Summary

1. The species abundance distribution (SAD) is an important concept in

ecology, and much work has focused on the SAD in a theoretical context.
However, less focus has been placed on the utility of SADs in applied
ecology and biodiversity management, which therefore forms the focus of
the present article. We illustrate that study of the SAD allows inferences
beyond those that flow from many simple diversity indices, enabling
workers to identify pat terns in the commonness and rarity of species in a
community.
2. First, we discuss how incorporating SAD analyses into the study of
ecological communi ties can generate useful information for the
management of biodiversity. In particular, we argue that deconstructing
ecological assemblages into various subsets and analysing how each
subset contributes to the overall SAD can reveal patterns of interest to
managers. Secondly, we review the many applications of SADs in applied
ecological fields, including disturbance ecology, conservation planning
and conservation biological control.
3. Using examples we show that the SAD can be useful in applied ecology
as it is visually intui tive, easy to implement in a broad variety of ecological
contexts, and does not require substan tial species-specific data. We
provide a summary of the various methods available for plotting the SAD
and illustrate how each method provides information of value for applied
ecologists.
4. Using empirical and simulated data, we show that the SAD can provide
early warning of the effects of disturbance on ecological communities and
that a number of SAD models rep resent useful tools for comparing
communities in a management context.
5. Synthesis and applications. Applied ecologists require tools that allow
for relatively quick assessments of ecosystem health and/or the success
of management prescriptions aimed at ameliorating the effects of
disturbance. We demonstrate that that the methods reviewed herein
provide such tools and that the species abundance distribution (SAD) has
additional applied uses beyond traditional applications in disturbance
ecology. We hope that this synthe sis will provide a catalyst for advancing
a more utilitarian SAD research agenda.

Key-words: community structure, conservation biogeography, disturbance

ecology, macroe cology, multimodal species abundance distribution, scaling,
species abundance distribution

Introduction
The species abundance distribution (SAD) characterizes the distribution of
abundances of all species within a
sample or ecological community. The observation that most species are
relatively rare with only a few being common is often described as one of the
few ecological laws (McGill et al. 2007). The SAD is an important con cept in
ecology and macroecology, being interesting in its own right as well as
providing the theoretical foundation for exploration of other ecological
patterns, such as the distance–decay relationship and the species–area
relation ship (SAR; Preston 1948; McGill et al. 2007; Whittaker & Fern andez-
Palacios 2007). Around 30 different SAD models have been proposed, with the
most commonly used being the log-normal (Preston 1948) and the logser ies
(Fisher, Corbet & Williams 1943). While SADs have been researched for over
seventy years (e.g. Fisher, Cor bet & Williams 1943), there has been a
resurgence of interest over the last decade (e.g. McGill et al. 2007). However,
much of this recent work has been on theoreti cal aspects of the pattern, and
less focus has been placed on the utility of SADs as a tool to provide useful
infor mation in applied ecology, conservation and manage ment. For instance,
a number of general SAD reviews have been published (e.g. May 1975;
Magurran 2004; McGill et al. 2007; McGill 2011), but few discuss the applied
uses of SADs (for an exception, see Dornelas, Soykan & Ugland 2011). Global
environmental change and disturbance to biotic communities resulting from
habitat loss, pollution and invasive species, amongst other drivers, represent
substantial pressures on biodiversity (Sala et al. 2000), in the light of which
applied ecologists and biodiversity managers require accurate, easy to use
and intuitive methods for measuring the impacts of the aforementioned drivers
of biodiversity decline (Mouillot et al. 2013). It is also necessary that these
methods allow workers to compare communities which differ in species
richness, an issue which plagues many traditional diver sity metrics. We
believe the SAD represents such a tool kit. Thus, the distinctive purpose of
this review is to synthesize the information and recent advances relating to the
use of the SAD in applied ecology. To achieve this aim, we have split the
review into two broad sections regarding the utility of SADs in applied
ecology: (1) a discussion on how incorporating the SAD into the study of
ecological communities can generate useful informa tion for the management
of biodiversity and (2) a review of the application, and potential application, of
SADs in biodiversity conservation and management. In the first section, we
review the recent literature on combining SADs with an assemblage
deconstruction approach in order to illustrate how the analysis of complete
assem blages (i.e. all sampled species) can obscure important abundance
patterns in species of conservation and man agement concern. We use recent
case studies of multi modal SADs to highlight the types of beneficial
information that can be derived from this approach. The second section
focuses more on the actual usage of SADs in conservation, including
application of SADs in mea suring ecosystem health, and for conservation
planning frameworks.
ON PLOTTING SPECIES ABUNDANCE DISTRIBUTIONS
Before progressing, it is necessary to review briefly the two main methods for
plotting SADs, as both are discussed throughout the paper. First, the SAD can
be visualized as a histogram of the number of individuals on the x-axis and the
number of species represented by a particular abundance on the y-axis. The
numbers of indi viduals are generally binned into octaves using a variety of
different methods (see Gray, Bjørgesæter & Ugland 2006; Matthews &
Whittaker 2014). The use of binning in SAD studies has been criticized as it
results in the loss of information (Gray, Bjørgesæter & Ugland 2006). Thus, a
number of studies use a different plotting method, termed rank abundance
plots (e.g. Foster & Dunstan 2010; see Matthews & Whittaker 2014). Rank
abundance plots/diagrams (RADs) are plots of abun dance (untransformed or
log-transformed) against rank order, where rank one corresponds to the
species with the highest abundance and so on. Generally, species with the
same abundance are assigned increasing ranks; for example, three species in
a sample represented by five individuals might be given the ranks of 10, 11 and
12. However, it may in fact be preferable to assign such spe cies the same
rank, in this case 10, 10 and 10. RADs are useful in that they can sometimes
reveal differences in model fits not apparent when using histograms (Fattorini
2005). Fattorini (2005) has recently shown that the geo metric series and
broken stick SAD models (see Matthews & Whittaker 2014) can be fitted using
linear regression and RAD plots. His analyses demonstrate that when the
abundance data are log-transformed, a linear relationship indicates that the
SAD follows a geometric series, while the broken stick model is supported in
cases where a linear relationship emerges following log-trans formation of the
rank axis.

Section 1: Using the species abundance distribution to derive information

on ecological community characteristics for applied purposes

The efficient management of ecological communities (gen erally a unit of

management interest, e.g. typical UK nat ure reserves) is reliant on detailed
and accurate information regarding particular community characteris tics, for
example the abundance of species of conservation interest (Newman 1993).
Typically, such information is presented in the form of simple counts of the
number of species, or occasionally a simple diversity index is calcu lated, such
as the Shannon–Wiener index. However, this type of information can only
reveal so much, and we would argue that looking at the full abundance
spectrum is a much more elucidative approach (below, Dornelas et al. 2009;
Dornelas, Soykan & Ugland 2011; Sæther, Engen & Grøtan 2013). We focus
this section of the review on a recent area of research interest which we feel
has particular potential for applied ecology, but which has not been discussed
 in recent SAD reviews: combining SADs with an assemblage deconstruction
approach.

Typically, empirical ecological and macroecological analy ses are based on lists of
all species encountered during a sampling exercise, for example all birds seen or
heard in a patch of forest. However, it has become increasingly apparent that the
aggregation in a sample of species with differences in key ecological properties,
such as body size, dispersal ability or habitat affinity and specialization, can act to
obscure patterns of interest for particular subsets of species, a theory we have
termed the amalgamation hypothesis in previous work (Matthews, Borges &
Whittaker 2014). Thus, a number of recent studies have focused on splitting
samples into various subsets prior to analysis and then exploring patterns of
interest in each subset separately (e.g. Magurran & Henderson 2003; Ulrich &
Zalewski 2006; Bommarco et al. 2010). For example, in a recent paper, we have
shown that the amal gamation of specialist and generalist bird species in forest
fragment data sets can act to mask the loss of specialist species, generally those
species of most conservation con cern, in response to habitat insularization (i.e.
generalists depress the slope of the island SAR; Matthews, Cottee Jones &
Whittaker 2014; see also Bommarco et al. 2010). To take another example, this
time in the context of SADs, Labra, Abades & Marquet (2005) used a decon
struction approach to look at invasive and native species and found that for US
birds, on average, invasive species obtain higher maximum abundances than
native species.

Species abundance distribution in applied ecology 445

This, it is argued, is due to the higher habitat generalization of many invasive
species, which allows them to reach higher abundances within a given
community (Labra, Abades & Marquet 2005). Failure to take into account the
effect on analyses of the amalgamation of different types of species in
samples, for instance in the above cases generalist and invasive species, may
result in inappropriate conservation plans being formulated (Matthews,
Cottee-Jones & Whittaker 2014). A decon struction approach for
macroecology in general is advocated and described in depth by Marquet et
al. (2004).

ANALYSIS OF MULTIMODAL SPECIES ABUNDANCE DISTRIBUTIONS WITHIN A

DECONSTRUCTION FRAMEWORK

While the possibility has long been recognized, it has become increasingly
apparent that many SADs may in fact be multimodal, that is characterized by
multiple dis tinct modes (e.g. Pielou 1969; Dornelas & Connolly 2008;
Vergnon, van Nes & Scheffer 2012; Matthews, Borges & Whittaker 2014). A
number of explanations have been put forward to explain multimodal SADs,
 including emer gent neutrality theory (Vergnon, van Nes & Scheffer 2012; but
see Barab as et al. 2013) and the possibility that they simply represent
statistical artefacts (Gaston 1994; McGill et al. 2007). A more detailed history
of multi modal SAD research is presented in Table 1. However, One of the first
studies to recognize the possibility of multimodal SADs, stating that increasing
a sample to include a broader range of taxa (e.g. expanding from a focus on
warblers to all birds), might result in the combination of several logseries
distributions, each with its own set of parameters. More recently, Sizling et al.
(2009) have discussed how incorporating multiple areas and a variety of taxa
within a sample can result in a SAD that is a proportional sum of different
partial SADs Ugland & Gray (1982) A problem with using the log-normal
distribution to model SADs is that it assumes equal density probabilities
across the species (symmetry), but in reality, there is a general pattern of
asymmetry as communities are comprised of three distinct abundance groups,
roughly translated as rare, intermediately abundant and common. Within each
group, the SAD is symmetric, which leads to a mixture of three log-normal
distributions when focusing on the whole assemblage Gray, Bjørgesæter &
Ugland (2005) Found strong evidence of bimodal log-normal distributions for
a mixture of marine and terrestrial data sets.
Dornelas & Connolly (2008) Fitted a combination of one- to four-mode
Poisson log-normal (PLN) distributions, in addition to the logseries, for a large
sample of coral communities, Australia, and compared the various
distributions using maximum-likelihood methods. At the scale of the entire
sample, the three-mode PLN had the greatest support, but as the four-mode
model was within an AIC of 2 of the three-mode model, this can also be
concluded to have strong support The SADs of arthropod communities on
Terceira Island, Azores, could be deconstructed into three distinct abundance
groups (abundant, intermediate and rare) Found that environmental
heterogeneity underpinned multimodal SADs of weed seed bank communities
Tested a number of data sets representing various taxa and found strong
evidence of multimodality. This indicates that multimodality may be more
common than previously thought: it has simply been overlooked
Using the methodology of Dornelas & Connolly (2008), it was found that a two-
mode PLN provided a better fit than the unimodal PLN and logseries for a number
of Azorean arthropod assemblages at a variety of spatial scales perhaps the most
widely accepted explanation is the afore mentioned amalgamation hypothesis (e.g.
Alonso, Ostling & Etienne 2008). In this regard, analysis of multimodal SADs often
reveals interesting information relating to community structure because the
different modes in the distribution have often been found to represent clusters of
different types of species (Ugland & Gray 1982; Borges et al. 2008; Matthews,
Borges & Whittaker 2014). That is, species are more similar, according to some
trait, to oth ers within clusters than between clusters. For instance, one
assemblage division which has proved enlightening in SAD studies has been into
core and satellite species sub sets. Conceptually, core species represent the
constituent members of any ecological community and are predicted to be
structured according to traditional niche-based mechanisms (Ulrich & Zalewski
2006). Satellite species are those species only occasionally found in any commu
nity (i.e. mostly immigrants from outside the local species pool) and are predicted
to be governed by stochastic pro cesses, largely random dispersal. Using a core–
satellite division, Magurran & Henderson (2003), who focused on an estuarine fish
community in the UK over a period of 21 sampling years, discovered that those
species which are relatively abundant throughout the record and possess
specialized estuarine habitat requirements (the ‘core’ spe cies) are characterized
by a log-normal distribution. In contrast, the species with low abundances,
infrequent records in the data set and different habitat requirements (the ‘satellite’
species) follow simple Poisson processes and are characterized by a logseries
distribution (Magurran & Henderson 2003). The two different sets of species leave
different signals within the SAD and when combined result in a distribution with
more rare species than pre dicted by the standard log-normal model (for further
examples, see Gray, Bjørgesæter & Ugland 2005; Ulrich & Zalewski 2006;
Unterseher et al. 2011). It is important to remember, however, that any division of
an assemblage into core and satellite species is a simplification of reality.
Empirical assemblages represent a continuum of species types, and workers must
be careful not to introduce arte facts into analyses when splitting species into
binary cate gories.
More recently, Matthews, Borges & Whittaker (2014) used a null model
approach to show that the SADs of Azorean forest arthropods were frequently
significantly bimodal, with the rarer mode of species predominantly
comprising satellite species and the common mode largely comprising core
species. Many of the satellite species in the rarer modes were as follows: (a)
species introduced to the Azores and (b) also classified as tourist species
(sensu Borges et al. 2008). Tourist species were defined as species present in
higher abundances in more anthropogenic land use types surrounding the
native forest patches (e.g. agri cultural pastures or exotic forest plantations)
and for which it was assumed that native forest was not primary habitat
(Matthews, Borges & Whittaker 2014). Thus, although occurring rarely in the
samples, they were not
species of conservation concern. The core–satellite dichot omy used in
conjunction with a multimodal SAD model in the above study was useful as it
allowed focus on the abundance of the different types of species and showed
that a significant proportion of the ‘rare’ species in the forest were in fact
species which managers would not want to conserve (e.g. introduced species;
see also Mat thews, Cottee-Jones & Whittaker 2014). Dornelas et al. (2009)
have also shown that fitting multimodal SAD mod els to weed communities in
agroecosystems can generate useful information about how weed species
respond to environmental heterogeneity (discussed below).
In sum, it has become increasingly apparent that the amalgamation of multiple
groups of species within a sam ple can mask patterns that may be of
considerable interest to managers. A natural next step then is to deconstruct
full assemblages/samples into different subsets and to explore patterns in the
subsets seperately. This does not have to add too much complexity to data
collection and analysis; simple divisions (e.g. core and satellite species,
specialist and generalists) can be informative.
Section 2: The species abundance distribution in biodiversity management
and conservation
Although not as widespread in the management and con servation literature as
other macroecological patterns (e.g. the SAR), SADs are a potentially useful
tool for conser vation scientists and managers as they can be used to ‘define’
rarity in a particular community. Conservation practices generally focus on
rare species within a delin eated protected area boundary (Gaston 1994), and
a SAD can provide evidence of the level of rarity of particular species of
interest relative to other species (McGill 2011) and thus extinction risk and
associated conservation action. In addition, changes in the empirical SAD can
act as an early warning for the effects of disturbance on bio diversity, as the
shape of the SAD can change markedly before any local extinctions occur (H
agvar 1994; Mouillot et al. 2013). Disturbance to ecological communities
arising from drivers such as habitat loss, invasive species and pol lution is a
pervasive feature of modern times (Sala et al. 2000), and there is thus an
exigent need for the develop ment of tools that allow for relatively quick
assessments of ecosystem health and/or the success of management
prescriptions aimed at ameliorating the effects of distur bance (Mouillot et al.
2013). We argue here that the SAD represents an example of such a tool, and
we use this sec tion to review a number of particularly promising areas of
application.

DISTURBANCE AND ECOSYSTEM HEALTH

An area of biodiversity conservation and management in which it has long
been argued that SADs represent a potentially useful tool is in acting as an
ecological indica tor to determine the effects of pollution, and
disturbance more generally, on biotic communities (Gray et al. 1979; Tokeshi
1993; Dornelas, Soykan & Ugland 2011). Distur bance plays a central role in
structuring communities, and the prevalence of human-induced disturbance
has resulted in wide-ranging effects on biodiversity and ecosystem
functioning and, in particular, species abundances. Char acterizing community
structure and comparing structure across communities are problematic as
ecological commu nities are complex and contain a large amount of infor
mation. Several nonparametric indices (e.g. the Simpson Diversity Index) have
been developed to condense this information and allow easier comparison
between com munities. However, such indices can oversimplify the com plex
structural nature of communities and most are not independent of sampling
 intensity (Mouillot & Lepretre 2000; Dornelas, Soykan & Ugland 2011). Thus, it
has been argued that the full SAD should be used to compare communities
(e.g. Mouillot & Lepretre 2000; Kim, Cho & Chon 2013; Sæther, Engen &
Grøtan 2013). Furthermore,ecological disturbance can have varying impacts
on differ ent parts of the SAD (e.g. common or rare species), a fact which is
obscured when focusing on individual diversity indices. K
urka, Sizling &
argued that the SAD can be used to identify the
Rosindell (2010) have also
impacts of disturbance in a delineated area. These authors show that
disturbance affects the spatial distribution of individ uals in an assemblage
and thus that disturbance alters the shape of the SAD through affecting either
or both of (a) the degree of spatial autocorrelation in the area and (b) the
degree of turnover between subplots within the area.
The log-normal distribution has been proposed as a means of accurately
modelling undisturbed communities and thus represents a measure by which
deviation from such ‘equilibrium’ can be measured (e.g. May 1975; Hill et al. 1995).
The suitability of the log-normal as a general SAD model derives from the central
limit theorem, in which a large number of biotic and abiotic factors affect ing
population sizes act multiplicatively to generate a log normal distribution of
abundances (May 1975). In con trast, SADs in disturbed communities have been
shown to follow distributions close to the logseries (e.g. Hill et al. 1995). Early
papers on the topic (e.g. Gray & Mirza 1979; Gray et al. 1979) focused on the
effects of organic pollu tion on marine benthic communities, largely in Norway and
Scotland, and used departure from log-normality to indicate the impact of
pollution on community structure (but see Tokeshi 1993 for criticisms of their
method). Fol lowing a mild pollution event, departure from log-normal ity was
argued to result as a few species became more abundant, while most species
became rarer or became extirpated from the system. Recent papers have provided
further evidence of the utility of this approach (e.g. Tang et al. 2010; Kim, Cho &
Chon 2013). For example, Kim, Cho & Chon (2013) found that the SAD of
macroinverte brates in non-polluted streams (a key target group for ecosystem
health) in Korea was best fitted by a log-nor mal model, while the SAD of
macroinvertebrates in pol
Species abundance distribution in applied ecology 447
luted streams was better fitted by a geometric model (a model with a similar
curve shape to the logseries, in which the abundance of a particular species is
proportional to the amount of limiting resource they have apportioned).
This transition from log-normal to logseries shaped curves is illustrated in Fig.
1. Here, we have simulated an undisturbed community by generating random
data to follow a log-normal distribution (Fig. 1a). We then mod elled a natural
perturbation to the system by randomly sampling 5% of the individuals (i.e.
removing 95% of the individuals) (Fig. 1b). Ecologically, this simulated
perturbation represents a situation in which populations of rare species are
most likely to go locally extinct. This simulation is also similar to the approach
 adopted by Green & Plotkin (2007), in which individuals were sam pled in a
spatially explicit manner from regional SADs to examine the scaling
relationships between sample and regional-scale SADs. Plotting the SAD of
the undis turbed and disturbed community (i.e. before and after) reveals the
shift from a log-normal to a logseries distri bution (Fig. 1). Comparing the fit of
the Poisson log-nor mal (here we have used the zero-truncated form) and
logseries distributions provides a more rigorous means of determining
deviation from the log-normal, and using Akaike’s information criterion (AIC) to
compare the two models reveals that in Fig. 1a the zero-truncated log-nor mal
provides a superior fit, while in Fig. 1b, the logseries is the better-fitting model
(AIC values are given in Fig. 1).
It is important to note that the use of the log-normal in such circumstances is
based on the assumption that it pro vides a good fit to the undisturbed data,
but other distri butions, including the logseries, have often been found to fit
data from undisturbed systems better (e.g. Syrek et al. 2006), while disturbed
communities have been shown to follow a log-normal distribution with similar
parameters to undisturbed communities (Nummelin 1998). Further more,
under acute stress, communities have been observed to go from logseries
back to log-normal, but with very different parameters (H agvar 1994). Due to
these issues, other SAD models have been proposed as more accurate
descriptors of undisturbed and disturbed communities, for example the Zipf–
Mandelbrot model (Mouillot & Lepre tre 2000), geometric series (Kim, Cho &
Chon 2013), bro ken stick and negative binomial distribution models (Syrek et
al. 2006), and power law and niche partitioning models (Tang et al. 2010).
Thus, fitting a suite of models to both undisturbed and disturbed communities
can be informative (e.g. Syrek et al. 2006). Additionally, instead of attempting
to detect deviation from log-normal to log series distributions, the
aforementioned multimodal SAD models can be used to measure the impact of
disturbance. For example, Dornelas et al. (2009) showed that for agri cultural
weed communities, homogenous environments were characterized by a
standard unimodal log-normal SAD. However, weed communities in
increasingly hetero geneous environments (e.g. due to tillage and
nitrogen fertilization) exhibited increasingly multimodal SADs (Dornelas et al.
2009). A further issue relates to the fact that the SAD of a poorly sampled
assemblage (i.e. a sam ple which represents a small proportion of the number
of individuals in the assemblage) may be mistaken for a dis turbed
community’s SAD. This is because in the situation where individuals are
randomly distributed across space, small samples from log-normal-like
assemblage SADs can result in truncated log-normal curve shapes (Green &
Plotkin 2007). Thus, it is important to attempt to keep sampling intensity
constant between the communities being studied.
In addition to analysing the fit of particular distribu tions, disturbance studies have
2
focused on particular model parameters. For instance, the r parameter of the log-
2
normal can be used to elucidate diversity information: large r means that relative
abundances are spread unevenly amongst species and thus diversity is low and
2
vice versa. Sæther, Engen & Grøtan (2013) showed that r significantly varied
according to the level of pollution for a macro-benthos community in a Scottish
lake (see also Dornelas et al. 2009; Tang et al. 2010). The a (shape) parameter of
the gambin model has also been shown to be an effective metric in such
circumstances (Ugland et al. 2007; Matthews et al. 2014). The gambin model is a
use ful SAD model in that it provides a good fit to a variety of empirical SADs, from
log-normal to logseries curve shapes (Matthews et al. 2014). Gambin is a single
param eter model, and this parameter (a) characterizes the shape of the SAD in a
single value. Low values indicate logser ies SAD shapes, and higher values
indicate log-normal curve shapes (Ugland et al. 2007). As such, a may have
significant potential in disturbance ecology. For instance, as described above, as a
community becomes increasingly disturbed, the SAD should shift from log-normal
to
logseries-like. Gambin’s a provides a simple tool with which to measure this
change. This process can be neatly illustrated using marine benthic
invertebrate data from the EKOFISK oil field, Norway (K.I. Ugland Personal
communication). The data consist of a number of unpol luted samples
(average distance of 3000 m from the oil platforms) and a smaller number of
highly polluted sam ples (100 m from the oil platforms). We have merged the
unpolluted and polluted samples and plotted the SAD of each in histogram
form (Fig. 2). It is clear that, as with simulated disturbance (Fig. 1), the SAD of
the unpolluted sample (Fig. 2a) approximates the log-normal, while the SAD of
the polluted sample (Fig. 2b) is closer to a logser ies distribution. Fitting the
gambin model to both data sets generates a = 2 2 (the mean of 100 iterations
of resampling the unpolluted sample to match the N of the polluted sample = 1
64), while a = 0 38 for the polluted samples (lower, indicating curve shapes
similar to a log series distribution). This exemplifies how SAD models provide a
clear and easily understood method, for exam ple to assess the effects of
disturbance on ecological communities, or to measure the impact of a
particular conservation action.
Disturbance as an ecological concept is wide ranging and need not be
confined to pollution. Habitat loss and fragmentation are also forms of
ecological disturbance that may impact on community properties, including
the SAD, which in turn may offer a rapid assessment tool (Hill et al. 1995; but
see Nummelin 1998). The aforemen tioned deconstruction approach would be
beneficial in this endeavour as the SAD should become more logseries-like
with increasing fragmentation. However, and as with the SAR (above), an influx
of generalist and matrix species into recently fragmented habitat may modify
or obscure changes in the shape of the SAD.
Fig. 1. Using the species abundance distribution to determine the effect of
disturbance on biotic community structure. Panel (a) represents an undisturbed
community, and the data were simulated by generating random data to follow a
log-normal distribution (N = 3000; S = 150). We then applied a perturbation (i.e. a
disturbance event) to this community, by randomly sampling 5% of the indi viduals
(i.e. randomly removing 95% of individuals). This resulted in a more logseries-like
distribution of abundances (panel b; N = 150; S = 73). In each instance, we fit the
Poisson log-normal distribution (PLN; zero-truncated version) and logseries
distribution to the data and calculated the Akaike’s information criterion (AIC) for
the fit of each distribution. To calculate AIC, the distributions were fitted to the raw
abundance data (i.e. not binned). The data were binned for graphical purposes
only. The predicted values on the plots (a,b) have been generated from fitting the
models to the binned data; the predicted value of the first octave from the
logseries distribution has been omitted to increase clarity. In (a), the PLN clearly
provides a superior fit, while in (b), the logseries has a lower AIC.

Fig. 2. The effect of pollution on a marine benthic invertebrate species

abundance distribution. The data are marine benthic inverte brates from the
EKOFISK oil field, Norway, and correspond to 30 unpolluted samples (average
distance of 3000 m from the oil plat forms) and two highly polluted samples
(100 m from the oil platforms). The 30 unpolluted samples, and the two
polluted samples, were merged to create (a) a single unpolluted, and (b) a
single polluted, sample. The Poisson log-normal (zero-truncated; triangles),
 logseries (squares) and gambin (circles) distributions were fit to both sets of
data (binned); the fit of the Poisson lognormal distribution to (b) is not shown
as the fitting algorithm produced a warning. The a value of the unpolluted
sample is 2 2 (the mean of 100 iterations of resam pling the unpolluted sample
to match the N of the polluted sample = 1 64), while a equals 0 38 for the
polluted samples. The data were obtained from K.I. Ugland (Personal
communication).

A summary of methods
Histograms are not the only method for plotting the SAD, and Fig. 3 provides a
graphical summary of the various SAD methods available for examining the
impact of disturbance and management actions on ecological communities.
For example, Lambshead, Platt & Shaw (1983) pioneered the K dominance plot
(Fig. 3a). This method plots cumulative abundance percentages against
species rank and has been argued to be a useful way of determining the effect
of disturbance on the SAD, but has been little explored. In K dominance plots,
if a curve lies completely below another curve (as community A does in Fig.
3a), then this community can be defined as being more diverse. A similar
method derived in the context of marine benthic communities uses K
dominance plots to compare the distribution of individuals amongst species,
with that of biomass (Fig. 3b; e.g. Warwick 1986). It is based on the theoretical
consideration that the standard individual-based SAD should behave
differently from the biomass-based SAD when affected by disturbance. Unpol
luted systems are characterized by the biomass curve appearing above the
abundance curve, as in Fig. 3b, and vice versa. Methods based on departures
from a log-nor mal SAD shape (using histograms) and comparisons of model
parameters (Fig. 3c,d) have been discussed above. The empirical cumulative
distribution function (ECDF) can be plotted instead of the commonly used
probability density function and can be useful for highlighting differ ences in
SADs between communities (Fig. 3e). For instance, three hypothetical SADs
(here, we have formu lated the data ourselves for effect) are plotted in ECDF
form in Fig. 3e, each relating to a river invertebrate com munity under different
pollution scenarios: high and mildly polluted, and unpolluted. In (e), the
unpolluted
curve lies below the two polluted curves at the left hand side of the plot,
indicating that there are higher propor tions of low abundance species in the
two polluted com munities. Finally, the slopes of SAD models plotted using
rank abundance diagrams (RAD) offer an alternative to histograms for
comparing SADs between communities (see Fig. 3f).
OTHER APPLICATIONS
Reviewing the literature reveals that SADs have been used in numerous other
subfields within applied ecology, albeit sometimes sporadically. However,
these uses have not been synthesized within a single review, and thus, the full
potential of SAD models in applied ecology may be underappreciated. As
such, we now review these different applications of SADs under five broad
headings.
Conservation planning
The use of abundance data, and in particular of SADs, has significant potential
in the field of conservation plan ning (e.g. Pearce & Ferrier 2001; Dunstan et
al. 2012). For instance, a study by Dunstan et al. (2012) explored the use of
SAD information in the selection of biodiver sity hotspots for the benthic fish
and invertebrates of the continental slope and shelf, south-west Australia.
Using a novel form of rank abundance diagram (see Foster & Dunstan 2010),
these authors found that incorporating SAD information greatly improved the
identification of biodiversity hotspots as it provided novel information on
which areas had a high proportion of rare species. The shape of RAD curves
can be used to determine the pro portion of relatively rare species and thus is
useful for comparing sites and selecting necessary sites for conserva tion. In a
further study, an analysis based on the logseries SAD predicted both the
number of tree species in Amazo nia and patterns of dominance and rarity (ter
Steege et al. 2013). These striking results were of particular conserva tion
importance as it was predicted that as many as half he individual trees in the
Amazon belong to a group of 227 ‘hyper-dominant’ species, meaning the
remaining 50% of individuals are distributed across over 10 000 species. This
indicates that the majority of the 10 000 species are very rare, a fact obscured
when only focusing on the spe cies richness of the region. Such findings are
interesting in themselves, but it is the general message of these papers that is
most relevant in the context of this review, namely that most biodiversity
conservation and land acquisition decisions are largely based on the use of
species richness as an index of biodiversity (Pearce & Ferrier 2001; Dunstan et
al. 2012), which may not reveal the whole pic ture. It is only when one
considers the SAD of a system that any idea of the distribution of rarity and
community structure can be inferred, and thus, ultimately conserva tion
planning could be greatly improved through broader appreciation of the
potential information content of SADs.
An integral part of conservation biogeography over the last 30 years has been
predicting extinctions related to habitat loss and fragmentation (Whittaker &
Fern andez Palacios 2007; Ladle & Whittaker 2011). This endeavour has largely
centred on the backwards use of the power law SAR model, a methodology
which has been criticized for being theoretically incorrect and producing
erroneous extinction estimates (see Whittaker & Fern andez-Palacios 2007;
He & Hubbell 2011). However, recent work has shown that incorporating the
SAD into extinction predic tions can greatly improve the accuracy of various
metrics (Kitzes & Harte 2014). For instance, the ‘extinction–area relationship’
and ‘probabilistic species–area relationship’ of Kitzes & Harte (2014) are
based on the logseries SAD and upper-truncated geometric spatial abundance
distri bution and have been shown to be more flexible and theo retically
 appropriate than the power law SAR.
Finally, focusing on changes to the SAD through time is likely to provide interesting
additional insights for bio diversity conservation. For instance, if we return to the
area of conservation planning, it can be seen that there has been an increasing
focus on temporal turnover and long-term persistence in reserve selection in the
last two decades (e.g. Rodrigues, Gregory & Gaston 2000; van Teeffelen, Cabeza
& Moilanen 2006). Specifically, it has been postulated that reserves and reserve
networks are likely to be more successful in their aim to conserve biodi versity in
the long term if the reserve sites are selected through incorporation of abundance
data of any species of conservation interest (Rodrigues, Gregory & Gaston 2000).
Sites should be located where these species are locally abundant, and thus, the
probability of long-term persistence is higher. SADs can help in this endeavour as
they give a broader perspective of relative abundances of a set of species in the
network or at particular sites. For instance, theoretical work has shown that
varying the assumptions on the processes affecting variations in popu lation size
through time leads to different SADs (see dis cussion in Sæther, Engen & Grøtan
2013).
Fig. 3. Species abundance distribution-based methods for determining the effect
of disturbance in applied ecology using simulated (a–e) and empirical (f) data. (a)
A K dominance plot (i.e. a plot of cumulative abundance percentages against
species rank; Lambshead, Platt & Shaw 1983). If a curve lies completely below
another curve [as community A does in (a)], then this community can be defined
as being more diverse. The vertical dashed lines in (a) indicate the species
richness of each community. (b) Abundance/biomass plots (i.e. a K dominance plot
that compares the distribution of individuals amongst species with that of
biomass; see Warwick 1986). Unpolluted sys tems are characterized by the
biomass curve appearing above the abundance curve, as in (b), and vice versa. (c)
The fit of the log-normal (PLN; here, the zero-truncated form) to an empirical SAD
(here, binned) is often used to assess the impact of disturbance on species
assemblages. (d) Variation in the logseries alpha along a hypothetical disturbance
gradient in which an increasing proportion of the indi viduals from a community
(simulated to follow a log-normal SAD; N = 2396, S = 52) are randomly lost
through sampling from the community (left to right on the x-axis represents
increasing hypothetical disturbance and loss of individuals). Sampling was
repeated 100 times (dots = median alpha values; shading = 95% confidence
intervals). (e) Three hypothetical SADs plotted using the empirical cumu lative
distribution function (ECDF). The data were created for this figure in order to
illustrate the method, but might for instance relate to river invertebrate
communities subjected to different degrees of pollution: heavily and mildly
polluted, and unpolluted. In (e), the unpolluted curve lies below the two polluted
curves at the left hand side of the plot, indicating that there are a higher
proportion of low abundance species in the polluted communities. (f) The SADs of
the pooled polluted (triangles) and unpolluted (circles) Norwegian mar ine
invertebrate samples (see Fig. 2. caption) are plotted in rank abundance form. The
fits of the truncated log-normal model to the unpolluted samples, and the
logseries model to the polluted data, are displayed. In the polluted sample, it is
clear that there are more observed rare species predicted even by the logseries.
The plots were constructed using the ggplot2 R package (Wickham 2009).

Conservation biological control

Conservation biological control (CBC) is ‘the manipula tion of the environment
to favour natural enemies by Species abundance distribution in applied
ecology 451
either removing or mitigating adverse factors, or provid ing requisites that are
lacking in natural enemies’ habi tat’ (Barbosa, Caldas & Riechert 2005, p.
345). In contrast to classic biological control, which generally focuses on one
or two predator–prey species interactions, CBC focuses on whole
assemblages. Thus, a key aspect is analysing how abundances are distributed
across dif ferent pest species and their natural enemies. In the con text of
CBC, Barbosa, Caldas & Riechert (2005) introduced a new method of
constructing and plotting SADs called the Robbins’s curve (see Fig. 4). As part
of this method, incremental percentage classes of rarity or dominance (i.e.
10%, 20%, 30% up to 100%) are plot ted on the x-axis, and the cumulative
percentage of abundance is plotted on the y-axis. For example, the first data
point along the x-axis in Fig. 4 relates to the number of individuals that belong
to the rarest 10% of species and the second point to the rarest 20% and so
on. The use of Robbins’s curves for visualizing the SAD is beneficial as (1) the
method is not biased by sample size, (2) it enables easy and accurate
identification of small differences in the dominance of species, and (3) it
provides a method for assessing the influence of manage ment prescriptions
on SADs, for example determining the effect of chemical pest control on the
abundance of natural predators in agroecosystems (Barbosa, Caldas &
Riechert 2005). Robbins’ curves have rarely been applied in practice and
represent an interesting avenue of future research. As we were unable to find
any software for constructing a Robbins’ curve, we provide the R code used to
construct Fig. 4, in Appendix S1 (Supporting information).
 Monitoring communities and assessing community recovery
The SAD has occasionally been employed in a monitoring capacity, for
example in the monitoring of fish communi ties as part of fisheries
management (e.g. Ambak & Mohsin 1986) and for measuring the recovery of
fungi communities following managed fire disturbance (e.g. Persiani & Maggi
2013). Within such contexts, it is gener ally the performance of rare species
that is of interest, that is rare species are the species of most conservation
and management concern. Analysis of the SAD over time allows managers to
observe changes in the abundance of rare species relative to the other
species in the community. Studies have also examined how the SAD changes
along succession gradients (e.g. Bazzaz 1975), and this type of application
could easily be extended to use in forestry management. For instance, it has
been shown that at the early stages of succession (e.g. following the felling of
trees for timber), deciduous forest plots in Illinois, USA, exhibit a SAD that
follows a geometric series, while the SAD of latter stage forest plots follow
log-normal distri butions (Bazzaz 1975). Upscaling and downscaling the
species abundance distribution
A large theoretical debate has ensued over the scaling properties of SADs,
that is determining whether SADs can be upscaled and/or downscaled (Green
& Plotkin 2007; Sizling et al. 2009; Borda-de-Agua et al. 2012). The ability
to accurately upscale SADs would be of great practical utility due to the
economic and logistical con straints of sampling large regions and would allow
for a deeper assessment of risk than just extrapolating species richness
estimates over broad scales (e.g. ter Steege et al. 2013).
Using measures other than abundance
Almost all SAD studies use the number of individuals as the measure of
abundance. However, other measures can be used, such as biomass (e.g.
Anderson, Chiarucci & Williamson 2012) and resource use (e.g. Morlon et al.
2009). It has been shown that these different measures are not equivalent, that is
individuals, biomass and resource use are not distributed amongst species in the
same way (see Morlon et al. 2009; Henderson & Magurran 2010). Thus, focusing
on SADs constructed using these other measures of abundance, or preferably
focusing on all in tandem, may be a more productive way to determine the
important processes driving community assembly and provides a more robust
framework for examining the response of species to disturbance (Henderson &
Magur ran 2010). Such an approach has been an applied context, however, with
the exception of the aforementioned abundance/biomass plots (Fig. 3b; War wick
1986). Another possibility, derived in the context of island biogeography, involves
ranking species according to their incidence, that is the number of sites occupied
is treated in a similar way to abundances in standard SAD models (S. Fattorini,
Personal communication). A distri bution of incidences can then be analysed in
RAD form. This approach is based on the frequently observed posi tive
interspecific abundance–occupancy relationship (see Gaston 1994) and may be
particularly useful in frag mented landscapes where abundance data are
unavailable.
Conclusions and management
recommendations
Much of the recent work on SADs has placed a strong focus on the theoretical
aspects of the pattern. This is a worthwhile and interesting endeavour, but we
would argue that the SAD also has significant applied utility in numerous fields
within ecology and biodiversity conserva tion. As the incidence and impact of
disturbance is pre dicted to increase in the future (Sala et al. 2000), the
development and use of tools that are visually intuitive, easy to implement, are
applicable in a broad variety of ecological contexts and do not require
substantial species specific data is vital (Mouillot et al. 2013). To this end, we
conclude with a set of recommendations regarding the use of SADs in applied
ecological contexts:
1. When examining the effect of disturbance and manage ment practices on
biodiversity, the full SAD is a better summary of ecological community
characteristics than simple diversity indices. The SAD can act as an early
warning analytic for the effects of disturbance on ecologi cal communities, as
well as providing an intuitive tool for measuring the performance of ecological
management pre scriptions.
2. When analysing ‘before and after’ data (e.g. before and after a management
treatment or disturbance event), plot ting the SADs is beneficial as it allows
the user to deter mine which part of the abundance spectrum (e.g. common
species or rare species) is most affected by the treatment. In this regard, Figs
3 and 4 provide a useful summary of the various methods available for plotting
the SAD, each providing unique information. Depending on the aim of the user,
multiple plotting methods will likely provide complementary information.
3. Species abundance distribution model parameters also represent useful
tools for comparing communities in a management context. For example, the a
parameter of the gambin model neatly summarizes the shape of the SAD and
can be used in comparative analysis and regression models.
4. Where the data allow, it can be enlightening to decon struct the full
assemblage into various subsets (e.g. based on habitat specialization) and to
examine the SAD pat terns in the subsets separately. This enables the user, for
example, to determine the impact of a specific treatment on the SAD of
specialist species, these species generally being of conservation concern.
Combining an assemblage deconstruction approach with the fitting of
multimodal SAD models can be particularly informative in this regard.
5. Species abundance distributions have uses in applied ecology beyond the
traditional applications in disturbance ecology. We have reviewed a number of
such applications here (e.g. fisheries management, conservation biological
control and conservation planning), and we urge workers and managers in
these fields to incorporate SADs into their work.
The underlying theory, mechanisms and general proper ties of SADs have
 been heavily debated (see McGill et al. 2007). However, a comprehensive and
unified SAD theory is not necessary for using the SAD in an applied ecologi cal
context, and there are many management situations where the SAD can
provide useful information. We hope this review acts as a catalyst for a greater
uptake and use of SAD-based methods in applied ecology.

Fig. 4. Exemplar species abundance distributions plotted using the Robbins’s

curve method. Data are for tabanids sampled using two different methods: Malaise
traps (number of species = 44) and aerial netting (number of species = 40). The
data are from Tallamy, Hansens & Denno (1976; in Barbosa, Caldas & Riechert
2005). For the R code to construct the plot, see Appendix S1 (Supporting
information).