vol. 163, no. 2 the american naturalist february 2004
On the Generality of the Latitudinal Diversity Gradient
Helmut Hillebrand*
Erken Laboratory, Department of Limnology, Evolutionary Biology
Centre, University of Uppsala, Norr Malma 4200, S-76173
Norrta¨lje, Sweden; and Leibniz-Institut fu¨r Meereswissenschaften,
University of Kiel, Marine Ecology, Du¨sternbrooker Weg 20, D-
24105 Kiel, Germany
Submitted May 19, 2003; Accepted August 8, 2003;
Electronically published January 15, 2004
Online enhancements: appendixes.
abstract: The decline of biodiversity with latitude has received
great attention, but both the concise pattern and the causes of the
gradient are under strong debate. Most studies of the latitudinal
gradient comprise only one or few organism types and are often
restricted to certain region or habitat types. To test for significant
variation in the gradient between organisms, habitats, or regions, a
meta-analysis was conducted on nearly 600 latitudinal gradients assembled
from the literature. Each gradient was characterized by two
effect sizes, strength (correlation coefficient) and slope, and additionally
by 14 variables describing organisms, habitats, and regions.
The analysis corroborated the high generality of the latitudinal diversity
decline. Gradients on regional scales were significantly
stronger and steeper than on local scales, and slopes also varied with
sampling grain. Both strength and slope increased with organism
body mass, and strength increased with trophic level. The body masseffect
size relation varied for ecto- versus homeotherm organisms
and for different dispersal types, suggesting allometric effects on
energy use and dispersal ability as possible mechanisms for the body
mass effect. Latitudinal gradients were weaker and less steep in freshwater
than in marine or terrestrial environments and differed significantly
between continents and habitat types. The gradient parameters
were not affected by hemisphere or the latitudinal range
covered. This analysis is the first to describe these general and significant
patterns, which have important consequences for models
aiming to explain the latitudinal gradient.
Keywords: macroecology, species richness, trophic level, body mass,
habitat.
* Present address: Leibniz-Institut fu¨r Meereswissenschaften, University of
Kiel, Marine Ecology, Du¨sternbrooker Weg 20, D-24105 Kiel, Germany; email:
hhillebrand@ifm.uni-kiel.de.
Am. Nat. 2004. Vol. 163, pp. 192­211. 2004 by The University of Chicago.
0003-0147/2004/16302-30195$15.00. All rights reserved.
Differences in the number of coexisting species already
fascinated early naturalists (Darwin 1859; von Humboldt
[1828] 1993) and remain a central aspect of ecological
research (Gaston and Blackburn 2000). No single pattern
of biodiversity has fascinated ecologists more than the increase
of richness toward the Tropics (Pianka 1966; Rohde
1992; Rosenzweig 1995; Gaston and Blackburn 2000). Still,
there is an astonishing lack of consensus about the mechanisms
leading to this spatial variation in diversity. An
obstacle to the search for a primary cause for this latitudinal
gradient has been the ever increasing number of
hypotheses (Pianka 1966; Rohde 1992), their interdependence
(Currie 1991; Gaston and Blackburn 2000), and
the lack of rigorous falsification (Currie et al. 1999).
The debate on causes for the latitudinal gradient of
diversity features biological and nonbiological explanations.
Mid-domain models use random placement of species
ranges in a domain (Earth) with hard boundaries (the
poles) to predict a peak in diversity in the middle of this
domain (equator) without invoking any ecological or evolutionary
processes (Colwell and Hurtt 1994; Colwell and
Lees 2000). Subsequent tests of the mid-domain model
revealed high predictive power of such models for global
(Lyons and Willig 1997; Jetz and Rahbek 2001; Koleff and
Gaston 2001) and regional data sets (Lees et al. 1999),
whereas others found strong differences between predicted
and observed diversity patterns (Bokma et al. 2001; Diniz
et al. 2002). For mid-domain models, latitude represents
the geometric constraint imposed on the range size of
species. All other models use latitude as a surrogate variable
for one or several factors co-varying with latitude.
Gradients of decreasing energy (and water) supply (Currie
1991; Allen et al. 2002) and decreasing biome area (Rosenzweig
1995) toward the poles have been proposed as
ultimate causes for the latitudinal diversity decline. There
is continuing debate about the relative importance of these
different covariates and their possible interactions (Currie
1991; Rohde 1997; Rosenzweig and Sandlin 1997).
Changes in the intensity or specificity of ecological interactions
(competition, predation, parasitism) with latitude
were also proposed as ultimate causes (Pianka 1966), but
several recent contributions failed to observe consistent
changes in interactions across latitude (Lambers et al. 2002;
Latitudinal Gradients 193
Ollerton and Cranmer 2002). The effective evolutionary
time hypothesis (Rohde 1992) finally assumes higher speciation
in the Tropics as the major process increasing diversity
at low latitudes. The higher diversification is based
on higher energy increasing mutation rates and decreasing
generation times (Cardillo 1999), larger area increasing
speciation (Losos and Schluter 2000), and higher temporal
stability on geological time scales enhancing clade persistence
(Jansson and Dynesius 2002).
Support for these major hypotheses is mainly based on
simple correlations between diversity and richness employing
a highly biased selection of organisms (Rohde
1992; Hillebrand and Azovsky 2001). Often, only one
group of organisms (or a limited number of related
groups) has been used to test the major hypotheses for
the latitudinal diversity gradient. Moreover, the main focus
has been on vertebrate taxa and higher plants, and a majority
of studies originated from the Americas. The predominance
of single-gradient studies counteracted the notion
of generality within the latitudinal diversity debate.
There is rather weak knowledge of how ecological and
evolutionary features of the organisms, geographic positions,
and habitat characteristics change the structure of
the latitudinal gradient.
Meta-analysis (Gurevitch and Hedges 1993; Rosenberg
et al. 2000) represents a suitable technique to analyze latitudinal
gradients across different biota and regions and
thus to generalize findings on the latitudinal distribution
of species richness. I assembled nearly 600 gradients from
the literature, and each gradient was characterized by the
standardized correlation coefficient (rz) and the slope (b)
of the relation between diversity and latitude. This database
was used to test the following hypotheses.
Hypothesis 1. The latitudinal gradient is a general pattern
across all organisms and habitat types. Generality is defined
here as the overall significance of the strength (rz)
and the slope (b) across the complete database as opposed
to invariability, which is defined as consistency of the magnitudes
rz and b across organisms and habitats.
Hypothesis 2. The magnitude of rz and b varies significantly
with characteristics of the measurement such as (a)
the spatial scale of diversity assessment, (b) the range of
latitudes covered, and (c) the diversity measure.
Hypothesis 3. The magnitude of rz and b varies significantly
with organism characteristics such as (a) body
mass, (b) dispersal type, (c) trophic level, and (d)
thermoregulation.
Hypothesis 4. The magnitude of rz and b varies significantly
between geographic regions such as (a) different
continents or oceans and (b) the hemispheres.
Hypothesis 5. The magnitude of rz and b varies significantly
(a) between terrestrial, freshwater, and marine
realms and (b) between habitat types.
Methods
Data
Four abstracting services were searched: Cambridge Biological
Abstracts (1988­2002), ISI Web of Science (1986­
2002), JSTOR (until 1998), and the Aquatic Science and
Fisheries Abstracts (1978­2002). The search strings were
"latitudinal gradient," "latitude AND diversity," and "latitude
AND species richness." More recent papers were
included if present and also papers derived from the bibliographies
of the papers screened. To broaden the basis
of this analysis and to reduce the impact of publication
bias, I also included studies that were not originally designed
to test for latitudinal gradients but that reported
diversity measures across different latitudes. The search
revealed more than 1,000 studies, which were checked for
the following criteria: measurement of diversity of a defined
target organism group at a minimum of three locations
spanning a minimum latitude range of 10 . Two
hundred thirty-two studies passed these criteria, reporting
a total of 581 gradients (see app. A in the online edition
of the American Naturalist). To calculate effect sizes, I obtained
the correlation coefficient r, the slope b, its standard
error (SEb), the intercept a, and the number of observations
(N) from each gradient.
Effect Sizes
Effect size is the general term for the parameter used to
measure the effect of a treatment or a variable within each
study, that is, in the present case the effect of latitude on
diversity. The choice of relevant effect sizes is a critical
step in meta-analysis (Osenberg et al. 1999). Meta-analyses
allow one to obtain quantitative central tendencies in effect
sizes over a large number of studies as well as the analysis
of differences in the effect size between studies categorized
with respect to grouping variables (Gurevitch and Hedges
1993; Rosenberg et al. 2000). The usefulness of metaanalysis
is increased by accounting for the variance within
studies, which is preferably done by a weighted analysis,
where each effect size is weighted by the inverse of its
sampling variance (Gurevitch and Hedges 1993).
Here, I described the latitudinal gradient of diversity by
two effect sizes reflecting different aspects of the relation,
the strength and the steepness. Throughout the article, I
use strength to indicate the degree of variation of diversity
around latitude, reflected by the correlation coefficient.
Steepness is used to indicate how rapidly diversity declines
with latitude, reflected by the slope. The gradient strength
was defined as Fisher's z-transform of the correlation coefficient
r, rz, which is an established effect size in metaanalyses
(Rosenberg et al. 2000) and can be weighted by
a sampling variance :Varrz
194 The American Naturalist
Table 1: Variables characterizing each gradient with respect to organism type, habitat, geographic position, and measurement
details
Variable name L Groups (k) k
Thermoregulation 2 Ectotherm (440), homeotherm (139) 579
Body mass (log g) CON Range: 9.41 to 4.71; mean: .47; median: .15 581
Dispersal type 6 Flying (170), mobility (136), pelagic larvae (119), seeds (73), passive (68), transfer (8) 574
Trophic level 7 Omnivores (280), autotrophs (87), herbivores (65), carnivores (58), suspension (34),
microbivores (41), parasite (16)
581
Realm 3 Terrestrial (305), marine (204), freshwater (69) 578
Habitat 16 All terrestrial (234), benthos coastal (102), pelagic ocean (39), lake (36), deep sea
(34), forest (31), all freshwater (23), coastal pelagic (20), nonforest (18), biota
(10), stream (10), soil (6), human (5), islands (5), estuary (3), cave (2)
578
Hemisphere 3 N (335), S (180), both (64) 579
Longitude 7 New World (183), Atlantic (112), Eurafrica (74), Australasia (43), Pacific (22), Indian
Ocean (19), worldwide (128)
581
Scale 2 Regional (349), local (222) 571
Grain (log km2
) CON Range: 7 to 9.5; mean: 3.1; median 4.3 370
Diversity type 3 a (223), ranges (182), g (150) 555
Range CON Range: 10­90; mean: 44.0; median: 44 571
Significance 2 Yes (342), no (206) 548
Global richness (log) CON Range: 1.6­6; mean: 3.8; median: 3.9 575
Measure 5 S (415), Sst (70), mean S (28), index (25), higher (33) 581
Note: Variable names and the number of group levels (L) are given for each variable, name of group levels, and the numbers of gradients (k) therein.
The last column gives the total number of gradients for which information was obtained. Some categories requiring explanations are discussed in
appendix B of the online edition of the American Naturalist. variable, for which range, mean, and median are given.CON p continuous
1 1 r
r p # ln ,z ( )2 1 r
1
Var p .rz
N 3
The slope b of the linear regression of diversity on latitude
was used for gradient steepness with SEb as variance
estimate (Hillebrand et al. 2001). The two effect sizes, rz
and b, are not completely independent mathematically (A.
P. Allen, personal communication), but they correlated
rather weakly ( , ). The use of effect sizesr p 0.17 N p 540
from linear regressions is arguably a simplification, but
more complex regression models would decrease the comparability
of the parameters and render the analysis
impossible.
Grouping Variables
Each gradient was classified with respect to 14 variables
characterizing the organism group, the habitat type, the
geography, the scale, and the analytical details of each
study. The definition of the categories aimed at a high level
of standardization and reproducibility combined with a
high level of retained information. The concise allocation
of categories to each gradient is revealed in the enhancement
material in the online edition of the American Naturalist
(app. A). Table 1 summarizes the 14 variables and
their group levels, whereas the concise definition for each
level is given in appendix B of the online edition of the
American Naturalist.
Organism types were characterized by their mode of
thermoregulation, their body mass, their dispersal type,
and their trophic level. Body mass estimates were obtained
from the original contributions or from the literature (Peters
1983; Wetzel and Likens 1991). The habitats were
characterized by realm and habitat type. Two categorical
variables characterized the geographic position of the gradient,
hemisphere and longitude (table 1).
Spatially, I characterized the scale as either regional or
local (see below) and noted the sampling grain (mean
sampled area) as continuous variable (log km2
). The scale
was local if diversity was assessed for certain sampling
locations (defined sampling site) or certain habitats (single
forests, single lakes), classically denoted a-diversity (Loreau
2000). The scale was regional if diversity was measured
for latitudinal grid squares (from up to1 # 1
) or for larger regions covering many habitats5 # 5
(countries, biomes). Whereas the latter represents classical
g-diversity, the superposition of range maps on a grid
system introduces an additional moment of spatial autocorrelation.
Therefore, a third spatial variable (diversity
type) was introduced to differentiate between a-diversity,
g-diversity, and range-based diversity (table 1). Local and
regional scales were characterized by widely differing sam-
Latitudinal Gradients 195
pling grains (log-transformed area; : local,mean SD
; regional, ).1.61 3.18 4.73 1.25
Methodological differences between studies were analyzed
in four different variables: the latitudinal range of
the study, the significance of the original correlation of
diversity to latitude, the global species richness of the defined
group, and the measure of diversity used (table 1;
see app. B for definitions). Global biodiversity of the organism
group investigated was given in original studies or
obtained from recent biodiversity assessments (mainly
from Levin 2001).
The intention to show general patterns of the latitudinal
gradient would be counteracted if a single level within one
variable would dominate the analysis. This was generally
not the case: although the number of studies per level
varied tremendously, no categorical variable was entirely
dominated by a single level (table 1). Defining dominance
as the ratio of the number of studies in the most numerous
level to the following level, the highest dominance is found
for the variables diversity measure and trophic level. For
the latter, many groups were treated as omnivores because
they contain species that are herbivores, carnivores, or just
omnivores. To test whether the observed pattern with
trophic level was an artifact of the large number of omnivore
studies, I analyzed suspension feeders as a separate
group and found that they fit well between herbivores and
omnivores (see "Results").
Meta-analysis
Weighted meta-analysis on rz and b was used (Gurevitch
and Hedges 1993; Rosenberg et al. 2000; Hillebrand et al.
2001) to calculate the overall magnitude of the parameters,
called grand mean effect sizes, and their 95% confidence
intervals (CIs) using the bootstrapping procedure in
MetaWin 2.0 (Rosenberg et al. 2000). For all categorical
variables (table 1), weighted meta-analyses were used to
calculate groupwise effect sizes and their CIs. A random
effect model was used (Gurevitch and Hedges 1993; Rosenberg
et al. 2000; Hillebrand et al. 2001) to test whether
the grouping variable explained significant heterogeneity
in the effect sizes, whereby the overall heterogeneity in the
effect sizes was divided into heterogeneity explained by the
variable Qb and into residual heterogeneity Qw. Significance
levels for the analysis of heterogeneity were obtained by
9,999 randomizations. Significant difference between levels
within variables was indicated by nonoverlapping CIs. For
the continuous variables (table 1), a weighted regression
analysis was conducted in MetaWin 2.0 (Rosenberg et al.
2000).
No factorial meta-analysis technique is available to date
to account for interactions between the different variables.
These interactions are obviously present between several
variables, such as between scale and body mass, between
body mass and many other organism characteristics, but
also between trophic level, thermoregulation, and dispersal
(e.g., autotrophs are all ectotherm and often have seed
dispersal). I reduced the problem by separating the regional
and local scales (see "Results") and by reanalyzing
the regression of gradient parameters on body mass for
different organism groups. Moreover, standard statistics
allowing factorial combinations cannot replace metaanalysis
because they do not account for the variability of
the effect size describing each gradient (Gurevitch and
Hedges 1993).
Multiple use of data required adjustment of significance
levels, which was done using a Dunn-Sidak adjustment of
P, with , where of1/x
P p 1 (1 P) x p numberadj
and level (.05 or .01). Sig-tests p 15 P p significance
nificance level adjustments are known to be highly conservative
(Sokal and Rohlf 1995); therefore I will discuss
not only significant differences but also trends with
.P ! .1adj
Results
Overall Effect Sizes
The grand mean effect size for rz was negative ( 0.729)
and significantly different from 0 (fig. 1A). The total heterogeneity
indicated significant structure in the effect sizes
( , ).Q p 756.95 P ! .001tot
The overall significance of the correlation between latitude
and diversity was observed, although more than onethird
of the original studies did not show significant
relationships (table 1). Most interestingly, even the nonsignificant
studies showed a general negative tendency,
which differed significantly from 0 (fig. 1B). However,
there was a strong difference in strength between significant
and nonsignificant studies ( ,Q p 278.7 Q pb w
, , ).631.4 df p 1, 522 P ! .01adj
For rz, the actual measure of diversity did not significantly
affect the effect size ( ). For b, however, theP 1 .05adj
diversity measure significantly influenced the effect size
( ). Species richness (S) resulted in steeper slopesP ! .01adj
than any other diversity measure. Therefore, I restricted
the analyses involving b to studies measuring species
richness.
As for strength, the grand mean effect size for gradient
slopes was negative ( 1.039) and differed significantly
from 0 (fig. 1A). The total heterogeneity indicated significant
structure in the effect sizes ( ,Q p 3,720.77 P !tot
). The slopes showed the same difference between ef-.001
fect sizes reported from originally significant studies versus
nonsignificant studies ( , ,Q p 238.8 Q p 2,906.7 df pb w
196 The American Naturalist
Figure 1: Mean effect size ( 95% confidence intervals) for strength rz
(triangles) and slope b (diamonds) of the latitudinal gradient of diversity.
A, Overall effect size. B, Effect sizes for studies originally reporting significant
or nonsignificant gradients. C, Effect sizes for gradients on local
or regional scales.
, ). Again, even nonsignificant studies1, 365 P ! .01adj
showed a significant negative effect size for b (fig. 1B).
The differentiation between regional and local studies
had very strong impact on the effect sizes, both for rz and
b (fig. 1C). Regional gradients were significantly stronger
(rz; , , , ) andQ p 114.4 Q p 780.3 df p 1, 545 P ! .01b w adj
steeper (b; , , ,Q p 151.5 Q p 3,529.9 df p 1, 370 P !b w adj
than local gradients. Still, the local gradients were.01)
significantly negative for both effect sizes (fig. 1C). The
strong difference between regional and local gradients and
the interference of other variables with scale suggested that
the regional and local gradients be analyzed separately,
which was done for all forthcoming analyses. Moreover,
only regional data are presented graphically, while regional
and local effect sizes can be found electronically (see app.
C in the online edition of the American Naturalist).
The Strength of the Latitudinal Gradient
Body mass had consistent and strong effects on rZ at both
spatial scales (table 2; fig. 2A). Please note that the contrast
between highly significant regressions (table 2) and observed
variance in the data (fig. 2A) comes from the use
of weighted regressions. At both spatial scales, rz became
more negative; that is, the gradient became stronger with
increasing body mass of the organisms. Additionally, the
slopes of the weighted regression were highly similar for
regional and local gradients (table 3).
The strength of the latitudinal gradient did not generally
differ with thermoregulation (table 2). Ectotherm and homeotherm
organisms had remarkably similar effect sizes
at regional (fig. 3A) and local scales, respectively. Because
body mass and thermoregulation may be negatively correlated
(ectotherms tending to be smaller), I analyzed
whether the mode of thermoregulation changed the regression
between body mass and effect size (table 3). Obviously,
the stronger and steeper gradients at higher body
mass were related to ectotherm organisms alone, whereas
the homeotherms, covering a smaller range of body
masses, did not reveal a negative relationship between effect
size and body mass. At regional scales, the homeotherms
showed even a strong positive relation between
body mass and rz (table 3).
The dispersal types were not significantly different from
each other at either scale (table 2; fig. 3B). However, the
relation between body mass and rz was interlinked with
dispersal type (table 3). Organisms with passively distributed
dispersal stages (seeds, pelagic larvae) or passively
transported adults tended to have steeper relationships
between weight and rz than organisms able to move on
their own force (flying, mobility).
At regional scales, a very strong impact of trophic level
on rz was found (table 2; fig. 3C). Gradient strength consistently
increased with higher trophic level. Significant
differences were observed between autotrophs and herbivores
on the one hand and omnivores and carnivores
on the other. At the local scale, the impact of trophic level
was less pronounced and nonsignificant but still with most
negative gradients confined to carnivores (app. C). Two
more trophic groups were present: the microbivores had
gradient strengths similar to those of herbivores, whereas
parasites differed strongly between regional and local scales
(app. C).
The longitudinal position of the gradient affected rz sig-
Latitudinal Gradients 197
Table 2: Analysis of heterogeneity for the strength (rz) of the latitudinal gradient of diversity at either regional
or local scales
Variable
Regional Local
dfb Qb dfw Qw Prand Padj dfb Qb dfw Qw Prand Padj
Body mass 1 8.2 334 537.3 .0011 ** 1 14.2 209 227.1 .0001 **
Thermoregulation 1 .03 332 542.5 .8848 NS 1 .1 209 234.5 .7436 NS
Dispersal type 5 19.2 328 539.6 .0337 NS 5 9.7 201 227.7 .1097 NS
Trophic level 6 60.7 329 508.7 .0001 ** 6 14.9 204 224.6 .0337 NS
Longitude 6 193.2 329 550.7 .0001 ** 6 11.8 204 235.7 .0835 NS
Hemisphere 2 8.1 332 536.5 .0782 NS 2 2.4 207 224.1 .3187 NS
Realm 2 34.3 331 531.0 .0001 ** 2 9.8 207 223.0 .0108 NS
Habitat 10 80.1 323 493.5 .0001 ** 14 58.3 194 231.6 .0002 **
Grain 1 2.5 264 464.2 .6858 NS 1 .1 89 83.8 .1398 NS
Range 1 37.9 329 575.3 .9999 NS 1 13.2 204 242.9 .8235 NS
Global richness 1 .08 272 449.8 .3693 NS 1 .15 120 138.5 .2407 NS
Note: For each categorical or continuous grouping variable, the results of the analysis are given with degrees of freedom (df) and
heterogeneity (Q) between (b) and within (w) group levels. Significance levels were obtained by 9,999 randomizations (Prand) and are
adjusted to correct for multiple use of data (Padj); significant.NS p not
** .P ! .01
nificantly (table 2). For marine habitats, regional gradients
were stronger in the Atlantic than any other ocean (fig.
4A), and local gradients were weaker in the Indian Ocean
than in the Atlantic (app. C). For terrestrial and freshwater
gradients, the New World harbored the strongest gradients
at regional scales (fig. 4A). In contrast to longitude, there
was no difference in gradient strength between the
Northern and Southern Hemispheres (table 2), which
showed almost identical effect sizes at both scales (fig. 4B).
Marine and terrestrial studies showed stronger gradients
than freshwater studies (fig. 4C), resulting in a significant
difference between realms at the regional scale and a similar
but nonsignificant trend at the local scale (table 2). A
finer typification of habitats explained significant heterogeneity
in effect sizes at both spatial scales (table 2). At
regional scales, pelagic marine habitats had very strong
gradients (fig. 5A), matched only by gradients from the
deep sea benthos. Weak regional gradients were confined
to freshwaters, lakes as well as streams, whereas there was
little variation between terrestrial habitats. The difference
between coastal and deep sea benthos was consistent at
local scales (app. C), where also significantly weaker gradients
in lakes than in streams were observed. In the terrestrial
realm, caves and biota (for parasites) revealed
weak gradients, with minor variation between the other
gradients.
The study grain (mean area used to assess diversity) did
not affect rz significantly, neither at regional nor at local
scales (table 2). However, the differences between local
and regional gradient strength were mirrored by the differences
between rz measured at small and large areas (fig.
6A). The latitudinal range of the study did not affect rz
(table 2; fig. 6B). The strength was not influenced by the
global species richness of the target organism group (table
2; fig. 6C).
The Slope of the Latitudinal Gradient
The slope of the gradient showed less significant variation
with grouping variables than the strength of the gradient
(table 4), due to larger variation (larger heterogeneity)
resulting in larger CIs for b compared to rz. However, many
patterns were similar for both effect sizes.
As for rz, significantly steeper (i.e., more negative) b
with increasing body mass was observed at both spatial
scales (table 4; fig. 2B). The parameters of the weighted
regression between body mass and effect size b were also
highly consistent at both spatial scales (table 3).
There was no difference in b between ectotherm and
homeotherm organisms (table 4; fig. 3D), but a strong
difference in the body mass­effect size regression, which
was strongly negative for ectotherms, but positive for homeotherms
(table 3). The reversal of the regression for
homeotherms was observed at both spatial scales.
Neither dispersal types (fig. 3E) nor trophic levels (fig.
3F) explained significant variation between slopes (table
4). The relation between effect size and trophic level for
slopes resembled the decreasing trend observed for gradient
strength (cf. fig. 3C, 3F) between autotrophs and
omnivores, but slopes were considerably flatter for carnivores.
Differences in the regression of effect size on body
mass were observed with different dispersal types (table
3), but the interaction changed strongly with spatial scale.
No clear difference between passively and actively dispersing
organisms was observed in b as for rz.
Longitude did not explain significant variation between
198 The American Naturalist
Figure 2: Effect size for (A) strength (rz) and (B) slope (b) of the latitudinal gradient of diversity in relation to body mass. Open symbols, local
gradients; solid symbols, regional gradients.
slopes (table 4), which is due to the high variability in
some groups (e.g., Indian Ocean, fig. 4D). It should be
noted, though, that the pattern on land resembles very
much the pattern found for gradient strength, with largest
effect sizes in the New World, and weak effect sizes in
Eurafrica and Australasia. For the oceans, however, the
pattern is different, with similar effect sizes for the Atlantic
and Pacific Oceans and steeper but very variable gradients
in the Indian Ocean. As for rz, the slope of the gradient
did not differ between hemispheres (table 4; fig. 4E).
Different realms did not explain significant variation in
b at regional scales (table 4), but the variation between
mean effect sizes showed the same trend as observed for
rz (fig. 4F). At local scales, terrestrial gradients were significantly
steeper than both other gradients (app. C). Different
habitat types were significantly different from each
other (table 4). For freshwaters and terrestrial gradient,
the pattern resembled the differences for gradient strength
(fig. 5B). Freshwater mean effect sizes for b were small at
regional scales, with a significant difference between stream
and lakes at local scales. Terrestrial gradients showed minor
variation at regional scales and small effect sizes for caves
and biota at local scales (app. C). The difference between
rz and b with regard to habitats became obvious in marine
environments. Both in the benthos and in the pelagial,
effect sizes were larger (more negative) in coastal areas
than in the open ocean or deep sea (fig. 5B).
Increasing the grain of the study consistently decreased
b, i.e., steepened the slope, at both spatial scales (fig. 6D).
The relation was significant at the local scale and marginally
nonsignificant at the regional scale (table 4). Additionally,
the grain variable well reflected the general difference
in effect size between local and regional studies
(fig. 6D). Also the range of latitudes studied affected the
gradient slope significantly (table 4; fig. 6E), but with different
signs at the different scales. Regionally, b became
steeper with increasing range (slope of weighted
), but locally b became less steepregression p 0.0017
(0.0061).
In contrast to gradient strength, b was closely related to
Latitudinal Gradients 199
Table 3: Regression results for the relation of effect size (rz or b) on body mass for different data
sets at either regional or local scale
Data and scale
rz b
Intercept Slope k Intercept Slope k
All:
Regional .908 (.032) .039 (.014) 335 1.330 (.045) .097 (.018) 254
Local .474 (.043) .043 (.012) 210 .598 (.048) .078 (.012) 116
Thermoregulation:
Ectotherm:
Regional .944 (.045) .059 (.019) 237 1.320 (.052) .133 (.022) 181
Local .532 (.050) .055 (.012) 178 .660 (.057) .093 (.014) 102
Homeotherm:
Regional 1.231 (.125) .146 (.055) 95 3.265 (.266) .683 (.105) 70
Local .283 (.470) .079 (.217) 31 .964 (.650) .152 (.303) 13
Dispersal type:
Pelagic larvae:
Regional 1.312 (.115) .205 (.056) 56 1.665 (.133) .275 (.072) 52
Local .675 (.129) .131 (.046) 51 .350 (.100) .073 (.027) 29
Flying:
Regional .846 (.050) .062 (.028) 114 2.063 (.119) .143 (.064) 83
Local .470 (.104) .022 (.052) 42 1.100 (.238) .032 (.113) 14
Mobility:
Regional .900 (.097) .076 (.038) 94 1.297 (.180) .023 (.067) 71
Local .421 (.109) .083 (.051) 33 .535 (.168) .146 (.076) 14
Passive:
Regional 2.829 (.664) .366 (.126) 14 .693 (.386) .067 (.092) 11
Local .857 (.294) .097 (.047) 50 .083 (.297) .048 (.051) 29
Seeds:
Regional .540 (.150) .059 (.068) 48 .202 (.388) .820 (.175) 28
Local .435 (.132) .114 (.080) 21 .871 (.300) .224 (.151) 9
Note: Table presents intercept and slope ( SE) from the weighted regression as well as the number of gradients
involved (k).
the global species richness of the target organism group.
For groups with higher global species richness, b was
steeper (fig. 6F) at both spatial scales. The relation was
significant only at regional scales (table 4).
Discussion
The Generality of the Latitudinal Gradient
The present analysis is the first truly general assessment
of the latitudinal gradient of diversity. The eukaryotic organisms
involved represent a large variety of functional
and taxonomic groups (app. A) and comprise a range of
14 orders of magnitude in body mass and of five orders
of magnitude in global diversity (table 1). The habitats
represent all major biotic realms on the planet and many
major habitat types, spread over all continents and oceans.
Moreover, the studies assessed local or regional diversity
on spatial scales ranging from less than 1 m2
to more than
1 million km2
and covered latitude ranges between 10
and 90 .
The first major conclusion from this general analysis is
therefore that the latitudinal decline of diversity is a ubiquitous
phenomenon (supporting hypothesis 1). The overall
effect sizes for both rz and b were significantly negative.
Similarly, most groupwise effect sizes for strength or slope
were negative and significantly different from 0, especially
at regional scales (figs. 3­5).
Even those studies not reporting significant declines of
diversity with latitude showed an overall significant negative
trend of diversity toward the pole. Obviously, the
majority of studies not detecting a diversity decline failed
to do so because of the limited statistical power of the
study. For the studies classified nonsignificant, the mean
number of observations was smaller than in significant
studies (mean N: 28 vs. 56) and the latitudinal range was
narrower (mean range: 37 vs. 47 ). It is an advantage of
meta-analyses that significant central tendencies can even
be observed even when there is limited statistical power
within single studies or effects are small (Francoeur 2001).
Nevertheless, there were some studies ( , or 4.3%k p 25
of all studies) showing significant positive relations be-
Figure 3: Mean effect size ( 95% confidence intervals) for (A­C) strength (rz) and (D­F) slope (b) of the regional latitudinal gradient of diversity.
A, D, Effect sizes for ectotherm and homeotherm organisms. B, E, Effect sizes for organisms with different dispersal types: pelagic larvae, flying,
mobility on ground or in water, passively transported adults, seeds, and parasitic transfer. C, F, Effect sizes for organisms with different trophic
levels: autotrophs, herbivores, suspension feeders, omnivores, carnivores.
Latitudinal Gradients 201
Figure 4: Mean effect size ( 95% confidence intervals) for (A­C) strength (rz) and (D­F) slope (b) of the regional latitudinal gradient of diversity.
A, D, Effect sizes for different continents or oceans: Atlantic, Pacific, Indian Ocean, New World, Australasia, Eurafrica, worldwide. B, E, Effect sizes
for different hemispheres: Northern, Southern, both. C, F, Effect sizes for different realms.
tween latitude and diversity. These studies mainly represent
exceptions in organism groups showing overall strong
latitudinal declines such as molluscs (Valdovinos et al.
2003) or birds (Rabenold 1979; see app. A for more details).
Aquatic macrophytes as a group tend to show significant
diversity increases toward the poles (Crow 1993;
Santelices and Marquet 1998). The gradients for ichneumonids,
which often were regarded as exceptional (Owen
202 The American Naturalist
Figure 5: Mean effect size ( 95% confidence intervals) for (A) strength (rz) and (B) slope (b) of the regional latitudinal gradient of diversity. Effect
sizes are for the following habitats: pe co, pelagic coast; pe oc, pelagic ocean; be co, benthic coastal; be ds, benthic deep sea; lake, lakes; str, streams;
all fw, all freshwater; for, forests; non fo, nonforest; all te, all terrestrial; and biota, biota.
and Owen 1974; Janzen 1981), were either positive or
negative and did not differ significantly from 0 (mean
gradient ; 95% tostrength p 0.1141 CI p 1.2407
1.0126). These few positive gradients remain exceptional,
though, and the latitudinal gradient of diversity is highly
general and arguably the most common statistical pattern
of spatial variation in diversity. Despite the generality, however,
the gradient is not uniform. The effect sizes show
high heterogeneity and significant variation in gradient
parameters between scales, geographic regions, realms,
habitat types, organism types, and with organism body
mass. I will discuss these variations before analyzing the
consequences of these patterns for models aiming to explain
the latitudinal decline of diversity.
Scale
Regional gradients of latitudinal diversity decline were
consistently stronger and steeper than local gradients (supporting
hypothesis 2a; fig. 7). On a highly general level,
this analysis corroborated findings for single groups such
as bats (Stevens and Willig 2002) or shallow-water gastropods
(Spight 1977). At the same time, the mean effect
sizes for b and rz were still significantly negative on local
scales. The strong scaling effect on both effect sizes has
two major implications. First, the decline of diversity toward
the pole mainly represents the decline in numbers
of regionally coexisting species (g-diversity). Second, the
weaker but still significant effect sizes on local scales indicates
that local diversity generally relates to the regional
diversity but that local interactions may modify diversity
at the community level (Stevens and Willig 2002).
These findings have important bearings for the discussion
of regional imprints on local species richness (Cornell
and Lawton 1992; Lawton 1999; Srivastava 1999; Hillebrand
and Blenckner 2002). Statistical relationship between
regional and local diversity were used to indicate
the relative importance of regional (historical) factors versus
local ecological interactions for local diversity (Cornell
and Lawton 1992; Lawton 1999; Srivastava 1999). These
Latitudinal Gradients 203
Figure 6: Effect size for (A­C) strength (rz) and (D­F) slope (b) of the latitudinal gradient of diversity in relation to (A, D) the grain of sampling,
(B, E) the range of latitudes studied, and (C, F) the global species richness of the organisms studied. Open symbols, local gradients; solid symbols,
regional gradients.
statistical relationships have been criticized for their strong
bias toward spatial autocorrelation (Hillebrand and
Blenckner 2002). Still, recent studies combining manipulation
of ecological interactions with analysis of the regional
species pool concluded on a simultaneous importance
of both factors (Shurin et al. 2000). On the global
scale of the present analysis, the patterns found for regional
and local diversity support this idea. The significant local
204 The American Naturalist
Table 4: Analysis of heterogeneity for the slope (b) of the latitudinal gradient of diversity at either regional or
local scales
Variable
Regional Local
dfb Qb dfw Qw Prand Padj dfb Qb dfw Qw Prand Padj
Body mass 1 27.5 253 2,871.6 .0001 ** 1 43.1 115 507.0 .0001 **
Thermoregulation 1 7.6 251 2,818.1 .2986 NS 1 2.9 115 570.9 .3817 NS
Dispersal type 5 81.1 247 2,621.0 .0477 NS 5 66.3 110 478.7 .0173 NS
Trophic level 6 113.9 248 2,640.8 .0210 NS 6 33.4 110 448.1 .1859 NS
Longitude 6 47.4 248 2,625.4 .2989 NS 6 49.4 110 424.9 .0922 NS
Hemisphere 2 17.3 251 2,923.7 .2834 NS 2 1.4 113 500.6 .8429 NS
Realm 2 42.3 250 2,859.3 .0440 NS 2 49.4 114 505.5 .0016 *
Habitat 10 150.2 242 2,641.1 .0425 NS 13 104.8 102 422.3 .0597 NS
Grain 1 .9 191 2,194.3 .0052 1 1.5 52 190.6 .0001 **
Range 1 .6 252 2,915.6 .0001 ** 1 8.1 115 532.0 .0001 **
Global richness 1 252.5 253 2,756.8 .0001 ** 1 16.7 112 485.1 .0086 NS
Note: For each categorical or continuous grouping variable, the results of the analysis are given with degrees of freedom (df) and
heterogeneity (Q) between (b) and within (w) group levels. Significance levels were obtained by 9,999 randomizations (Prand) and are
adjusted to correct for multiple use of data (Padj); significant.NS p not
.P ! .1
* .P ! .05
** .P ! .01
gradients reflect the effect of regional diversity on the number
of locally coexisting species, whereas the differences
in gradient strength and slopes reflect the modification of
local richness patterns by ecological interactions. A central
future task in macroecology will therefore be to understand
systematic variation in the relation between a- and gdiversity
(b-diversity).
A possible systematic difference between local and regional
gradients is the use of range maps to assess diversity
at regional scales. The actual method to convert range
maps into regional diversity differed between studies, but
often gaps in range distributions were interpolated, increasing
the spatial autocorrelation in the diversity data
(W. Jetz, personal communication). The difference between
scales could thus be due to the difference in diversity
assessment rather than real scale differences. However, this
concern is probably negligible: regional diversity was assessed
not only by range maps ( gradients) butk p 182
also by independent estimates of diversity within regions,
such as countries or provinces ( ). Restricting thek p 150
analysis of scale to the latter regional studies still reveals
weaker local ( , mean effect , 95%k p 201 size p 0.39
to 0.3068) than regional gradientsCI p 0.4623
( , mean effect , 95% CI pk p 150 size p 0.6535
0.7584 to 0.5484). The difference was significant for
rz ( ) and also for the difference between regionalP ! .01adj
and local slopes ( ).P ! .01adj
The grain of the studies, measured as log-transformed
"sampling" area, varied tremendously and mirrored the
difference between regional and local scales. Within these
predefined scale groups, however, the strength of the gradient
was not affected by sampling grain, whereas larger
sampling areas resulted in steeper gradient slopes (fig. 7).
Previous studies found either invariance of the gradient
for bats and marsupials at regional scales ranging from
1,000 to 25,000 km2
(Lyons and Willig 1999) or less obvious
gradients at increasing spatial scales for hummingbirds
in South America (Rahbek and Graves 2000). Compared
to both these studies, the present analysis covers a
much broader range of sampling areas, almost 12 orders
of magnitude at the local scale and 6 orders of magnitude
at the regional scale. It may thus be concluded that the
sampling grain has no general effect on the strength of
local or regional latitudinal gradients, although it may be
important for single organism types at certain scales (Rahbek
and Graves 2000).
The range of latitudes covered had no effect (rz) or weak
effects (b) on the structure of the latitudinal gradient (refuting
hypothesis 2b; fig. 7). For b, the effects of range
width were moreover opposing for local and regional gradients.
Studies covering less than 10 latitude were excluded
from my analysis, although latitudinal differences
in diversity may be visible at even smaller range size (Gotelli
and Ellison 2002). Beyond the limit of 10 , no systematic
change in effect sizes occurred, which is astonishing
in light of the debate over how the spillover of high
tropical richness into adjacent subtropical areas (Blackburn
and Gaston 1997; Ruggiero 1999) or the thorough
inclusion of high polar diversity (Gray 2001) might affect
the latitudinal gradient.
The global species richness of the organism group influenced
the slope (fig. 7) but not the strength of the
Latitudinal Gradients 205
Figure 7: Synoptic summary of variable effects on the strength and slope of the latitudinal gradient. The graphs depict the trend of species richness
with latitude (from the equator to the poles) and the effects of 11 variables on the slope (shown as change in slope) and on the strength (shown
as change in line thickness). The slope of the latitudinal gradient was affected by the scale of the analysis, the sampling grain, the global richness
of the group investigated, the body mass of the organisms, and the realm. Moreover, the slope changed with latitudinal range, but with contrasting
signs at different spacial scales. The strength of the gradient differed with scale, trophic level, body mass, longitude, and the realm. Dispersal type
and thermoregulation indirectly influenced the body mass­effect size relations (arrows).
latitudinal gradient (partly supporting, partly refuting hypothesis
2c). Defining the target organism group more
broadly with respect to global species richness must obviously
affect the slope of the gradient because the global
species richness sets a limit for the steepness of the gradient.
More importantly, though, the correlation between
latitude and diversity was not significantly different if diversity
is analyzed for a small group of species (a family
or genus with less than 100 species) or for large classes or
phyla representing 10,000 to hundreds of thousands of
species.
Organism Features
Organism body mass had consistent impact on both
strength and slope of the gradient (supporting hypothesis
3a). The relation was significant at both spatial scales and
revealed almost identical quantitative relations between
body mass and effect sizes (fig. 7). The importance of body
mass for rz had previously been shown with a much smaller
database involving 150 gradients (Hillebrand and Azovsky
2001).
Despite the high generality of the pattern, it is less obvious
how it is generated. Before discussing possible mechanisms,
it should be noted that systematic errors such as
lower taxonomic knowledge of smaller organisms are not
able to explain the systematic variation of gradient parameters
with body mass. First, the decrease of gradient
strength and slopes is visible along the complete gradient
and remains significant when excluding unicellular and
small metazoan organisms. This was shown by reanalyzing
the body mass regressions for certain dispersal groups such
as flying metazoans (including only bats, birds, and insects),
seed-dispersing multicellular plants, or metazoans
with pelagic larvae. Second, many gradients for small organisms
originate from single authors assessing diversity
around the world, reducing the risk of systematic underestimation
of tropical diversity. Nevertheless, new tools
and refined taxonomy may have an impact on gradient
structure for least known organisms. It should also be
noted that the body mass impact on the gradient is significant
only over a wide range of sizes, whereas inverse
tendencies can be found within constrained body mass
ranges (Cardillo 2002).
It is well known, both from the renowned "Bergmann's
rule" found for mammals (Ashton et al. 2000; Freckleton
et al. 2003) and in the systematic variation of invertebrate
body size distributions with latitude (Gillooly and Dodson
2000; Roy et al. 2001), that body size within organism
groups can vary with latitude. By contrast, the present
analysis shows that the structure of the latitudinal gradient
varies systematically with body mass across organism
groups. Body mass clearly has major physiological and
ecological implications (Peters 1983) and is under strong
206 The American Naturalist
evolutionary control (Blanckenhorn 2000). Two aspects
systematically related to body mass are dispersal rates and
population density. Small organisms should have higher
dispersal by chance due to their high population sizes,
high transportability (Finlay et al. 1996; Hillebrand et al.
2001), and short generation times (Peters 1983; Gillooly
et al. 2002). The lower weight and generally higher abundance
of small organisms and their fast population growth
increase the chance to reach new habitats and establish
populations therein. For unicellular eukaryotes, local
abundance patterns mirror global abundance patterns
(Finlay 2002), and local community composition is less
affected by geographic distance than for larger organisms
(Hillebrand et al. 2001). These patterns suggest that small
organisms may show even distributions of diversity due
to rapid dispersal and random sampling of the global diversity
pool. Additionally, body mass may influence net
diversification due to less geographic separation and, often,
vegetative reproduction strategies, limiting the global species
pool (Fenchel 1993).
The dispersal hypothesis is partially supported by the
variation of the body mass regression with dispersal type.
Dispersal type had no direct effect on b and rz (refuting
hypothesis 3b; fig. 7), but for organisms with passive dispersal,
the regression between body mass and rz was steeper
than for organisms dispersing by own flight or mobility
(table 3). Dispersal type and body mass were thus clearly
interlinked, strengthening the idea that the body mass impact
on the latitudinal gradient is due to dispersal processes.
However, the variation was not consistent for both
spatial scales and both parameters b and rz.
Alternatively, body mass allometrically scales to metabolic
rates (Peters 1983; Enquist et al. 1998), development
time (Gillooly et al. 2002), and to population density (Allen
et al. 2002; Belgrano et al. 2002). Allen et al. (2002)
described the relation of mass-corrected population density
of ectotherms to temperature and the relation of temperature-corrected
population density of ecto- and homeotherms
to body mass.
In spite of the strong effect of thermoregulation on the
above mentioned relations, the present analysis showed
that gradient parameters for ectotherm organisms did not
differ from homeotherm organisms on any scale (refuting
hypothesis 3d; fig. 7). Still, possible differences in ectotherms
and endotherms may have been masked by covarying
differences in body mass. Reanalyzing the body
mass impact on the gradient parameters for both groups
separately revealed that ectotherms alone were responsible
for the pattern. Even a reversal of the regression between
body mass and b was seen for homeotherms (lower b with
increasing body mass). These results indicate an important
role of energy-related factors for the latitudinal gradients.
Allometric relations of population density to temperature
and body mass might indeed have strong consequences
for global species richness (Allen et al. 2002), but it remains
to be analyzed whether these relations can explain different
gradient structure between organisms with different body
mass.
An unprecedented result was the strong relation of regional
gradient strength to trophic level (supporting hypothesis
3c; fig. 7). The relation is not due to systematic
changes in body mass with trophic levels because autotrophs
included trees as well as algae, and carnivores included
chaetognaths as well as vertebrates. Several alternative
explanations are possible: First, the pattern could
be directly generated by the ecological interactions between
the trophic levels. However, the pattern was confined to
regional gradients, whereas the effect of trophic interactions
should be visible especially at local scales (Huston
1999). Second, the pattern may derive from a positively
reinforced correlation between diversity in adjacent
trophic levels, where higher diversity at basal levels founds
higher diversity in the next higher trophic levels. If this
were true, the "founding" effect has to be disproportionate;
that is, an increase in producer diversity must lead to an
even larger increase in consumer diversity. It has been
shown that diversity in adjacent trophic levels in fact might
be positively related (Siemann et al. 1999; Hawkins and
Porter 2003) but it has also been questioned whether this
correlation is causal (Hawkins and Porter 2003). To my
knowledge, a disproportionate effect of basal on top diversity
has not been shown so far. Third, the restriction
of the pattern to regional gradients points at variation in
range-size distributions and thus beta diversity over
trophic levels as a possible proximate explanation. The
final explanation involves statistical patterns of the food
web itself. Two seemingly disparate patterns are known:
higher species richness in food webs results in a higher
proportion of top-level species and lower proportion of
intermediate species (Martinez 1994), and the slope of the
species-area curve becomes steeper with increasing trophic
level (Holt et al. 1999). Both factors do not necessarily
hold on the scale of macroecology, and the general validity
of the scaling of top-level predators with food web size is
not without debate (Havens 1992; Martinez 1994). Still,
these findings indicate two processes possibly reinforcing
the latitudinal gradient at higher trophic levels, which
would disproportionally increase with the higher overall
richness and the higher biome area of the Tropics.
Geographic Position
The strength and slope of the latitudinal gradient varied
significantly with longitude (i.e., between continents or
oceans, supporting hypothesis 4a) but not between hemispheres
(refuting hypothesis 4b; fig. 7). On land, New
Latitudinal Gradients 207
World gradients were stronger (and slightly steeper) than
those in Eurafrica or Australasia. Interestingly, the narrow
Central American isthmus interrupting the American land
masses did not reduce gradient strength in contrast to the
predicted importance of area for the latitudinal gradient
(Rosenzweig 1995). Australia represents a diversity hotspot
for some aquatic organisms such as bivalves (Crame 2000)
but reveals often weak latitudinal gradients on land due
to adverse climatic conditions at low latitude (Schall and
Pianka 1978; Abensperg Traun and Steven 1997). Similarly,
Eurafrican longitudes are also characterized by arid areas
on the Southern Hemisphere. Additionally, the different
geological background of different oceans (Poore and Wilson
1993) or terrestrial regions (Qian and Ricklefs 2000)
has been evoked to explain their different regional species
richness. However, the present analysis showed weaker and
more variable gradients for the Pacific, especially at regional
scales, although the Pacific-Indian Oceans are older
than the Atlantic. Still, postglacial history may contribute
to the continental differences because different orientation
of major dispersal barriers in Europe versus America or
Asia have largely influenced the effect of and the recovery
from glaciation (Blondel and Mourer-Chauvire´ 1998). In
general terms, the Americas may represent uninterrupted
biome gradients in that north-south direction, whereas
transverse deserts and alpine barriers increase discontinuity
in Europe, Africa, and Australia.
Several authors postulated differences in diversity patterns
between Southern and Northern Hemispheres because
of different biome areas, different land : ocean size
ratios, different geological age, and different oceanography
(Gaston and Williams 1996; Wilson 1998; Gray 2001). On
a general level, no such difference was found in the present
analysis (fig. 7). Gradients on the Northern Hemisphere
are neither stronger nor steeper than on the Southern
Hemisphere. This does not contradict the findings that
gradients on the two hemispheres differ within single organism
groups, among others described for New World
birds (Gaston and Williams 1996) and for foraminiferans
(Culver and Buzas 2000). However, it is clearly not warranted
to envision latitudinal gradients as a primarily
northern phenomenon.
Realms and Habitats
The strength of the gradient was significantly different
between the major realms (partly supporting hypothesis
5a), and both strength and slope differed with habitat types
(supporting hypothesis 5b). Latitudinal diversity gradients
in freshwater were weaker and flatter than their terrestrial
or marine counterparts (fig. 7). Weak gradients in lakes
have been proposed quite early (Hubendick 1962), but the
difference had not been shown on this highly general level.
Possible explanations are manifold: in contrast to both
marine and terrestrial environment, the freshwater realm
is geographically isolated in small "freshwater islands"; that
is, even by joining local habitats into regions, the region
consists of rather isolated habitats. Furthermore, the body
mass distribution is skewed toward small organisms in
freshwater (this database, mean log g wet weight: 2.4)
compared to marine ( 1.8) or terrestrial (0.9) environments.
Additionally, as biome area presumably contributes
to the latitudinal gradient (Rosenzweig 1995), freshwaters
differ less in tropical versus temperate biome size because
the majority of large lakes and many large rivers are in
temperate regions (Wetzel 2001). The last argument becomes
interesting when comparing streams and lakes at
the local scale (app. C). Seemingly, the global distribution
of area for large rivers has a larger tropical peak than the
distribution for large lakes (Wetzel 2001), and actually on
local scales the gradients for running waters were both
stronger and steeper than for lakes.
In contrast to freshwater, marine gradients were as
strong or slightly stronger compared to terrestrial gradients.
This result is remarkable because Clarke (1992) asked
only a decade ago if general latitudinal gradients can be
found in the sea (a more detailed account of marine gradients
will be published separately). The marine gradients
also showed substantial variation between habitats, with
strong gradients in the pelagic ocean and in the deep sea.
The strong pelagic gradients oppose the idea that higher
spatial heterogeneity in the Tropics is crucial for the latitudinal
diversity gradient. Also within the terrestrial
realm, habitats with high architectural complexity (forests)
did not show gradients differing from grasslands, wetland,
or other less structured habitats. Recent analyses emphasized
the importance of heterogeneity (O'Brien et al. 2000;
Diniz et al. 2002), with special emphasis on topography
or elevational heterogeneity (Rahbek and Graves 2001;
Ruggiero 2001).
Conclusion and Outlook
This article constitutes the first general test for consistent
quantitative variation in latitudinal gradients across a large
variety of organisms and habitats. The analysis corroborated
that the latitudinal gradient of diversity is a highly
general spatial pattern of diversity with very few notable
exceptions. At the same time, the strength and the slope
of the gradient show significant variation with the scale
of the analysis and with factors describing features of the
organism (body mass, trophic level) and the habitat (geographic
position, realm, and habitat type). Latitudinal gradients
are mainly a regional phenomenon, which are less
strong in freshwaters than in terrestrial or marine environments.
The strength and the slope of the gradient in-
208 The American Naturalist
crease with body mass, and the strength increases with
trophic level. The relation between effect size (strength or
slope) and body mass differs with dispersal mode and
between homeotherm and ectotherm organisms.
These results do not serve to directly falsify or support
the models proposed to explain the latitudinal diversity
gradients, but the presence of strong statistical patterns
has important implications for the assumptions and predictions
made by the different models.
The observed generality of the latitudinal gradient and
the variation with scale do not conform to explanations
involving many different local processes, which mainly act
at local scales (Huston 1999). Ecological interactions such
as competition or consumption (including parasitism)
clearly can regulate local diversity (Huston 1999; Shurin
et al. 2000; Worm et al. 2002), but the importance for the
latitudinal gradient has to be questioned. This analysis thus
adds to recent studies failing to find a consistent change
in ecological interactions with latitude (Lambers et al.
2002; Ollerton and Cranmer 2002) or concluding that
large-scale diversity patterns are not due to variations in
community richness (Collins et al. 2002; Stevens and Willig
2002).
Mid-domain models predict the generality of the gradient
without involving biological factors (Colwell and
Hurtt 1994; Colwell and Lees 2000). The generality of the
gradient observed in the present analysis supports the idea
of an inevitable poleward decline of diversity. The middomain
models were criticized for the structure of the null
model (Bokma et al. 2001; Diniz et al. 2002) and for the
restricted importance of mid-domain effects for narrowranging
species (Jetz and Rahbek 2002), which may account
for a large proportion of the tropical diversity. The
present analyses additionally pose the question whether
the observed differences in gradient strengths and slopes
are congruent with null model predictions. Does random
placement of geographic ranges only set a general tendency
toward higher tropical species richness whereas the
strength and slope of the trend are controlled by differences
in organisms and habitats? To answer this question,
mid-domain models would have to be calculated for different
frequency distributions of species ranges and range
shapes, which are related to organism and habitat characteristics,
and compared to the real data set published
here.
The global distribution of area (Terborgh 1973; Rosenzweig
1995), energy (Wright 1983; Currie 1991; Allen et
al. 2002), and spatial heterogeneity (topography; O'Brien
et al. 2000; Rahbek and Graves 2001; Ruggiero 2001; Diniz
et al. 2002) varies tremendously with latitude. These factors
have all been proposed to explain (parts of) the latitudinal
diversity decline, resulting in an ongoing debate
on the causal mechanism (Rohde 1992; Rosenzweig 1995;
Huston 1999; Gaston and Blackburn 2000). The debate
has not produced a consensus whether there is a single
primary cause (Rohde 1992) or whether these factors act
in combination (Gaston and Blackburn 2000). Partly, this
failure is due to a lack of consistency in defining these
factors, in defining the processes by which these factors
operate, and in defining the scales on which they operate.
From the present analysis, an interaction of energy- and
area-related processes is supported, because many significant
factors relate to differences in energy acquisition and
processing (body mass, trophic level) or in area (realms
and habitats) or both (continents). Additionally, the generality
of the pattern strongly suggests that the main acting
principles are very basic, giving priority to simple allometric
rules such as relations of species richness to temperature
(Allen et al. 2002).
The effective evolutionary time hypothesis assumes that
tropical areas accumulated most species because speciation
increases with the gradients in area, energy, and spatial
heterogeneity, reinforced by the longer geological stability
of the Tropics (Rohde 1992). Concurrently, it has been
observed that long-term climatic oscillations such as Milankovich-cycles
reduce persistent cladogenesis at higher
latitudes (Jansson and Dynesius 2002). The hypothesis fits
well with observations from fossil records showing that
the tropical areas are centers of evolutionary novelty
(Crane and Lidgard 1989; Jablonski 1993) with higher evolutionary
speed than temperate areas (Cardillo 1999) and
a higher proportion of species without fossil records (i.e.,
from recent diversification) and with restricted distribution
(Buzas and Culver 1999). Moreover, higher tropical
diversity of rhizopods could be related to higher increase
in diversity through geological time in the Tropics than in
temperate regions (Buzas et al. 2002). The study by Buzas
et al. (2002) suggests the importance of a general comparison
of the recent gradients analyzed here to information
on fossil gradients. The nonequilibrium approach
of the effective evolutionary time hypothesis also requires
mechanisms preventing the equalization of diversity during
evolutionary time. The body mass impact observed in
this analysis might indicate the importance of dispersal
limitation for the maintenance of the diversity gradient.
Acknowledgments
This article profited from interactions with many colleagues
commenting, discussing, and correcting earlier
drafts: A. P. Allen, A. I. Azovsky, A. Belgrano, P. Eklo¨v, A.
Fitter, W. Jetz, H. K. Lotze, B. Malmqvist, U. Sommer, L.
Tranvik, and B. Worm. The article was improved by comments
from R. K. Colwell and one anonymous reviewer.
S. Floeder and U. Sommer helped with the synoptic figure.
Financial support is acknowledged from a Marie Curie
Latitudinal Gradients 209
Individual Fellowship, the Institute for Marine Research
Kiel, and the Swedish National Research Council.
Literature Cited
Abensperg Traun, M., and D. Steven. 1997. Latitudinal
gradients in the species richness of Australian termites
(Isoptera). Australian Journal of Ecology 22:471­476.
Allen, A. P., J. H. Brown, and J. F. Gillooly. 2002. Global
biodiversity, biochemical kinetics, and the energeticequivalence
rule. Science 297:1545­1548.
Ashton, K. G., M. C. Tracy, and A. de Queiroz. 2000. Is
Bergmann's rule valid for mammals? American Naturalist
156:390­415.
Belgrano, A., A. P. Allen, B. J. Enquist, and J. F. Gillooly.
2002. Allometric scaling of maximum population density:
a common rule for marine phytoplankton and terrestrial
plants. Ecology Letters 5:611­613.
Blackburn, T. M., and K. J. Gaston. 1997. The relationship
between geographic area and the latitudinal gradient in
species richness in New World birds. Evolutionary Ecology
11:195­204.
Blanckenhorn, W. U. 2000. The evolution of body size:
what keeps organisms small? Quarterly Review of Biology
75:385­407.
Blondel, J., and C. Mourer-Chauvire´. 1998. Evolution and
history of the western Palaearctic avifauna. Trends in
Ecology & Evolution 13:488­492.
Bokma, F., J. Bokma, and M. Mo¨nkko¨nen. 2001. Random
processes and geographic species richness patterns: why
so few species in the north? Ecography 24:43­49.
Buzas, M. A., and S. J. Culver. 1999. Understanding regional
species diversity through the log series distribution
of occurrences. Diversity and Distributions 8:
187­195.
Buzas, M. A., L. S. Collins, and S. J. Culver. 2002. Latitudinal
difference in biodiversity caused by higher tropical
rate of increase. Proceedings of the National Academy
of Sciences of the USA 99:7841­7843.
Cardillo, M. 1999. Latitude and rates of diversification in
birds and butterflies. Proceedings of the Royal Society
of London B 266:1221­1225.
------. 2002. Body size and latitudinal gradients in regional
diversity of New World birds. Global Ecology and
Biogeography 11:59­65.
Clarke, A. 1992. Is there a latitudinal diversity cline in the
sea? Trends in Ecology & Evolution 7:286­287.
Collins, S. L., S. M. Glenn, and J. M. Briggs. 2002. Effect
of local and regional processes on plant species richness
in tallgrass prairie. Oikos 99:571­579.
Colwell, R. K., and G. C. Hurtt. 1994. Nonbiological gradients
in species richness and a spurious Rapoport effect.
American Naturalist 144:570­595.
Colwell, R. K., and D. C. Lees. 2000. The mid-domain
effect: geometric constraints on the geography of species
richness. Trends in Ecology & Evolution 15:70­76.
Cornell, H. V., and J. H. Lawton. 1992. Species interactions,
local and regional processes, and limits to the
richness of ecological communities: a theoretical perspective.
Journal of Animal Ecology 61:1­12.
Crame, J. A. 2000. Evolution of taxonomic diversity gradients
in the marine realm: evidence from the composition
of recent bivalve faunas. Paleobiology 26:188­
214.
Crane, P. R., and S. Lidgard. 1989. Angiosperm diversification
and paleolatitudinal gradients in Cretaceous floristic
diversity. Science 246:675­678.
Crow, G. E. 1993. Species diversity in aquatic angiosperms:
latitudinal patterns. Aquatic Botany 44:229­258.
Culver, S. J., and M. A. Buzas. 2000. Global latitudinal
species diversity gradient in deep-sea benthic foraminifera.
Deep-Sea Research Part I Oceanographic Research
Papers 47:259­275.
Currie, D. J. 1991. Energy and large-scale patterns of
animal- and plant-species richness. American Naturalist
137:27­49.
Currie, D. J., A. P. Francis, and J. T. Kerr. 1999. Some
general propositions about the study of spatial patterns
of species richness. Ecoscience 6:392­399.
Darwin, C. 1859. The origin of species by means of natural
selection. J. Murray, London.
Diniz, J. A. F., C. E. R. de Sant'Ana, M. C. de Souza, and
T. F. L. V. B. Rangel. 2002. Null models and spatial
patterns of species richness in South American birds of
prey. Ecology Letters 5:47­55.
Enquist, B. J., J. H. Brown, and G. B. West. 1998. Allometric
scaling of plant energetics and population density.
Nature 395:163­165.
Fenchel, T. 1993. There are more small than large species?
Oikos 68:375­378.
Finlay, B. J. 2002. Global dispersal of free-living microbial
eukaryote species. Science 296:1061­1063.
Finlay, B. J., G. F. Esteban, and T. Fenchel. 1996. Global
diversity and body size. Nature 383:132­133.
Francoeur, S. N. 2001. Meta-analysis of lotic nutrient
amendment experiments: detecting and quantifying
subtle responses. Journal of the North American Benthological
Society 20:358­368.
Freckleton, R. P., P. H. Harvey, and M. Pagel. 2003. Bergmann's
rule and body size in mammals. American Naturalist
161:821­825.
Gaston, K. J., and T. M. Blackburn. 2000. Pattern and
process in macroecology. Blackwell Scientific, Oxford.
Gaston, K. J., and P. H. Williams. 1996. Spatial patterns
in taxonomic diversity. Pages 202­229 in K. Gaston, ed.
210 The American Naturalist
Biodiversity: a biology of numbers and difference.
Blackwell Scientific, Cambridge, Mass.
Gillooly, J. F., and S. I. Dodson. 2000. Latitudinal patterns
in the size distribution and seasonal dynamics of New
World, freshwater cladocerans. Limnology and Oceanography
45:22­30.
Gillooly, J. F., E. L. Charnov, G. B. West, V. M. Savage,
and J. H. Brown. 2002. Effects of size and temperature
on developmental time. Nature 417:70­73.
Gotelli, N. J., and A. M. Ellison. 2002. Biogeography at a
regional scale: determinants of ant species density in
New England bogs and forests. Ecology 83:1604­1609.
Gray, J. S. 2001. Antarctic marine benthic biodiversity in
a world-wide latitudinal context. Polar Biology 24:633­
641.
Gurevitch, J., and L. V. Hedges. 1993. Meta-analysis: combining
the results of independent experiments. Pages
378­398 in S. M. Scheiner and J. Gurevitch, eds. Design
and analysis of ecological experiments. Chapman &
Hall, New York.
Havens, K. 1992. Scale and structure in natural food webs.
Science 257:1107­1109.
Hawkins, B. A., and E. E. Porter. 2003. Does herbivore
diversity depend on plant diversity? the case of California
butterflies. American Naturalist 161:40­49.
Hillebrand, H., and A. I. Azovsky. 2001. Body size determines
the strength of the latitudinal diversity gradient.
Ecography 24:251­256.
Hillebrand, H., and T. Blenckner. 2002. Regional and local
impact on species diversity: from pattern to processes.
Oecologia (Berlin) 132:479­491.
Hillebrand, H., F. Watermann, R. Karez, and U. G. Berninger.
2001. Differences in species richness patterns
between unicellular and multicellular organisms. Oecologia
(Berlin) 126:114­124.
Holt, R. D., J. H. Lawton, G. A. Polis, and N. D. Martinez.
1999. Trophic rank and the species-area relationship.
Ecology 80:1495­1505.
Hubendick, B. 1962. Aspects on the diversity of freshwater
fauna. Oikos 13:249­261.
Huston, M. A. 1999. Local processes and regional patterns:
appropriate scales for understanding variation in the
diversity of plants and animals. Oikos 86:393­401.
Jablonski, D. 1993. The tropics as a source of evolutionary
novelty through geological time. Nature 364:142­144.
Jansson, R., and M. Dynesius. 2002. The fate of clades in
a world of recurrent climatic change: Milankovitch oscillations
and evolution. Annual Review of Ecology and
Systematics 33:741­777.
Janzen, D. H. 1981. The peak in North American ichneumonid
species richness lies between 38 N and 42 N.
Ecology 63:532­537.
Jetz, W., and C. Rahbek. 2001. Geometric constraints explain
much of the species richness pattern in African
birds. Proceedings of the National Academy of Sciences
of the USA 98:5661­5666.
------. 2002. Geographic range size and determinants of
avian species richness. Science 297:1548­1551.
Koleff, P., and K. J. Gaston. 2001. Latitudinal gradients in
diversity: real patterns and random models. Ecography
24:341­351.
Lambers, J. H. R., J. S. Clark, and B. Beckage. 2002.
Density-dependent mortality and the latitudinal gradient
in species diversity. Nature 417:732­735.
Lawton, J. H. 1999. Are there general laws in ecology?
Oikos 84:177­192.
Lees, D. C., C. Kremen, and L. Andriamampianina. 1999.
A null model for species richness gradients: bounded
range overlap of butterflies and other rainforest endemics
in Madagascar. Biological Journal of the Linnean
Society 67:529­584.
Levin, S. A. 2001. Encyclopedia of biodiversity. Academic
Press, San Diego, Calif.
Loreau, M. 2000. Are communities saturated? on the relationship
between a, b and g diversity. Ecology Letters
3:73­76.
Losos, J. B., and D. Schluter. 2000. Analysis of an evolutionary
species-area relationship. Nature 408:847­850.
Lyons, S. K., and M. R. Willig. 1997. Latitudinal patterns
of range size: methodological concerns and empirical
evaluations for New World bats and marsupials. Oikos
79:568­580.
------. 1999. A hemispheric assessment of scale dependence
in latitudinal gradients of species richness. Ecology
80:2483­2491.
Martinez, N. D. 1994. Scale-dependent constraints on
food-web structure. American Naturalist 144:935­953.
O'Brien, E. M., R. Field, and R. J. Whittaker. 2000. Climatic
gradients in woody plant (tree and shrub) diversity:
water-energy dynamics, residual variation and topography.
Oikos 89:588­600.
Ollerton, J., and L. Cranmer. 2002. Latitudinal trends in
plant-pollinator interactions: are tropical plants more
specialised? Oikos 98:340­350.
Osenberg, C. W., O. Sarnelle, S. D. Cooper, and R. D.
Holt. 1999. Resolving ecological questions through
meta-analysis: goals, metrics and models. Ecology 80:
1105­1117.
Owen, D. F., and J. Owen. 1974. Species diversity in temperate
and tropical Ichneumonidae. Nature 249:583­
584.
Peters, R. H. 1983. The ecological implications of body
size. Cambridge University Press, Cambridge.
Pianka, E. R. 1966. Latitudinal gradients in species diversity:
a review of concepts. American Naturalist 100:33­
46.
Latitudinal Gradients 211
Poore, G. C. B., and G. D. F. Wilson. 1993. Marine species
richness. Nature 361:597­598.
Qian, H., and R. E. Ricklefs. 2000. Large-scale processes
and the Asian bias in species diversity of temperate
plants. Nature 407:180­182.
Rabenold, K. N. 1979. A reversed latitudinal diversity gradient
in avian communities of eastern deciduous forests.
American Naturalist 114:275­286.
Rahbek, C., and G. R. Graves. 2000. Detection of macroecological
patterns in South American hummingbirds
is affected by spatial scale. Proceedings of the Royal
Society of London B 267:2259­2265.
------. 2001. Multiscale assessment of patterns of avian
species richness. Proceedings of the National Academy
of Sciences of the USA 98:4534­4539.
Rohde, K. 1992. Latitudinal gradients in species diversity:
the search for the primary cause. Oikos 65:514­527.
------. 1997. The larger area of the tropics does not explain
latitudinal gradients in species diversity. Oikos 79:
169­172.
Rosenberg, M. S., D. C. Adams, and J. Gurevitch. 2000.
MetaWin: statistical software for meta-analysis. Version
2.0. Sinauer, Sunderland, Mass.
Rosenzweig, M. L. 1995. Species diversity in space and
time. Cambridge University Press, Cambridge.
Rosenzweig, M. L., and E. A. Sandlin. 1997. Species diversity
and latitudes: listening to area's signal. Oikos 80:
172­176.
Roy, K., D. Jablonski, and J. W. Valentine. 2001. Climate
change, species range limits and body size in marine
bivalves. Ecology Letters 4:366­370.
Ruggiero, A. 1999. Spatial patterns in the diversity of mammal
species: a test of the geographic area hypothesis in
South America. Ecoscience 6:338­354.
------. 2001. Size and shape of the geographical ranges
of Andean passerine birds: spatial patterns in environmental
resistance and anisotropy. Journal of Biogeography
28:1281­1294.
Santelices, B., and P. A. Marquet. 1998. Seaweeds, latitudinal
diversity patterns, and Rapoport's rule. Diversity
and Distribution 4:71­75.
Schall, J. J., and E. R. Pianka. 1978. Geographical trends
in numbers of species. Science 201:679­686.
Shurin, J. B., J. E. Havel, M. A. Leibold, and B. Pinel Alloul.
2000. Local and regional zooplankton species richness:
a scale-independent test for saturation. Ecology 81:
3062­3073.
Siemann, E., J. Haarstad, and D. Tilman. 1999. Dynamics
of plant and arthopod diversity during old field succession.
Ecography 22:406­414.
Sokal, R. R., and F. J. Rohlf. 1995. Biometry. W. H. Freeman,
New York.
Spight, T. M. 1977. Diversity of shallow-water gastropod
communities on temperate and tropical beaches. American
Naturalist 111:1077­1097.
Srivastava, D. S. 1999. Using local-regional richness plots
to test for species saturation: pitfalls and potentials.
Journal of Animal Ecology 68:1­16.
Stevens, R. D., and M. R. Willig. 2002. Geographical ecology
at the community level: perspectives on the diversity
of new world bats. Ecology 83:545­560.
Terborgh, J. 1973. On the notion of favorableness in plant
ecology. American Naturalist 107:481­501.
Valdovinos, C., S. A. Navarrete, and P. A. Marquet. 2003.
Mollusk species diversity in the Southeastern Pacific:
why are there more species towards the pole? Ecography
26:139­144.
von Humboldt, A. (1828) 1993. Alexander von Humboldt
u¨ber das Universum: die Kosmosvortra¨ge 1827/28 in
der Berliner Singakademie. Insel, Frankfurt.
Wetzel, R. G. 2001. Limnology. Academic Press, San Diego,
Calif.
Wetzel, R. G., and G. E. Likens. 1991. Limnological analyses.
Springer, New York.
Wilson, G. D. F. 1998. Historical influences on deep-sea
isopod diversity in the Atlantic Ocean. Deep-Sea Research
Part II Topical Studies in Oceanography 45:279­
301.
Worm, B., H. K. Lotze, H. Hillebrand, and U. Sommer.
2002. Consumer versus resource control of species diversity
and ecosystem functioning. Nature 417:848­851.
Wright, D. H. 1983. Species-energy theory: an extension
of species-area theory. Oikos 41:496­506.
Associate Editor: Chris D. Thomas