25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

Medidas de diversidad funcional basadas en la distancia y

su descomposición: un marco basado en números de Hill
Chun-Huo Chiu, ana chao

Publicado: 7 de julio de 2014 https://doi.org/10.1371/journal.pone.0100014

Corrección
14 de noviembre de 2014:
El personal de PLOS ONE (2014)
Corrección: Medidas de diversidad funcional basadas en la
distancia y su descomposición: un marco basado en números de colina. PLOS UNO 9(11): e113561.
https://doi.org/10.1371/journal.pone.0113561 Ver corrección
|

Resumen
Los números de colinas (o el “número efectivo de especies”) se utilizan cada vez más para caracterizar la diversidad de especies
de un conjunto. Este trabajo amplía los números de Hill para incorporar las distancias funcionales por pares de especies
calculadas a partir de los rasgos de las especies. Derivamos una clase paramétrica de números de Hill funcionales, que
cuantifican "el número efectivo de especies igualmente abundantes y (funcionalmente) igualmente distintas" en un conjunto.
También proponemos una clase de diversidad funcional media (por especie), que cuantifica la suma efectiva de las distancias
funcionales entre una especie fija y todas las demás especies. El producto del número de Hill funcional y la diversidad funcional
media cuantifica así la diversidad funcional (total), es decir, la distancia total efectiva entre las especies del conjunto. Las tres
medidas (números de Hill funcionales, la diversidad funcional media y la diversidad funcional total) cuantifican diferentes aspectos
del espacio de características de las especies, y todos se basan en la abundancia de especies y las distancias funcionales por
pares de especies. Cuando todas las especies son igualmente distintas, nuestros números de Hill funcionales se reducen a
números de Hill ordinarios. Cuando no se consideran las abundancias de las especies o las especies son igualmente abundantes,
nuestra diversidad funcional total se reduce a la suma de todas las distancias por pares entre las especies de un conjunto. Los
números de Hill funcionales y la diversidad funcional media satisfacen un principio de replicación, lo que implica que la diversidad
funcional total satisface un principio de replicación cuadrático. Cuando hay conjuntos múltiples definidos por el investigador, cada
una de las tres medidas del conjunto combinado (gamma) se puede descomponer multiplicativamente en componentes alfa y beta,
y los dos componentes son independientes. El componente beta resultante mide la diferenciación funcional pura entre
ensamblajes y puede transformarse aún más para obtener varias clases de medidas de similitud (o diferenciación) funcional
normalizadas, que incluyenN -generalizaciones funcionales de ensamblaje de los índices de similitud clásicos de Jaccard,
Sørensen, Horn y Morisita-Horn. Las medidas propuestas se aplican a datos artificiales y reales a modo ilustrativo.

Cita: Chiu CH, Chao A (2014) Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco
basado en los números de Hill. PLoS ONE 9(7): e100014. https://doi.org/10.1371/journal.pone.0100014

Editor: Francesco de Bello, Instituto de Botánica, Academia Checa de Ciencias, República Checa

Recibido: 15 de diciembre de 2013; Aceptado: 21 de mayo de 2014; Publicado: 7 de julio de 2014

Derechos de autor: © 2014 Chiu, Chao. Este es un artículo de acceso abierto distribuido bajo los términos de la licencia de
atribución Creative Commons , que permite el uso, la distribución y la reproducción sin restricciones en cualquier medio,
siempre que se acredite al autor original y la fuente.

Financiamiento: Consejo Nacional de Ciencias, Taiwán. Número de contrato: 100-2118-M007-006-MY3 (

http://web1.nsc.gov.tw ). Los patrocinadores no tuvieron ningún papel en el diseño del estudio, la recopilación y el análisis de
datos, la decisión de publicar o la preparación del manuscrito.

Conflicto de intereses: Los autores han declarado que no existen conflictos de intereses.

Introducción
La diversidad funcional cuantifica la diversidad de los rasgos de las especies en las comunidades biológicas y se considera
ampliamente como una clave para comprender los procesos de los ecosistemas y el estrés o la perturbación ambiental [1] - [11] .
Una mayor diversidad funcional significa mayores diferencias entre los valores de los rasgos de las especies, funciones ecológicas
más distintas y, por lo tanto, una estabilidad funcional potencialmente mejor frente a las perturbaciones causadas por los impactos
humanos o el estrés ambiental [12] - [15] . Por lo tanto, es fundamental para cuantificar la diversidad funcional correctamente [16] .

La diversidad funcional generalmente se cuantifica mediante el uso de medidas basadas en los valores de las características de
las especies y la abundancia de las especies (o cualquier medida de la importancia de las especies, por ejemplo, la cobertura o la
biomasa). En la literatura se ha desarrollado una amplia gama de medidas de diversidad funcional [4] , [17] - [22] ; ver [23] – [26]

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 1/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

para revisiones. Existen tres enfoques principales para construir medidas de diversidad funcional: basadas en valores de rasgos
[27] , basadas en dendrogramas [4] , [28] - [31] y basadas en la distancia [17] , [32] - [35]. Para el enfoque basado en el valor de
los rasgos, las medidas se calculan directamente a partir de los valores de los rasgos de las especies. En el enfoque basado en
dendogramas, se construye un dendograma funcional aplicando un algoritmo de agrupamiento a la matriz de distancia por pares
de especies. Sin embargo, se ha demostrado que diferentes métodos de agrupamiento pueden conducir a diferentes conclusiones
[21] , [30] , [31] . Un problema inevitable en el enfoque basado en dendogramas es cómo seleccionar un algoritmo de
agrupamiento para construir un dendograma funcional.

Este documento se centra en el enfoque basado en la distancia que no requiere un dendrograma. Por lo tanto, se puede evitar la
selección del algoritmo de agrupamiento. Un índice de diversidad funcional comúnmente utilizado en el enfoque basado en la
distancia es el FAD (Diversidad de Atributos Funcionales), la suma de distancias por pares entre especies [17] . Sin embargo, FAD
no tiene en cuenta la abundancia de especies. La entropía cuadrática Q de Rao y sus transformaciones también se han aplicado
ampliamente para cuantificar la diversidad funcional [32] - [35] . La medida Q, una generalización del índice Gini-Simpson
tradicional, incorpora distancias por pares de especies y abundancias de especies. Sin embargo, hereda propiedades matemáticas
del índice de Gini-Simpson que son inapropiadas para una medida de diversidad [35] , [36] – [38] . Los problemas de interpretar Q
como una medida de diversidad se discutirán brevemente e ilustrarán con ejemplos más adelante en este documento. Ricotta y
Szeidl [35] y de Bello et al. [39] resolvió estos problemas convirtiendo Qa “especies equivalentes”; vea el texto posterior para más
detalles. Sin embargo, mostramos aquí que su solución en su forma original no se comporta correctamente para matrices de
distancia no ultramétricas. (Una distancia métrica d es ultramétrica si satisface el criterio d ( x , y )≤max{ d ( x , z ), d ( y , z )} para
todo x , y y z.) En muchas aplicaciones, las matrices de distancia calculadas a partir de los rasgos de las especies no satisfacen
este criterio y, por lo tanto, no son ultramétricas. Por ejemplo, las matrices de distancia de Gower de uso común calculadas a partir
de tres hábitats en nuestros datos reales (consulte Ejemplos y aplicaciones ) no son ultramétricas. Además, las medidas basadas
en la entropía cuadrática dan mucho más peso a las especies comunes que a su fracción de población. Sería más informativo
tener un parámetro para controlar la sensibilidad de la medida a la abundancia de especies. Por lo tanto, estábamos motivados
para derivar una nueva clase de medidas paramétricas que son válidas tanto para matrices ultramétricas como no ultramétricas.

Nuestro marco se basa en los números de Hill, una familia de índices de diversidad de un parámetro (que difieren entre sí solo por
un parámetro q que determina la sensibilidad a las abundancias relativas) que incorporan la riqueza de especies y abundancias
relativas. Los números de Hill incluyen la riqueza de especies, la diversidad de Shannon (el exponencial de la entropía) y la
diversidad de Simpson (inverso del índice de Simpson). Fueron utilizados por primera vez en ecología por MacArthur [40] ,
desarrollados por Hill [41] , y recientemente reintroducidos a los ecologistas por Jost [42] , [43] . A continuación se proporciona una
descripción muy breve de los números de Hill.

Los números de colinas se utilizan cada vez más para caracterizar la diversidad de especies de un conjunto basada en la
abundancia; véase una serie de artículos en un foro reciente [44] . Una ventaja importante de usar números de Hill es que los
números de Hill obedecen a un principio de replicación intuitivo , una propiedad matemática esencial que captura la noción intuitiva
de diversidad de los biólogos [40] , [41] ; ver Conclusión y Discusión para más detalles. El principio de replicación requiere que si
tenemos N conjuntos igualmente diversos e igualmente grandes sin especies en común, la diversidad del conjunto combinado
debe ser N veces la diversidad de un solo conjunto. Nos referimos al caso especial deN  = 2 como una "propiedad de duplicación"
como se define en [36] . Los números de Hill se ampliaron recientemente para incorporar la distancia filogenética y la distancia
funcional basada en dendogramas entre especies [45] sin dejar de satisfacer el principio de replicación.

Este trabajo primero generaliza los números de Hill a números de Hill funcionales basados ​en la distancia , que cuantifican "el
número efectivo de especies igualmente abundantes y (funcionalmente) igualmente distintas". A lo largo de este documento, las
especies son igualmente distintas si todas las distancias funcionales por pares de especies son una constante fija. Para
caracterizar completamente la diversidad funcional basada en la distancia, también necesitamos medidas en unidades de
"distancia". El producto de nuestro número de Hill funcional y la entropía cuadrática de Rao (la distancia media ponderada por
abundancia entre especies) cuantifica la diversidad funcional media(por especie), es decir, la suma efectiva de las distancias
funcionales entre una especie fija y todas las demás especies (más la distancia intraespecífica, si existe). El producto del número
de Hill funcional y la diversidad funcional media cuantifica así la diversidad funcional total (o simplemente diversidad funcional ), la
distancia total efectiva entre las especies del conjunto. Cuando todas las especies son igualmente distintas, nuestros números de
Hill funcionales se reducen a números de Hill ordinarios. Cuando no se consideran las abundancias de especies o las especies
son igualmente abundantes, nuestra diversidad funcional total se reduce a FAD [17] . Por lo tanto, nuestro enfoque también se
extiende FADpara incorporar abundancias de especies. En [46] , [47] , [48] y [48] , respectivamente, se pueden encontrar
diferentes perspectivas sobre los enfoques basados ​en la distancia y el principio de replicación .

Cuando hay conjuntos múltiples definidos por el investigador, los números de Hill pueden dividirse multiplicativamente en
componentes alfa y beta independientes (o no relacionados) [43] , [49] . Una ventaja de nuestro enfoque es que cada una de las
tres medidas de diversidad funcional que proponemos (números de Hill funcionales, diversidad funcional media y diversidad
funcional total) se puede utilizar para una partición multiplicativa completa. El componente beta se puede transformar aún más en
el rango [0, 1] para obtener medidas normalizadas de similitud funcional (o diferenciación), incluido N-Generalizaciones
funcionales de ensamblaje de los índices de similitud clásicos de Jaccard, Sørensen, Horn y Morisita-Horn. Nuestro marco, por lo
tanto, une medidas de diversidad funcional y similitud funcional (o diferenciación) entre ensamblajes. Nuestro trabajo anterior
sobre los números de Hill cubrió la diversidad taxonómica y la diversidad filogenética. Con el desarrollo actual de los números de
Hill funcionales, ahora tenemos un enfoque unificado para cuantificar y dividir las diversidades taxonómicas, filogenéticas y
funcionales. La mayoría de las medidas de diversidad propuestas anteriormente se pueden transformar en este marco y se
pueden entender mejor a través de él. Véase Chao et al. [50] para un marco integrado.

Números de la colina

En la diversidad de especies tradicional, solo se consideran la riqueza de especies y la abundancia de especies. Suponga que hay
S especies en un conjunto y las especies están indexadas por i  = 1, 2, …, S . Sea p i la abundancia relativa de la i -ésima especie.
Hill [41] integró la riqueza de especies y la abundancia de especies en una clase paramétrica de medidas de diversidad que luego

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 2/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

se denominó números de Hill, o el número efectivo de especies, definido para q ≠1 como

(1a)

El parámetro q determina la sensibilidad de la medida a las abundancias relativas. Cuando q  = 0, 0 D es simplemente riqueza de
especies. Para q  = 1, la ecuación. 1a no está definido, pero su límite cuando q tiende a 1 es la exponencial de la conocida entropía
de Shannon, y se denomina diversidad de Shannon en [51] :

(1b)

La medida 1 D pesa las especies en proporción a sus abundancias. Cuando q  = 2, la ecuación. 1a produce el inverso de la
concentración de Simpson, que se conoce como diversidad de Simpson [51] :

(1c)

q
Esta medida pone más peso en las especies abundantes y descarta fuertemente las especies raras. Para todo q , si D  =  u , la
diversidad (de orden q ) del conjunto real es la misma que la de un conjunto idealizado con u especies igualmente abundantes.
Esta es la razón por la que los números de Hill se denominan números efectivos de especies o equivalentes de especies.

Una caracterización completa de la diversidad de especies basada en la abundancia tradicional de un conjunto con especies S y
abundancias relativas se transmite mediante un perfil de diversidad que traza q D versus q de q  = 0 a q  = 3 o 4 (más
allá de esto cambia poco) [52] . Aunque se pueden calcular los números de Hill para q <0, están dominados por la abundancia de
especies raras y tienen malas propiedades estadísticas de muestreo. Por lo tanto, nos limitamos al caso q ≥0 a lo largo del
artículo.

Hill [41] proved a weak version of replication principle for Hill numbers: if two equally large assemblages with no species in common
have identical relative abundance distributions, then the Hill number of the pooled assemblage is doubled. Chiu et al. (Appendix B
of [36]) recently proved a strong version as given in Introduction: the assumptions needed are that N assemblages with no species
in common are equally large and equally diverse (relative abundance distributions may be different, unlike the weak version).
Species richness is a Hill number (with q = 0) and obeys both versions of the replication principle, but most other traditional diversity
indices do not obey even the weak version. The replication principle has been discussed for characterizing abundance-based
species diversity measures [40]–[43], [53]–[56]. This replication principle will be generalized to functional diversity measures in later
sections.

Previous Distance-based Functional Diversity and Differentiation Measures

A large number of functional diversity measures have been proposed in the literature, and each measure quantifies a different
aspect of species trait space. Here we mainly review those distance-based functional indices and differentiation measures that are
related to our functional generalization of Hill numbers. The FAD measure is defined as [17]

(2a)

where dij is the functional distance between the ith and jth species, dij = dji≥0. However, this measure does not take into account the
abundances of the species, which may play an important role in the functioning of ecosystems; see [19], [57]–[63].

Functional diversity measures combining both functional distance and species abundance have been proposed [20], [32], [33], [34],
[64], [65]. Rao's quadratic entropy for an assemblage with species relative abundances is the most widely used
measure [32]:

(2b)

The measure Q is interpreted as the mean distance between any two randomly sampled individuals in the assemblage. It can also
be interpreted as the abundance-weighted mean distance between two species. Here the weighting factor for a distance dij is the
product of the relative abundances, pipj. This measure is an extension of the Gini-Simpson index. Although this measure has wide
applications in many disciplines, Q, like the Gini-Simpson index, is not linear with respect to the addition of new species and thus
does not obey the replication principle, causing counterintuitive results in ecological applications [35], [36], [66]. For an additive
decomposition, another problem arises when the species functional distance matrix does not result in the concavity of Q [5], [67]
(e.g. for the Gower distance matrix, in general). Then Q in the pooled assemblage (gamma quadratic entropy) may be smaller than

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 3/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

the Q of the average of local assemblages (alpha quadratic entropy), implying Q in this situation could not to be used for additive
decomposition [67], [68], [69]. When additive decomposition is feasible, the associated differentiation measure that has been used
in the literature is the quadratic entropy excess normalized by the gamma quadratic entropy [70], [71]:

(2c)

where and denote respectively the alpha and gamma quadratic entropy. However, when alpha quadratic entropy is high, the
differentiation measure always tends to zero (implying no differentiation) regardless of distance matrices and differences in
species abundances across assemblages [36]. This behavior leads to severe interpretational problems.

To fix the problems with Q, Ricotta and Szeidl [35] and de Bello et al. [39] made an advance by transforming Q to the “species
equivalents”, which is the effective number of equally distinct species with a constant distance dmax for all different-species pairs;
here dmax denotes the maximum value in the distance matrix. Their transformation is expressed as

(2d)

We refer to this number as “the effective number of (equally distinct) species with maximum distance”. Equivalently, they scale all
distances so they are between 0 and 1, by dividing each distance by its maximum value in the distance matrix prior to all analyses.
de Bello et al. [39], Villéger et al. [72] and Escalas et al. [73] applied the above formula to gamma and alpha quadratic entropies
and obtained the corresponding effective number of species for gamma (denoted here by ) and alpha (denoted by ), where
the subscript “e” denotes “effective”. The resulting beta based on a multiplicative decomposition is  =  . de Bello [39]
further transformed this beta diversity into a normalized differentiation measure so that the resulting measure is in the unit interval
[0, 1]:

(2e)

where N denotes the number of assemblages. Villéger et al. [72] also proposed a normalized differentiation measure:

(2f)

However, as we will show by examples, Eq. 2d and the two associated differentiation measures (given in Eqs. 2e and 2f) might
yield un-interpretable results when they are applied to non-ultrametric distance matrices. This motivates our new approach which is
valid for both ultrametric and non-ultrametric matrices.

Guiasu and Guiasu [38], [74] proposed a class of distance-weighted Gini-Simpson index as follows:

(2g)

They also proposed the corresponding measure for a multiplicative decomposition. We will show that the three measures (FAD, Q
and GSD) are closely related to our proposed measures. Leinster and Cobbold [75] derived a parametric class of measures
sensitive to species similarity. Scheiner [63] also proposed a metric that integrates abundance, phylogeny, and function. Since both
approaches are also based on a framework of Hill numbers, it is important to distinguish these two previous approaches from ours;
see Conclusion and Discussion for more details. Neither Leinster and Cobbold's approach nor Scheiner's metric have been
developed to construct normalized similarity (or differentiation) measures that can be applied to analyze datasets such as those
discussed in Examples and Applications.

Proposed Functional Diversity Measures

A Simple Framework for Ordinary Hill Numbers.

We first present a simple conceptual framework for ordinary Hill numbers. Then we extend it to obtain our proposed functional Hill
numbers. The intuitive interpretation of the “effective number of species” implies that if an assemblage with S species and species
abundance vector has diversity D, then the diversity of this actual assemblage is the same as that of an idealized
reference assemblage with D species and species abundance (1/D, 1/D, …, 1/D).

Now we construct the q-th power sum (q≠1) of the abundances with unity weight for each species, i.e., ; see
Table 1. Taking the same function for the idealized reference assemblage, i.e., replacing S and by D and (1/D, 1/D, …,
1/D) respectively, we obtain . Equating the two sums shows that D is the Hill number of order q:

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 4/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

This provides a simple and intuitive derivation of Hill numbers. This derivation facilitates the extension of Hill numbers to incorporate
functional distances.

Table 1. A framework for Hill numbers, functional Hill numbers, mean functional diversity and (total) functional diversity of a single assemblage.
https://doi.org/10.1371/journal.pone.0100014.t001

Functional Diversity Measures of an Assemblage.

Let dij denote the functional distance between the ith and jth species, with dij≥0, and dij = dji. Denote the S×S symmetric pairwise
distance matrix by [dij]. In our approach, species functional distance, which quantifies the proximity of species in functional trait
space, can be any type of symmetric matrix. To extend Hill numbers to incorporate functional distances between species, we
consider a framework based on pairs of species [38], [74]. That is, we consider a collection of all S2 pairs of species: {(1, 1), (1, 2),
(1, 3), …, (S, S)}. The joint “relative abundance” or joint probability for each species-pair (i, j) is pipj. Consider the matrix ,
where the (i, j) element of the matrix is pipj (Table 1). Note that the mean distance between any two species weighted by their joint
probability is Rao's quadratic entropy defined in Eq. 2b.

Analogous to the derivation of Hill numbers, we consider the q-th power sum (q≠1) of all elements of the matrix with
weight dij for species pair (i, j), i.e., . A similar concept of the “effective number of equally abundant and equally
distinct species” as in ordinary Hill numbers can be applied to the functional version as follows. When species are equally distinct
with a constant pairwise distance, the quadratic entropy Q must be equal to this constant. An assemblage with the effective number
of species means that this assemblage has the same diversity as an idealized reference assemblage having equally common
and equally distinct species with a constant distance Q for all S2 pairs of species. Here we have S2 pairs because same-species
pairs are included so that intraspecific variability can be considered when trait values are available at the individual level [25], [76].
(If there is no intraspecific variability, then the distance for a same-species pair is set to be 0 and a common distance
is set for different-species pairs; see Table 1. All measures derived in the following are still valid when intraspecific
distance is zero, and all interpretations can be adapted to the case when there is no intraspecific variability.) For simplicity, our
derivation and interpretations are mainly based on S2 pairs of species.

Taking the same q-th power sum function (q≠1) for the idealized reference assemblage with a constant weight Q for all species
pairs, we obtain . Equating the two sums from the actual and the idealized reference assemblages leads to

Then we can solve and the solution given below is denoted by :

(3)

For q = 1, we define the following limit as our measure:

The measure is a function of the distance matrix [dij] and the joint probability matrix . Here we express it as a
function of the quadratic entropy Q to emphasize the important role of Q in the construction of other measures (see Eqs. 4a and 4b)
and in the proof of the replication principle (discussed later). The measure is the dimension (the number of columns or rows)

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 5/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

of the distance matrix of the idealized reference assemblage in Table 1. We refer to it as the functional Hill number of order q. The
measure can be interpreted as “the effective number of equally abundant and (functionally) equally distinct species” with a
constant distance Q for all species pairs. Thus if  = v, then the functional Hill number of order q of the actual assemblage is the
same as that of an idealized assemblage having v equally abundant and equally distinct species with a constant distance Q for all
species pairs; see Table 1 for illustration.

To derive measures in units of “distance”, note that in the idealized reference assemblage, all columns and all rows have identical
sums. We define the column (or row) sum as our proposed measure of mean functional diversity (per species), qMD(Q), of order q:

(4a)

which quantifies the effective sum of pairwise distances between a fixed species and all other species (plus intraspecific distance if
exists). In other words, qMD(Q) measures the dispersion per species in the functional trait space [18]. The product of the functional
Hill numbers and the mean functional diversity thus quantifies the total functional diversity (or simply functional diversity), qFD(Q), in
the assemblage:

(4b)

This functional diversity quantifies the effective total distance between species of the assemblage. If qFD(Q) = u, then the effective
total distance between species of the actual assemblage with quadratic entropy Q is the same as that of an idealized assemblage
having (u/Q)1/2 equally abundant and equally distinct species with a constant distance Q for all species pairs.

Consider the following special cases to intuitively understand the meaning of our functional diversity measures and their
relationships with FAD (Eq. 2a) or GSD (Eq. 2g):

a. When all species in the assemblage are equally distinct (i.e., for all species pairs (i, j), for i, j = 1, 2, …, S), the functional Hill number reduces
to ordinary Hill number.

b. For q = 0,  = (FAD/Q)1/2, 0MD(Q) = (FAD×Q)1/2, and 0FD(Q) = FAD, where FAD is defined in Eq. 2a. Thus, our measures have a direct link to FAD.

c. If all species are equally abundant, then for any distance matrix (dij), we have  = S, and qFD(Q) = FAD for all orders of q. Therefore, when species
abundances are not considered (q = 0) or species are equally abundant, our total functional diversity reduces to FAD. In the equally abundant case, we have
q
MD(Q) , implying that our mean functional diversity is conceptually similar to the Modified Functional Attribute Diversity (MFAD) proposed by
Schmera et al. [18].

d. When q = 2, we have the following link to the weighted Gini-Simpson index defined in Eq. 2g [38], [74]:

(4c)

As with the diversity profile for Hill numbers, a profile which plots , qMD(Q) or qFD(Q) with respect to the order q completely
characterizes the information each measure gives for an assemblage. As proved in Appendix S1, all three measures ,
q
MD(Q) and qFD(Q) are Schur-concave with respect to the product of relative abundances, implying these measures satisfy a
functional version of “weak monotonicity” [45], [77], [78]. That is, if a rarest new species is added to an assemblage, then the
measure qFD(Q) does not decrease regardless of distance matrices. Also, if a rarest new species is added to an assemblage such
that the quadratic entropy remains unchanged, then all three measures do not decrease.

Functional Diversity Measures for a Pair of Assemblages.

We next extend Rao's quadratic entropy, FAD, functional Hill number, mean functional diversity and total functional diversity to a
pair of assemblages (say, I and II). Assume that there are S1 species in Assemblage I and S2 species in Assemblage II. Let the two
sets of species relative abundances be denoted by and for Assemblage I and II respectively.

We first extend Rao's quadratic entropy to a pair of assemblages. Assume that an individual is randomly selected from each of the
assemblages. Then the probability that the individual from Assemblage I belongs to species i and the individual from Assemblage II
belongs to species j is pi1pj2, i = 1, 2, …, S1, j = 1, 2, …, S2. The mean distance between these two randomly selected individuals is

(5a)

This measure can also be interpreted as the abundance-weighted mean distance between a species from Assemblage I and a
species from Assemblage II, and the weighting factor is the product of their relative abundances. For simplicity, we refer to Q12 as
the mean distance between species of Assemblage I and Assemblage II. Clearly, we have Q12 = Q21. The traditional Rao's

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 6/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

quadratic entropy for Assemblage I is simply Q11 for the same-assemblage pair (I, I) and the quadratic entropy for Assemblage II is
simply Q22 for the same-assemblage pair (II, II).

We can apply a similar approach to that in Table 1 by conceptually thinking that there are two idealized assemblages, and each
assemblage includes equally abundant and equally distinct species such that the two actual assemblages and the two
idealized assemblages have the same value of a given diversity measure. Replacing the joint probability matrix in Table 1
with the S1×S2 matrix and using parallel derivations, we obtain the following functional Hill number for Assemblage I
and Assemblage II:

(5b)

This measure is interpreted as “the effective numbers of equally abundant and equally distinct species in each of two assemblages,
with a constant distance Q12 between species of Assemblage I and Assemblage II”. We also define the mean functional diversity of
Assemblages I and II as , which quantifies the effective sum of pairwise distances between a fixed
species in one assemblage and all species in the other assemblage. Then the product of and quantifies the total
functional diversity (or simply functional diversity) of Assemblage I and Assemblage II as

(5c)

In the special case of q = 0, the above total functional diversity reduces to the total sum of all pairwise distances between species of
Assemblage I and Assemblage II. Since Q12 is not involved in the measure for q = 0, we denote 0FD(Q12)≡FAD12, which represents
an extension of Walker's FAD to a pair of assemblages. Thus FAD11 is identical to FAD for Assemblage I and FAD22 is identical to
FAD for Assemblage II. Also, we have the following relationship:

(5d)

Replication Principle.

We generalize the concept of the replication principle to a functional version and show that the proposed functional Hill numbers
and the mean functional diversity both satisfy the replication principle. Consequently, the product of these two measures (i.e., our
proposed total functional diversity) satisfies a quadratic replication principle (i.e., the total functional diversity of the pooled
assemblage is N2 times that of any individual assemblage.) A general proof of the replication principle for N completely distinct
assemblages is given in Appendix S1. Throughout this paper, N assemblages are completely distinct if there are no shared species
(and thus no shared species pairwise distances).

To simplify the concept, here we present the replication principle only for two assemblages. Assume that two equally large and
completely distinct assemblages are pooled. Let Q11, Q12, Q21, and Q22 denote respectively the mean distance between species of
the four pairs of assemblages, (I, I), (I, II), (II, I) and (II, II). Assume that the functional Hill number of order q for all of the four pairs
of assemblages is a constant . When the two assemblages are combined, the quadratic entropy in the pooled assemblage
becomes and the functional Hill number of order q in the pooled assemblage is doubled.
Consequently, if we further assume that the four mean distances (Q11, Q12, Q21 and Q22) are identical, then the mean functional
diversity in the pooled assemblage is also doubled, and the total functional diversity is quadrupled; see Appendix S1 for a general
proof for N assemblage.

In Guiasu and Guiasu's work on the quadrupling property [74], they proved a weak version of the quadrupling property for their
proposed weighted Gini-Simpson type index (Eq. 4c) when two equally large and completely distinct assemblages (I and II) are
pooled. They assume that the joint probability matrices for the four pairs of assemblages, (I, I), (I, II), (II, I) and (II, II), are identical,
and also assume that the species distance matrices for the four pairs of assemblages are also identical. The latter assumption
implies the FAD for the four pairs is a constant (say, A), i.e., FAD11 = FAD12 = FAD21 = FAD22≡A. This weak version can be directly
used to understand why the functional diversity of order zero (i.e., FAD) satisfies a quadrupling property. In this simple case,
consider the distance matrix of the pooled assemblage when the two actual assemblages have no species shared. It is readily seen
that the total distance between species in the pooled assemblage is quadrupled because the FAD in the pooled assemblage is
FAD11+FAD12+FAD21+FAD22 = 4×A. As shown in the proof (Appendix S1), our replication principle is a strong version in the sense
that there are no restrictions on the joint probability matrices and on the distance matrices.

Partitioning Functional Diversity Measures

Assume that there are N assemblages defined by the investigator. The functional Hill number (Eq. 3), mean functional
diversity qMD(Q) (Eq. 4a) and total functional diversity qFD(Q) (Eq. 4b) of the pooled assemblage can each be decomposed into
independent alpha and beta components. Here we adopt the decomposition method developed by Chiu et al. [36]. We first briefly
review Chiu et al.'s method for partitioning Hill numbers in order to provide necessary backgrounds. To calculate the gamma Hill
numbers, species abundances are pooled over assemblages; the gamma diversity is Hill numbers computed from the species
relative abundances in the pooled assemblage. However, as noted by de Bello et al. [39], how to define species relative
abundances in the pooled assemblage depends on how we pool data over assemblages. The pooling scheme depends on the

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 7/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

objective of the study. If the objective is to compare absolute abundances among assemblages (given the total abundance in the
pooled assemblage), we should pool individuals over assemblages; if the objective is to compare relative abundances among
assemblages, we should pool relative abundances of individual assemblages. These two kinds of pooling schemes lead to different
relative abundances in the pooled assemblage. (The former is equivalent to an assemblage-size-weighted pooling scheme for
relative abundances, whereas the latter naturally reduces to the equal-assemblage-weighted pooling scheme.) See Chao et al. [50,
their Appendix 2] for a simple example for illustration.

Suppose in the pooled assemblage there are S species indexed by 1, 2, …, S. To incorporate both kinds of pooling schemes into
our framework, we define zik as any measure of species “abundance” of the ith species in the kth assemblage, i = 1, 2, …, S, k = 1,
2, …, N. Some of the zik may be zero. The measure zik can be absolute abundances, relative abundances, incidence, biomasses,
cover areas or any other species importance measure. Define as the size of the kth assemblage. Let
be the total abundance in pooled assemblage and be the total abundances of the ith species in
the pooled assemblage. Then the species relative abundance set in the pooled assemblage for both pooling schemes can be
expressed as { ; i = 1, 2,…, S}. Note that if we pool relative abundances over assemblages instead of absolute abundance,
we have the special case that z+k = 1 and z++ = N.

The abundance-based gamma diversity is Hill numbers computed from the species relative abundances { ; i = 1, 2,…, S}
and is interpreted as the effective number of species in the pooled assemblage. The traditional definition of alpha diversity is “the
mean of the diversities of individual assemblages”. Routledge [55] and Jost [43] each derived a mathematical formula for alpha
diversity based on this traditional definition and obtained the corresponding multiplicative beta component. As Chiu et al. [36]
indicated, this traditional approach to alpha diversity based on Hill numbers leads to a beta that can only be used to produce
differentiation measures to compare species relative abundances, but not absolute abundances. This is because in the framework
of Hill numbers, diversity is a function of relative abundances only, and thus “the mean of individual diversities” lose information
about absolute abundances. Chiu et al. [36] expanded the conventional concept of alpha and proposed a modified definition for
abundance-based alpha diversity: “alpha diversity is the effective number of species per assemblage” so that the resulting beta can
be transformed to quantify the differentiation (or similarity) among N sets of vectors , k = 1, 2, …, N, for any measure
of species importance zik, including absolute abundances. Based on this expanded definition, Chiu et al. derived a new formula for
abundance-based alpha diversity.

When the data represent species relative abundances (i.e., equal-weight for assemblages), all three alpha formulas (Routledge,
Jost and Chiu et al.) are identical. They differ, however, when the data represent species absolute abundances (i.e., assemblage-
size as weight); Routledge's beta may exceed N and Jost's beta may be less than unity (i.e., gamma may be less than alpha) if
q≠1. Also, for q = 0, Routledge's alpha and beta are not independent [49]. Chiu et al.'s new formula of the abundance-based alpha
diversity has the following advantages: (1) it leads to a beta that can be applied to compare any types of data (zik), depending on
the investigator's objective; (2) gamma is always greater than or equal to alpha for all orders q≥0; (3) beta is always between unity
(when all assemblages are identical in species absolute abundances) and N (when the N assemblages have no species in
common); and (4) alpha and beta components are independent for all orders q≥0.

Based on species abundance (zik) and a species pairwise distance matrix, we now extend Chiu et al. approach to decompose the
functional diversity qFD(Q) of the pooled assemblage into within-assemblage component (functional alpha diversity) and between-
assemblage component (functional beta diversity). As with the partition of Hill numbers, the functional gamma diversity of order q is
based on the distance matrix [dij] and the relative abundance in the pooled assemblage. It then follows from Eq. 4b,
the functional gamma diversity of order q can be written as

(6a)

where is the quadratic entropy in the pooled assemblage. The limit when q approaches unity exists and is
equal to

(6b)

The functional gamma diversity is interpreted as the effective total distance between species in the pooled assemblage with a
constant distance Q for all species pairs.

We follow Chiu et al.'s definition of alpha diversity to define the functional alpha diversity as the effective total distance between
species of a pair of individual assemblages. Then we obtain (details of derivation are provided in Appendix S2) the functional alpha
diversity of order q:

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 8/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

(7a)

(7b)

Note that in our alpha formula, Q still refers to the quadratic entropy for the pooled assemblage. When relative abundances are the
relevant quantities for the investigative question, we simply define the measure zik as the ith species relative abundance in the kth
assemblage. Then and thus z++ is replaced by N in all of the above formulas, Eqs. 6a, 6b, 7a and 7b.

As with ordinary Hill numbers [36], [43], the complete partitioning of functional gamma diversity into independent within- and
between-assemblage (alpha and beta) components is multiplicative. That is, the functional beta diversity is the ratio of functional
gamma to functional alpha diversities:

(8)

The decomposition procedures for the other two measures are generally parallel and interpretations are similar. A summary of the
decomposition of the three measures with interpretations is given in Table 2 and the formulas for decomposing functional Hill
numbers are provided in Appendix S2. Note that for functional Hill number and mean functional diversity, their beta components are
identical, i.e.,  =  . Also, we have .

Table 2. Decomposition of the functional Hill number (Eq. 3), the mean functional diversity qMD(Q) (Eq. 4a) and the (total) functional
diversity qFD(Q) (Eq. 4b) along with interpretations.
https://doi.org/10.1371/journal.pone.0100014.t002

For each of the three measures ( , qMD(Q) and qFD(Q)), the gamma value is always greater than or equal to the
corresponding alpha component for all orders q≥0 and all distance matrices; see Appendix S2 for a proof. When N assemblages
are identical in species identities and abundance, the beta components of all three measures take their minimum value of unity.
When all assemblages are completely distinct (no shared species and thus no shared pairwise distances), we have  = 
and both attain the maximum value of N, and attains the maximum value of N2. The functional beta Hill number,
, thus quantifies the effective number of equally large and completely distinct assemblages. The functional beta diversity,
, quantifies the effective number of equally large and completely distinct pairs of assemblages. In Appendix S2, we show
that is always between unity and N; and is always between unity and N2. Thus, the range of each beta component
is independent of the corresponding alpha component, implying that the alpha and beta components based on the multiplicative
partitioning for each of the three functional diversity measures ( ) are unrelated (or independent).

We also note the following properties:

1. When all species are equally distinct, the functional beta Hill numbers reduce to the beta diversity for ordinary Hill numbers, and the functional beta diversity
reduces to the squared beta diversity of ordinary Hill numbers [36].

2. When q = 0, we have , and , where denotes the sum of pairwise distances in the pooled assemblage. For the alpha components, we have , and , where
FADpair is the sum of FADs over all possible pairs of assemblages (there are N2 pairs of assemblages). So the functional alpha diversity for q = 0 is the
average of FAD per pair of assemblages. Therefore, we have and  = .

3. For the special case q = 1 and q = 2, we will present formulas for some special transformations of the functional beta diversity in the next section.

Four Classes of Normalized Functional Similarity Measures

Our functional beta components, ( = ) and , all quantify pure functional differentiation among the N assemblages and their ranges
depends only on N. Thus, each can be transformed to obtain the normalized similarity and differentiation measures in [0, 1] so that
the dependence on N can be removed [36], [43], [49], [79]. As stated in the preceding section and proved in Appendix S2, the
range of is different from that of , so the transformations to normalized similarity measures in [0, 1] are thus different for these two
beta components. The similarity measures based on quantify species-overlap from different perspectives, whereas the similarity

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 9/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

measures based on quantify distance-overlap from different perspectives. In most applications of functional diversity, we suggest
using the distance-overlap similarity measures and their corresponding differentiation measures. We now describe the two major
classes of normalized distance-overlap measures based on ; see Table 3 for all formulas and Appendix S3 for details.

Table 3. Two major classes of distance-overlap (or similarity) measures and their special cases based on the functional beta diversity .
https://doi.org/10.1371/journal.pone.0100014.t003

(1) A class of local distance-overlap measures from the perspective of a pair of local assemblages

(9a)

Here “local” refers to a property of a pair of individual assemblage. This measure gives the effective average proportion of the
species pairwise distances in a pair of local assemblages that are shared with all other pairs of assemblages. See Appendix S3 for
the interpretation of the “effective” average proportion.

We first give the special case of q = 0 to intuitively explain its properties: since and , the measure reduces to

(9b)

where and FADpair are defined in the preceding section. In this expression, the denominator is the zero-order functional alpha
diversity, which is the average of FADs over all N2 assemblage pairs; the numerator is the average of all repeated pairwise
distances in the N2−1 pairs of assemblages (excluding the assemblage-pair in which a pairwise distance is first counted). The
measure thus quantifies the proportion of repeated distances in a pair of local assemblages. This interpretation is conceptually
similar to the traditional Sørensen similarity index. The difference is that here we consider “assemblage-pairs” in functional
distance-overlap measures rather than “individual assemblage” as in the traditional measure. Thus, this index can be regarded as
an extension of the Sørensen index to functional similarity. Therefore, the measure is referred to as “func-Sørensen” in Table 3. For
q = 1, this local distance-overlap measure is called “func-Horn” in Table 3 because when all distances are identical and (zik)
represents species relative abundance within each assemblage, it reduces to the classical Horn measure [80]. For q = 2, this
measure is called “func-Morisita-Horn” in Table 3 because its interpretation is generally similar to the classic Morisita-Horn measure
[81]. See Appendix S3 for more details.

(2) A class of regional distance-overlap measures in the pooled assemblage

(10a)

Here “regional” refers to a property of the pooled assemblage. This class of measures differs from the local distance-overlap
measures by taking a regional perspective. It quantifies the effective proportion of the species pairwise distances in the pooled
assemblage that are shared with all pairs of local assemblages (Appendix S3).

For the special case of q = 0, the measure reduces to

(10b)

Again, the interpretation is similar to the measure (in Eq. 9b) except that the denominator is replaced by the gamma FAD. Thus,
our index can be regarded as an extension of the Jaccard index to functional similarity. This is why the measure is referred to as
the N-assemblage “func-Jaccard” in Table 3. Since  = , the measure for q = 1 is also called “func-Horn” measure. For q = 2, this
measure is referred to as “func-regional-overlap” measure; see Appendix S3.

As shown in Chiu et al. [36], we can also define two additional classes of functional distance-overlap measures. First, a class of
functional distance-homogeneity measures which is a linear function of the inverse of the functional beta diversity:

(11)

Second, we have a class of measures which is a linear function of the functional beta diversity:

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 10/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

(12)

Its complement quantifies the functional distance-turnover rate.

All the four classes of similarity measures are continuous in q≥0, so a functional similarity or differentiation profile as a function of q
can be made for any of them. We suggest using this method for conveying complete information about the functional similarity or
differentiation of a set of assemblages. It is thus sufficient to focus on the two major classes ( and ) because they include and as
special cases for q = 0 and 2. See Example 3 for illustrative profiles.

Our decomposition presented above is based on the multiplicative scheme. We can also apply the additive decomposition to each
of the three measures, (Eq. 3), qMD(Q) (Eq. 4a) and (Eq. 4b). For example, we can define the “functional diversity excess” as .
The excess quantifies the effective total distances between species in the pooled assemblage not contained in a typical pair of local
assemblages. As with ordinary Hill numbers [49], the functional diversity excess depends not only on the number of assemblages
N, but also on the functional alpha diversity. Consequently, the excess measure cannot be directly applied to compare the similarity
or differentiation among assemblages across multiple sets of assemblages even if the numbers of assemblages in these multiple
regions are the same. We can eliminate these dependences by using appropriate normalizations [49]. In Appendix S4, we show
that after proper normalizations, the multiplicative approach and additive approach both lead to the same four classes of normalized
functional similarity and differentiation measures presented above. Thus, a consensus can be achieved on functional similarity and
differentiation measures, including those measures given in Table 3.

Examples and Applications

To examine the performance of our functional diversity measures and to compare our proposed similarity and differentiation
measures with previous indices, we use both artificial distance matrices (Examples 1 and 2) and real data (Example 3) for
illustration. Although the distance matrices considered in our artificial examples are simple, they provide transparent answers so
that we can clearly examine the performance of measures. Any meaningful differentiation measure should work properly for all
matrices. If a functional diversity or similarity/differentiation measure cannot yield logical and sensible results for simple matrices,
we would not expect it to work for complicated cases. The more complicated distance matrix calculated from real species traits is
used in Example 3 for illustration.

In our comparisons, we consider various functional differentiation measures: (1) a differentiation measure (Eq. 2c) based on the
traditional additive decomposition of quadratic entropy; (2) two differentiation measures (Eqs. 2e and 2f) based on the effective
number of species with maximum distance; and (3) the proposed distance-based differentiation measures derived from our
functional beta diversity (in Table 3 of this paper). Appendix S5 presents a simple example to show that the traditional measure
based on the additive partitioning of the quadratic entropy (Eq. 2c) cannot work properly even for a very simple functional distance
matrix; see Chiu et al. [36] for theoretic discussions and more examples.

Example 1: Effect of Functional Distances on Differentiation Measures.

Consider two assemblages (I and II). Each assemblage contains 20 species, with 12 shared species and 8 non-shared species.
There are 28 species in the pooled assemblage. For each assemblage, we first consider the equally abundant case in order to
examine how differentiation measures vary with functional distances. (Two non-equally-abundant cases are given in Appendix S5.)
The classical Sørensen-type dissimilarity index (the proportion of non-shared species in an individual assemblage) is 8/20 = 0.4.
(The abundance-based local differentiation measure based on Hill numbers is 0.4 for all q≥0; see [36].) The classical Jaccard-type
dissimilarity index (the proportion of non-shared species in the pooled assemblage) is 1–12/28 = 0.571; see Table 4 for abundance-
based regional differentiation measure based on Hill numbers [36]. For functional differentiation measures, the quantifying target is
shifted to the proportion of the total non-shared distances (incorporating abundances if q>0) in an individual assemblage () or in the
pooled assemblage ().

Table 4. Comparison of various differentiation measures for Matrix I (with  = 0.48,  = 0.47) and Matrix II (with  = 0.167,  = 0.102) based on abundance
and function (A&F), on function (F) only, and abundance (A) only.
https://doi.org/10.1371/journal.pone.0100014.t004

We generated two contrasting types of distance matrices (Matrix I and Matrix II). Both matrices are displayed in Appendix S6. For
easy presentation, species are indexed by 1, 2, …, 28 in the pooled assemblage. Assemblage I includes Species 1–20, and
Assemblage II includes Species 9–28 (Species 9–20 are shared). In Matrix I, the distances for two species within an assemblage
follow the same distribution as those for species from the pooled assemblage so that the alpha quadratic entropy Qα (the average
distance between any two individuals within an assemblage) is close to the gamma quadratic entropy (the average distance
between any two individuals in the pooled assemblage). In this case, we expect that any meaningful functional differentiation
measure is largely determined by species abundances. In Matrix II, the gamma quadratic entropy is much higher than the alpha
quadratic entropy Qα, as described below. Consequently, we expect that functional distances should play an important role in
characterizing functional differentiation.

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 11/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

a. Matrix I. All the species pairwise distances in the 28×28 distance matrix of the pooled assemblage were generated from a beta (4, 4) distribution, which is a
symmetric distribution with respect to 0.5. In this case, the alpha quadratic entropy (Qα = 0.47) is close to the gamma quadratic entropy ( = 0.48).

b. Matrix II. We constructed the 28×28 distance matrix by generating substantially larger distances for pairs of “non-shared species” (s1, s2), where the first
species s1 is a non-shared species in Assemblage I, and the second species s2 is a non-shared species in Assemblage II. The distances for such pairs of
non-shared species were generated from a uniform (0.8, 1) distribution whereas the distances for other species pairs were generated from a uniform (0, 0.2)
distribution. We have Qα = 0.102 and  = 0.167. There is large relative difference between Qα and , as reflected by the high relative difference (with respect to the
alpha) of 63.7%.

In Table 4, we first compare separately for Matrix I and Matrix II the differentiation measures incorporating both abundance and
function (A&F), function (F) only, and abundance (A) only. The measures considering both (A&F) are based on our proposed
measures and (with formulas in Table 3) derived from the functional beta diversity. The measure based only on function only (F)
does not consider abundance, so it is identical to the zero-order of the measure considering A&F. The measures considering
abundance only (A) refer to the abundance-based local differentiation measure (1−CqN) and regional differentiation measure
(1−UqN) based on partitioning Hill numbers ([36], p. 31).

Comparing the column under A& F and the column under A within Matrix I, we find for each fixed order of q = 0 and q = 2 that there is
appreciable difference between these two values (A& F and A) but the difference is limited to some extent (relatively to the
corresponding difference for Matrix II); the difference is very little for q = 1. This is valid for both differentiation measures and . Thus,
for Matrix I (with similar distributional pattern of distances for all species pairs), functional differentiation is largely determined by
species abundance pattern and function plays a minor factor.

In contrast, for Matrix II, the impact of function on our differentiation measures is clearly seen for both measures and by noting that
our measure considering both (A&F) is much higher than the corresponding measure considering A for all orders q = 0, 1 and 2.
This is because the functional distances for pairs of non-shared species are substantially larger than those of other species pairs,
leading to a large increase in the proportion of non-shared distances in an assemblage (as reflected in our local distance-
differentiation measure ), and also in the pooled assemblage (as reflected in our regional distance-differentiation measure ). In this
case, function has profound effect on characterizing functional differentiation. Since the two measures (A&F and A) of q = 1 differ
little for Matrix I whereas they differ substantially for Matrix II, their difference is a potentially useful indicator for the effect of
function. All the above findings not only hold for equally abundant species as the example presented here but also are generally
valid if species abundances are heterogeneous; see Appendix S5 for two heterogeneous cases.

For both matrices the proposed measures exhibit moderate differentiation between the two assemblages for Matrix I and moderate
to high differentiation for Matrix II. For example, our proposed measure, , yields values 0.324 (for q = 0), 0.408 (for q = 1) and 0.491
(for q = 2) for Matrix I. The corresponding three values for Matrix II are 0.579 (for q = 0), 0.628 (for q = 1) and 0.678 (for q = 2). Table 4
reveals that the differentiation measure based on the additive partitioning of the quadratic entropy exhibits an unreasonably low
differentiation value of 0.002 for Matrix I. As shown in reference [36], this measure does not properly quantify functional
differentiation; also see the example in Appendix S5. The two measures based on the effective number of species with maximum
distance (Eqs. 2e and 2f) for both matrices also show unreasonably low differentiation. For Matrix I, the measure in Eq. 2e gives a
value of 0.004 and the measure in Eq. 2f gives a value of 0.002, implying that there is almost no differentiation among the two
assemblages. These are counter-intuitive and unexpected values because function has almost no effect and thus all measures for
Matrix I should yield close results to those based on abundances only (the column under A in Table 4). This example also helps
show that the measures in Eqs. 2e and 2f cannot be applied to non-ultrametric cases, as the two matrices are both non-ultrametric
(Appendix S6). Similar findings about substantially low functional differentiation are also revealed in other papers [82], [83]. For
Matrix II, each of the two previously developed measures (Eqs. 2e and 2f) is also substantially lower than our proposed
differentiation measure considering both (A&F). More evidence from other perspectives is provided in Example 2 below.

Example 2: Ultrametric vs. Non-ultrametric Distance Matrices.

In this example, we compare the performance of various differentiation measures when they are applied to an ultrametric matrix
(Case I in Table 5) and a non-ultrametric matrix (Case II in Table 5). Each matrix represents a distance matrix for a pooled
assemblage of four species. In each case, there are two completely distinct assemblages (no species shared). There are two
equally common species (a, b) in the first assemblage, and two equally common species (c, d) in the other assemblage. We use
this simple example to show that the effective approach based on the effective number of species with maximum distance (Eq. 2d)
and the associated differentiation measures (Eqs. 2e and 2f) may lead to un-interpretable conclusions if they are applied to non-
ultrametric distance matrices.

Table 5. Comparison of various differentiation measures between two assemblages for an ultramteric distance matrix (Case I below) and a non-
ultrametric distance matrix (Case II below).
https://doi.org/10.1371/journal.pone.0100014.t005

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 12/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

Comparing the two distance matrices, we see that the two matrices are identical except for the distances for the two pairs, (a, c)
and (b, d). The distance between Species a and Species c is 0.2 in Case I but it is increased to 0.9 in Case II; the distance for
Species b and Species d is 0.2 in Case I but it is increased to 0.8 in Case II. Thus, when the matrix is changed from Case I to Case
II, the distance for any two species in different assemblages is either increased or kept as the same, whereas all the distances for
species in the same assemblage are kept the same. By intuition and by theory for our measures (Proposition S2.2 in Appendix S2),
any sensible differentiation measure should not decrease.

In Table 5, we compare various differentiation measures between the two assemblages separately for Case I and Case II. The
measures based on Eqs. 2e and 2f both produce a maximum differentiation of unity for Case I. This is intuitively understandable
because the two assemblages are completely distinct and all distances for two species in different assemblages are higher than the
distances for two species within an assemblage. In both Case 1 and Case II, the proposed differentiation measures, and , attain
the maximum differentiation of unity for all orders of q, showing the differentiation does not decrease from Case I to Case II.
However, the two differentiation measures (Eqs. 2e and 2f) for Case II give unexpectedly lower differentiation than that of Case I.
This example shows why application of Eq. 2d and the associated differentiation measures (Eqs. 2e and 2f) to non-ultrametric
cases might be misleading. Although the measure based on additively partitioning quadratic entropy (Eq. 2c) yields higher
differentiation for Case II, we have demonstrated its counter-intuitive behavior in Appendix S5 and in Example 1.

In this example, we specifically use the extreme case that two assemblages are completely distinct (no shared species) for
illustrative purpose. A more general property of monotonicity is proved in Appendix S2 (Proposition S2.2): any differentiation
measure based on our functional beta diversity is a non-decreasing function with respect to the distance of any non-shared species
pair regardless of species abundance distributions. This property of monotonicity implies that the differentiation measure including
and do not decrease if the distance for a non-shared species pair becomes larger even if the two assemblages are not completely
distinct. In Appendix S5, we provide a supplementary example in which there are shared species between assemblages; our
proposed measures yield the expected property of monotonicity, while the two previous differentiation measures (Eqs. 2e and 2f) do
not.

Example 3: A Real Functional Distance Matrix for Dune Vegetation.

We apply our proposed measures to the real data discussed by Ricotta et al. in [84]. The data contain a total of 43 vascular plant
species collected from 272 random vegetation plots of 2×2 m in size during the period 2002–2009 in three successively less
extreme fore dune habitats: embryo dunes (EM; 17 species in 70 plots), mobile dunes (MO; 39 species in 131 plots) and transition
dunes (TR; 42 species in 71 plots) along the Tyrrhenian coast, where EM is closest to the sea, MO is between EM and TR, and TR
is farthest from the sea; see [85], [86], [87] for details. There are 17 shared species (out of a total of 39 species) between EM and
MO, 16 shared species (out of a total of 43 species) between EM and TR, and 38 shared species (out of a total of 43 species)
between MO and TR. In each habitat, we pooled species abundance data over plots and applied various diversity and
differentiation measures based on the species relative abundances (Table S5.4 in Appendix S5) in the three type habitats.

All species were described by a set of sixteen functional traits which include seven quantitative variables: plant height, leaf size,
leaf thickness, seed mass, seed shape, leaf dry mass and specific leaf area, together with nine categorical variables: life form,
growth form, leaf texture, dispersal mode, leaf persistence, plant life span, pollination system, clonality and flowering phenology.
Based on these sixteen traits, the species distance matrix in the pooled assemblage was calculated by a Gower mixed-variables
coefficient of distance with equal weights for all traits [71]. The Gower species pairwise distance matrix of the pooled assemblage is
provided in Appendix S6. The matrix and the three sub-matrices (corresponding to those of three habitats) are all non-ultrametric.
The two idealized examples (Example 1 and Example 2) just given showed that previously-proposed functional differential
measures led to unexpected conclusions when applied to non-ultrametric matrices. This real example shows how such
mathematical problems can lead to misinterpretation of important ecological patterns.

For each of the three habitats, we present four diversity measures: ordinary Hill numbers qD (Eq. 1a), our functional Hill number
(Eq. 3), mean functional diversity qMD(Q) (Eq. 4a) and functional diversity qFD(Q) (Eq. 4b). The diversity profiles for the four
diversity measures as a function of order q are shown in Fig. 1. A consistent pattern is revealed in Fig. 1: EM has the lowest
diversity, MO is intermediate, and TR has the highest diversity. This pattern is valid for all orders of q, and is expected from
ecologists' perspectives [84]. The EM is closest to the sea, and hence exposed to wind disturbance, flooding, salt spray, and other
harsh environmental factors. Therefore, the assemblage in the EM is mainly composed of a few specialized pioneer species with
similar functional traits (as reflected by the value of quadratic entropy, which is respectively 0.513, 0.556, and 0.561 in EM, MO and
TR) to adapt the extreme environment, leading to the lowest functional diversity in this habitat. The vegetation of the MO is less
affected by harsh environment factors, so the vegetation presents more diverse species composition, resulting in larger functional
distances and thus higher functional diversity. The species richness and evenness in the TR are the highest among the three
habitats and the vegetation of TR is even more weakly constrained by these environmental factors, supporting an even higher
functional diversity. The diversity pattern for Hill numbers is similar to those based on functional diversity measures, as will be
discussed later. In each of the three functional diversity profiles (the two middle panels and the right panel of Fig. 1), the initial value
(i.e., the value for q = 0) represents the diversity when only function is considered.

Figure 1. Diversity profiles as a function of order q for ordinary Hill numbers qD (left panel), functional Hill numbers (the second panel from the
left), mean functional diversity qMD(Q) (the third panel from the left) and (total) functional diversity qFD(Q) (right panel) for three habitats (TR,
MO, and EM).
All the profiles show a consistent diversity pattern about the ordering of the three habitats: TR>MO>EM.
https://doi.org/10.1371/journal.pone.0100014.g001

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 13/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

The formula in Eq. 2d produces much lower values of species equivalents: 2.94 (EM), 3.39 (MO) and 2.95 (TR), substantially lower
than the corresponding functional Hill numbers (q = 2): 7.72 (EM), 15.27 (MO), 19.42 (TR); see the second panel of Fig. 1.
Moreover, the number of species equivalents from Eq. 2d give a diversity ordering MO>TR≈EM, which does not conform to
ecologists' expectation.

In Fig. 2, we show the differentiation profiles of the two proposed measures and as a function of order q for q between 0 and 5. In
Table 6, we compare various differentiation measures between any two habitats (EM vs. MO, EM vs. TR and MO vs. TR). In the
same table, as we did in Table 4, we also show the differentiation values incorporating both abundance and function (A&F), function
(F) only, and abundance (A) only. Table 6 reveals that in any pair of assemblages, we have a pattern similar to that in Table 4 for
Matrix I. That is, our differentiation measures considering both (A&F) yield comparable results to those considering abundance only
(A) for q = 0 and for q = 2, and yield very close results for q = 1. As with Example 1, this may be explained by the fact that the gamma
quadratic entropy in each pair of assemblage is only slightly higher than the alpha quadratic entropy. The relative differences
between gamma and alpha quadratic entropies is respectively 2.8%, 4.5% and 2.7% for EM vs. MO, EM vs. TR and MO vs. TR.
Therefore, abundance is the major factor that determines the differentiation between any two habitats, implying that the four
measures incorporating abundances with or without considering function exhibit very similar patterns in Fig. 1.

Figure 2. Differentiation profiles for the functional differentiation measures (left panel) and (right panel) as a function of order q for three pairs
of habitats (EM vs. MO, EM vs. TR and MO vs. TR.)
https://doi.org/10.1371/journal.pone.0100014.g002

Table 6. Comparison of various differentiation measures for three pairs of habitats in the real data analysis based on abundance and function
(A&F), on function (F) only, and abundance (A) only.
https://doi.org/10.1371/journal.pone.0100014.t006

Our proposed differentiation measures, and (Table 6 and Fig. 2) implies that EM vs. TR has the highest functional differentiation,
MO vs. TR has the lowest differentiation, and EM vs. MO is somewhat in between for any fixed order q between 0 and 5. This
pattern is anticipated. As discussed above, the vegetation within EM is composed by few specialized plants with similar ecological
functions to adapt the extreme environmental stress. However, these traits are unique to species in EM when compared with
species in the other two habitats. There are also fewer shared species between EM and TR (also EM and MO). In contrast, the
vegetation in MO and TR is similarly diverse and most species in these two habitats are shared. These explain why MO vs. TR
exhibits the lowest functional differentiation, whereas EM vs. TR (also EM vs. MO) exhibit higher functional differentiation.

Table 6 and Fig. 2 further reveal that the two measures and for the three pairs of habitats give moderate to high differentiation. For
example, for q = 2, our differentiation measure for the three pairs (EM vs. MO, EM vs. TR and MO vs. TR) is respectively 0.658,
0.885 and 0.539, and the corresponding differentiation measure is respectively 0.324, 0.659 and 0.226. In sharp contrast, the three
previous measures based on the quadratic entropy (Eqs. 2c, 2e and 2f) show substantially lower differentiation. For these data, the
differentiation measure based on the additive decomposition of quadratic entropy (Eq. 2c) for EM vs. MO, EM vs. TR and MO vs.
TR is respectively 0.028, 0.042 and 0.026. This wrongly implies substantially low differentiation between any two habitats. For the
differentiation measure based on Eq. 2f are also low (0.034, 0.054 and 0.035). These values also give an unexpected ordering in
that EM vs. MO exhibits the lowest functional differentiation, which is counter-intuitive. Similarly, the measure given in Eq. 2e gives
a wrong ordering. All three examples demonstrate that our functional diversity measures and their associated differentiation
measures yield the expected results and ecologically sensible interpretations.

Conclusion and Discussion

We have extended ordinary Hill numbers to the distance-based functional Hill number to take into account the pairwise functional
distance between species (see Eq. 3, in units of effective number of equally abundant and equally distinct species). Here Q (Rao's
quadratic entropy) plays an important indirect role, even though the measure Q itself cannot be directly used to measure functional
diversity as noted by several authors [26], [35], [36], [39]. We have also proposed a class of mean functional diversity qMD(Q) = ;
see Eq. 4a. The product of the functional Hill number and the mean functional diversity quantifies the (total) functional diversity
q

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 14/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

q
FD(Q) = , i.e., the effective total distance between species of the assemblage. See Fig. 1 for an example comparing ordinary Hill
numbers and the three functional diversity measures. The three proposed measures quantify different aspects of species trait
space. Our approach is valid not only for any symmetric distance matrices in ecology, but also for all types of symmetric matrices in
other disciplines.

Since the pioneering work by MacArthur [40] and Hill [41], the replication principle has been identified as an essential property for
characterizing abundance-based species diversity. As we reviewed in this paper, Hill numbers obey the replication principle. Hill
numbers have been extended to phylogenetic Hill numbers (in units of “species equivalent”) and related branch diversity (in units of
“branch length”); both satisfy a phylogenetic generalization of the replication principle [36], [45]. In this paper, we have proved that
the functional Hill numbers (in units of “species equivalent”) and the mean functional diversity (in units of “functional distance”) both
satisfy a functional version of the replication principle, and also proved that the functional diversity (in units of “functional distance”)
satisfies a quadratic replication principle. Therefore, we think replication principle is an essential property for measures in units of
species equivalents, but for other related measures this property may be valid (e.g., branch diversity, mean functional diversity), or
may be replaced by a quadratic (or a power function of N) property; see [48] for a different perspective.

Recently, Chao et al. [50] integrated species diversity, phylogenetic diversity and functional diversity into a unified framework of
“attribute diversity” based on Hill numbers. Both Leinster and Cobbold [75] and Scheiner [63] derived their integrative metrics under
a framework of Hill numbers and their metrics are also in units of “species equivalents”. In Appendix S5, we provide detailed
comparison to distinguish these two previous approaches from ours. Generally, we find that Leinster & Cobbold's measure may not
be sensitive to species abundances when species similarity matrix is computed from species traits in functional analysis. If species
similarity matrix deviates greatly from a naïve identity matrix, then their measure typically yields very low diversity values especially
for assemblages with many species; this causes problems for the interpretation of “species equivalents” in their approach. Reeve et
al. [88] recently proposed a diversity partition based on Leinster & Cobbold's measure. In the same Appendix, we show by an
example that their gamma diversity may be less than their proposed alpha diversity even in equal weight case. Scheiner's approach
and our measures have different meanings of “species equivalents” and thus quantify different aspects and properties of
ecosystems. Scheiner's measure cannot be directly linked to most of the previous commonly used phylogenetic diversity (e.g.,
Faith's PD) and functional measures (e.g., FAD).

Except for Rao's quadratic entropy, the decomposition of other functional diversity measures is rarely discussed in the literature. In
this paper, we have developed the decomposition of the proposed three functional diversity measures of any order q; see Table 2
for a summary. In the decomposition of each of the three measures, the alpha and beta components are unrelated (or
independent), and thus each beta component measures pure functional differentiation among assemblages and can be
transformed onto the range [0,1] to obtain the normalized distance-overlap measures (from the beta component based on the
functional diversity) or species-overlap measures (from the beta components based on decomposing the functional Hill numbers
and the mean functional diversity). In most applications, we recommend applying the distance-overlap measures (given in Table 3
for the two major classes of similarity measures). An important advantage of using the framework of Hill numbers is that there is a
direct link between functional diversity measures and functional similarity (or differentiation) among assemblages. To convey the
information about functional differentiation among multiple assemblages, we suggest plotting the differentiation profiles for two
differentiation measures, (from the perspective of a pair of local assemblages) and (from the perspective of the pooled
assemblage) with respect to q; see Fig. 2 for examples. For the special case of q = 0, the measure (Eq. 9b and Table 3) in terms of
FAD represents the functional generalization of the Sørensen similarity index, and the measure (Eq. 10b and Table 3) represents
the functional generalization of the classic Jaccard similarity index. Conceptually different approach to diversity decomposition was
proposed by Kosman [48].

Villéger et al. [20] proposed that trait-based functional diversity should include three aspects: functional richness, functional
evenness and functional divergence. In ordinary Hill numbers, zero-order diversity represents species richness, and Hill numbers
(or their transformations) of different orders can be used to construct various functional evenness measures as those proposed by
Jost [89]. Jost used partitioning theory to derive Hill's useful class of evenness measures, the ratios of Hill numbers qD and species
richness, qD/S for q>0, and he showed that the ratio of the logarithms of Hill numbers and logarithm of richness, log(qD)/log(S),
expresses the corresponding relative evenness. Applying this idea to our framework, we can construct measures of functional
richness and functional evenness based on any of our three functional diversity measures. For example, we can regard the
effective total distance between species of order q = 0 as a measure of functional richness and use /[] and log[]/log[] as measures of
functional evenness.

Finally, we mention a potential application of our proposed measures. In genetics, the nucleotide diversity [90] is based on a
nucleotide distance matrix. The elements of this distance matrix are obtained as the nucleotide differences between any two DNA
sequences. Since our approach can be applied to any type of symmetric distance matrix, we expect our proposed measures would
be useful in genetics. The nucleotide diversity in genetics is equivalent to Rao's quadratic entropy, and the corresponding
differentiation measure NST is identical to the traditional differentiation measure in Eq. 2c. We have shown in this paper by
hypothetical and real examples that the measure in Eq. 2c does not provide legitimate measure of differentiation; see [36] for
theoretical discussion. We have proposed unified and rigorous distance-overlap measures and their corresponding differentiation
measures (in Table 3), and thus those measures merit investigation for applications to genetics.

Supporting Information
Appendix S1.
Some properties of the proposed functional diversity measures.
https://doi.org/10.1371/journal.pone.0100014.s001
(PDF)

Appendix S2.
Decomposition of the proposed functional diversity measures.
https://doi.org/10.1371/journal.pone.0100014.s002

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 15/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

(PDF)

Appendix S3.
Four classes of functional similarity/differentiation measures.
https://doi.org/10.1371/journal.pone.0100014.s003
(PDF)

Appendix S4.
Functional beta diversity and functional diversity excess lead to the same classes of similarity and differentiation measures.
https://doi.org/10.1371/journal.pone.0100014.s004
(PDF)

Appendix S5.
Supplementary examples and comparisons.
https://doi.org/10.1371/journal.pone.0100014.s005
(PDF)

Appendix S6.
Distance matrices used in Example 1 and Example 3.
https://doi.org/10.1371/journal.pone.0100014.s006
(XLSX)

Acknowledgments
The original data for calculating the species relative abundances in Appendix S5 and the Gower species pairwise distance matrix
used in our real data analysis were kindly provided by Carlo Ricotta and Alicia T. R. Acosta. We thank them for permission to use
their data. The authors also thank Lou Jost, Carlo Ricotta, the Academic Editor (Francesco de Bello), and two reviewers (Samuel
Scheiner and Evsey Kosman) for carefully reading and editing an earlier version and providing very helpful and insightful
comments, which substantially improved this paper. C. H. Chiu is supported by a post-doctoral fellowship, National Tsing Hua
University, Taiwan.

Author Contributions
Conceived and designed the experiments: CHC AC. Performed the experiments: CHC AC. Analyzed the data: CHC AC.
Contributed reagents/materials/analysis tools: CHC AC. Wrote the paper: CHC AC. Figures: CHC.

References
1. Tilman D, Knops J, Wedin D, Reich P, Ritchie M, et al. (1997) The influence of functional diversity and composition on ecosystem processes. Science 277:
1300–1302.
View Article Google Scholar

2. Tilman D (2001) Functional diversity. In: Levin AS, editor. Encyclopedia of Biodiversity, Vol. 3. San Diego: Academic Press. pp. 109–120.

3. Diaz S, Cabido M (2001) Vive la différence: plant functional diversity matters to ecosystem processes. Trends Ecol Evol 16: 646–655.
View Article Google Scholar

4. Petchey OL, Gaston KJ (2002) Functional diversity (FD), species richness and community composition. Ecol Lett 5: 402–411.
View Article Google Scholar

5. Ricotta C (2005) A note on functional diversity measures. Basic Appl Ecol 6: 479–486.
View Article Google Scholar

6. Swenson NG (2011) Phylogenetic beta metrics, trait evolution and inferring the functional beta diversities of communities. Plos One 6: e21264.
View Article Google Scholar

7. Swenson NG, Erickson DL, Mi X, Bourg NA, Forero-Montaña J, et al. (2012) Phylogenetic and functional alpha and beta diversity in temperate and
tropical tree communities. Ecology 93: 112–125.
View Article Google Scholar

8. Norberg J, Swaney DP, Dushoff J, Lin J, Casagrandi R, et al. (2001) Phenotypic diversity and ecosystem functioning in changing environments: a
theoretical framework. Proc Natl Acad Sci USA 98: 11376–11381.
View Article Google Scholar

9. Suding KN, Lavorel S, Chapin FS, Cornelissen JH, Diaz S, et al. (2008) Scaling environmental change through the community-level: a trait-based
response-and-effect framework for plants. Glob Change Biol 14: 1125–1140.
View Article Google Scholar

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 16/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

10. Cadotte MW, Cavender-Bares J, Tilman D, Oakley TH (2009) Using phylogenetic, functional and trait diversity to understand patterns of plant community
productivity. PLoS One 4: e5695.
View Article Google Scholar

11. Flynn DF, Mirotchnick N, Jain M, Palmer MI, Naeem S (2011) Functional and phylogenetic diversity as predictors of biodiversity-ecosystem-function
relationships. Ecology 92: 1573–1581.
View Article Google Scholar

12. Dukes JS (2001) Productivity and complementarity in grassland microcosms of varying diversity. Oikos 94: 468–480.
View Article Google Scholar

13. Folke C, Carpenter S, Walker B, Scheffer M, Elmqvist T, et al. (2004) Regime shifts, resilience, and biodiversity in ecosystem management. Annu Rev
Ecol Evol Syst 557–581.
View Article Google Scholar

14. Cardinale BJ, Duffy JE, Gonzalez A, Hooper DU, Perrings C, et al. (2012) Biodiversity loss and its impact on humanity. Nature 486: 59–67.
View Article Google Scholar

15. Mouillot D, Graham NA, Villéger S, Mason NW, Bellwood DR (2013) A functional approach reveals community responses to disturbances. Trends Ecol
Evol 28: 167–177.
View Article Google Scholar

16. Petchey OL, Gaston KJ (2009) Dendrograms and measures of functional diversity: a second installment. Oikos 118: 1118–1120.
View Article Google Scholar

17. Walker B, Kinzig A, Langridge J (1999) Plant attribute diversity, resilience, and ecosystem function: The nature and significance of dominant and minor
species. Ecosystems 2: 95–113.
View Article Google Scholar

18. Schmera D, Erös T, Podani J (2009) A measure for assessing functional diversity in ecological communities. Aquat Ecol 43: 157–167.
View Article Google Scholar

19. Díaz S, Lavorel S, de Bello F, Quétier F, Grigulis K, et al. (2007) Incorporating plant functional diversity effects in ecosystem service assessments. Proc
Natl Acad Sci USA 104: 20684–20689.
View Article Google Scholar

20. Villéger S, Mason NW, Mouillot D (2008) New multidimensional functional diversity indices for a multifaceted framework in functional ecology. Ecology 89:
2290–2301.
View Article Google Scholar

21. Poos MS, Walker SC, Jackson DA (2009) Functional-diversity indices can be driven by methodological choices and species richness. Ecology 90: 341–
347.
View Article Google Scholar

22. Mouchet MA, Villéger S, Mason NWH, Mouillot D (2010) Functional diversity measures: an overview of their redundancy and their ability to discriminate
community assembly rules. Funct Ecol 24: 867–876.
View Article Google Scholar

23. Cadotte MW, Carscadden K, Mirotchnick N (2011) Beyond species: functional diversity and the maintenance of ecological processes and services. J Appl
Ecol 48: 1079–1087.
View Article Google Scholar

24. Weiher E (2011) A primer of trait and functional diversity. In: Magurran AE, McGill BA, editors. Biological diversity: frontiers in measurement and
assessment: pp.175–193.

25. Pla L, Casanoves F, Di Rienzo J. (2012) Functional diversity indices. In: Quantifying Functional Biodiversity, London: Springer. pp. 27–51.

26. Mouillot D, Graham NAJ, Villéger S, Mason NWH, Bellwood DR (2013) Functional approach reveals community responses to disturbances. Trends Ecol
Evol 28: 167–177.
View Article Google Scholar

27. Mason NW, MacGillivray K, Steel JB, Wilson JB (2003) An index of functional diversity. J Veg Sci 14: 571–578.
View Article Google Scholar

28. Petchey OL, Gaston KJ (2006) Functional diversity: back to basics and looking forward. Ecol Lett 9: 741–758.

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 17/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

View Article Google Scholar

29. Petchey OL, Gaston KJ (2007) Dendrograms and measuring functional diversity. Oikos 116: 1422–1426.
View Article Google Scholar

30. Podani J, Schmera D (2006) On dendrogram-based measures of functional diversity. Oikos 115: 179–185.
View Article Google Scholar

31. Podani J, Schmera D (2007) How should a dendrogram-based measure of functional diversity function? A rejoinder to Petchey and Gaston. Oikos 116:
1427–1430.
View Article Google Scholar

32. Rao CR (1982) Diversity and dissimilarity coefficients: A unified approach. Theor Popul Biol 21: 24–43.
View Article Google Scholar

33. Botta-Dukát Z (2005) Rao's quadratic entropy as a measure of functional diversity based on multiple traits. J Veg Sci 16: 533–540.
View Article Google Scholar

34. Ricotta C, Szeidl L (2006) Towards a unifying approach to diversity measures: bridging the gap between the Shannon entropy and Rao's quadratic index.
Theor Popul Biol 70: 237–243.
View Article Google Scholar

35. Ricotta C, Szeidl L (2009) Diversity partitioning of Rao's quadratic entropy. Theor Popul Biol 76: 299–302.
View Article Google Scholar

36. Chiu C-H, Jost L, Chao A (2014) Phylogenetic beta diversity, similarity, and differentiation measures based on Hill numbers. Ecol Monogr 84: 21–44.
View Article Google Scholar

37. Hardy OJ, Jost L (2008) Interpreting and estimating measures of community phylogenetic structuring. J Ecol 96: 849–852.
View Article Google Scholar

38. Guiasu RC, Guiasu S (2011) The weighted quadratic index of biodiversity for pairs of species: a generalization of Rao's index. Nat Sci 3: 795–801.
View Article Google Scholar

39. de Bello F, Lavergne S, Meynard CN, Lepš J, Thuiller W (2010) The partitioning of diversity: showing Theseus a way out of the labyrinth. J Veg Sci 21:
992–1000.
View Article Google Scholar

40. MacArthur RH (1965) Patterns of species diversity. Biol Rev 40: 510–533.
View Article Google Scholar

41. Hill M (1973) Diversity and evenness: A unifying notation and its consequences. Ecology 54: 427–432.
View Article Google Scholar

42. Jost L (2006) Entropy and diversity. Oikos 113: 363–375.

View Article Google Scholar

43. Jost L (2007) Partitioning diversity into independent alpha and beta components. Ecology 88: 2427–2439.
View Article Google Scholar

44. Ellison AM (2010) Partitioning diversity. Ecology 91: 1962–1963.

View Article Google Scholar

45. Chao A, Chiu C-H, Jost L (2010) Phylogenetic diversity measures based on Hill numbers. Phil Trans R Soc B 365: 3599–3609.
View Article Google Scholar

46. Kosman E (1996) Difference and diversity of plant pathogen populations: a new approach for measuring. Phytopathology 86: 1152–1155.
View Article Google Scholar

47. Kosman E, Leonard KJ (2007) Conceptual analysis of methods applied to assessment of diversity within and distance between populations with asexual
or mixed mode of reproduction. New Phytol 174: 683–696.
View Article Google Scholar

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 18/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

48. Kosman E (2014) Measuring diversity: from individuals to populations. Eur J Plant Pathol 138: 467–486.
View Article Google Scholar

49. Chao A, Chiu C-H, Hsieh TC (2012) Proposing a resolution to debates on diversity partitioning. Ecology 93: 2037–2051.
View Article Google Scholar

50. Chao A, Chiu C-H, Jost L (2014) Unifying species diversity, phylogenetic diversity, functional diversity and related similarity/differentiation measures
through Hill numbers. Annu Rev Ecol Evol Syst (in press).
View Article Google Scholar

51. Chao A, Gotelli NJ, Hsieh TC, Sander E, Ma KH, et al. (2014) Rarefaction and extrapolation with Hill numbers: a framework for sampling and estimation in
species diversity studies. Ecol Monogr 84: 45–67.
View Article Google Scholar

52. Tóthmérész B (1995) Comparison of different methods for diversity ordering. J Veg Sci 6: 283–290.
View Article Google Scholar

53. MacArthur RH (1972) Geographical Ecology: patterns in the distribution of species. Princeton: Princeton University Press. 288 p.

54. Whittaker RH (1972) Evolution and measurement of species diversity. Taxon 12: 213–251.
View Article Google Scholar

55. Routledge R (1979) Diversity indices: which ones are admissible? J Theor Biol 76: 503–515.
View Article Google Scholar

56. Jost L, DeVries P, Walla T, Greeney H, Chao A, et al. (2010) Partitioning diversity for conservation analyses. Divers Distrib 16: 65–76.
View Article Google Scholar

57. Grime J (1998) Benefits of plant diversity to ecosystems: immediate, filter and founder effects. J Ecol 86: 902–910.
View Article Google Scholar

58. Vile D, Shipley B, Garnier E (2006) Ecosystem productivity can be predicted from potential relative growth rate and species abundance. Ecol Lett 9:
1061–1067.
View Article Google Scholar

59. Stuart-Smith RD, Bates AE, Lefcheck JS, Duffy JE, Baker SC, et al. (2013) Integrating abundance and functional traits reveals new global hotspots of fish
diversity. Nature 501: 539–542.
View Article Google Scholar

60. Devictor V, Mouillot D, Meynard C, Jiguet F, Thuiller W, et al. (2010) Spatial mismatch and congruence between taxonomic, phylogenetic and functional
diversity: the need for integrative conservation strategies in a changing world. Ecol Lett 13: 1030–1040.
View Article Google Scholar

61. Stegen JC, Hurlbert AH (2011) Inferring ecological processes from taxonomic, phylogenetic and functional trait β-diversity. PloS one 6: e20906.
View Article Google Scholar

62. Swenson NG (2011) Phylogenetic beta diversity metrics, trait evolution and inferring the functional beta diversity of communities. PloS one 6: e21264.
View Article Google Scholar

63. Scheiner SM (2012) A metric of biodiversity that integrates abundance, phylogeny, and function. Oikos 121: 1191–1202.
View Article Google Scholar

64. Mason NW, Mouillot D, Lee WG, Wilson JB (2005) Functional richness, functional evenness and functional divergence: the primary components of
functional diversity. Oikos 111: 112–118.
View Article Google Scholar

65. Laliberté E, Legendre P (2010) A distance-based framework for measuring functional diversity from multiple traits. Ecology 91: 299–305.
View Article Google Scholar

66. Jost L (2008) GST and its relatives do not measure differentiation. Mol Ecol 17: 4015–4026.
View Article Google Scholar

67. Champely S, Chessel D (2002) Measuring biological diversity using Euclidean metrics. Environ Ecol Stat 9: 167–177.

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 19/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

View Article Google Scholar

68. Pavoine S, Dufour AB, Chessel D (2004) From dissimilarities among species to dissimilarities among communities: a double principal coordinate analysis.
J Theor Biol 228: 523–537.
View Article Google Scholar

69. Pavoine S, Ollier S, Pontier D (2005) Measuring diversity from dissimilarities with Rao's quadratic entropy: Are any dissimilarities suitable? Theor Popul
Biol 67: 231–239.
View Article Google Scholar

70. Hardy OJ, Senterre B (2007) Characterizing the phylogenetic structure of communities by an additive partitioning of phylogenetic diversity. J Ecol 95: 493–
506.
View Article Google Scholar

71. Pavoine S, Vallet J, Dufour AB, Gachet S, Daniel H (2009) On the challenge of treating various types of variables: application for improving the
measurement of functional diversity. Oikos 118: 391–402.
View Article Google Scholar

72. Villéger S, Miranda JR, Hernandez DF, Mouillot D (2012) Low functional β-diversity despite high taxonomic β-diversity among tropical estuarine fish
communities. PloS One 7: e40679.
View Article Google Scholar

73. Escalas A, Bouvier T, Mouchet MA, Leprieur F, Bouvier C, et al. (2013) A unifying quantitative framework for exploring the multiple facets of microbial
biodiversity across diverse scales. Environ. Microbiol. 15: 2642–2657.
View Article Google Scholar

74. Guiasu RC, Guiasu S (2012) The weighted Gini-Simpson index: revitalizing an old index of biodiversity. International J Ecol 2012, Article ID 478728, 10
pages.
View Article Google Scholar

75. Leinster T, Cobbold CA (2012) Measuring diversity: the importance of species similarity. Ecology 93: 477–489.
View Article Google Scholar

76. de Bello F, Lavorel S, Albert CH, Thuiller W, Grigulis K, et al. (2011) Quantifying the relevance of intraspecific trait variability for functional diversity.
Methods Ecol Evol 2: 163–174.
View Article Google Scholar

77. Marshall A, Olkin I (1979) Inequalities: Theory of Majorization and its Applications. New York: Academic Press.

78. Weikard HP, Punt M, Wessler J (2006) Diversity measurement combining relative abundances and taxonomic distinctiveness of species. Divers Distrib 12:
215–217.
View Article Google Scholar

79. Chao A, Jost L, Chiang SC, Jiang Y-H, Chazdon RL (2008) A two-stage probabilistic approach to multiple-community similarity indices. Biometrics 64:
1178–1186.
View Article Google Scholar

80. Horn HS (1966) Measurement of “overlap” in comparative ecological studies. Am Nat 419–424.
View Article Google Scholar

Morisita M (1959) Measuring of interspecific association and similarity between communities. Memoires of the Faculty of Science, Kyushu University,
81.
Series E (Biology) 3: 65–80.
View Article Google Scholar

82. Bernard-Verdier M, Flores O, Navas ML, Garnier E (2013) Partitioning phylogenetic and functional diversity into alpha and beta components along an
environmental gradient in a Mediterranean rangeland. J Veg Sci 24: 877–889.
View Article Google Scholar

83. Villéger S, Miranda JR, Hernandez DF, Mouillot D (2012) Low functional β-diversity despite high taxonomic β-diversity among tropical estuarine fish
communities. PloS one 7: e40679.
View Article Google Scholar

84. Ricotta C, Pavoine S, Bacaro G, Acosta ATR (2012) Functional rarefaction for species abundance data. Methods Ecol Evol 3: 519–525.
View Article Google Scholar

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 20/21
25/4/22, 20:26 Medidas de diversidad funcional basadas en la distancia y su descomposición: un marco basado en números de Hill

85. Carboni M, Santoro R, Acosta A (2010) Are some communities of the coastal dune zonation more susceptible to alien plant invasion? J Plant Ecol 3: 139–
147.
View Article Google Scholar

86. Carboni M, Santoro R, Acosta AT (2011) Dealing with scarce data to understand how environmental gradients and propagule pressure shape fine-scale
alien distribution patterns on coastal dunes. J Veg Sci 22: 751–765.
View Article Google Scholar

87. Carboni M, Acosta AT, Ricotta C (2013) Are differences in functional diversity among plant communities on Mediterranean costal dunes driven by
phylogenetic history? J Veg Sci 24: 932–941.
View Article Google Scholar

88. Reeve R, Matthews L, Cobbold CA, Leinster T, Thompson J, Brummitt N (2014) How to partition diversity. Cornell University Library, arXiv:1404.6520,
assessed on May 28, 2014

89. Jost L (2010) La relación entre uniformidad y diversidad. Diversidad 2: 207–232.

Ver artículo Google Académico

90. Nei M, Li WH (1979) Modelo matemático para estudiar la variación genética en términos de endonucleasas de restricción. Proc Natl Acad Sci USA 76:
5269–5273.
Ver artículo Google Académico

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0100014 21/21