Copyrighted Material

III.1
Biodiversity: Concepts, Patterns,
and Measurement
Robert K. Colwell

OUTLINE evenness. A measure of the homogeneity of abun-

dances in a sample or a community
1. What is biodiversity?
functional diversity. The variety and number of species
2. Relative abundance: Common species and rare
that fulfill different functional roles in a community
ones
or ecosystem
3. Measuring and estimating species richness
rarefaction curve. The statistical expectation of the
4. Species diversity indices
number of species in a survey or collection as a func-
5. The spatial organization of biodiversity
tion of the accumulated number of individuals or
6. Estimating b and g diversity from samples
samples, based on resampling from an observed
7. Species–area relations
sample set
relative abundance. The quantitative pattern of rarity
Life on Earth is diverse at many levels, beginning with genes
and commonness among species in a sample or a
and extending to the wealth and complexity of species, life
community
forms, and functional roles, organized in spatial patterns
richness estimator. A statistical estimate of the true
from biological communities to ecosystems, regions, and
species richness of a community or larger sampling
beyond. The study of biodiversity encompasses the dis-
universe, including unobserved species, based on
covery, description, and analysis of the elements that un-
sample data
derlie these patterns as well as the patterns themselves.
species accumulation curve. The observed number of
The challenge of quantifying patterns of diversity at the
species in a survey or collection as a function of the
species level, even when the organisms are known to sci-
accumulated number of individuals or samples
ence, is complicated by the problem of detecting rare
species–area relation. The generally decelerating but
species and the underlying complexity of the environmental
ever-increasing number of species as sampling area
template.
increases
species richness. The number of species in a commu-
nity, in a landscape or marinescape, or in a region
GLOSSARY
a, b, and c diversity. The species diversity (or richness)
1. WHAT IS BIODIVERSITY?
of a local community or habitat (a), the difference in
diversity associated with differences in habitat or Although E. O. Wilson first used the term biodiversity
spatial scale (b), and the total diversity of a region or in the literature in 1988, the concept of biological di-
other spatial unit (g) versity from which it arose had been developing since
biodiversity. The variety of life, at all levels of organi- the nineteenth century and continues to be widely used.
zation, classified both by evolutionary (phyloge- Biodiversity encompasses the variety of life, at all levels
netic) and ecological (functional) criteria of organization, classified both by evolutionary (phy-
diversity index. A mathematical expression that com- logenetic) and ecological (functional) criteria. At the
bines species richness and evenness as a measure of level of biological populations, genetic variation among
diversity individual organisms and among lineages contributes
 Copyrighted Material
258 Communities and Ecosystems
to biodiversity as both the signature of evolutionary as by source of nutrients. In microbial communities,
and ecological history and the basis of future adap- microbial taxa that depend on and transform differ-
tive evolution. Species that lack substantial genetic ent chemical substrates represent distinct functional
variation are thought to be more vulnerable to extinc- groups.
tion from natural or human-caused changes in their At the level of landscapes, marinescapes, or ecosys-
environment. tems, biodiversity is conceived on a landscape or larger
It is at the species level that the term biodiversity is scale, often in terms of the number, relative frequency,
most often applied by ecologists and conservation bi- and spatial arrangement of distinguishable ecosystem
ologists, although higher levels of classification (gen- types, or ecoregions.
era, families, orders) or patterns of evolutionary di-
versification are sometimes also considered, especially
2. RELATIVE ABUNDANCE: COMMON SPECIES
in paleontology. Species richness is the number of spe-
AND RARE ONES
cies of a particular taxon (e.g., birds or grasses) or life
form (e.g., trees or plankton) that characterize a par- The species that characterize any natural community
ticular biological community, habitat, or ecosystem differ in relative abundance, usually with a few species
type. When data are not available at the community, quite common and most species much less so. Another
habitat, or ecosystem level, political units (counties, way of looking at it is that most individuals belong to
states or provinces, countries) are often used as the the few common species in a typical community. For
basis of statements about species richness. example, in a study of the soil ‘‘seed bank’’ in a Costa
Within biological communities and ecosystems, Rican rainforest, by B. J. Butler and R. L. Chazdon, the
functional diversity refers to the variety and number of 952 seedlings that germinated from 121 soil samples
species that fulfill different functional roles. A food included 34 species. The most common single species
web and some measure of its complexity and connec- was represented by 209 seedlings, and the next most
tivity is one way to depict the functional diversity of common had 109. In contrast, the least common 15
a community. Another is the classification and enu- species each had 10 or fewer seedlings.
meration of species representing different functional One way to plot such species abundance data
groups, such as primary producers, herbivores, and (an approach originated by R. H. Whittaker) is a rank-
carnivores. Within forest communities, for example, abundance curve, in which each species is represented
plant functional groups that are often distinguished by a vertical bar proportional to its abundance. Fig-
include fast-growing pioneer species that quickly col- ure 1 shows such a plot for the seed bank data. Notice
onize disturbed habitats, slower-growing species that the long ‘‘tail’’ of rarer species. A community with such
characterize mature forests, and plants that fill special striking disparities in abundance among species is said
functional roles, such as those that fix atmospheric to have low evenness. A rank-abundance plot for a hy-
nitrogen. A marine biologist working on soft-bottom pothetical community with perfect evenness would be
communities might categorize benthic organisms by flat instead of declining, indicating that every species
the physical effect they have on the substrate as well had the same abundance.

250

200
Number of individuals

150

100

50

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 Figure 1. A rank-abundance
Rank of abundance curve.
 Copyrighted Material
Biodiversity 259

12
3. MEASURING AND ESTIMATING SPECIES
RICHNESS
10
On first consideration, measuring species diversity
might seem an easy matter: just count the number of
Number of species

8
species present in a habitat or study area. In practice,
however, complications soon arise. With the exception
6 of very well-known groups in very well-known places
(for which we already have good estimates of total
4 richness anyway), species richness must generally be
estimated based on samples. First of all, even for
2
groups as well known as birds or flowering plants, not
all species that are actually present are equally easy to
detect. Although size, coloration, and—for animals—
0 behavior can affect the detectability of individuals,
1 2-3 4-7 8-15 16-31 32-63 64- 128-
127 255 relative abundance is the most important influence on
Abundance category
the effort required to record a species. As every be-
Figure 2. A log abundance plot. ginning stamp or coin collector soon discovers, the
common kinds of coins or stamps are usually the first
to be found. As the collection grows, the rate of dis-
covery of kinds new to the collection declines steadily,
Another way to plot the same species abundance as rarer and rarer kinds remain to be found.
data is to count up the number of species in each abun- For species richness, this process can be depicted as
dance category, starting with the rarest species, and a species accumulation curve, sometimes called a col-
plot these frequencies against abundance categories, as lector’s curve. The jagged line in figure 3 shows a
in figure 2. It is customary to use abundance categories species accumulation curve for the seed bank data of
in powers of two, which gives a log abundance plot figure 1, as the 121 soil samples were added one at a
(originated by F. W. Preston). When relative abundance time to the total. Because the order in which the soil
distributions approximate a normal (bell-shaped) samples were added to the collection was arbitrary, a
curve in a log abundance plot (the seed bank data in smoothed version of such a curve, called a rarefaction
figure 2 come close), the statistical distribution is called curve, makes more sense. Conceptually, a rarefaction
lognormal. Lognormal distributions of relative abun- curve can be produced by drawing 1, 2, 3,. . .N sam-
dance are common for large, well-inventoried natural ples (or individuals) at a time (without replacement)
communities. Many other statistical distributions have from the full set of samples, then plotting the means
been used to describe relative abundance distributions, of many such draws. Fortunately, this is not necessary,
including the log-series distribution, which is described as the mathematics of combinations allows rarefac-
later in the context of diversity indices. tion curves to be computed directly, along with 95%
Conservation biologists are concerned with relative confidence intervals (the dashed lines in figure 3),
abundance because rare species are more vulnerable to based on work by C. X. Mao and colleagues. Rar-
extinction. Some species that are rare in one commu- efaction curves are especially useful for comparing
nity are common in another (e.g., gulls are rare in many species richness among communities that have not
inland areas, but common along coasts), but some been fully inventoried or have been inventoried with
species are scarce everywhere they occur (e.g., most unequal effort.
large raptors). In a classic paper, D. Rabinowitz clas- Richness estimation offers an alternative to rare-
sified species by three factors: (1) size of geographic faction for comparing richness among incompletely
range (not localized versus localized); (2) habitat spec- inventoried communities. Instead of interpolating
ificity (not habitat specific versus habitat specific); and ‘‘backward’’ to smaller samples as in rarefaction, rich-
(3) local population density (not sparse versus sparse). ness estimators extrapolate beyond what has been re-
She pointed out that there are seven ways to be rare, by corded to estimate the unknown asymptote of a species
this classification, but only one way to be common: not accumulation curve. Simple (regression-based) or so-
localized, not habitat specific, not sparse. Species that phisticated (mixture model) curve-fitting methods of
are rare by all three criteria (localized, habitat specific, extrapolation can be used, or nonparametric richness
and sparse), such as the ivory-billed woodpecker in the estimators can be computed. The latter depend on the
United States, are the most vulnerable to extinction. frequencies of the rarest classes of observed species to
 Copyrighted Material
260 Communities and Ecosystems

40 40

35 35

30 30

Number of species
Number of species

25 25

20 20

15 15

10 10

5 5

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Number of samples Number of samples
Figure 3. Species accumulation and rarefaction curves. Figure 4. Estimated species richness and rarefaction curves.

4. SPECIES DIVERSITY INDICES

estimate the number of species present but not detected The concept of diversity, including biodiversity itself as
by the samples. The simplest nonparametric estimator, well as the narrower concept of species diversity, is a
Chao1, augments the number of species observed (Sobs) human construct without any unique mathematical
by a term that depends only on the observed number of meaning. The simplest measure of species diversity is
singletons (a, species each represented by only a single species richness, but a good case can be made for giving
individual) and doubletons (b, species each represented some weight to evenness as well. For example, the sub-
by exactly two individuals): jective sense of tree species richness is likely to be
greater for a naturalist walking through a forest com-
a2 posed of 10 species of trees, each equally represented,
Sest ¼ Sobs þ :
2b than a forest of 10 species in which one species con-
tributes 91% of the individuals and the others each 1%.
For the seed bank example of figures 1 and 2, Diversity indices are mathematical functions that
when all samples are considered, 34 species were combine richness and evenness in a single measure, al-
observed. Of these species, two were singletons, and though usually not explicitly. Although there are many
two were doubletons, so that the estimated true others, the most commonly used diversity indices in
richness is 35 species, confirming the visual evidence ecology are Shannon diversity, Simpson diversity, and
from the rarefaction curve that the inventory was Fisher’s a. If species i comprises proportion pi of the
virtually complete. The real utility of estimators, total individuals in a community of S species, the
however, lies in their potential to approximate as- Shannon diversity is
ymptotic species richness from much smaller sam-
ples. Figure 4 shows the same rarefaction curve (solid X
s

line) as in figure 3, with the estimated (asymptotic) H¼
pi ln pi or, preferably, eH

i¼1
species richness (shown by the dashed line) for the
Chao1 estimator, which begins to approximate true and Simpson diversity is
richness with as few as 20 samples. (The estimator
curve shows the mean of 100 random draws for each !
1
X
s X
s
number of samples.) It should be noted that richness D¼1
p2i or, preferably, D ¼0
p2i :
estimators are not a panacea for problems of un- i¼1 i¼1
dersampling. Hyperdiverse communities with large
numbers of very rare species, such as tropical ar- Both Shannon and Simpson diversities increase as rich-
thropods, have so far resisted efforts to provide re- ness increases, for a given pattern of evenness, and
liable nonparametric richness estimators. increase as evenness increases, for a given richness, but
 Copyrighted Material
Biodiversity 261

they do not always rank communities in the same or- ‘‘biotic mosaic’’ of variably discontinuous assemblages
der. Simpson diversity is less sensitive to richness and of species. On land, the discontinuities are driven in the
more sensitive to evenness than Shannon diversity, shorter term by topography, soils, hydrology, recent
which, in turn, is more sensitive to evenness than is a disturbance history, dispersal limitation, species inter-
simple count of species (richness, S). At the other ex- actions, and human land use patterns, and in the longer
treme, a third index in this group, the Berger-Parker term and at greater spatial scales by climate and Earth
index, depends exclusively on evenness; it is simply the history. The same or analogous factors structure bio-
inverse of the proportion of individuals in the com- diversity in the sea.
munity that belong to the single most common species, If you were to keep track of the plant or bird species
1 ⁄ pi (max). Because rare species tend to be missing from encountered, in the form of a species accumulation
smaller samples, the sensitivity of these indices to curve, during a long walk in a forest followed by a long
sampling effort depends strongly on their sensitivity to walk in an adjacent grassland, the curve would first rise
richness. In practice, which measure of diversity to use quickly, as the common forest species were recorded,
depends on what one wishes to focus on (pure richness leveling off (if the walk is long enough) as the rarest
or a combination of richness and evenness), the relative forest species are finally included. The number of spe-
abundance pattern of the data, comparability to pre- cies accumulated at that point (or a species diversity
vious studies, and the interpretability of the results. index computed for the accumulated data) is called the
These four diversity measures (richness, the exponen- a diversity (or local diversity) for a habitat or com-
tial form of Shannon diversity, the reciprocal form of munity, a concept originated by R. H. Whittaker. (Note
Simpson diversity, and the Berger-Parker index) can be that a diversity has nothing to do with Fisher’s a, in
shown to be specific points on a diversity continuum terms of the names, although the latter may be used as
defined by a single equation based on the classical one measure of the former.) As you leave the forest and
mathematics of Rényi entropy, as first shown in the enter the grassland, the curve will rise steeply again, as
ecology literature by M. O. Hill in 1972 and periodi- common grassland species are added to the list. Once
cally rediscovered since then. L. Jost, in 2005, reviewed rarer grassland species are finally included, the curve
these relationships and provided compelling arguments begins to level off at a new plateau. The increment in
for preferring the exponential version of Shannon index total species (or the change in a diversity index) caused
and the reciprocal (D0 ) version of the Simpson index. by the change in habitat is one measure of b diversity,
Fisher’s a is mathematically unrelated to the Rényi in Whitaker’s terminology (sometimes called differen-
family of indices. It is derived from the log-series dis- tiation diversity), although there are many ways to
tribution, proposed by R. A. Fisher as a general model quantify b diversity and little agreement about which is
for relative abundance: best. The total richness or diversity for both habitats
combined (the second plateau in the species accumu-
ax, ax2 ⁄ 2, ax3 ⁄ 3, ax4 ⁄ 4, . . . axn ⁄ n, lation curve) is the g diversity (regional diversity) for
this hypothetical forest–grassland landscape.
where successive terms represent the number of species The forest-to-grassland example presents a classic
with 1, 2, 3,. . .n individuals, and a is treated as an illustration of b diversity, as originally conceived by
index of species diversity. Estimating a from an em- Whittaker, but the concept has been generalized to
pirical relative abundance distribution, however, de- include spatial differentiation of biotas within large
pends only on S (the total number of species) and N expanses of continuous, environmentally undifferenti-
(the total number individuals) but nevertheless requires ated habitat as well as between isolated patches of simi-
substantial computation because iterative methods lar habitat. Within expanses of homogeneous habitat,
must be used. Fisher’s a is relatively insensitive to rare b diversity is usually considered to be the result of
species, and the relative abundance distribution need dispersal limitation—the failure of propagules (fruits,
not be distributed as a log-series. seeds, juveniles, dispersive larval stages, migrants, etc.)
to mix homogeneously over the habitat—but in prac-
tice, it is often hard to rule out subtle differences in
5. THE SPATIAL ORGANIZATION OF BIODIVERSITY
environment as a cause of biotic differentiation.
Imagine walking through a forest into a grassland or
snorkeling across a coral reef beyond the reef edge
6. ESTIMATING b AND c DIVERSITY FROM SAMPLES
toward the open sea. The testimony of our own eyes
confirms that the biosphere is not organized as a set of Estimating b or g diversity for a region or landscape,
smooth continua in space but rather as a complex from samples, is a daunting prospect for any but the
 Copyrighted Material
262 Communities and Ecosystems
best-known groups of organisms. Over larger spatial or types within a region, it would be simple to determine
climatic scales, the ‘‘patches’’ of the mosaic can be better the total biota for two, three,. . .all types combined,
viewed as ordered along gradients, in either physical or computing some measure of (average or pair-specific) b
multivariate environmental space. Unfortunately, the richness (species turnover) along the way. For sampling
geometry of the biotic mosaic is remarkably idiosyn- data, the problem is much more difficult. Undetected
cratic (although it may be properly fractal for some species within patch types are not only undetected, they
organisms at some scales), which means that designing are unidentified, so that that we do not know whether
a scheme for estimating richness at large spatial scales the same or different species remain undetected in
is likely to require many ad hoc decisions—it is more different patch types.
like designing trousers for an elephant than finding Nonetheless, it is possible in principle to estimate
yourself a hat that fits. lower and upper bounds for g (regional) richness. The
A common approach to coping with idiosyncratic union of detected species lists for all patch types, pooled,
biotic patterns is to take advantage of biotic dis- provides a lower-bound estimate of total domain rich-
continuities to define ‘‘patch types’’ in the mosaic for ness, on the assumption that every species undetected
sampling purposes. For example, the vegetation of in one patch type is detected in at least one other patch
treefalls in a forest might be distinguished from the type. The sum of total richness estimates over all
riparian (streamside) vegetation and from the mature patch types (including undetected species from each
forest matrix. Or the fish fauna of isolated patch reefs patch type, using nonparametric estimators or extrap-
might be distinguished from the fish fauna of fringing olation techniques), adjusted for the number of ob-
reefs. An alternative is to select sampling sites along served shared species, is an approximate upper-bound
explicit gradients, such as elevational transects on land estimate of total regional richness, assuming that un-
or depth and substrate gradients in the sea. Both detected species included in the estimates are entirely
strategies represent forms of stratified sampling in different for each patch type and were detected in none.
which the strata are the patch types or gradient sites, The truth inevitably lies between these bounds, for
and multiple samples within them are treated as ap- data from nature. To estimate the true regional rich-
proximate replicates, meaning, in practice, that sam- ness, we need information about the true pattern of
ples within patch types or gradient sites are expected to shared species among patch types. Statistical tools for
be more similar than samples from different types or estimating the true number of species shared by two
sites. sample sets, including species undetected in one or both
Any particular definition of patch types and the scale sets, are scarce, and this is an area in which much more
that underlies them is inevitably somewhat arbitrary. A work is needed. Many studies have attempted to ad-
seemingly less arbitrary alternative would be spatially dress the problem of estimating b diversity, or pooling
random sampling over the entire region of interest, samples (between patch types or random samples) by
analyzed using a multivariate approach to assess the re- using similarity indices, such as the Sørensen or Jaccard
lationship of richness and species composition to un- indices. Unfortunately, the number of observed, shared
derlying environmental and historical factors. But, species is almost always an underestimate of the true
given limited resources (are they ever otherwise?), number of shared species because of the undersampling
random sampling over heterogeneous domains is often of rare species. This means that species lists based on
highly inefficient because of the uneven relative abun- samples generally appear proportionally more distinct
dance of patch types: the biota of common patch types than they ought to be, similarity indices are routinely
are oversampled compared to the biota of rarer patch biased downward, and slope estimates for the decline
types, which may even be missed entirely. If one ac- in similarity with distance (‘‘distance decay of simi-
cepts a within- and between-patch-type design frame- larity’’) are likely to be overestimated. Recently, A.
work, the definition of patch types (or sample spacing Chao and others have developed estimation-based
on gradients) is best made at the design phase based on similarity indices that greatly reduce undersampling
expert advice and whatever prior data exist, with the bias and promise to help correct this longstanding di-
possibility of later iterative adjustment. lemma. These indices are based on the probability that
Although comparisons of a diversity among patch two randomly chosen individuals, one from each of
types by rarefaction are interesting in their own right, two samples, both belong to species shared by both
they fail to provide the information needed to estimate samples (but not necessarily to the same shared spe-
g diversity because some species are likely to be shared cies). The estimators for these indices take into account
among patch types and some species may be missed by the contribution to the true value of this probability
the sampling in all patch types. If we had full knowl- made by species actually present at both sites but not
edge of the biota (complete species lists) for all patch detected in one or both samples.
 Copyrighted Material
Biodiversity 263

tion, so that larger areas contain more species. (2)

7. SPECIES–AREA RELATIONS
Larger areas are more likely to include a greater
Ecologists and biogeographers have long documented a number of habitat types or ecoregions, each with its
striking regularity in the pattern of increase in the spe- own distinct or partially distinct biota. (3) For very
cies count as larger and larger geographic areas are large areas, on continental scales, ecologically similar
considered. When the number of species or its logarithm biotas may have very different evolutionary histories.
(depending on the case) is plotted against the logarithm For example, the lizard fauna of coastal Chile and
of area, an approximately linear relationship is revealed. coastal California share many ecological similarities
With either plot (a log-log power curve or a semilog but have no species (or even genera) in common. Such
exponential curve), the pattern on arithmetic axes is a cases could be viewed as an extreme form of dispersal
decelerating but ever-increasing number of species as limitation, as we discover to our dismay when alien
area increases. This pattern, known as the species–area species from similar biomes on other continents become
relation (SAR), has been called one of the few universal local invasives (e.g., California poppy, Eschscholzia
patterns in ecology, but its causes are not simple. californica, in Chile, and the Chilean ice plant, Car-
There are many variants on SARs, but the primary pobrotus chilensis, in California).
dichotomy separates plots based on nested sampling
schemes from plots in which the areas of increasing size FURTHER READING
are distinct places, such as islands in lakes or seas,
habitat islands on land, or simply political units (states, Chao, A. 2004. Species richness estimation. In N. Bala-
countries) of different areas. There are two important krishnan, C. B. Read, and B. Vidakovic, eds., Ency-
causes for the increase in species count with increasing clopedia of Statistical Sciences. New York: Wiley. A
comprehensive review of the statistical methods for esti-
area. The first cause is undersampling. Especially in the
mating richness.
case of nested sampling schemes, in which smaller ar- Magurran, A. E. 2004. Measuring Biological Diversity. Ox-
eas lie within larger ones, the smaller units may be too ford: Blackwell Publishing. The standard reference for
small or too poorly sampled to reveal all species char- conceptual and quantitative aspects of diversity mea-
acteristic of the habitat(s) they represent. In this case, surement.
the supposed SAR for the smaller areas is better de- Rosenzweig, M. 1995. Species Diversity in Space and Time.
scribed as a species accumulation curve or rarefac- New York: Cambridge University Press. An overview and
tion curve. B. D. Coleman and colleagues pointed out analysis of the geography of species richness, especially
that, even for a completely homogeneous species pool, species–area relations.
larger areas will have more species because they con- Soulé, M. E., ed. 1986. Conservation Biology: The Science of
Scarcity and Diversity. Sunderland, MA.: Sinauer As-
tain more individuals; the model they proposed is vir-
sociates. A classic collection of papers focusing on the
tually indistinguishable from a rarefaction curve. conservation of biodiversity.
The second cause of increasing species count with Wilson, E. O., and F. M. Peter, eds. 1988. Biodiversity.
area is b diversity, in all its varieties. (1) Within large Washington, DC: National Academy Press. An important
expanses of homogeneous habitat, species composition collection of papers that launched public awareness of
may vary spatially simply because of dispersal limita- biodiversity and its importance.