Blackwell Publishing LtdOxford, UKBOJBotanical Journal of the Linnean Society0024-4074The Linnean Society of London, 2006?

2006
151?
103111
Review Article

MORPHOMETRICS IN PALM SYSTEMATICS

A. HENDERSON

Botanical Journal of the Linnean Society, 2006, 151, 103–111.

The Palms
Guest edited by William J. Baker and Scott Zona

Traditional morphometrics in plant systematics and its

Downloaded from https://academic.oup.com/botlinnean/article-abstract/151/1/103/2420544 by guest on 13 February 2020

role in palm systematics
ANDREW HENDERSON*
New York Botanical Garden, Bronx, NY 10458, USA

Received April 2005; accepted for publication November 2005

A review is given of the role of traditional morphometrics in plant systematics. The three most commonly used tech-
niques of data analysis – Cluster Analysis, Principal Component Analysis and Discriminant Analysis – are dis-
cussed. The kinds of data that can be taken from palm specimens and the problems of using specimens as data
sources are outlined. Published systematic studies of palms using traditional morphometrics are reviewed. More
recent studies indicate that: hybrid zones between species may be common; infraspecific diversity is greater than
previously suspected; there may be more than double the currently accepted number of species; and our current
knowledge of morphological variation in palms is superficial. A procedure for scientific systematics is given, which
incorporates traditional morphometric methods. © 2006 The Linnean Society of London, Botanical Journal of the
Linnean Society, 2006, 151, 103–111.

ADDITIONAL KEYWORDS: Arecaceae – Cluster Analysis – Discriminant Analysis – geometric morphomet-

rics – numerical taxonomy – Principal Component Analysis – Palmae.

INTRODUCTION are size and shape variables (i.e. distances or angles

between homologous landmarks) and also include
Morphometrics can be defined as the quantitative
qualitative variables, usually taken from herbarium
analysis of biological form. It has been widely used in
specimens. I emphasize that my perspective is that of
a variety of disciplines, including systematics. The
a herbarium systematist and this bias means I am
field has developed rapidly over the last 20 years, to
interested in both size and shape variables. Other
the extent that we now distinguish between tradi-
morphometricians consider that the field concerns
tional morphometrics (e.g. Marcus, 1990) and the
only the analysis of shape (e.g. Rohlf, 1990).
more recent geometric morphometrics (e.g. Rohlf &
I use the term variable for any qualitative (binary
Marcus, 1993; Adams, Rohlf & Slice, 2004). Geometric
or multistate) or quantitative (continuous, meristic)
morphometrics, in which information about the rela-
attribute. Although I distinguish between qualitative
tive spatial arrangement of landmarks is preserved, is
and quantitative variables, the distinction is not
not widely used in plant systematics (Jensen, 2003)
always clear (Stevens, 1991; Thiele, 1993). Qualitative
and has not yet found its way into palm systematics. It
variables can also be referred to as characters or
will not be discussed further here.
traits, in the sense of the Phylogenetic Species Con-
In this paper I shall be concerned solely with tradi-
cept (PSC) of Nixon & Wheeler (1990). Characters are
tional morphometrics as used by plant systematists,
qualitative variables that are found in all comparable
and with the descriptive, as opposed to the phyloge-
individuals within a terminal lineage (i.e. species);
netic, component of systematics. The data employed
traits are qualitative variables that are not distri-
buted universally amongst comparable individuals
*E-mail: ahenderson@nybg.org within such a lineage. I return to this topic later.

© 2006 The Linnean Society of London, Botanical Journal of the Linnean Society, 2006, 151, 103–111 103
104 A. HENDERSON

The methods of analysis in traditional morphomet- MOST COMMONLY USED MORPHOMETRIC

rics are primarily multivariate statistics. Although METHODS
the terms morphometrics and multivariate statistics
are sometimes used interchangeably – and some There are two important discussions of morphometric
authors speak of multivariate morphometrics – it is methods as applied to botanical systematics: those of
possible to analyse morphometric data without multi- Pimentel (1981) and Crisp & Weston (1993). More gen-
variate statistical methods (Rohlf, 1990). I give a brief eral reviews are provided by Marcus (1990) and James
review of traditional morphometrics in plant systema- & McCulloch (1990). An excellent and detailed review
tics, and of some of the most commonly used methods of the field of multivariate statistics, from an ecologi-

Downloaded from https://academic.oup.com/botlinnean/article-abstract/151/1/103/2420544 by guest on 13 February 2020

of analysis. I then concentrate on palm systematics cal perspective, is that of Legendre & Legendre (1998).
and discuss the kinds of data used, and give examples The computer program NTSYSpc (Rohlf, 2000) is
of traditional morphometric analyses used to address designed specifically for morphometric analysis of sys-
systematic problems in palms. I conclude by outlining tematic data.
what I take to be a scientific methodology for descrip- Here I discuss morphometric techniques, at least as
tive systematics, and show how morphometrics is an far as systematics is concerned, under three headings,
integral part of this method. depending on the data used and the aims of the study.
(1) Cluster Analysis of qualitative data with no a pri-
ori knowledge of specimen groupings. (2) Ordination of
HISTORICAL OVERVIEW OF TRADITIONAL
quantitative data (sometimes with qualitative data)
MORPHOMETRICS IN PLANT SYSTEMATICS
with no a priori knowledge of specimen groupings. (3)
Some of the earliest papers in plant systematics using Discriminant Analysis of quantitative data with a pri-
morphometrics were associated with the school of ori knowledge of specimen groupings. The first two are
numerical taxonomy. The development of this school exploratory analyses (i.e. they are not necessarily
in the late 1950s and its expanding influence in the inferential but merely look for patterns in the data),
1960s and 1970s (e.g. Sneath & Sokal, 1973) led to an while the third is confirmatory, with inferential anal-
increase in the use of morphometrics and multivariate ysis. First, I discuss some of the assumptions of these
statistics. This coincided with the increased availabil- methods, and then give a brief outline of them as
ity of computers capable of rapid analysis of large applied in plant systematics.
datasets. However, most of the effort in numerical tax- Difficult problems for systematists using multivari-
onomy, and the source of its conflict with other schools ate statistics are the underlying assumptions of the
of systematics, concerned the classification of taxa (i.e. methods, particularly if they are being used inferen-
phylogeny), rather than the descriptive component of tially. The most important requirement for statistical
systematics. The unfortunate legacy of numerical tax- inference is that the sample be random. Although
onomy is that many researchers associate morphomet- some authors insist on this (e.g. Marcus, 1990), sam-
ric methods with the demise of that school of ples of herbarium specimens cannot be random in the
systematics. Similarly, many associate the term phe- strict sense – ‘a sample collected from a population in
netic with numerical taxonomy, and phenetics or such a manner that every individual in the population
numerical phenetics both refer to numerical taxon- has the same probability of being sampled’ (Pimentel,
omy. I prefer therefore not to use the term phenetic. 1979). A second requirement for statistical inference is
In the last 30 years, independently of numerical tax- that data be distributed in a multivariate normal fash-
onomy, there have been numerous studies using mor- ion. This again is a difficult requirement to meet. Mul-
phometrics, approached from a number of different tivariate normality means that each and all variables
perspectives. Commonly, these methods are used in combined are normally distributed (Tabachnik &
the study of species complexes (e.g. Chandler & Crisp, Fidell, 2001). This is not readily tested, although some
1998; Naczi, Reznicek & Ford, 1998; Raulings & Ladi- multivariate procedures are relatively insensitive to
ges, 2001; Gengler-Nowak, 2002) and hybrids (e.g. deviations from normality (Tabachnik & Fidell, 2001).
Murrell, 1994; Campbell & Wright, 1996; Marhold Tabachnik and Fidell give discussions of assumptions
et al., 2002). Recently, researchers have combined for each multivariate procedure.
molecular and morphological data (e.g. Hansen, Elven Generally, authors of systematic papers using mul-
& Brochman, 2000; Lihová et al., 2004). In most cases, tivariate statistical methods have ignored the require-
the methods used in these studies have solved ments of random sampling and other assumptions.
problems that proved intractable using the methods Perhaps the problem is solved by using only noninfer-
of traditional herbarium systematics. Traditional ential methods. Pielou (1984), for example, makes the
monographs and revisions, published in such series as distinction between interpreting the data at hand
Flora Neotropica and Systematic Botany Monographs , using cluster analysis and ordination, and using mul-
very seldom employ morphometric methods. tivariate statistics based on inference. She considers

© 2006 The Linnean Society of London, Botanical Journal of the Linnean Society, 2006, 151, 103–111
 MORPHOMETRICS IN PALM SYSTEMATICS 105

the latter activity to be a separate field of enquiry (see tions of the original variables. Rotation of axes is rigid,
also James & McCulloch, 1990). On the other hand, so that the data points retain their positions relative
Tabachnik & Fidell (2001) place much less emphasis to one another. Most of the variance is usually sum-
on the importance of random sampling. They write, marized by the first few components, and PCA thus
‘Use of inferential and descriptive statistics is rarely reduces a larger number of variables to fewer vari-
an either-or proposition’. ables, which are often easier to interpret. PCA is thus
described as a dimension reducing method. Scores of
each specimen on the principal components, usually
CLUSTER ANALYSIS the first two, can be plotted on bivariate scattergrams,

Downloaded from https://academic.oup.com/botlinnean/article-abstract/151/1/103/2420544 by guest on 13 February 2020

Cluster Analysis (CA) is an exploratory tool for classi- allowing visualization of the relative positions of the
fying objects. It is an example of a Q-mode type of specimens. Some authors (e.g. James & McCulloch,
analysis, in which the association amongst objects 1990) consider that PCA should not be used for mul-
(specimens) is being assessed (Legendre & Legendre, tiple samples – a restriction which would entirely
1998). There are no statistical assumptions about the eliminate its usefulness in systematics. This opinion is
data. The most common procedure is for similar not shared by others (e.g. Humphries et al., 1981).
objects to be placed in groups and these in more inclu- Examples of the use of PCA in plant systematics are
sive groups, in a hierarchical manner. Results are usu- provided by Boyd (2002), Chandler & Crisp (1998),
ally presented in the form of trees or dendrograms. Crisp & Weston (1993) and Naczi et al. (1998).
The greatest development of CA in systematics was If a dataset contains both quantitative and qualita-
associated with the school of numerical taxonomy (e.g. tive variables, and is thus unsuitable for analysis with
Sneath & Sokal, 1973), and there are numerous dif- coefficients of dependence as in PCA, Principal Coor-
ferent procedures available. dinates Analysis (PCoA) can be used. PCoA begins
The two preliminary decisions in CA concern the with any kind of distance matrix amongst specimens
choice of an association coefficient and of a clustering (or a similarity matrix transformed to a distance
algorithm. Association matrices for this kind of anal- matrix) and is thus a Q-mode type of analysis. The dis-
ysis are produced either by similarity or distance coef- tance matrix is often based on Gower’s coefficient
ficients. The simplest similarity coefficient is the (Legendre & Legendre, 1998). PCoA extracts eigenval-
simple matching coefficient. This can be used only for ues and eigenvectors from the distance matrix. Dis-
qualitative data (binary or multistate) and CA works tances amongst specimens are maximized on the first
better with these kinds of data (Crisp & Weston, 1993; principal coordinate, and then on the second coordi-
James & McCulloch, 1990). There are many different nate, and so on. PCoA thus provides a geometrical rep-
clustering algorithms available. The most commonly resentation of the distances amongst specimens, and it
used algorithm in systematics is the hierarchical, is the visualization of specimens in two- or three-
agglomerative algorithm using averages – UPGMA dimensions that is its main purpose.
(unweighted pair-group method using arithmetic
averages). Examples of the use of CA in plant system-
atics are provided by Binns, Baum & Arnesen (2002), DISCRIMINANT ANALYSIS
Crisp & Weston (1993), Gengler-Nowak (2002). Also known as Discriminant Function Analysis or
In contrast to cluster analysis, R-mode analysis Canonical Variate Analysis, Discriminant Analysis
assesses the association amongst variables, and (DA) comprises a group of methods rather than a sin-
begins with matrices produced by coefficients of gle procedure (Pimentel, 1979). The kinds of data used
dependence amongst variables. Principal Component are the same as for PCA, quantitative variables, but
Analysis (PCA) is an R-mode type of analysis. the specimens are identified a priori to groups (i.e.
taxa). The first stage in DA is usually a multivariate
analysis of variance (MANOVA), which tests the
PRINCIPAL COMPONENT ANALYSIS hypothesis that group centroids are the same. If the
PCA is usually used as an exploratory tool in system- MANOVA supports the alternate hypothesis, that
atics. It is a method for rotating the axes of a coordi- group centroids are significantly different, then DA
nate system such that the first axis (first principal proceeds with two operations – classification and dis-
component) passes through the greatest dimension of crimination. Some authors (e.g. James & McCulloch,
the swarm of data points, and thus accounts for the 1990) regard the initial, inferential stage, MANOVA,
greatest amount of variance of any possible axis. The as a separate procedure, so that DA itself becomes a
second principal component, orthogonal to the first, noninferential method.
accounts for the greatest amount of residual variance, For classification, DA produces classification, or
and so on. There are as many components as original identification functions that are used to determine to
variables, and these components are linear combina- which group specimens belong. Results of this may be

© 2006 The Linnean Society of London, Botanical Journal of the Linnean Society, 2006, 151, 103–111
106 A. HENDERSON

presented in the form of a classification matrix in never been collected for any plants, let alone palms
which each specimen is classified according to the clas- (Henderson, 1995). The consequences of uneven collec-
sification functions, either correctly according to the tion density are that distribution gaps may be artefac-
original grouping, or into another group. The percent- tual, leading to erroneous conclusions.
age of correct classifications is given and this gives an Secondly, palms are woody plants and are often
indication of the validity of the original grouping. large and/or spiny and do not fit into the general col-
These functions can also be used for identification, and lector’s main purpose – to collect as many specimens
can predict group membership of an unidentified spec- as possible. Consequently there are relatively few
imen (assuming it belongs to one of the groups). palm collections in herbaria, and those that do exist

Downloaded from https://academic.oup.com/botlinnean/article-abstract/151/1/103/2420544 by guest on 13 February 2020

For discrimination, DA finds discriminant functions are often fragmentary or incomplete. Poorly collected,
(i.e. linear combinations of variables) that best pressed and mounted specimens are often useless as
discriminate amongst the predefined groups, by data sources. For example, collectors routinely cut off
maximizing the differences amongst groups while the base of the inflorescence, making it impossible to
minimizing variation within groups. Discriminant see the important variables of peduncle length and
functions can be used to assess the relative impor- bract number or type of insertion. Leaves are often
tance of the original variables for discriminating folded and pressed in such a way that it is impossible
amongst groups. to score such variables as number, angles, lengths and
DA is well suited to systematics. The first stage widths of leaflets. In general, the larger the size of the
(MANOVA) may not seem at first sight to be of much palm, the more incomplete the specimen. The conse-
interest, because the groups have been identified a quences of poor specimens are missing data. These are
priori on some grounds. However, MANOVA of groups the bane of multivariate analysis.
which have been delimited by some other dataset, e.g. Even when complete organs are present on the spec-
qualitative data, can be tested by DA using an inde- imen, there may still be problems in their measure-
pendent dataset in the form of quantitative data. If ment. Many organs are distorted as a result of
these groups represent species, then the initial pressing and drying. Some organs continue to change
MANOVA of DA is a test of species as hypotheses (see in size as they age. For example, petioles continue to
later). DA can be used to show which variables are the elongate after the leaf opens (Chazdon, 1991) and
most discriminatory, to classify specimens, and to rachillae thicken as the fruits develop. Such variables
identify unknown specimens. Examples of the use of should be used with some caution. The extent of onto-
DA in plant systematics are provided by Battaglia & genetic allometry in palms (at least beyond the juve-
Patterson (2001), Binns et al. (2002), Boyd (2002) and nile stage), and the extent to which this may confound
Murrell (1994). systematic studies is not yet clear. Do younger or
shorter adult palms have smaller organs than older or
taller adults of the same species? Nor have the poten-
TRADITIONAL MORPHOMETRICS IN PALM
tial size differences between sexes of dioecious species
SYSTEMATICS
been investigated. The various methods to remove the
In this section I first review the kinds of data used in effects of size in morphometric studies have been
palm morphometrics, and some of the problems asso- reviewed by Humphries et al. (1981), but these have
ciated with their collection. I then review all papers in not been attempted in palm studies. Indeed, some
which traditional morphometrics has been used in would be difficult to apply because we have no mea-
palm systematics. sure of overall size (e.g. weight, volume or length) of
palms, nor a reasonable surrogate.
Despite these problems, one can usually recover a
SOURCES OF DATA reasonable amount of data for analytical purposes
The data used in traditional morphometric studies of using simple tools such as a protractor, digital cali-
palms are morphological or anatomical, taken from pers and a ruler. Typically, 20–40 variables can be
herbarium specimens. Less often data are taken measured or counted from herbarium specimens, and
directly from living plants. There are several problems there may be several hundred specimens in the sam-
associated with using specimens as a data source. ple. Some variables can be taken from specimen
First, palms are tropical plants, and there are still labels, such as stem height or stem diameter,
numerous gaps in the collection coverage of tropical although these are seldom measured accurately in
countries. Even in small, relatively well-collected the field.
countries such as Panama, there is uneven coverage. The use of leaf anatomical data circumvents the
These kinds of gaps and the problems they cause are problem of missing data, because virtually all speci-
magnified many times in countries such as Brazil, mens have leaves. The drawback is that it requires
where enormous areas of the Amazon region have much more time and labour to make anatomical sec-

© 2006 The Linnean Society of London, Botanical Journal of the Linnean Society, 2006, 151, 103–111
 MORPHOMETRICS IN PALM SYSTEMATICS 107

tions and extract the data, and such data have seldom results. One of the problems Bayton encountered was
been used. that leaf sheath spines provided a suite of six qualita-
tive variables, but spines were lacking on some
specimens. Bayton discussed the problem of scoring
SYSTEMATIC STUDIES OF PALMS USING such inapplicable variables. However, Bayton’s use of
MORPHOMETRICS Strong & Lipscomb’s (1999) coding regimes is not
The first morphometric study of a group of palms was appropriate. Strong and Lipscomb’s study is orien-
that of Madulid (1981). This concerned variation in tated to phylogenetic analysis and variable coding,
leaf morphology in Calamus javensis Blume, and the using such programs as PAUP and HENNIG86.

Downloaded from https://academic.oup.com/botlinnean/article-abstract/151/1/103/2420544 by guest on 13 February 2020

purpose was to find morphologically based groups Clearly, in a morphometric study these scoring con-
within this complex species. Madulid scored 125 spec- ventions and programs are not applicable. A better
imens for 35 leaf variables. Almost half of these were approach to the problem of inapplicable variables is to
quantitative but were scored as multistate qualitative treat them in terms of characters and traits. Based on
variables. Madulid used CA and PCA to analyse the Bayton’s results, leaf sheath armature seems more
data, but found few discernible groupings. likely to be a trait, rather than a character, and thus
Nauman & Sanders (1991) carried out a study of the should not be used to delimit species. I return to this
classification of species of Coccothrinax. Based prima- topic later.
rily on a literature survey, rather than on specimens, In four of the six studies discussed above, the aim
they found 22 variables. All of these, including quan- was to classify species (Nauman & Sanders, 1991;
titative ones, were scored as qualitative (binary or Mogea, 1999; Pinheiro, 1997; Rustiami, 1999), and in
multistate). Nauman and Sanders used PCoA and CA the other two the purpose was to delimit taxa within a
to analyse the data, and found three main clusters of species complex (Madulid, 1981; Bayton, 2001). How-
species. ever, all six studies were infused with the philosophy
Pinheiro (1997) studied generic boundaries in the of numerical taxonomy and all suffer because of that
subtribe Attaleinae. He used 87 qualitative and quan- to a greater or lesser extent. Procedures that were
titative, leaf anatomical variables from 75 herbarium commonly used in numerical taxonomy would not be
specimens, representing 30 species. Also included used today. For example, quantitative variables were
were a number of putative hybrids. He used PCA and scored as qualitative, qualitative variables were used
CA to analyse the data. Although Pinheiro found sev- in PCA, and CA was used to classify species.
eral reasonably well-defined groups of species, he con- Borchsenius’s (1999) paper was something of a turn-
sidered that extensive hybridization made generic ing point in palm systematics, because it was the first
resolution difficult. However, the observed variation to apply morphometric methods free from the influ-
appears typical of closely related genera and may not ence of numerical systematics. Borchsenius studied
necessarily be indicative of hybridization. Pinheiro’s variation within the Geonoma cuneata H. Wendl. ex
study is remarkable in that he was the first to use leaf Spruce species complex. He measured 12 quantitative
anatomy and was able to score many variables and variables from stems and leaves from 105 individual
produce a dataset free of missing data. living plants at one site in western Ecuador. At this
Rustiami (1999) analysed species classification in site he found four different varieties of G. cuneata.
section Piptospatha of Daemonorops. She used 58 Using PCA and DA, he could clearly separate the four
specimens to represent 11 previously recognized spe- varieties. However, differences between the varieties
cies. She scored 39 vegetative and reproductive vari- broke down when he included plants from other sites
ables, both qualitative and quantitative. She used CA in western Ecuador in the analysis. Borchsenius con-
and PCA to analyse the data, and the result was a cluded that the current varietal classification was not
reduction in the number of species in the section from applicable in western Ecuador, much less throughout
11 to five. the total range of G. cuneata, from western Ecuador to
Mogea (1999) scored 51 morphological, qualitative Nicaragua.
variables from 22 species of Arenga. He used CA to Another species complex in Geonoma, the Amazo-
produce a classification of species, and this was con- nian G. stricta (Poit.) Kunth, was studied by Hender-
sidered to represent a phylogeny of the genus. At an son & Martins (2002). They measured 19 quantitative
arbitrarily chosen level of similarity, four groups of variables from 238 herbarium specimens and analy-
species were recognized. sed the data with PCA, CA and DA. Although their
Bayton (2001) analysed the Calamus hollrungii approach was different from that of Borchsenius
Becc. species complex. Using 40 vegetative variables (1999), i.e. they used herbarium specimens as a data
scored from 137 specimens, he tested an informal clas- source across the range of the species, as opposed to
sification that recognized seven species within the Borchsenius who used living plants from a limited
complex. Bayton used PCoA, but got inconclusive part of the range, their conclusions were similar. The

© 2006 The Linnean Society of London, Botanical Journal of the Linnean Society, 2006, 151, 103–111
108 A. HENDERSON

current varietal classification of G. stricta was unreal- absence of systematic conclusion can be considered a
istic, and one was probably not possible based on the shortcoming of these studies. In Henderson (2004) and
data and methods employed. The main problem was Henderson (2005a), multivariate analysis of specimen
found to be in variation in leaf size and shape. data was combined with delimitation of taxa.
Loo et al. (2001) carried out an analysis of the vari- In Hyospathe (Henderson, 2004), 31 variables from
ation within Licuala glabra Griff. in Peninsular 428 specimens were scored. CA was used to divide
Malaysia. They studied 74 herbarium specimens and qualitative variables into characters or traits. Charac-
scored 43 vegetative and reproductive variables, both ters only were used to delimit species, and six species
qualitative and quantitative. They used PCoA with were recognized, based on groups of specimens with

Downloaded from https://academic.oup.com/botlinnean/article-abstract/151/1/103/2420544 by guest on 13 February 2020

qualitative and quantitative variables, and PCA with unique combinations of character states. This is an
quantitative variables. Results showed three clear application of the Phylogenetic Species Concept (PSC;
groupings of specimens, and these corresponded with Nixon & Wheeler, 1990). Variation within each species
certain mountain ranges in the region. Like that of was then analysed using PCA and DA. One species
Borchsenius (1999), this paper is important in palm was found to be widespread and complex, and was
systematics because of its use of appropriate method- divided into six subspecies. Henderson hypothesized
ology and conclusive results. Loo et al. (1999) used the existence of a hybrid zone between two species
molecular data to analyse three populations of along eastern Andean slopes.
L. glabra, and used both CA and PCA. Henderson (2005a) analysed variation within
Kahn & Gluchy (2002) measured 4393 pistillate Calyptrogyne, for which 35 variables, both qualitative
flowers from 135 living plants of Astrocaryum and quantitative, were scored from 563 specimens.
urostachys Burret in Amazonian Ecuador. Using CA, PCA and DA were used to analyse the data. As a
univariate statistics, they found that pistillate result of the analysis, the number of recognized spe-
flower morphology varied quantitatively but not cies increased from eight to 18, these with 13 subspe-
qualitatively. cies. Regression showed evidence of clinal variation of
In 2002, Henderson and coworkers began a series of some variables with longitude, and there was also evi-
morphometric studies on palm genera, mostly from dence of the existence of a hybrid zone between two
Mesoamerica. Henderson & Ferreira (2002) studied species in Costa Rica.
variation between and within the two species of Syn- The four papers on Synechanthus, Reinhardtia,
echanthus. They used 23 qualitative and quantitative Calyptrogyne and Hyospathe show the development of
variables from 355 herbarium specimens, and ana- my own thinking on methods in herbarium systemat-
lysed the data with CA, PCA and DA. They found that ics. A shortcoming of the Synechanthus and Reinhard-
qualitative but not quantitative variables clearly sep- tia papers is that they have no systematic basis: they
arated the two species. Within S. warscewiczianus H. are merely analyses of morphological variation. In the
Wendl., there were populations of very small plants Calyptrogyne study there is a systematic basis, but
from isolated mountains in central Panama. Within characters are not distinguished from traits, and spec-
S. fibrosus (H. Wendl.) H. Wendl., there were three imen groups were delimited before a species concept
separate populations, one in Mexico, a second in Bel- was applied. In Hyospathe, qualitative variables were
ize, Guatemala, Honduras and Nicaragua, and a third divided into characters and traits, and the PSC
in Costa Rica. PCA showed that S. warscewiczianus applied.
varied mostly in size, and regression showed that
some of this variation was associated with elevation.
In contrast, S. fibrosus varied more in shape and CONCLUSIONS FROM RECENT STUDIES
showed no association with elevation. A review of these recent papers using morphometric
Henderson (2002) analysed 22 qualitative and methods on palms leads to several conclusions. These
quantitative variables from 476 specimens of Rein- derive from the complexity of variation uncovered by
hardtia, using CA, PCA and DA. In marked contrast to morphometric analysis, a complexity that is not
other studies of palms, PCA of quantitative variables revealed, or is glossed over, in traditional systematic
clearly separated the six species. Using regression, studies.
Henderson found clinal variation in some variables
with elevation. One species, R. gracilis (H. Wendl.) 1 Hybrid zones may be common phenomena in palms.
Burret, exhibited considerable variation and seven Henderson (2004, 2005a) postulated the existence of
different populations were distinguished. zones of intermediacy between two species in both
The two studies discussed above, on Synechanthus Hyospathe and Calyptrogyne. In studying Geonoma,
and Reinhardtia, lacked any systematic content in I have found evidence of a hybrid zone between
that they only discussed variation discernible by G. deversa (Poit.) Kunth and G. leptospadix Trail.
morphometric analysis of data from specimens. The Hybrid zones will have important implications

© 2006 The Linnean Society of London, Botanical Journal of the Linnean Society, 2006, 151, 103–111
 MORPHOMETRICS IN PALM SYSTEMATICS 109

for systematics, but will need to be investigated diversity. Recently I have put forward what I
further using ecological and genetic data and consider to be a scientific method for herbarium
methodologies. systematics, at least its descriptive component
2 Subspecific variation in many species of palms is (Henderson, 2005b). This method involves several
greater than previously suspected, and is seldom separate yet sequential stages, two of which involve
adequately documented by traditional methods. morphometric methods. I give a brief review of this
Species such as Synechanthus warscewiczianus and method here.
Reinhardtia gracilis, as well as many species of
The first stage in a systematic study is to choose a
Geonoma, are widespread and exhibit marked
species concept (Henderson, 2005b). This should have

Downloaded from https://academic.oup.com/botlinnean/article-abstract/151/1/103/2420544 by guest on 13 February 2020

morphological variation, and this often correlates
an appropriate theoretical background and an appli-
with geographical disjunction. In such situations,
cable discovery operation – and these two parts of the
subspecies can be recognized. I think it is useful to
concept are closely linked. Herbarium systematists
distinguish here between these kind of variable,
are therefore constrained in their choice of concept. By
or polytypic, species and polymorphic, or ochlo-,
taking into account theoretical background, discovery
species (Cronk, 1998). Variation in the former can
operation and data (i.e. primarily morphological data
readily be understood using morphometric methods.
taken from specimens), I consider that the PSC is the
Variation in the latter is not so easily resolved.
most appropriate for herbarium systematics. Phyloge-
Ochlospecies exhibit great variation that does not
netic species are: ‘the smallest aggregation of popula-
correlate with geography, and are not amenable to
tions . . . diagnosable by a unique combination of
systematic treatment (Cronk, 1998). There are sev-
character states in comparable individuals’ (Nixon &
eral examples from Geonoma, where ochlospecies
Wheeler, 1990).
have been termed species complexes (G. cuneata:
The next stage is to assemble a sample of specimens
Borchsenius, 1999; G. stricta: Henderson & Mar-
and to state an initial hypothesis: that a certain
tins, 2002). However, I think that ochlospecies may
number of groups are present in the sample. Data are
be relatively uncommon in palms; most species com-
gathered from the specimens and the matrix is
plexes are polytypic and may be understood using
constructed. Qualitative attributes are divided into
morphometric analysis. I have found examples of
characters and traits using either Population Aggre-
polytypic species in Geonoma, e.g. G. interrupta
gation Analysis (Davis & Nixon, 1992) if applicable, or
(Ruiz & Pav.) Mart.and G. maxima (Poit.) Kunth.
CA and the sequential removal of suspected traits
3 Morphometric studies suggest the number of spe-
from the analysis (e.g. Henderson, 2004). Characters
cies of palms may be underestimated. The number
alone are used to produce groups of specimens with
of species recognized in Hyospathe has increased
unique combinations of character states. At this stage
from two to six, these with six subspecies (Hender-
the initial hypothesis may be accepted or rejected, and
son, 2004). In Calyptrogyne the number has
in the latter case a new hypothesis proposed. This may
increased from eight to 18, these with 13 subspecies
be in the form of a null hypothesis – for example, that
(Henderson, 2005a). Similar increases are found in
group centroids are different – and such a hypothesis
Geonoma (A. Henderson, unpubl. data). I estimate,
can be tested inferentially using DA of quantitative
from this admittedly small sample, that there may
data. The PSC is applied to these groups. Quantitative
be double the currently accepted number of 2300
and geographical variation within species may then be
species of palms.
analysed and subspecies recognized. It is this stage at
4 Following directly from this, our knowledge of spe-
which morphometrics is most useful. Descriptions of
cies-level variation in general is poor. The founda-
taxa are given in the form of summary statistics of the
tion of our systematic knowledge of palms at all
sample. Ranges, means or medians, confidence inter-
levels rests on our knowledge of species, and yet we
vals or coefficients of variation, and sample size, taken
have hardly scratched the surface when it comes to
directly from the data matrix, contain informative,
understanding morphological variation in palms. I
unambiguous statements about the sample. For
consider the main impediment to understanding
qualitative data in descriptions, all that is needed
species-level variation is not so much the data
are the relevant states of the characters. Finally,
source, i.e. specimens, but rather flawed methodol-
names are applied to species according to the rules of
ogy. Although there are limitations on the knowl-
nomenclature.
edge that can be gained from herbarium specimens,
it is apparent that when we use morphometric
methods we recover far more information than we
FUTURE PROSPECTS
do with traditional methods. Analysis of this infor-
mation, combined with a scientific systematic Considering the total number of systematic papers on
method, gives a much improved estimate of species palms published over the last 30 years, the role of mor-

© 2006 The Linnean Society of London, Botanical Journal of the Linnean Society, 2006, 151, 103–111
110 A. HENDERSON

phometric analysis has been minor. I consider this is Binns SE, Baum BR, Arnason JT. 2002. A taxonomic
due in part to the legacy of numerical taxonomy. As revision of Echinacea (Asteraceae: Heliantheae). Systematic
shown above, the few palm papers based on this philo- Botany 27: 610–632.
sophical background have not been overly successful. Borchsenius F. 1999. Ecology and systematics of the
Nevertheless, traditional systematics has not, in my Geonoma cuneata complex. Memoirs of the New York Botan-
opinion, been entirely successful either. Palm system- ical Garden 83: 131–139.
atists have, for the most part, ignored the problems of Borchsenius F, Skov F. 1997. Ecological amplitudes of Ecua-
dorian palms. Palms 41: 179–183.
species concepts and species delimitation, and pro-
Boyd A. 2002. Morphological analysis of sky island popula-
duced monographs and revisions without any recourse
tions of Macromeria viridiflora (Boraginaceae). Systematic

Downloaded from https://academic.oup.com/botlinnean/article-abstract/151/1/103/2420544 by guest on 13 February 2020

to a scientific methodology. The result has been wildly
Botany 27: 116–126.
different estimates of the numbers of species of palms
Campbell CS, Wright WA. 1996. Apomixis, hybridization,
and an unstable systematics that changes with every
and taxonomic complexity in eastern North American
revision. Traditional herbarium systematics of palms Amelanchier (Rosaceae). Folia Geobotanica Phytotaxonomica
is untenable because of the lack of stated species con- 31: 345–354.
cepts and of a scientific methodology. Recent system- Chandler GT, Crisp MD. 1998. Morphometric and phylo-
atic revisions employing an explicit methodology give genetic analysis of the Daviesia ulicifolia complex
a scientific estimate of species diversity. (Fabaceae, Mirbelieae). Plant Systematics and Evolution
I think that if we can move toward a more scientific 209: 93–122.
systematics of palms, as outlined above, incorporating Chazdon RL. 1991. Plant size and form in the understory
morphometric methods, then we will have a more sta- palm genus Geonoma: are species variations on a theme?
ble systematics. We will also have a sound basis on American Journal of Botany 78: 680–694.
which to study other problems, such as subspecific Crisp MD, Weston PH. 1993. Geographic and ontogenetic
variation, hybrids and hybrid zones, species com- variation in morphology of Australian Waratahs (Telopea:
plexes and biogeographical patterns. Furthermore, we Proteaceae). Systematic Biology 42: 46–76.
should complete descriptive systematic studies before Cronk Q. 1998. The ochlospecies concept. In: Huxley CR, Lock
we attempt phylogenetic analysis, although these JM, Cutler DF, eds. Chorology, taxonomy and ecology of the
things are often carried out the wrong way round floras of Africa and Madagascar. Kew: Royal Botanic Gar-
(Wheeler, 2004). Geometric morphometric methods dens, 155–170.
should also be applied to systematic problems in Davis JI, Nixon KC. 1992. Populations, genetic variation,
and the delimitation of phylogenetic species. Systematic
palms. Geographic Information Systems (GIS) tech-
Biology 41: 421–435.
nology will become increasingly important in palm
Gengler-Nowak K. 2002. Phenetic analyses of morphological
systematics, and this will be especially informative
traits in the Malesherbia humilis complex (Malesherbi-
with morphometric datasets. We will begin to under-
aceae). Taxon 51: 281–293.
stand biogeographical patterns and the association
Hansen KT, Elven R, Brochmann C. 2000. Molecules and
between morphological variables and environmental morphology in concert: tests of some hypotheses in Arctic
variables. A group of researchers at the University of Potentilla (Rosaceae). American Journal of Botany 87: 1466–
Aarhus is already active in this field (e.g. Borchsenius 1479.
& Skov, 1997; Skov & Borchsenius, 1997). Henderson AJ. 1995. The palms of the Amazon. New York:
Oxford University Press.
ACKNOWLEDGEMENTS Henderson AJ. 2002. Phenetic and phylogenetic analysis of
Reinhardtia (Palmae). American Journal of Botany 89:
I thank the organizers of this conference for the oppor- 1491–1502.
tunity to present this paper. Henderson AJ. 2004. A multivariate analysis of Hyospathe
(Palmae). American Journal of Botany 91: 953–965.
Henderson AJ. 2005a. A multivariate study of Calyptrogyne
REFERENCES
(Palmae). Systematic Botany 30: 60–83.
Adams DC, Rohlf FJ, Slice DE. 2004. Geometric morpho- Henderson AJ. 2005b. The methods of herbarium taxonomy.
metrics: ten years of progress following the ‘revolution’. Ital- Systematic Botany 30: 453–456.
ian Journal of Zoology 71: 5–16. Henderson AJ, Ferreira E. 2002. A morphometric study of
Battaglia RE, Patterson R. 2001. A morphometric analy- Synechanthus (Palmae). Systematic Botany 27: 693–702.
sis of the Leptosiphon androsaceus complex (Polemoni- Henderson AJ, Martins R. 2002. Classification of specimens
aceae) in the central and south coast ranges. Madroño 49: in the Geonoma stricta (Palmae) complex: the problem of leaf
62–78. size and shape. Brittonia 54: 202–212.
Bayton RP. 2001. A multivariate study of the Calamus holl- Humphries JM, Bookstein FL, Chernoff B, Smith GR,
rungii Becc. complex (Palmae) in New Guinea. MSc Thesis, Elder RL, Poss SG. 1981. Multivariate discrimination by
University of Reading. shape in relation to size. Systematic Zoology 30: 291–308.

© 2006 The Linnean Society of London, Botanical Journal of the Linnean Society, 2006, 151, 103–111
 MORPHOMETRICS IN PALM SYSTEMATICS 111

James FC, McCulloch CE. 1990. Multivariate analysis in studies in Coccothrinax (Palmae: Coryphoideae). Selbyana
ecology and systematics: panacea or Pandora’s box? Annual 12: 91–101.
Review of Ecology and Systematics 21: 129–166. Nixon KC, Wheeler QD. 1990. An amplification of the phy-
Jensen RJ. 2003. The conundrum of morphometrics. Taxon logenetic species concept. Cladistics 6: 211–223.
52: 663–671. Pielou EC. 1984. The Interpretation of Ecological Data. New
Kahn F, Gluchy D. 2002. Variation in morphology of the York: John Wiley & Sons.
pistillate flower of Astrocaryum urostachys (Palmae) in Pimentel RA. 1979. Morphometrics. The multivariate analy-
Amazonian Ecuador. Nordic Journal of Botany 22: 353– sis of biology data. Dubuque, IA: Kendall/Hunt.
360. Pimentel RA. 1981. A comparative study of data and ordina-
Legendre P, Legendre L. 1998. Numerical ecology. Amster- tion techniques based on a hybrid swarm of sand verbenas

Downloaded from https://academic.oup.com/botlinnean/article-abstract/151/1/103/2420544 by guest on 13 February 2020

dam: Elsevier. (Abronia Juss.). Systematic Zoology 30: 250–267.
Lihová J, Marhold K, Tribsch A, Stuessy TF. 2004. Mor- Pinheiro CUB. 1997. Systematic and agro-ecological studies
phometric and AFLP re-evaluation of tetraploid Cardamine in the Attaleinae (Palmae). PhD Thesis, City University of
amara (Brassicaceae) in the Mediterranean. Sysematic Bot- New York.
any 29: 134–146. Raulings EJ, Ladiges PY. 2001. Morphological variation
Loo AHB, Tan HTW, Kumar PP, Saw LG. 1999. Population and speciation in Stylidium graminifolium (Stylidiaceae),
analysis of Licuala glabra var. glabra (Palmae) using RAPD description of S. montanum and reinstatement of S. armeria.
profiling. Annals of Botany 84: 421–427. Australian Systematic Botany 14: 901–935.
Loo AHB, Tan HTW, Kumar PP, Saw LG. 2001. Intraspe- Rohlf JF. 1990. Morphometrics. Annual Review of Ecology
cific variation in Licuala glabra Griff. (Palmae) in Peninsular and Systematics 21: 299–316.
Malaysia – a morphometric analysis. Biological Journal of Rohlf JF. 2000. NTSYSpc. Numerical Systematics and Mul-
the Linnean Society 72: 115–128. tivariate Analysis System, v.2.1. New York: Exeter Software.
Madulid DA. 1981. A morpho-numerical analysis of the leaf Rohlf JF, Marcus LF. 1993. A revolution in morphometrics.
variation in Calamus javensis complex. Sylvatrop Philippine Trends in Ecology and Evolution 8: 129–132.
Forest Research Journal 6: 129–162. Rustiami H. 1999. Phenetic study of Dragon’s Blood species of
Marcus LF. 1990. Traditional morphometrics. In: Rohlf JF, Daemonorops section Piptospatha (family Arecaceae). MSc
Bookstein FL, eds. Proceedings of the Michigan Morphomet- Thesis, Reading University.
rics Workshop. Special Publication no. 2. Ann Arbor, MI: Uni- Skov F, Borchsenius F. 1997. Predicting plant species dis-
versity of Michigan Museum of Zoology, 77–122. tribution patterns using simple climatic parameters: a case
Marhold K, Lihová J, Perny M, Grupe R, Neuffer B. 2002. study of Ecuadorian palms. Ecography 20: 347–355.
Natural hybridization in Cardamine (Brassicaceae) in the Sneath PHA, Sokal RR. 1973. Numerical taxonomy. San
Pyrenees: evidence from morphological and molecular data. Francisco: W.H. Freeman.
Botanical Journal of the Linnean Society 139: 275–294. Stevens PS. 1991. Character states, morphological variation,
Mogea JP. 1999. Relationships and phylogeny of the species and phylogenetic analysis. Systematic Botany 16: 553–583.
of the genus Arenga (Palmae) based on morphology using the Strong EE, Lipscomb D. 1999. Character coding and inap-
polarity method and the NTSYS program. Memoirs of the plicable data. Cladistics 15: 363–371.
New York Botanical Garden 83: 169–177. Tabachnik BG, Fidell LS. 2001. Using multivariate statis-
Murrell ZE. 1994. Dwarf dogwoods: intermediacy and the tics, 4th edn. Boston: Allyn and Bacon.
morphological landscape. Systematic Botany 19: 539–556. Thiele K. 1993. The holy grail of the perfect character: the
Naczi R, Reznicek A, Ford B. 1998. Morphological, geo- cladistic treatment of morphometric data. Cladistics 9:
graphical,and ecological differentiation in the Carex willde- 275–304.
nowii complex (Cyperaceae). American Journal of Botany 85: Wheeler QD. 2004. Taxonomic triage and the poverty of
434–447. phylogeny. Philosophical Transactions of the Royal Society,
Nauman CE, Sanders RW. 1991. Preliminary classificatory London B 359: 571–583.

© 2006 The Linnean Society of London, Botanical Journal of the Linnean Society, 2006, 151, 103–111