You are on page 1of 12

Journal of Theoretical Biology 386 (2015) 147–158

Contents lists available at ScienceDirect

Journal of Theoretical Biology


journal homepage: www.elsevier.com/locate/yjtbi

Analytical formulae for computing dominance


from species-abundance distributions
Tak Fung a,n, Laura Villain b, Ryan A. Chisholm a,c
a
National University of Singapore, Department of Biological Sciences, 14 Science Drive 4, 117543 Singapore
b
Institut National des Sciences Appliquées de Lyon, 20 Avenue Albert Einstein, Villeurbanne cedex, 69621 Lyon, France
c
Smithsonian Tropical Research Institute, Balboa, Ancón, Republic of Panama

H I G H L I G H T S

 Ecosystem stability and functioning critically dependent on species evenness.


 Evenness quantified as proportion of species that dominate total abundance.
 Formulae for dominance derived for different species-abundance distributions (SADs).
 High dominance found for SADs with a log-series, lognormal or exponential form.
 Formulae used to show how dominance increases with environmental variance.

art ic l e i nf o a b s t r a c t

Article history: The evenness of an ecological community affects ecosystem structure, functioning and stability, and has
Received 15 April 2015 implications for biodiversity conservation. In uneven communities, most species are rare while a few
Received in revised form dominant species drive ecosystem-level properties. In even communities, dominance is lower, with
6 August 2015
possibly many species playing key ecological roles. The dominance aspect of evenness can be measured
Accepted 12 September 2015
Available online 25 September 2015
as a decreasing function of the proportion of species required to make up a fixed fraction (e.g., half) of
individuals in a community. Here we sought general rules about dominance in ecological communities by
Keywords: linking dominance mathematically to the parameters of common theoretical species-abundance dis-
Biodiversity tributions (SADs). We found that if a community’s SAD was log-series or lognormal, then dominance was
Evenness
almost inevitably high, with fewer than 40% of species required to account for 90% of all individuals.
Gamma
Dominance for communities with an exponential SAD was lower but still typically high, with fewer than
Lognormal
Log-series 40% of species required to account for 70% of all individuals. In contrast, communities with a gamma SAD
only exhibited high dominance when the average species abundance was below a threshold of
approximately 100. Furthermore, we showed that exact values of dominance were highly scale-depen-
dent, exhibiting non-linear trends with changing average species abundance. We also applied our for-
mulae to SADs derived from a mechanistic community model to demonstrate how dominance can
increase with environmental variance. Overall, our study provides a rigorous basis for theoretical
explorations of the dynamics of dominance in ecological communities, and how this affects ecosystem
functioning and stability.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction sustained and ongoing loss of biodiversity (Chapin et al., 2000;


Cardinale et al., 2012; Naeem et al., 2012). Since biodiversity is a
Human populations have caused massive and widespread fundamental determinant of ecosystem functioning, its loss affects
changes to the structure of natural communities (Vitousek et al., functioning and associated ecosystem services (Cardinale et al.,
1997; Chapin et al., 2000; Steffen et al., 2007). This has resulted in 2012; Naeem et al., 2012). Research over the last few decades has
focused on the negative effects of species loss on functioning
(Balvanera et al., 2006; Cardinale et al., 2006, 2012; Naeem et al.,
n
Corresponding author.
2012), but the number of species is just one component of biodi-
E-mail addresses: tfung2000@gmail.com (T. Fung), versity. Another important component is species evenness, which
villain.laura@gmail.com (L. Villain), ryan.chis@gmail.com (R.A. Chisholm). describes the relative abundances of species in a community.

http://dx.doi.org/10.1016/j.jtbi.2015.09.011
0022-5193/& 2015 Elsevier Ltd. All rights reserved.
148 T. Fung et al. / Journal of Theoretical Biology 386 (2015) 147–158

Evenness is arguably more directly related than richness to the 2. Methods


trait diversity on which the level of functioning depends (Reiss
et al., 2009). Recent experiments have demonstrated substantial 2.1. Deriving analytical formulae for calculating dominance: discrete
positive effects of evenness on biomass (Wilsey and Potvin, 2000; SADs
Mattingly et al., 2007; Zhang et al., 2012; Orwin et al., 2014), water
retention (Orwin et al., 2014), resistance to invasion (Wilsey and To derive a formula specifying dominance given an SAD for a
community, the minimum fraction of species (i.e., the most
Polley, 2002) and nutrient retention after leaching (Orwin et al.,
abundant ones) required to account for a fraction u of the indivi-
2014). Moreover, evenness is expected to change quicker than
duals in the community, denoted by pu , must first be computed.
richness in response to human stressors, since changes in abun-
We consider the case u Z 0:5, for which the minimum set of spe-
dance occur before species go extinct. Therefore, quantitative
cies is called the set of dominant species, making up at least half of
assessments of the causes and consequences of changing evenness all individuals. As pu decreases, fewer species are required to
are essential for a thorough understanding of ecosystem structure account for a fraction u of all individuals, such that dominance
and function. increases. Thus, dominance is defined as Du ¼ 1  pu .
One metric of evenness is dominance, which has been used to To calculate pu , we introduce a quantity nu , which is defined as
quantify the degree to which the total abundance of a community the abundance of the least abundant dominant species. Three
is dominated by a subset of species (e.g., Grime, 1998; Barker et al., simple examples are now given to illustrate clearly the meaning of
2002; Binkley, 2004; Davidar et al., 2005; Hillebrand et al., 2008; pu and nu , together with some difficulties with interpretation in
Dornelas et al., 2011; ter Steege et al., 2013). Dominant species specific cases. First, consider a community with S ¼ 8 species and
tend to have large effects on flows of energy and matter in an abundances f1; 1; 3; 4; 4; 7; 8; 12g. In this example, p0:5 ¼ 0:25
ecological network (Hillebrand et al., 2008; Berlow et al., 2009). In because we can account for half of the individuals with just one
particular, empirical evidence suggests that in autotrophic and quarter of all species (i.e., the two most abundant ones). We also
suspension-feeding invertebrate assemblages, the contribution of have n0:5 ¼ 8, because species with abundance greater than or
a species to functioning can be predicted largely by its biomass – equal to 8 comprise half of the community. Second, consider
another community with eight species but with abundances
the “mass ratio hypothesis” (Grime, 1998; Davies et al., 2011).
f1; 4; 4; 4; 5; 5; 5; 8g. Here we have p0:5 ¼ 0:375 because we can
Anthropogenic stressors such as targeted exploitation (e.g., stocks
account for half of the individuals with the three most abundant
of Atlantic cod, Gadus morhua; Lotze and Worm, 2009) may dis-
species. However, we note that the value of n0:5 is biologically not
proportionately affect dominant species, resulting in heavy loss of
well-defined here. It might seem logical to take n0:5 ¼ 5, but this is
functioning (Gaston and Fuller, 2008). Ecosystems with high problematic because to get 50% of all individuals we only want
dominance also typically have many rare species, posing chal- two of three species with abundance 5. Thus, the quantity nu is
lenges for biodiversity conservation. generally not as useful as pu , although it is necessary to calculate
Studies have quantified dominance for particular communities nu in the derivation of pu . As a final simple example, consider yet
(e.g., Grime, 1998; Barker et al., 2002; Binkley, 2004; Davidar et al., another community with eight species, this time with abundances
2005; Hillebrand et al., 2008; Dornelas et al., 2011; ter Steege et al., f1; 3; 4; 6; 6; 7; 8; 15g. In this case, the top three species are required
2013), but as yet there is no general theory for how dominance to reach at least 50% of individuals, but we actually overshoot and
emerges from standard ecological models. In this study, we first get 60% of individuals. Such overshoots arise from the funda-
derive mathematical formulae expressing dominance as a function mental discreteness of individuals and pose problems of inter-
of the parameters of four standard species-abundance distributions pretation. Thus, in our mathematical treatment, we will work in
(SADs) – the log-series, exponential, gamma and lognormal. Spe- the high-diversity, large-community size limit in which overshoots
cifically, for each SAD we derive formulae specifying dominance as a are negligible, with the understanding that in practical applica-
decreasing function of the number of species accounting for a tions, theoretical values for pu S, where S is the expected total
number of species, will need to be rounded to the nearest integer
proportion u of the total number of individuals (dominance is
to get a whole number of species.
defined in this way because fewer dominant species equates to
We first detail a method for deriving exact formulae for pu and
higher dominance). As u increases, more species are required for
hence Du for discrete SADs, in the high-diversity, large-community
dominance of numerical abundance in the community considered.
size limit. Consider a discrete SAD defined by a probability dis-
We examine four SAD forms that have been commonly studied and tribution ψ ðnÞ, specifying the probability that a randomly selected
fitted to empirical data: the log-series (Fisher et al., 1943; Hubbell, species will have abundance n. We have:
2001), exponential (Cohen, 1968), gamma (Brian, 1953; Plotkin and
sðnÞ sðnÞ
Muller-Landau, 2002; Forster and Warton, 2007) and lognormal ψ ðnÞ ¼ ¼ 1
P
; ð1Þ
S
(Preston, 1948; Engen et al., 2002; Volkov et al., 2003; Forster and sðnÞ
n¼1
Warton, 2007; Engen et al., 2011) forms. After deriving the dom-
inance formulae, we proceed to apply them to (i) gauge the levels of where sðnÞ is the expected number of species with abundance n. To
dominance expected in natural communities under biologically obtain nu , the equation
plausible scenarios; (ii) assess dominance in the metacommunity of P
1 P
1
nsðnÞ nψ ðnÞ
tree species found in Amazonia, using the data of ter Steege et al.
FðnÞ ¼ n ¼ m
¼n¼ m
¼u ð2Þ
(2013); and (iii) quantify how dominance is expected to change P1 P
1
nsðnÞ nψ ðnÞ
with increasing environmental variance, using the dynamic model n¼1 n¼1
of Engen and Lande (1996a). These three applications of our for- can be solved analytically for m. Here, FðmÞ is the fraction of
mulae highlight their three-fold utility: they allow dominance to be individuals belonging to species with abundance greater than or
quantified quickly for discrete SADs, they permit dominance to be equal to m.
quantified for continuous SADs, and they can be mathematically In general, the value of nu obtained would not be an integer, so
analyzed further to derive general biological insights that cannot be it is rounded up to the nearest integer, ⌈nu ⌉. Species with an
obtained from simulations alone. abundance equal to or greater than ⌈nu ⌉ are insufficient to make
T. Fung et al. / Journal of Theoretical Biology 386 (2015) 147–158 149

up a fraction u of all individuals, so a proportion η of species with analytical tractability, which is why we used 1, 0 and 0 for the
abundance ⌈nu ⌉  1 are required. The quantity η satisfies three distributions, respectively. For the continuous exponential,
X
1 X
1 all three values of δ (0, 0.5 and 1) are examined since analytical
nψ ðnÞ þ ηð⌈nu ⌉  1Þψ ð⌈nu ⌉  1Þ ¼ u nψ ðnÞ: ð3Þ formulae for nu and pu can be derived for all three δ values.
n ¼ ⌈nu ⌉ n¼1

Rearranging gives 2.3. Applying the formulae to quantify the typical level of dominance
in ecological communities
P
1 P
1
u nψ ðnÞ  nψ ðnÞ
n¼1 n ¼ ⌈nu ⌉ The formulae derived using the methods described in Subsec-
η¼ : ð4Þ
ð⌈nu ⌉  1Þψ ð⌈nu ⌉  1Þ tion 2.1 and 2.2, for two types of discrete SAD and four types of
The quantity pu , which is the minimum fraction of species continuous SAD, are applied to quantify dominance across a broad
required to make up a fraction u of all individuals, can then be range of biologically possible scenarios. Specifically, parameter
calculated as values are specified for each type of SAD examined to give a broad
range of average species abundances ðJ=SÞ, from small to very
P
1 P
1
u nψ ðnÞ  nψ ðnÞ large, and then dominance values calculated for these parameter
X
1
n¼1 n ¼ ⌈nu ⌉ X
1
pu ¼ ηψ ð⌈nu ⌉  1Þ þ ψ ðnÞ ¼ þ ψ ðnÞ: values using the formulae. To facilitate comparison of dominance
⌈nu ⌉  1
n ¼ ⌈nu ⌉ n ¼ ⌈nu ⌉ values between different types of SAD, dominance for each type is
ð5Þ plotted against J=S, which is defined for all SADs. It is noted that
J=S for an SAD is equal to the mean of the underlying distribution;
The exact formula for dominance is then Du ¼ 1  pu . Usually,
this can be seen for discrete SADs because
the formula for pu cannot be simplified further because of the
ceiling function. However, simpler analytical approximations for P
1 P
1
nsðnÞ nψ ðnÞ X
1
pu can be derived by retaining the generally non-integer but J n¼1
¼ 1 ¼ n ¼11 ¼ nψ ðnÞ ð9Þ
analytical form of nu and then using S P P
sðnÞ ψ ðnÞ n¼1
X
1 n¼1 n¼1
pu  pu ¼ ψ ðnÞ: ð6Þ
and it can be seen for continuous SADS using analogous calcula-
n ¼ nu
tions. In addition, we mathematically analyze the dominance for-
We derive pu and pu according to Eqs. (5) and (6), respectively, mulae for the continuous log-series, continuous exponential and
when ψ ðnÞ follows a discrete log-series distribution and when it lognormal SADs to gain general insights into how they change
follows a discrete exponential distribution. We then derive Du and with J=S.
Du for these two distributions as 1 pu and 1  pu , respectively. For the discrete log-series SAD with parameter x, we examine
the parameter range ½4:65  10  10 ; 0:980, which gives J=S from
2.2. Deriving analytical formulae for calculating dominance: con- 1.01 (corresponding to the maximum x) to 108 (corresponding to
tinuous SADs the minimum x). 1.01 is close to the theoretical minimum of
1 whereas 108 is on the order of magnitude of the median number
Dominance formulae for continuous SADs can be derived ana- of individuals found for 207 dominant tree species in Amazonia,
logously to those for discrete SADs. For these distributions, nu can which covers an area of 6.29 million km2 (ter Steege et al., 2013).
take non-integer positive values and can be found by solving Thus, the range of J=S that we examine is a wide range encom-
R1 passing most values likely to be encountered in reality. The range
nψ ðnÞdn
GðmÞ ¼ Rn1¼ m ¼u ð7Þ of λ for the discrete exponential SAD is chosen in the same way as
n ¼ δ ψ ðnÞdn
n
½1:00  10  8 ; 4:62, with the maximum value corresponding to J=
for n, where ψ ðnÞ is the probability density corresponding to S ¼ 1:01 and the minimum value corresponding to J=S ¼ 108 . For
abundance n and δ is the lower limit of the domain of the corre- each of the two types of discrete SAD, we calculate exact and
sponding probability distribution. An analytical formula specifying approximate dominance values, Du and Dnu, respectively, over the
pu can then be derived using range of parameter values considered. The values of u (fraction of
Z 1 individuals accounted for by the dominant species) examined
pu ¼ ψ ðnÞdn; ð8Þ range from 0.5 to 0.9 in increments of 0.1. We also quantify the
n ¼ nu
error Du-Dnu for each combination of u and x or λ (i.e. each com-
allowing the corresponding dominance formula to be derived bination of u and J/S).
using Du ¼ 1 pu . We follow this method to derive dominance Parameter values for the continuous SADs are chosen using the
formulae for four continuous distributions that approximate dis- same rationale as for the discrete SADs. Thus, the range of x for the
crete SADs: the continuous log-series, exponential, gamma and continuous log-series SAD is chosen to be ½4:79  10  10 ; 1:00,
lognormal distributions. Values of Du for the continuous log-series corresponding to a range of J=S of ½1:01; 108 . Similarly, the ranges
and exponential distributions can be considered as approxima- of λ for the continuous exponential SAD are chosen to be
tions to the exact discrete cases because the distributions permit ½1:00  10  8 ; 0:990, ½1:00  10  8 ; 1:96 and ½1:00  10  8 ; 100 for
non-integer species abundances, which are not observed in reality. lower abundance limits of δ ¼ 0, 0.5 and 1 respectively, to corre-
Because the abundance n is assumed to be continuous, there are spond to the same range of J=S. For the gamma SAD with shape
conceptual issues with choosing δ, the lowest value of n for a and rate parameters α and β, it was found that Du is independent
continuous distribution. With δ ¼ 1, values between 0.5 and 1 are of β and hence scale (θ ¼ 1=β ). This is expected because the scale
omitted, but these values become 1 when rounded to the nearest parameter does not change the shape of the gamma distribution.
integer, so could be considered biologically meaningful. With However, J=S depends on both α and β. Therefore, we (i) calculate
δ ¼ 0, values between 0 and 0.5 are included, but these values are a range of β values from previous studies on plant and insect
rounded to 0, which is not biologically meaningful since it repre- communities (Brian, 1953; Plotkin and Muller-Landau, 2002; For-
sents species with no individuals. Thus, it may be better to use ster and Warton, 2007); (ii) calculate the minimum and maximum
δ ¼ 0:5 instead. However, using δ ¼ 0:5 for the continuous log- values of β and for each, calculate the corresponding range of α
series, gamma and lognormal distributions results in loss of that would give a range of J=S of ½1:01; 108 ; (iii) calculate values of
150 T. Fung et al. / Journal of Theoretical Biology 386 (2015) 147–158

Du corresponding to the two ranges of α found; and (iv) plot 2.5. Applying the formulae to quantify how dominance changes with
curves of Du against J=S, one for each of the two ranges of α. The environmental variance
minimum and maximum β values found were 0.000321 and
0.00761 respectively, with the corresponding ranges of α being In the dynamic community model of Engen and Lande (1996a),
½0:000324; 32100 and ½0:00769; 761000 respectively. Similarly, the abundance ðn Z 1Þ of each species changes according to
for the lognormal SAD with log-scale and shape parameters μ and dn dBðtÞ
¼ rn  ngðnÞ þ σ r ðnÞ ; ð10Þ
σ , it was found that Du is independent of μ, but that J=S depends dt dt
on both parameters. Therefore, we calculated the minimum and where r is the per-capita growth rate, gðnÞ is the density-
maximum values of μ from previous studies on plant and bird dependent per-capita mortality rate, and BðtÞ is a Wiener process
communities (Preston, 1948; Volkov et al., 2003; Forster and representingstochastic  variation in the growth rate, with variance
Warton, 2007) as 1.39 and 3.47 respectively, and for each value σ 2r ðnÞ ¼ σ 2e þ σ 2d =n . This variance is the sum of two components,
calculated a corresponding range of σ that would give a range of the first representing environmental variance and the second
J=S of ½32:2; 108 . Values of J=S lower than 32.2 are not considered demographic variance. Under a Gompertz growth curve gðnÞ ¼ γ ln
ðn þ εÞ with ε ¼ σ 2d =σ 2e (there is a typo on p. 175 of Engen and Lande
because they correspond to negative values of σ when the max-
(1996a) – they mistakenly stated that ε ¼ σ 2e =σ 2d ), the SAD at the
imum value of μ ¼ 3:47 is used. For each range of σ , we calculate
steady state is given by
corresponding values of Du and plot them against J=S. The values "   2 #
of u examined range from 0.5 to 0.9 in increments of 0.1, as for the A 1 lnðn þ εÞ  r=γ
SðnÞ ¼ pffiffiffiffiffiffiffiffiffiexp  ; ð11Þ
discrete SADs. In addition, for each of the log-series and expo- ðn þ εÞσ e π =γ 2 σ 2e =2γ
nential SADs, the error in Du arising from using the continuous
where A is a scaling constant (Engen and Lande, 1996a). Explicitly,
version is quantified as the difference between Du derived using
pffiffiffiffiffiffiffiffiffi (   )
the exact formula for the discrete version minus Du derived using 2ω π =γ γ r 2
A¼ exp lnð1 þ ε Þ  ; ð12Þ
the formula for the continuous version, for each combination of u σe 2σe γ
and J=S.
where ω is the per-capita speciation rate. Using the transformation
z ¼ n þ ε, SðzÞ follows a truncated lognormal distribution
pffiffiffiffiffiwith
ffi log-
scale parameter μ ¼ r=γ and shape parameter σ ¼ σ e = 2γ over the
2.4. Applying the formulae to assess dominance for tree species in range ½1 þ ε; 1Þ. This distribution can be approximated by the full
Amazonia lognormal distribution with range ½0; 1Þ if the density of the full
distribution is approximately zero in the range ½0; 1 þ εÞ, which is
To estimate the total number of tree species in Amazonia (the the case for large communities. In this case, we apply our dom-
Amazon and Guiana Shield), ter Steege et al. (2013) fitted a dis- inance formula for the lognormal distribution to explicitly relate
crete log-series SAD to species population abundances estimated dominance to environmental variance. We mathematically analyze
from their sample data and then extrapolated from this distribu- this relationship to show how dominance changes with increasing
tion. Using this method, they estimated that Amazonia has environmental variance, and illustrate the trend found with
approximately 16,000 tree species. However, their low sampling example parameter values.
intensity of 0.0002% (not 0.002% as ter Steege et al. state on p. 4 of
their Supplementary material) means that the distribution of rare
species is essentially unknown. Thus, other theoretical SADs, not 3. Results
considered by ter Steege et al. (2013), provide equally good fits to
3.1. Analytical dominance formulae for SADs
their estimated population abundances and give virtually the same
sample SAD (Supplementary material 1). These other theoretical
For the six types of SAD that we considered, explicit formulae
SADs lead to vastly different estimates of the total Amazon tree were derived specifying dominance, measured as 1 minus the
species richness S. The examples in Supplementary material 1 minimum proportion of species required to account for a propor-
show plausible Amazon-scale discrete log-series SADs truncated tion u of all individuals. This measure of dominance is denoted by
such that S¼ 12,000, 9,000 and 6,000, with removal of increasingly Du and varies between 0 and 1. In addition, for the two types of
more of the least abundant species. To calculate corresponding discrete SADs, we derived formulae specifying Du , which is an
dominance values for these alternative distributions, we adapt our approximation to the exact dominance measure Du . Table 1 lists
dominance formulae for log-series SADs and apply them to the the formulae obtained, together with formulae specifying the
truncated log-series SADs. This provides an idea of the range probability distributions underlying the SADs. Appendix A pro-
within which the true dominance may vary. Dominance is vides details of how the formulae were derived following the
expected to decrease with greater truncation because the domi- methods detailed in Subsection 2.1 and 2.2. It is seen that the
formulae for the (continuous) gamma and lognormal SADs depend
nant species would likely form a higher proportion of a smaller
only on the shape parameter, that is, α for the gamma distribution
total number of species.
and σ for the lognormal distribution.
Given the fundamental uncertainty in the distribution of rare
species in poorly sampled species-rich communities, a more
3.2. Quantifying dominance typically found in ecological
informative approach could be to fit only the observed portion of
communities
the SAD, calculate the dominance of the fitted SAD and use it as an
estimate of a lower dominance bound. We apply this approach to In general, Du values obtained from the formula for the discrete
the Amazonian tree metacommunity by calculating dominance for log-series were high, indicating that numerical abundance was
a log-series SAD fitted to the sample data and truncated to give dominated by relatively few species (Fig. 1). When u increased
S  5; 000, corresponding approximately to the number of from 0.5 to 0.9 (in increments of 0.1), Du remained above 0.7 for
observed species. average species abundances (J=S) greater than 100. As J=S
T. Fung et al. / Journal of Theoretical Biology 386 (2015) 147–158 151

Table 1
Formulae specifying the six types of species-abundance distributions (SADs) examined and the corresponding dominance. Each SAD is specified by a probability mass
function or probability density function (for discrete and continuous SADs, respectively), ψ ðnÞ, describing the probability or probability density of a species having abundance
n, respectively. Dominance is measured as Du ¼ 1 pu , where pu is the minimum proportion of species required to obtain a proportion u of all individuals. For each of the two
discrete SADs, an approximation to Du is also given as Du , which uses an approximation to pu that is specified by Eq. (6).

Probability distribution underlying SADa Dominance formula(e)

n h i
Discrete log-series ð1  xÞ
ψ ðnÞ ¼ nlog ; ð1  xÞ u  ð1  xÞ⌈nu ⌉  1 ⌈nu ⌉  1
ð1xÞ Du ¼ 1  log ð1=xÞ xð⌈nu ⌉  1Þ þ ð1  xÞ Φð1  x; 1; ⌈nu ⌉Þ ;
nZ1 log ðuÞ
nu ¼ 1 þ
log ð1  xÞ

uð1  xÞ log ðuÞ


Du ¼ 1  Φ 1  x; 1; 1 þ
log ð1=xÞ log ð1  xÞ

  h i
Discrete exponential ψ ðnÞ ¼ eλ  1 e  λn ; u  e  λ⌈nu ⌉ ð1  ⌈nu ⌉ þ ⌈nu ⌉eλ Þ
Du ¼ 1  eλ ðeλ  1Þð⌈nu ⌉  1Þ þ e  λ⌈nu ⌉ ;
nZ1 1 W  1 ðuye  y Þ
nu ¼  ;
1  eλ λ
λ

1 eλ   
Du ¼ 1  exp λ  yþ W  1 uye  y

n
Continuous log-series ψ ðnÞ ¼ nΓð0; ð1  xÞ
 log ð1  xÞÞ
Du ¼ 1  liðliuð1  xÞÞ
ð1  xÞ
,
nZ1

h i
Continuous exponential ψ ðnÞ ¼ λeλð  n þ δÞ ; Du ¼ 1  exp 1 þ δλ þ W  1  uð1 þ δλÞ
eð1 þ δλÞ
nZδ

Gamma (continuous) α
nα  1 e  βn Γðα; Q  1 ð1 þ α;uÞÞ
ψ ðnÞ ¼ β ΓðαÞ ; Du ¼ 1  ΓðαÞ
nZ0


Lognormal (continuous) exp
2
 ðμ  log2ðnÞÞ
1
Du ¼ 1  12 erfc pσffiffi2  erfc ð2ð1  uÞÞ
ψ ðnÞ ¼ pffiffiffiffi ;

nσ 2π
nZ0

a
Special functions used in the SAD and dominance formulae are the Gamma function ΓðzÞ, the incomplete gamma function Γða; zÞ, the Lerch transcendent Φðz; s; aÞ, the
Lambert W function W  1 ðzÞ, the inverse regularized gamma function Q  1 ða; zÞ, the complementary error function erfcðzÞ, the inverse complementary error function
1
erfc ðzÞ, and the logarithmic integral liðzÞ (Abramowitz and Stegun, 1972).

decreased below 100, Du decreased non-linearly to low values, S3). Thus, dominance trends for continuous log-series and
dropping below 0.2 when u ¼ 0:9 (Fig. 1). Values of Du for the exponential SADs closely follow those for their discrete coun-
discrete exponential were smaller than for the discrete log-series terparts. In addition, we showed
  analytically that for the con-
(Fig. 1), indicating that numerical abundance was dominated by tinuous log-series, dDu =d J=S 40 for small and large J=S
more species. Nonetheless, Du was generally quite high: when u (Appendix B). These analytical results hold for all u, thus gen-
increased from 0.5 to 0.7, it remained above 0.6 for J=S 4 100 eralizing the trends found from simulations for particular values
(Fig. 1a–c), only dropping below 0.6 for J=S 4 100 when u of u (Fig. 1). Furthermore, we proved that for the continuous

increased further to 0.8 and 0.9 (Fig. 1d and e). As for the discrete exponential with lower abundance limit δ 4 0, dDu =d J=S 4 0 for
log-series, Du for the discrete exponential decreased non-linearly all u (Appendix B), again generalizing the simulated dominance
when J=S decreased below 100, to below 0.2 when u ¼ 0:9 (Fig. 1). trends (Fig. 1).
The Du values for both the discrete log-series and exponential Values of Du for the lognormal distribution followed a similar
were found to be very close approximations of the exact dom- trend to those for the log-series and exponential, exhibiting high
inance values (Du values) (Fig. 2). The absolute error between the values greater than 0.6 for J=S above 200, regardless of the value of
approximate and exact values was almost zero except for very low u tested, and decreasing non-linearly as J=S decreased below 200
values of J=S below 10, when the absolute error increased but still (Fig. 1). For large J=S above 10,000, Du reached values close to 1 for
remained below 0.1 (Fig. 3). Therefore, trends in Du closely fol- all values of u tested, exceeding Du for both the log-series and
lowed trends in Du . exponential (Fig. 1). Values of Du for the gamma distribution
Values of Du for the continuous log-series are close approx- showed the opposite trend to that of the other three distributions,
imations of the corresponding values for the discrete log-series decreasing rather than increasing with J=S (Fig. 1). For small J=S
(Fig. 2a), with absolute errors typically o 0.05 and always o 0.11 below 50, Du attained high values above 0.6, but as J=S increased
(Fig. 3a). Similarly, values of Du for the continuous exponential above 50, Du decreased according to a non-linear sigmoidal shape,
were typically close approximations of corresponding values for to values below 0.6 (Fig. 1). We mathematical analyzed the form of
 
the discrete exponential, regardless of whether a lower limit of Du for the lognormal and found that dDu =d J=S 4 0 for all u
δ ¼ 0, 0.5 or 1 was used (Figs. 2b and S2 in Supplementary (Appendix B).
material 2); absolute errors were virtually zero for J=S above 100
(Figs. 3b and S3 in Supplementary material 2). The absolute 3.3. Assessing dominance of tree species in Amazonia
errors were typically higher as J=S decreased below 100, reaching
values close to 0.4 with δ ¼ 0 (Figs. 3b and S2). However, with In their study of tree species in Amazonia, ter Steege et al.
δ ¼ 0:5, the errors remained virtually zero down to J=S ¼ 10 and (2013) fitted a discrete log-series SAD with a parameter value of
remained below 0.2 for lower J=S values (Fig. 3b), and with δ ¼ 1, x ¼ 1:93  10  9 to species abundances estimated from sampled
the errors remained below 0.1 as J=S decreased below 100 (Fig. data. This parameter value corresponds to J=S ¼ 2:58  107 and
152 T. Fung et al. / Journal of Theoretical Biology 386 (2015) 147–158

Fig. 1. Dominance values ðDu Þ for the discrete log-series, discrete exponential, (continuous) gamma and (continuous) lognormal SADs, calculated using the formulae in
Table 1. Panels (a)–(e) show how dominance changes across average species abundance (total number of individuals of all species, J, divided by total number of species, S) for
five different values of u, which is the fraction of individuals accounted for by the set of dominant species.

S  16; 000. Application of our exact dominance formula (Table 1) and


shows that for this SAD, 1.89% of species is required to account for
uΦ 1  x; 1; nmin þ loglogð1ðuÞ
50% of all tree individuals, i.e. Du ¼ 0:981 for u ¼ 0:5. However, a Du ¼ 1
 xÞ
ð14Þ
truncated discrete log-series fits the data just as well (Supple- Φð1  x; 1; nmin Þ
mentary material 1). Following the method used to derive the
exact and approximate dominance formulae for the full log-series, respectively. In these formulae, nmin is the species abundance at
which the SAD is truncated, such that species with a lower abun-
we derived corresponding formulae for a truncated log-series as  
dance are removed; also, in (13), nu ¼ nmin þ log ðuÞ=log ð1  xÞ . The
" #
1 u  ð1  xÞ⌈nu ⌉  nmin dominance formula for the continuous version of the truncated log-
Du ¼ 1  þ ð1  xÞ⌈nu ⌉  nmin Φð1  x; 1; ⌈nu ⌉Þ
Φð1  x; 1; nmin Þ xð⌈nu ⌉  1Þ series can be derived analogously to that for the continuous log-
ð13Þ
T. Fung et al. / Journal of Theoretical Biology 386 (2015) 147–158 153

Fig. 2. (a) Dominance values for the discrete log-series SAD using the exact and approximate formulae, compared with dominance values for the continuous log-series SAD
(formulae shown in Table 1). The panel shows how dominance changes across average species abundance (total number of individuals of all species, J, divided by total
number of species, S) for different values of u, which is the fraction of individuals accounted for by the set of dominant species. (b) is the same as (a) but for the discrete
exponential SAD and the continuous exponential SAD with a lower limit of δ ¼ 0:5.

Fig. 3. Errors in approximating the exact dominance values for (a) the discrete log-series SAD and (b) the discrete exponential SAD when using the approximate formulae
and the formulae for the continuous versions (formulae shown in Table 1). An error is calculated as the exact value minus the approximate value. Each panel shows how the
errors change across average species abundance (total number of individuals of all species, J, divided by total number of species, S) for different values of u, which is the
fraction of individuals accounted for by the set of dominant species. In (b), the continuous exponential SAD has a lower limit of δ ¼ 0.5.

series, as (13) and (14) because evaluation of the Lerch transcendent is slow
  with small x.
li uð1  xÞnmin
Du ¼ 1    : ð15Þ A lower bound for the dominance of tree species in Amazonia
li ð1  xÞnmin
was estimated by applying dominance formulae (13)–(15) with
Application of (13) with u ¼ 0:5 and nmin ¼ 84, 3,646 and nmin ¼ 550; 755, corresponding to S  5; 000, where 5000 is the
157,027, corresponding to S  12; 000, 9,000 and 6,000 respec- approximate number of observed species. All three formulae gave
tively, gave Du ¼ 0:975, 0.966 and 0.949 respectively. Application Du ¼ 0:940 to 3 decimal places. By plotting Du against nmin from
of (14) and (15) gave dominance values that are the same to 1 to 550,755, we confirmed our expectation that Du decreases with
3 decimal places; however, application of (15) is much faster than nmin , such that Du ¼ 0:940 can be taken as a lower bound.
154 T. Fung et al. / Journal of Theoretical Biology 386 (2015) 147–158

Fig. 4. (a) Lognormal distributions underlying (stationary) SADs produced by the dynamic community model by Engen and Lande (1996a), for three different values of the
square root of environmental variance (environmental standard deviation), σ e (measured in units of yr  1/2, not shown on graphs to improve their clarity). Values of the other
parameters are r ¼ 0:16yr  1 , γ ¼ 0:02 yr  1 , σ d ¼ 0:447 yr  1=2 and ω ¼ 10  4 yr  1 (for parameter definitions, see Subsection 2.5). (b) Dominance values for the lognormal
SADs, as calculated by applying the formula in Table 1 for five different values of u (fraction of individuals accounted for ffiffiffiffiffi the set of dominant species) and σ e ranging from
pby
0.224–0.380 yr  1/2. The log-scale parameter for the SADs is μ ¼ r=γ ¼ 8 and the shape parameter varied from σ ¼ σ e = 2γ ¼ 1:12 to 1.90.

3.4. Quantifying how dominance changes with increasing environ- (Table 1). We also examined the discrete broken-stick (MacArthur,
mental variance 1957) and Zipf-Mandelbrot (Frontier, 1985) SADs, but due to the
complexity of the SAD formulae, we were unable to derive cor-
We applied the dominance formula for the lognormal SAD responding analytical dominance formulae. However, in the limit
(Table 1) to the lognormal SADs arising from the dynamic com- of large S, the broken-stick SAD converges to a gamma distribution
munity model by Engen and Lande (1996a) (as described in (Longuet-Higgins, 1971), which we examined (Table 1). The dom-
Subsection 2.5). According to this formula, dominance increases inance formulae derived facilitate quantitative investigation of
with the shape parameter σ , and hence the average species dominance by speeding up calculation of dominance for discrete
abundance J=S (Subsection 3.2). SADs from the dynamic model SADs (compared with simple summation, the formulae can be
have a shape parameter that is proportional to σ e , which is the orders of magnitude quicker); allowing dominance to be calcu-
square root of environmental variance. Thus, application of the lated for continuous SADs; and permitting further mathematical
formula shows that dominance increases with σ e in the model. analyses that yield general insights into how dominance changes
The relationship between dominance and environmental variance with key ecological variables. We also derived analytical formulae
is illustrated in Fig. 4 for example parameterpsets ffiffiffiffiffiffiffiffiffithat
ffi are the same to approximate Du for the discrete log-series and exponential
except for σ e increasing by a factor of 1.7 from 0:05 ¼ 0:224yr  1=2 to SADs, and found that these approximations gave almost zero error
0.380 yr  1/2. This gave a range of σ 2e values that are in accordance when the average number of individuals per species was greater
with empirical values collated by Lande et al. (2003; Table 1.2), typi- than about 10. In addition, under these circumstances, dominance
cally on the order of 0.01–0.1 yr  1. Similarly, demographic variance formulae for the continuous versions of the SADs gave values that
σ 2d was set at 0.2, giving a ratio ε ¼ σ 2d =σ 2e from 1.38–4, in accordance were very similar to those from the exact formulae for the discrete
with typical empirical ratios of 1 to 10 (Table 1.2 of Lande et al., 2003). versions. Thus, the formulae for the continuous versions can be
The dynamic model explicitly models speciation, which operates over used to give accurate estimates of dominance for any natural
large spatial scales. Thus, values of the other parameters were chosen community of reasonable size that is distributed according to one
to give large values of J and J=S, commensurate with communities at of these two discrete SADs. The errors when using the approx-
large spatial scales. These quantities together with S can be calculated imate formulae for the discrete SADs were typically lower than
using Eqs. (11) and (12), after translation of the abundance variable those when formulae for the continuous SADs were used (Fig. 3).
and assuming that the transformed variable starts at zero (a reason- This result is intuitive because the continuous SADs assume that
able assumption for large communities). For the example shown in species abundance changes continuously, whereas in reality it is
Fig. 4, dominance increased with σ e by 12–68% for values of u ranging discrete. However, despite this continuity assumption, the result-
from 0.5 to 0.9 in increments of 0.1. The sharpest increase was seen for ing continuous SADs largely retain the dominance structure of
the highest u tested. In addition, as σ e increased, S decreased from their discrete counterparts (Fig. 2). Given lower errors, the discrete
139,000 to 7.52; J decreased from 7:76  108 to 136,000; and J=S formulae would be preferred in practice, except for the discrete
increased from 5570 to 18,100. log-series SAD with low values of the parameter x. In this case, the
approximate discrete formula is slow to compute because of the
Lerch transcendent function, and so the continuous formula would
4. Discussion be preferred.
A striking result from application of our formulae is that
In this study, we examined four standard forms of SADs that dominance was typically very high for reasonably large commu-
together have commonly been fitted to empirical data (e.g., Fisher nities with individuals distributed according to a log-series,
et al., 1943; Preston, 1948; Brian, 1953; Cohen, 1968; Hubbell, exponential or lognormal distribution. When the average abun-
2001; Engen et al., 2002; Plotkin and Muller-Landau, 2002; Volkov dance of a species exceeded 200 individuals, fewer than 40% of
et al., 2003; Forster and Warton, 2007; Engen et al., 2011) and species were required to account for 90% of all individuals for the
successfully derived analytical formulae quantifying dominance log-series or lognormal case (Du 40:6 for J=S 4 200 and u ¼ 0:9),
according to Du ¼ 1  pu , where pu is the minimum proportion of and 70% of all individuals for the exponential case (Du 4 0:6 for
species required to account for a proportion u of all individuals J=S 4 200 and u ¼ 0:7). Given that many empirical SADs can be
T. Fung et al. / Journal of Theoretical Biology 386 (2015) 147–158 155

fitted by these three distributions (Fisher et al., 1943; Preston, formula for discrete log-series SADs, that this result follows
1948; Cohen 1968; Hubbell, 2001; Engen et al., 2002; Volkov et al., inevitably from fitting a log-series: the formula predicts high
2003; Forster and Warton, 2007; Engen et al., 2011; the lognormal dominance for values of the parameter x corresponding to large
distribution is typically mixed with a Poisson sampling distribu- communities with high average species abundances ðJ=SÞ. How-
tion), the concentration of abundance in so few species that we ever, whether the actual distribution of Amazon trees is a log-
typically found has important implications for community func- series is highly uncertain, given a low sampling intensity of
tioning and stability. Communities with high dominance attain a 0.0002% (ter Steege et al., 2013). Alternative types of SADs can be
higher level of functioning if there is a positive correlation fitted to the Amazonian data equally well, and one of these is the
between numerical dominance of a species and its level of func- truncated discrete log-series (Supplementary material 1). We
tioning (Hillebrand et al., 2008) – this is the classic positive adapted our calculations to derive dominance formulae for a
selection effect that has been demonstrated to be an important truncated log-series and used it to show that dominance in the
mechanism underlying positive relationships between richness Amazonian tree flora can vary between 0.940 and 0.981, i.e., the
and functioning for plant and invertebrate communities (Huston, percentage of species making up 50% of individuals could be
1997; Wilsey and Potvin, 2000; Hooper et al., 2005; Larsen et al., anywhere between 1.9% and 6.0%. Thus, dominance is high, but the
2005; Orwin et al., 2014). However, another implication of high exact degree of dominance currently remains uncertain. The true
dominance is a decrease in the strength of multispecies processes, dominance may even exceed the upper limit of 0.981 if there are
thus reducing the prevalence of niche complementarity and more rare species than predicted by a log-series. Furthermore, the
facilitation, both of which are important drivers of functioning true level of dominance is probably fundamentally unknowable
(Loreau and Hector, 2001; Cardinale et al., 2002; Hooper et al., from small sample plots, because it is not possible to calculate an
2005; Carey and Wahl, 2010). In contrast to the selection effect, unbiased estimate of the total number of species from a small
this acts to decrease community functioning. Therefore, the net sample (Gotelli and Chao, 2013; Supplementary material 1). A
effect of higher dominance depends on the balance of two coun- more justifiable approach in such situations may thus be to cal-
tervailing forces – the selection effect and weaker multispecies culate dominance from a fitted distribution that covers only the
processes. A meta-analysis of 54 studies suggests that the selection empirical range of species abundances and use it as an estimate of
effect is typically stronger in forest communities, resulting in a lower bound for dominance – for the Amazonian tree flora, such
greater community functioning with higher dominance (Zhang an estimate would be 0.940.
et al., 2012). In addition to these effects on functioning, an increase We also applied our dominance formula for lognormal SADs to
in dominance results in an increase in intraspecific and corre- the dynamic community model by Engen and Lande (1996a),
sponding decrease in interspecific interaction strengths, causing which predicts lognormal SADs as a function of parameters
changes in community stability. Studies on model and empirical representing the ecological mechanisms modeled. This allowed a
food webs suggest that weak interspecific interactions can quantitative, process-based assessment of how dominance is
increase stability (May, 1972; McCann et al., 1998; Neutel et al., predicted to change with variations in the ecological mechanisms.
2002; Emmerson and Yearsley, 2004), such that higher dominance In our application, we focused on how dominance is expected to
is expected to result in more stable communities. change with environmental variance, which is modeled as sto-
In contrast to the log-series, exponential and lognormal dis- chastic fluctuations in the growth rate. We showed analytically
tributions, dominance values for the gamma distribution were that the model predicts increases in dominance with the degree of
typically high only for communities with small average species environmental variance, and illustrated this with an example
abundances. When the average species abundance was below 50 whereby dominance increased by up to 68%. It is also possible to
individuals, fewer than 40% of species were required to account for show analytically that the model predicts a decline in species
90% of all individuals (Du 40:6 for J=S o 50 and u ¼ 0:9). However, richness with environmental variance, which was the case in the
as the average species abundance increases beyond 50 individuals, example shown. Therefore, the model predicts that greater
dominance decreases non-linearly, typically reaching values below environmental fluctuations homogenize communities by causing
0.5. Thus, our formulae give opposite trends for how dominance in extinctions and promoting dominance of abundance by pro-
gamma- and lognormally-distributed communities scale with portionally fewer species, in effect pushing species abundances
community size. This result is surprising given that these two towards the extremes (Fig. 4a). This prediction is particularly
distributions can have similar shapes. However, it can be explained relevant in the context of forecasted increases in environmental
by the tail of the lognormal distribution becoming heavier as the variance driven by climatic change, for example greater intra-
value of the shape parameter, which correlates with community annual variation in precipitation resulting in more severe floods
size, increases, whereas the tail of the gamma distribution and droughts (Knapp et al., 2008). Our predictions complement
becomes less heavy. Therefore, for communities that are fit well by previous theoretical predictions on how environmental variance
a gamma distribution (typically mixed with a Poisson sampling can increase extinction risk and decrease species richness (e.g.,
distribution; e.g., Brian, 1953), it is advisable to rigorously test Adler and Drake, 2008; Ovaskainen and Meerson, 2010). Impor-
whether a lognormal distribution can give a better fit or not, since tantly, our predictions can be tested quantitatively using empirical
this can have important implications for determining dominance data in future studies. In addition, the homogenizing effect of
and assessing its community effects. On the other hand, this is environmental variance has possible knock-on effects for func-
difficult given that statistical estimates of the rare species tail of an tioning and stability, as discussed above. This can be further
SAD are highly uncertain unless a community is intensively sam- investigated in future studies by explicitly quantifying functioning
pled. This is also evident from our analysis of the SAD underlying and stability using model outputs and possibly other theory. Fur-
the tree metacommunity in Amazonia (discussed below), and thermore, future studies could use our dominance formulae in
together these results emphasize the sensitivity of dominance to conjunction with other stochastic community models (e.g., Dennis
sampling uncertainty in general. and Costantino, 1988; Engen and Lande, 1996b; Hubbell, 2001) to
A recent assessment of dominance among trees in Amazonia further investigate the mechanistic links between evenness and
involved fitting a discrete log-series SAD to sample plot data and ecosystem functioning and stability. Predictions from these studies
concluded that dominance is very high, in the sense that very few would facilitate understanding of the ongoing effects of anthro-
species are required to account for 50% of all individuals (ter pogenically driven biodiversity loss (Cardinale et al., 2012), vital
Steege et al., 2013). We now see, from application of our analytical for conservation and sustainable management.
156 T. Fung et al. / Journal of Theoretical Biology 386 (2015) 147–158

 
Acknowledgments X
1
eλ  λ⌈nu ⌉ 1  ⌈nu ⌉ þ ⌈nu ⌉eλ
nψ ðnÞ ¼ ; ðA:11Þ
n ¼ ⌈nu ⌉
eλ 1
We would like to thank two anonymous reviewers, whose
comments have helped to substantially improve the clarity and and
content of our work. TF and RAC are supported by the National X
1
University of Singapore, Singapore start-up Grant WBS R-154-000- ψ ðnÞ ¼ eλ  λ⌈nu ⌉ : ðA:12Þ
551-133. n ¼ ⌈nu ⌉

(A.10) and (A.11) can be derived using the identity


P1 2
k ¼ a kz ¼ ½z ða þ ð1  aÞzÞ=ð1  zÞ , which holds for 0 o z o 1.
k a
Appendix A. Details on derivation of dominance formulae
Thus,
 
This Appendix contains details of how the dominance formulae ueλ  eλ  λ⌈nu ⌉ 1  ⌈nu ⌉ þ ⌈nu ⌉eλ
pu ¼   þ eλ  λ⌈nu ⌉
in Table 1 were derived. eλ  1 ð⌈nu ⌉  1Þ
"   #
Discrete log-series λ u  e  λ⌈nu ⌉ 1  ⌈nu ⌉ þ ⌈nu ⌉eλ  λ ⌈n ⌉
¼e   þe u
: ðA:13Þ
eλ  1 ð⌈nu ⌉ 1Þ
In this case, FðmÞ ¼ u is
Using Du ¼ 1  pu gives the formula for Du in Table 1. Also,
ð1  xÞm  1 ¼ u; ðA:1Þ
X
1   
which can be solved for m to give pu ¼ ψ ðnÞ ¼ eλð1  nu Þ ¼ exp λ  y þ W  1 uye  y : ðA:14Þ
n ¼ nu
log ðuÞ
nu ¼ 1 þ : ðA:2Þ Using Du ¼ 1  pu gives the formula in Table 1.
log ð1 xÞ
In addition,
Continuous log-series
X
1
1x
nψ ðnÞ ¼  ; ðA:3Þ
n¼1
xlog 1=x In this case, GðmÞ ¼ u is

X
1 ð1  xÞm  1 ¼ u; ðA:15Þ
ð1  xÞ⌈nu ⌉
nψ ðnÞ ¼  ; ðA:4Þ
n ¼ ⌈nu ⌉
xlog 1=x such that

and log ðuÞ


nu ¼ 1 þ ðA:16Þ
log ð1  xÞ
X
1
ð1 xÞ ⌈nu ⌉
ψ ðnÞ ¼   Φð1  x; 1; ⌈nu ⌉Þ: ðA:5Þ (this is the same as in the discrete case (Eq. (A.2)). Therefore,
n ¼ ⌈nu ⌉
log 1=x
Z 1  
li ð1  xÞnu liðuð1  xÞÞ
Thus, using Eq. (5), pu ¼ ψ ðnÞdn ¼ ¼ : ðA:17Þ
n ¼ nu lið1  xÞ lið1  xÞ
uð1  xÞ  ð1  xÞ⌈nu ⌉ ð1  xÞ⌈nu ⌉
pu ¼ þ Φð1  x; 1; ⌈nu ⌉Þ Using Du ¼ 1  pu gives the formula in Table 1.
xlog ð1=xÞð⌈nu ⌉  1Þ log ð1=xÞ
" #
ð1  xÞ u  ð1  xÞ⌈nu ⌉  1
¼ þ ð1  xÞ⌈nu ⌉  1 Φð1  x; 1; ⌈nu ⌉Þ : Continuous exponential
log ð1=xÞ xð⌈nu ⌉  1Þ
ðA:6Þ GðmÞ ¼ u becomes
Using Du ¼ 1  pu gives the formula for Du in Table 1. Further-  
eλð  m þ δÞ 1 þ λm
more, ¼ u: ðA:18Þ
1 þ δλ
X
1
ð1 xÞnu
pu ¼ ψ ðnÞ ¼  Φð1  x; 1; nu Þ Solving for m gives
n ¼ nu log 1=x

 
uð1 xÞ log ðuÞ  1  W  1  e  1  δλ uð1 þ δλÞ
¼  Φ 1  x; 1; 1 þ : ðA:7Þ nu ¼ ; ðA:19Þ
log 1=x log ð1  xÞ λ
Using Du ¼ 1  pu gives the formula for Du in Table 1. such that
Z 1 

uð1 þ δλÞ
Discrete exponential
pu ¼ ψ ðnÞdn ¼ eλð  nu þ δÞ ¼ exp 1 þ δλ þ W  1  ð1 þ δλÞ :
n ¼ nu e
FðmÞ ¼ u ðA:20Þ

is Using Du ¼ 1  pu gives the formula in Table 1.



e  λm 1 þ eλ  1 m ¼ u; ðA:8Þ Continuous gamma
such that
Here, GðmÞ ¼ u is equivalent to
1 W  1 ðuye  y Þ  
nu ¼  ; ðA:9Þ Γ 1 þ α; β m
1e λ λ ¼ u; ðA:21Þ
  Γð1 þ αÞ
where y ¼ λ= 1  eλ . In addition,
X
1
eλ which on rearranging gives
nψ ðnÞ ¼ λ ; ðA:10Þ  
n¼1
e 1 Γ 1 þ α; βm ¼ uΓð1 þ αÞ; ðA:22Þ
T. Fung et al. / Journal of Theoretical Biology 386 (2015) 147–158 157

such that Continuous exponential


Q 1
ð1 þ α ; uÞ
nu ¼ : ðA:23Þ The dominance formula for the continuous exponential dis-
β tribution is:
Thus, h i
Du ¼ 1  exp 1 þ δλ þW  1  uð1 þ δλÞe  ð1 þ δλÞ : ðB:6Þ
Z  
Γ α; βnu Γ α; Q ð1 þ α; uÞ
1
1
pu ¼ ψ ðnÞdn ¼ ¼ : ðA:24Þ With δ 4 0,
n ¼ nu ΓðαÞ ΓðαÞ   
d 1 þ δλ þ W  1  uð1 þ δλÞe  ð1 þ δλÞ
Using Du ¼ 1  pu gives the formula in Table 1. dλ
(
  )
δλ W  1  uð1 þ δλÞe  ð1 þ δλÞ
Continuous lognormal ¼ δ 1   
1 þ δλ 1 þ W  1  uð1 þ δλÞe  ð1 þ δλÞ
(  )
In this case, GðmÞ ¼ u is δ 1 þ δλ þ W  1  uð1 þ δλÞe  ð1 þ δλÞ
¼   : ðB:7Þ

1 þ δλ 1 þ W  1  uð1 þ δλÞe  ð1 þ δλÞ
1 μ þ σ 2  log ðmÞ
1  erfc pffiffiffi ¼ u; ðA:25Þ
2 σ 2 The denominator of this derivative is thus negative, so that the
derivative has the opposite sign to 1 þ δλ þW  1 ð  uð1
such that
h i þ δλÞe  ð1 þ δλÞ Þ. Also, because δ 4 0, the derivative is non-zero and
pffiffiffi 1
nu ¼ exp μ þ σ 2  σ 2erfc ð2ð1  uÞÞ : ðA:26Þ therefore its sign does not change  with λ. For small
 positive λ,
1 þ δλ þ W  1 uð1 þ δλÞe  ð1 þ δλÞ  1 þ W  1  ue  1 o0 and the
Therefore, derivative is positive; hence the derivative is positive for all λ 4 0.
 
Z 1

This implies that dDu =dλ o 0, i.e. dDu =d J=S 40.


1 μ  log ðnu Þ
pu ¼ ψ ðnÞdn ¼ 2 erfc pffiffiffi
n ¼ nu 2 σ 2 Continuous lognormal

1 σ 1
¼ erfc pffiffiffi  erfc ð2ð1  uÞÞ : ðA:27Þ
2 2 Consider the dominance formula for the continuous lognormal
distribution:
Using Du ¼ 1  pu gives the formula in Table 1.

1 σ 1
Du ¼ 1  erfc pffiffiffi erfc ð2ð1  uÞÞ : ðB:8Þ
2 2
Appendix B. Details on mathematical analyses of dominance The complementary error function erfcðzÞ is a decreasing
 
formulae function of z, so dDu =dσ 4 0, i.e. dDu =d J=S 4 0.

Continuous log-series
Appendix C. Supplementary material
Consider the dominance formula for the continuous log-series
distribution: Supplementary data associated with this article can be found in
liðuð1  xÞÞ the online version at http://dx.doi.org/10.1016/j.jtbi.2015.09.011.
Du ¼ 1  : ðB:1Þ
lið1  xÞ

Thus, References
!
d liðuð1  xÞÞ

dDu lið1  xÞ 1 ulið1  xÞ liðuð1 xÞÞ


¼ ¼  2  : Abramowitz, M., Stegun, I.A., 1972. Handbook of Mathematical Functions with
dx dx lið1 xÞ log ðuÞ þ log ð1  xÞ log ð1  xÞ Formulas, Graphs and Mathematical Tables. Dover Publications, New York.
Adler, P.B., Drake, J.M., 2008. Environmental variation, stochastic extinction, and
ðB:2Þ competitive coexistence. Am. Nat. 172, E186–E195. http://dx.doi.org/10.1086/
591678.
For small x oo1, Balvanera, P., et al., 2006. Quantifying the evidence for biodiversity effects on
!
ecosystem functioning and services. Ecol. Lett. 9, 1146–1156. http://dx.doi.org/
dDu 1 ulið1  xÞ liðuÞ 10.1111/j.1461-0248.2006.00963.x.
  2  : ðB:3Þ
dx lið1  xÞ log ðuÞ log ð1 xÞ Barker, J.R., Ringold, P.L., Bollman, M., 2002. Patterns of tree dominance in con-
iferous riparian forests. For. Ecol. Manage. 166, 311–329. http://dx.doi.org/
10.1016/S0378-1127(01)00683-1.
The two terms in the second brackets in (B.3) have derivative Berlow, E.L., Dunne, J.A., Martinez, N.D., Stark, P.B., Williams, R.J., Brose, U., 2009.
Simple predictions of interaction strengths in complex food webs. Proc. Natl.
u liðuÞ Acad. Sci. USA 106, 187–191. http://dx.doi.org/10.1073/pnas.0806823106.
  4
40; ðB:4Þ
log ðuÞlog ð1  xÞ ðlog ð1 xÞÞ2 ð1  xÞ Binkley, D., 2004. A hypothesis about the interaction of tree dominance and stand
production through stand development. For. Ecol. Manage. 190, 265–271. http:
such that the sum of the two terms increases quickly with x and //dx.doi.org/10.1016/j.foreco.2003.10.018.
Brian, M.V., 1953. Species frequencies in random samples from animal populations.
hence decreases quickly with decreasing x. Therefore, for small x, J. Anim. Ecol. 22, 57–64.
the sum is negative, such that dDu =dx o 0. Thus, Du decreases with Cardinale, B.J., Palmer, M.A., Collins, S.L., 2002. Species diversity enhances ecosys-
x when it is small, i.e. increases with J=S when it is large. For large tem functioning through interspecific facilitation. Nature 415, 426–429. http:
//dx.doi.org/10.1038/415426a.
x  1, Cardinale, B.J., et al., 2006. Effects of biodiversity on the functioning of trophic
!
groups and ecosystems. Nature 443, 989–992. http://dx.doi.org/10.1038/
dDu 1 liðuð1 xÞÞ nature05202.
  2 o0: ðB:5Þ
dx lið1  xÞ log ð1  xÞ Cardinale, B.J., et al., 2012. Biodiversity loss and its impact on humanity. Nature 486,
59–67. http://dx.doi.org/10.1038/nature11148.
Carey, M.P., Wahl, D.H., 2010. Fish diversity as a determinant of ecosystem prop-
So Du decreases with x when it is large, i.e. increases with J=S erties across multiple trophic levels. Oikos 120, 84–94. http://dx.doi.org/
when it is small. 10.1111/j.1600-0706.2010.18352.x.
158 T. Fung et al. / Journal of Theoretical Biology 386 (2015) 147–158

Chapin, F.S., et al., 2000. Consequences of changing biodiversity. Nature 405, Lande, R., Engen, S., Sæther, B.E., 2003. Stochastic Population Dynamics in Ecology
234–242. http://dx.doi.org/10.1038/35012241. and Conservation. Oxford University Press, Oxford.
Cohen, J.E., 1968. Alternate derivations of a species-abundance relation. Am. Nat. Larsen, T.H., Williams, N.M., Kremen, C., 2005. Extinction order and altered com-
102, 165–172. munity structure rapidly disrupt ecosystem functioning. Ecol. Lett. 8, 538–547.
Davidar, P., Puyravaud, J.P., Leigh Jr, E.G., 2005. Changes in rain forest tree diversity, http://dx.doi.org/10.1111/j.1461-0248.2005.00749.x.
dominance and rarity across a seasonality gradient in the Western Ghats, India. Longuet-Higgins, M.S., 1971. On the Shannon-Weaver Index of diversity, in relation
J. Biogeogr. 32, 493–501. http://dx.doi.org/10.1111/j.1365-2699.2005.01165.x. to the distribution of species in bird censuses. Theor. Popul. Biol. 2, 271–289.
Davies, T.W., Jenkins, S.R., Kingham, R., Kenworthy, J., Hawkins, S.J., Hiddink, J.G., http://dx.doi.org/10.1016/0040-5809(71)90020-7.
2011. Dominance, biomass and extinction resistance determine the con- Loreau, M., Hector, A., 2001. Partitioning selection and complementarity in biodi-
sequences of biodiversity loss for multiple coastal ecosystem processes. PLoS versity experiments. Nature 412, 72–76. http://dx.doi.org/10.1038/35083573.
One 6, e28362. http://dx.doi.org/10.1371/journal.pone.0028362. Lotze, H.K., Worm, B., 2009. Historical baselines for large marine animals. Trends
Dennis, B., Costantino, R.F., 1988. Analysis of steady-state populations with the Ecol. Evol. 24, 254–262. http://dx.doi.org/10.1016/j.tree.2008.12.004.
gamma abundance model: application to Tribolium. Ecology 69, 1200–1213. MacArthur, R.H., 1957. On the relative abundance of bird species. Proc. Natl. Acad.
Dornelas, M., Phillip, D.A.T., Magurran, A., 2011. Abundance and dominance become Sci. 43, 293–295.
less predictable as species richness decreases. Glob. Ecol. Biogeogr. 20, Mattingly, W.B., Hewlate, R., Reynolds, H.L., 2007. Species evenness and invasion
832–841. http://dx.doi.org/10.1111/j.1466-8238.2010.00640.x. resistance of experimental grassland communities. Oikos 116, 1164–1170. http:
Emmerson, M., Yearsley, J.M., 2004. Weak interactions, omnivory and emergent //dx.doi.org/10.1111/j.0030-1299.2007.15406.x.
food-web properties. Proc. R. Soc. Lond. B 271, 397–405. http://dx.doi.org/ May, R.M., 1972. Will a large complex system be stable? Nature 238, 413–414. http:
10.1098/rspb.2003.2592. //dx.doi.org/10.1038/238413a0.
Engen, S., Lande, R., 1996a. Population dynamic models generating the lognormal McCann, K., Hastings, A., Huxel, G.R., 1998. Weak trophic interactions and the
species abundance distribution. Math. Biosci. 132, 169–183. http://dx.doi.org/ balance of nature. Nature 395, 794–798. http://dx.doi.org/10.1038/27427.
10.1016/0025-5564(95)00054-2. Naeem, S., Duffy, J.E., Zavaleta, E., 2012. The functions of biological diversity in an
Engen, S., Lande, R., 1996b. Population dynamic models generating species abun- age of extinction. Science 336, 1401–1406. http://dx.doi.org/10.1126/
dance distributions of the gamma type. J. Theor. Biol. 178, 325–331. http://dx. science.1215855.
doi.org/10.1006/jtbi.1996.0028. Neutel, A.-M., Heesterbeek, J.A.P., de Ruiter, P.C., 2002. Stability in real food webs:
Engen, S., Lande, R., Walla, T., DeVries, P.J., 2002. Analyzing spatial structure of weak links in long loops. Science 296, 1120–1123. http://dx.doi.org/10.1126/
communities using the two-dimensional poisson lognormal species abundance science.1068326.
model. Am. Nat. 160, 60–73. Orwin, K.H., Ostle, N., Wilby, A., Bardgett, R.D., 2014. Effects of species evenness and
Engen, S., Aagaard, K., Bongard, T., 2011. Disentangling the effects of heterogeneity, dominant species identity on multiple ecosystem functions in model grassland
stochastic dynamics and sampling in a community of aquatic insects. Ecol. communities. Oecologia 174, 979–992. http://dx.doi.org/10.1007/
Model. 222, 1387–1393. http://dx.doi.org/10.1016/j.ecolmodel.2011.01.008. s00442-013-2814-5.
Fisher, R.A., Corbet, A.S., Williams, C.B., 1943. The relation between the number of Ovaskainen, O., Meerson, B., 2010. Stochastic models of population extinction.
species and the number of individuals in a random sample of an animal Trends Ecol. Evol. 25, 643–652. http://dx.doi.org/10.1016/j.tree.2010.07.009.
population. J. Anim. Ecol. 12, 42–58. Plotkin, J.B., Muller-Landau, H.C., 2002. Sampling the species composition of a
Forster, M.A., Warton, D.I., 2007. A metacommunity-scale comparison of species- landscape. Ecology 83, 3344–3356. http://dx.doi.org/10.1890/0012-9658(2002)
abundance distribution models for plant communities of eastern Australia. 083[3344:STSCOA]2.0.CO;2.
Ecography 30, 449–458. http://dx.doi.org/10.1111/j.0906-7590.2007.04899.x. Preston, F.W., 1948. The commonness, and rarity, of species. Ecology 29, 254–283.
Frontier, S., 1985. Diversity and structure in aquatic ecosystems. Oceanogr. Mar. Reiss, J., Bridle, J.R., Montoya, J.M., Woodward, G., 2009. Emerging horizons in
Biol. 23, 253–312. biodiversity and ecosystem functioning research. Trends Ecol. Evol. 24,
Gaston, K.J., Fuller, R.A., 2008. Commonness, population depletion and conservation 505–514. http://dx.doi.org/10.1016/j.tree.2009.03.018.
biology. Trends Ecol. Evol. 23, 14–19. http://dx.doi.org/10.1016/j. Steffen, W., Crutzen, P.J., McNeill, J.R., 2007. The Anthropocene: are humans now
tree.2007.11.001. overwhelming the great forces of nature? Ambio 36, 614–621.
Gotelli, N.J., Chao, A., 2013. Measuring and estimating species richness, species ter Steege, H., et al., 2013. Hyperdominance in the Amazonian tree flora. Science
diversity, and biotic similarity from sampling data, second ed. In: Levin, S. (Ed.), 342, 1243092. http://dx.doi.org/10.1126/science.1243092.
Encyclopedia of Biodiversity, vol. 5. Elsevier, Amsterdam, pp. 195–211. Vitousek, P.M., Mooney, H.A., Lubchenco, J., Melillo, M., 1997. Human domination of
Grime, J.P., 1998. Benefits of plant diversity to ecosystems: immediate, filter and earth’s ecosystems. Science 277, 494–499. http://dx.doi.org/10.1126/
founder effects. J. Ecol. 86, 902–910. science.277.5325.494.
Hillebrand, H., Bennett, D.M., Cadotte, M.W., 2008. Consequences of dominance: a Volkov, I., Banavar, J.R., Hubbell, S.P., Maritan, A., 2003. Neutral theory and relative
review of evenness effects on local and regional ecosystem processes. Ecology species abundance in ecology. Nature 424, 1035–1037. http://dx.doi.org/
89, 1510–1520. http://dx.doi.org/10.1046/j.1365-2745.1998.00306.x. 10.1038/nature01883.
Hooper, D.U., et al., 2005. Effects of biodiversity on ecosystem functioning: a con- Wilsey, B.J., Potvin, C., 2000. Biodiversity and ecosystem functioning: importance of
sensus of current knowledge. Ecol. Mongr. 75, 3–35. http://dx.doi.org/10.1890/ species evenness in an old field. Ecology 81, 887–892. http://dx.doi.org/
04-0922. 10.1890/0012-9658(2000)081[0887:BAEFIO]2.0.CO;2.
Hubbell, S.P., 2001. The Unified Neutral Theory of Biodiversity and Biogeography. Wilsey, B.J., Polley, H.W., 2002. Reductions in grassland species evenness increase
Princeton University Press, Princeton. dicot seedling invasion and spittle bug infestation. Ecol. Lett. 5, 676–684. http:
Huston, M.A., 1997. Hidden treatments in ecological treatments: re-evaluating the //dx.doi.org/10.1046/j.1461-0248.2002.00372.x.
ecosystem function of biodiversity. Oecologia 110, 449–460. http://dx.doi.org/ Zhang, Y., Chen, H.Y.H., Reich, P.B., 2012. Forest productivity increases with even-
10.1007/s004420050180. ness, species richness and trait variation: a global meta-analysis. J. Ecol. 100,
Knapp, A.K., et al., 2008. Consequences of more extreme precipitation regimes for 742–749. http://dx.doi.org/10.1111/j.1365-2745.2011.01944.x.
terrestrial ecosystems. Bioscience 58, 811–821. http://dx.doi.org/10.1641/
B580908.

You might also like