Conservation
50:199–218.
Rahbek, C. 1995. The elevational gradient of species richness: a uni-
form pattern? Ecography 18:200–205.
Rapoport, E. H. 1982. Aerography: geographical strategies of species.
Pergamon, Oxford, United Kingdom.
Reid, W. V. 1998. Biodiversity hotspots. Trends in Ecology and Evolu-
tion 13:275–280.
Complementary Hotspots and Range Margins 1719
Soulé, M. E., and M. A. Sanjayan. 1998. Conservation targets: do they
help? Science 279:2060–2061.
Stevens, G. C. 1992. The elevational gradient in altitudinal range: an
extension of Rapoport’s latitudinal rule to altitude. The American
Naturalist 140:893–911.
Turpie, J. K. 1995. Prioritising south African estuaries for conservation:
a practical example using waterbirds. Biological Conservation 74:
175–185.
Vane-Wright, R. I., C. J. Humphries, and P. H. Williams. 1991. What to
protect? Systematics and the agony of choice. Biological Conserva-
tion 55:235–254.
Williams, P. H. 1998. Key sites for conservation: area-selection meth-
ods for biodiversity. Pages 211–249 in G. Mace, A. Balmford, and J.
R. Ginsberg, editors. Conservation in a changing world: integrating
process into priorities for action. Cambridge University Press, Cam-
bridge, United Kingdom.
Williams, P. H. 1999. Biodiversity and WORLDMAP. Natural History
Museum, London. Available at http://www.nhm.ac.uk/science/
projects/worldmap (accessed 26 January 2001).
Williams, P. H., and M. A. Araújo. 2000. Using probabilities of persis-
tence to identify important areas for conservation. Proceedings of
the Royal Society London B 267:1959–1966.
Williams, P. H., D. Gibbons, C. Margules, A. Rebelo, C. Humphries, and
R. Pressey. 1996. A comparison of richness hotspots, rarity
hotspots and complementary areas for conserving diversity using
British birds. Conservation Biology 10:15–74.
Williams, P. H., H. de Klerk, and T. M. Crowe. 1999. Interpreting bio-
geographical boundaries among Afrotropical birds: spatial patterns
in richness gradients and species replacement. Journal of Biogeog-
raphy 26:459–474.
Williams, P. H., C. J. Humphries, M. B. Araújo, R. Lampinen, W. Hage-
meijer, J. P. Gasc, and A. Mitchell-Jones. 2000. Endemism and im-
portant areas for conserving European biodiversity: a preliminary
exploration of atlas data for plants and terrestrial vertebrates. Bel-
gian Journal of Entomology 2:21–46.
Whittaker, R. H. 1967. Gradient analysis of vegetation. Biological Re-
views 42:207–269.
Wolf, C. M., B. Griffith, C. Reed, and S. A. Temple. 1996. Avian and
mammalian translocations: update and reanalysis of 1987 survey
data. Conservation Biology 10:1142–1154.
Conservation Biology
Volume 15, No. 6, December 2001
1720 Complementary Hotspots and Range Margins Araújo & Williams

Appendix
Rules for selecting a given number (n) of hotspots of complementarity to provide a near-maximum coverage of species.*

Step Rule or procedure

1 Select all areas for species that are more restricted than the representation goal (for representing all species at least once, this
means selecting all areas with unique species records).
2 Apply the following rules repeatedly until all species are represented:
a select areas with the greatest complementary richness in only the rarest species (ignoring less rare species); if there are ties
(areas with equal scores), then
b select areas among ties with the greatest complementary richness in the next-rarest species, and so on; if there are persistent
ties, then
c select areas among persistent ties with the lowest grid-cell number (arbitrary rule used rather than random choice among
ties to ensure repeatability in tests; other criteria, such as proximity to previously selected cells or number of records in
surrounding cells, can be substituted);
d repeat steps a–c until all species are represented.
3 Identify and reject any areas that in hindsight are unnecessary to represent all species.
4 Repeat steps 1–3 for representing every species at least once, twice and so on, until the required number of areas, n, is attained
or exceeded, disregarding all areas selected in one iteration before moving to the next.
5 Reorder areas by complementary richness if, before all areas are re-ordered, complementary richness reaches all species or if
the maximum complementary richness increment declines to 0. Continue to resequence areas by resetting the cumulative
richness to 0 (ignoring previously reordered areas, species more restricted than a particular multiple representation target,
and species already represented the required number of times within smaller sets) and starting again with scoring
complementary richness from the current position on the area list. Repeat this resetting as often as necessary to reorder all
areas.
*In this context, “areas” are grid cells, and the “rarest species” is the one with the fewest grid-cell records.

Conservation Biology
Volume 15, No. 6, December 2001