Annual Reviews Annual Review of Ecology and Systematics

Continuum Concept, Ordination Methods, and Niche Theory
Author(s): M. P. Austin
Source: Annual Review of Ecology and Systematics, Vol. 16 (1985), pp. 39-61
Published by: Annual Reviews
Stable URL: https://www.jstor.org/stable/2097042
Accessed: 28-01-2019 19:30 UTC
REFERENCES
Linked references are available on JSTOR for this article:
https://www.jstor.org/stable/2097042?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Annual Reviews is collaborating with JSTOR to digitize, preserve and extend access to Annual
Review of Ecology and Systematics
This content downloaded from 132.248.9.8 on Mon, 28 Jan 2019 19:30:01 UTC
All use subject to https://about.jstor.org/terms
Ann. Rev. Ecol. Syst. 1985. 16:39-61
Copyright (? 1985 by Annual Reviews Inc. All rights reserved
CONTINUUM CONCEPT,
ORDINATION METHODS,
AND NICHE THEORY
M. P. Austin
Department of Environmental Biology, Research School of Biological Sciences, Aus-

tralian National University, and Commonwealth Scientific and Industrial Research
Organization, Division of Water and Land Resources, GPO Box 1666, Canberra, ACT
2601, Australia
INTRODUCTION
Major similarities exist between the role of the continuum concept in plant
community ecology and that of niche theory in animal community ecology.
Identical techniques have been used to study each idea. This article reviews the
adequacy of the current use of the continuum concept and associated ordination
techniques and briefly discusses the implications for niche studies.
The concept of vegetation as a continuum with a changing species composi-
tion along environmental gradients arose in antithesis to the community-unit
theory which stated that plant communities are natural units of coevolved
species populations forming homogeneous, discrete, and recognizable units.
Goodall (49) provides a very clear statement of the issues. Two reviews (76,
117) and responses (29) summarize the controversy well. Ecologists now
accept the continuum concept and incorporate it into textbooks (e. g. 68, 81, 97,
118), though questions remain as to its adequacy (e.g. 68).
The need for quantitative procedures for examining vegetation as a con-
tinuum (27, 114) prompted development of ordination techniques. Goodall
(48) introduced the term "ordination" for methods that arrange samples (or
species) in relation to "a multidimensional series." Excellent reviews survey the
history of ordination techniques and the theory and application of current
methods (37, 54, 119). However, problems of interpreting the results of these
methods continue, as well as questions about whether their mathematical
assumptions are compatible with ecological theory (8, 87).
39
0066-4162/85/1120-0039$02.00
40 AUSTIN
Research on niche theory (75) often ignores ordination stud

continua. In fact one recent review (94) does not mention veg
all. Here it will be argued that many ideas associated with the vegetation
continuum concept are complementary and relevant to the concept of niche in
animal studies. Ordination methods -are used in both vegetation analysis and
niche studies, though animal ecologists only recently appear to have become
aware of the limitations of these methods (78, 79, 96). This review draws
attention to problems of vegetation theory, the possibility of experimental
verification of the continuum, and the question of whether the continuum
concept and ordination methods lead to testable hypotheses.
CONTINUUM CONCEPT
A statement of the principle of community continuity is found in Whittak

(1 18) textbook: "The broad overlap and scattered centres of species popula
along a gradient imply that most communities intergrade continuously al
environmental gradients, rather than forming distinct, clearly separated zo
This principle is usually linked by ecologists with Gleason's (45) individual
hypothesis: "Each species is distributed in its own way, according to its own
genetic, physiological, and life-cycle characteristics and its way of relating to
both physical environment and interactions with other species; hence no two
species are alike in distribution" (118).
The individualistic hypothesis and the continuum concept of vegetation
composition are not necessarily linked; vegetation composition may vary
continuously, with nonrandom distributions of species optima and boundaries
(49).
General discussion of the continuum concept provoked considerable confu-
sion because those unfamiliar with the concept have equated position on an
environmental gradient with physical location on a transect. There is no
necessary spatial relationship between sites with similar values on a gradient.
The gradients are the abstract dimensions of an ecological space, where the
relative positions of sites reflect their similar environments or floristic composi-
tion, depending on the ordination method used.
Extensive debate on the continuum versus community concept in the mid-
1960s (29, 30, 76, 117) focused on two central issues: (a) whether the data
collected by supporters of the continuum were from climax (equilibrium) forest
or disturbed sites, and (b) whether the continuum was an artifact of the methods
used. Evidence (68, 119) was drawn from the continuum index approach
developed by the Wisconsin school (18, 19, 26, 27) and the gradient analysis
approach of Whittaker (115, 116, 122). The index approach orders sites using
only their vegetation composition; the gradient approach uses direct measure-
ments of an environmental factor e.g. elevation. The graphical evidence from
CONTINUUM, ORDINATION, AND NICHE 41
these methods showed extensive overlap of species distributions, with no

obvious limits in common. This was taken as support for the continuum
concept, and a general consensus has arisen that accepts the continuum concept
in preference to the community concept.
Current Issues
The modern argument for the existence of the continuum is presented graphical-
ly (118) for a single environmental gradient (Figure la, b). The community-unit
(Figure la) is composed of dominant species, e.g. trees, plus other species
(shrubs, herbs) that through selection have adapted to live in association with
the dominants and with each other. Competition between dominants results in
sharp boundaries between the communities. The individualistic continuum
(Figure lb) has neither sharp boundaries between species nor well-defined
groups of species with similar distributions. In contrast to Whittaker (118),
Gleason in his original paper (45) made almost no mention of competition as
determining species distribution.
Resource partitioning and associated competition, as usually considered in
niche studies (94), would be expected to lead to an even distribution of species
along an environmental gradient (Figure ic). If species belong to different
strata or functional guilds, then each stratum may partition the gradient in-
dependently of the other strata (Figure Id). With more than three independent
strata, such an arrangement would be difficult to distinguish from the in-
dividualistic continuum (Figure lb).
Whittaker examined his data for evidence of environmental niche partition-
ing (120; Gauch & Whittaker 39). He suggested that where species are domi-
nant, few in number, and close ecological equivalents to each other, they may
be regularly distributed along an environmental gradient. He quoted an ex-
ample of a replacement series for coniferous tree species in relation to elevation
on north-facing slopes in the Santa Catalina and Pinaleno Mountains, Arizona.
Where the species departed from even partitioning of the elevational gradient,
they were distinguishable on a second environmental gradient, "topographic
moisture" (120, 122).
The current statement of the continuum concept supported by the "Cornell
school" (37, 118, 119) is set out in a paper published in 1972 (39). Gauch &
Whittaker (39) put forward nine propositions regarding species behavior along
environmental gradients. The propositions, based on an informal graphical
examination of vegetation data, without any statistical analysis, were used in an
algorithm for generating artificial data sets for comparative studies of ordina-
tion methods. The two major propositions are: 1. Species response curves
approximate normal (Gaussian) curves, i.e. symmetric bell-shaped curves
similar in appearance to a normal error probability distribution function, and 2.
42 AUSTIN
(a)
C.)
(C)
Environmental gradient
Figure I Alternative models for vegetation organization along an environmental gradient. a)

community concept b) individualistic continuum c) resource-partitioned continuum d) resource-
partitioned continuum within strata, individualistic relationships between strata.
Modes of minor species' abundance are randomly distributed along an environ-

mental gradient, while modes of major species' abundance may have a more
uniform distribution.
Propositions 3-5 deal with the statistical distributions shown by maximum
importance values, habitat ranges, and the lack of correlation between species
modes, maximum abundance, and habitat range. Proposition 6 states that
competition can be incorporated by scaling importance values to a constant
environmental carrying capacity, while 7-9 are rules for simulation in order to
(a) truncate normal curves at the tails, (b) introduce variation in species
richness, and (c) provide for sampling error (37, 39, 121).
The central assumption concerns the shape of the curve relating species'
abundance to an environmental gradient, i.e. whether it is a bell-shaped
Gaussian curve or is perhaps bimodal. If the curves are not Gaussian then many
of the propositions become inapplicable. The species response curve to an
environmental gradient in the presence of other species is equivalent to the
realized utilization function for habitat or environmental niche (57, 123). Some
evidence from vegetation studies suggests that the realized habitat-niche re-
sponse for one dimension approximates to a so-called Gaussian curve (39,
120). However, skewed and other non-Gaussian curves predominate in pub-
lished empirical studies (5). Bimodal curves occur more frequently than sym-
metric bell-shaped curves, a fact which may reflect authors' decisions to
publish the more unusual curves. The occurrence of bimodal curves, however,
seems well established (5, 8, 26, 37, 81, 115, 116, 122). The variety of
response shapes may be an artifact of studying one-dimensional gradients
projected from the multidimensional ecological hyperspace (15, 94). Data
averaged from different regions of the environmental hyperspace (i.e. con-
founding data from regions both inside and outside the species niche along an
axis not considered in the graph) can produce spurious bimodal curves (15).
Rigorous evidence is currently insufficient to suggest a conclusion on the
relative frequency of shapes of species response curves.
Others have presented similar descriptive models (5, 13, 81, 108, 110), but
none of the models is mechanistic. The continuum concept and associated
graphical models (8, 39, 81) remain solely phenomenological. Alternative
hypotheses about the organization of vegetation in relation to environment (56,
57, 109) present interesting ideas regarding patterns of species richness in
relation to stress and productivity (56), persistence of species due to the
regeneration niche (57), and the role of resource competition in determining
dominance of species (109). Unfortunately the hypotheses do not provide clear
statements about the shape of species response curves or indicate whether the
curves might show continuum- or community type patterns along environmen-
tal gradients.
44 AUSTIN
Present Position
No formal statistical or experimental tests of either the continuum concept or
the individualistic hypothesis have been attempted (8, 37, 39, 48, 54, 119).
Acceptance appears to have been based on accumulating circumstantial evi-
dence provided by ordination methods (15, 26, 39, 76, 90, 91, 117). The
individualistic hypothesis has been qualified by Gauch & Whittaker's (39)
proposition that major species are regularly distributed along environmental
gradients (120).
A continuing source of difficulty in testing these models will be the nature of
the gradient used to examine the species response. Is it appropriate to use a
compositional gradient based on multivariate ordination methods the ecological
meaning of which may be obscure? What credence can be given to an environ-
mental gradient when its relevance to the growth of the species concerned is
unlikely to be causal or even remotely related to the proximate factors (8)? If an
axiomatic definition of an appropriate environmental gradient and the units in
which it is to be measured are provided, then each of the propositions (39) can
potentially be falsified. The continuum concept makes far fewer assumptions
about species behavior than does the community concept, but it requires
additional assumptions about environment.
The continuum concept has played an important part in changing plant
ecologists' perception of vegetation. Modem statements of the concept do not
provide adequate mechanisms to explain the occurrence of a continuum. The
methods quoted to provide evidence for the concept require greater scrutiny.
Higher standards of hypothesis testing, beyond that of graphical interpretation
of one- or two-dimensional gradients are required. Multidimensional analyses
cannot be accepted unless they account for the major proportion of the variation
in species abundance; otherwise there is no guarantee that the dimensions are
relevant to the species populations studied.
ORDINATION METHODS
In plant ecology, ordination is the process of arranging vegetation sampl

species) in relation to one or more ecological gradients, or to abstract axes
may represent such gradients (5, 48, 49). More generally, ordination m
seen as an exploratory data-analysis technique that seeks pattern (trend
clusters or outliers) in a multivariate data set. The aim is to generate hypoth
and/or summarize complex data in fewer dimensions (5, 28, 87).
Two distinct approaches are used. The first is direct gradient analysis
119), a more descriptive term for which is environmental ordination (4). In
vegetation measurements are examined either graphically or mathematica
relation to selected environmental factors-gradients (115), or scalars (70
Second is indirect gradient analysis (1 17, 119), or vegetational ordination (4),

in which mathematical methods are used to express the major underlying
structure of a data matrix of species by sites in a few abstract compositional
dimensions. Vegetational ordination usually requires definition of a distance or
similarity between samples based on their multivariate species composition
(53), and then the application of an eigenvalue analysis or other mathematical
algorithm to the distance matrix (54). Many exploratory multivariate ordination
techniques developed in other disciplines have been utilized by ecologists-
Principal component analysis (PCA) (48, 89); canonical correlation (4);
parametric mapping (85); reciprocal averaging (RA) (59). However, consider-
able difficulties prevent the acceptance of the compositional dimensions pro-
duced by these methods as unequivocal evidence for the continuum concept.
Inappropriate ecological assumptions implicit in the mathematics of the method
may produce an apparent continuum as an artifact of the method (5, 6, 8, 87).
Vegetational Ordination
The appeal of indirect compositional techniques is that they make no assump-
tion about the important environmental gradients. The results from these
techniques should indicate graphically which samples occupy the extremes of
the major underlying gradients determining the vegetation composition and
which samples are intermediate. Trial and error plotting of possible environ-
mental variables or use of more explicit statistical techniques may then indicate
which environmental variables are important and are correlated with the com-
positional factors (54).
EARLY ORDINATION METHODS The continuum index method (27) was soon
superseded by the Bray & Curtis (or polar) ordination technique (18), as this
allowed more than a single compositional gradient to be estimated. Early
interpretation of the Wisconsin ordination results (18, 19, 26, 27) did not
clearly distinguish whether the principal variation was due to environmental
factors or whether it was due to successional trends. It was only later, when a
clearer perception of the methodology was available, that Peet & Loucks (93)
were able to show the relative importance of environment and succession in
determining forest composition. This confounding could contribute to the
appearance of a continuum. Controversy has continued over the original Bray-
Curtis or polar ordination method. Beals (17) provides an extensive review,
arguing that the method is still useful. Earlier work (12) that criticized the use of
the Bray-Curtis coefficient and suggested the use of Euclidean distance instead
was incorrect, but other features of the method (e.g. axis positioning) remain
controversial.
Principal components analysis superseded polar ordination because it made
use of all the information contained in the similarity matrix to determine the
46 AUSTIN
component axes (12, 89). PCA has since become widely used
method in ecology in areas very different from the vegeta
was originally introduced (48, 54, 89). See Green (52) and Gauch (37) for
examples. The components, axes or dimensions, are orthogonal mathematical
constructs. This does not mean they are necessarily ecologically independent
(4, 13). The orientation of the components is sensitive to the distribution of
samples in the multidimensional species space; thus, there is no necessary
ecological interpretation for components (4, 54).
The mathematical methods used in vegetational ordination imply particular
vegetation models concerning the shape of species response curves. This was
first shown by Swan (108) for Bray-Curtis ordination using artificial data, then
for PCA by Noy-Meir & Austin (86). The problem had previously been
anticipated by Goodall (48). PCA specifically assumes a linear model of
relationships measured in terms of covariance or correlation coefficients. If the
species response curves are bell-shaped as indicated in earlier continuum
studies, then a linear model is inapplicable and a one-dimensional gradient as
shown in Figure 1 becomes distorted into a "horse-shoe" in two or more
dimensions (5, 6, 54, p. 265). This has since been confirmed by many workers
(e.g. 35, 39,42, 87, 121). Some workers (1, 36, 84, see also 54) still claim that
PCA is a viable procedure for vegetation studies. This may be appropriate under
special conditions (13, 87, 99), e.g. for short compositional gradients or for
structural or other attributes which are monotonic along gradients. However,
two- or three-dimensional data using bell-shaped species response curves can
produce complex distorted "flasks" (13, 54, 121) in which the underlying
gradient structure is impossible to recognize without prior knowledge. Thus,
although no assumptions are made about the number or environmental nature of
the underlying gradients, the biological implications of the mathematical model
mean that neither the number nor the orientation of these gradients may be
detectable.
Other linear models-principal coordinate analysis (PCoA) (51), canonical
variates analysis (CVA), and canonical correlation analysis (CCA)-suffer
from analogous problems of curvilinear distortion (4, 38, 42, 67, 121) and other
unrealistic statistical assumptions (125). In a monographic treatment of the use
of CCA in ecology, Gittins (44) has argued that ecological information can be
usefully derived from CCA, even though it is a linear method. Animal ecolo-
gists often use PCA and other linear techniques (22, 52, 65, 101, 102, 106)
without discussion of the ecological implications of the linear model (cf
78, 96). There appears to be no justification for using PCA or Bray-Curtis
ordination in preference to the newer, more robust techniques of detrended
correspondence analysis (DCA) or multidimensional scaling (MDS) (cf 37,
52, 54, 80).
CORNELL TECHNIQUES Reciprocal averaging (59, 60) and its recent variant,
detrended correspondence analysis (61, 62), are currently the most popular
ordination methods. It has been claimed that DCA constitutes a near-optimal
ordination method (37). RA is equivalent to a principal coordinates analysis of a
X2 distance matrix where sites are weighted according to their totals (20). The
ecological model equivalent to this mathematical formulation is not clear to the
reviewer. Analysis using simulated vegetation data has shown that RA pro-
duces an "arch" in two dimensions when there is a one-dimensional gradient
(42). Mathematically, a polynomial relationship exists between higher di-
mensions (axes) and the first (42, 60). When there exist more than one major
ecological axis, the spurious polynomial axes interact with the ecological axes
and confound interpretation. This is a major weakness of RA.
DCA (37, 61, 62) uses a complex algorithm to remove the expected arch by
detrending, i.e. adjusting the higher axes' scores to an average of zero within
segments of the lower axes (37). The adjustment makes no allowance for
interaction between a two- or three-dimensional set of gradients (coenoplane)
and the mathematical distortion (32). A compression problem associated with
the arch distortion at the ends of gradients (37) is further allowed for in DCA by
assuming the species abundances have certain normal-distribution-like proper-
ties and adjusting their values accordingly, in effect assuming species appear
and disappear at a constant rate along the gradient. Numerous applications of
DCA are now being published (e.g. 104, 112), but further evaluation appears
needed (80).
MULTI-DIMENSIONAL SCALING Multi-dimensional scaling (MDS) is an

alternative to DCA and RA that many workers have used (2, 6, 21, 35, 37, 41,
54,72,73,74, 87, 88, 98, 99, 100, 113). MDS refers to a group of methods for
deriving a spatial configuration of observations for a specified number of
dimensions, such that the distances between observations correspond, in some
defined sense, to the measured similarities (resemblance measures) between the
observations (21, 98, 99). Various criteria are used to define the correspon-
dence between the ordination distances and the observed similarities. There are
numerous options, many of which have not been explored in an ecological
context (e.g. multidimensional unfolding, (100)). Clymo (21) provides a useful
commentary on many of the issues associated with these techniques. Interest
has concentrated on nonmetric (nonparametric, 21) multidimensional scaling
(NM-MDS) rather than metric MDS, because NM-MDS assumes a less restric-
tive model of the relationship between the similarity measure (proximity
measure, see 21) and the separation along the underlying gradient, i.e. ecologi-
cal distance. It assumes only that there is a monotonic (rank order) relationship
between similarity and ecological distance. A wide range of ecological
48 AUSTIN
models can be accommodated by this weak assumption. Gauch (37, p. 162)

incorrectly states that NM-MDS assumes that species response curves are
monotonic. One weakness of MDS methods is that no objective criterion exists
for determining the number of dimensions needed to summarize the variation.
COMPARATIVE STUDIES Numerous comparative studies of RA and PCA

have been made (e.g. 5, 6, 20, 21, 35, 41, 42, 88), which show RA to be less
sensitive to curvilinear distortion than PCA. RA is sensitive to extreme disjunct
plots, although this can be turned to advantage for regional analysis of vegeta-
tion (92). Comparisons between RA, DCA, and NM-MDS (41) have suggested
that DCA is to be preferred on grounds of computational efficiency. Gauch et al
(41) considered that DCA and RA performed better than either global or local
versions of NM-MDS (21, 98) with simulated Gaussian coenoclines (one-
dimensional gradients). For models with two or more gradients NM-MDS
performed better than RA if the correct dimensionality was specified, though
DCA was judged the best. This preference for DCA has been contested recently
by Minchin (80, see also 21), who found that NM-MDS consistently out-
performed DCA for two-dimensional simulated data sets. More research is
needed on this topic.
Recent assessments of ordination methods have relied on the ability of the
methods to recover gradients from data sets generated by simulation algorithms
based on explicit models of vegetation composition. Most evaluations have
used the "Cornell school" program (39, 40) for generating artificial data sets
(41, 121). The assumptions of these programs have not been tested. Recently
Minchin (80) has undertaken an extensive investigation of the use of artificial
data sets. He identifies other important weaknesses of this approach. First,
simulation studies should examine a reasonable number of randomly generated
data sets in order to evaluate the behavior of a method. Early work (86, 108)
was based on single populations of each type of artificial continuum. More
recent ordination evaluations (41, 42) are vague on how many random sets of
each model were generated. Limited replication can provide misleading con-
clusions (80). Second, the distribution of samples in two- or three-dimensional
space has usually been assumed to be of a simple type, e.g. rectangular (13, 41,
42). Other sample distributions lead to different conclusions regarding the
suitability of methods such as RA, DCA, and NM-MDS (80). Neither DCA or
NM-MDS can be recommended without reservation at the present time.
Vegetational ordination studies depend on correlation. It is not true to say
that "the methodology of correlation is intrinsically without theoretical content
about the real world" (69). The linear correlation as used in PCA ordination
implies a very special and incorrect theoretical model of the real world (78).
DCA and MDS use more complex correlation functions, but they remain
exploratory, hypothesis-generating methods. Few ecologists proceed to an
hypothesis-testing phase. Two studies have used experiments to test ordination

interpretations that soil nitrogen determines limestone grassland composition
(43, 44) and that the differential effect of salt spray on species' competitive
abilities determines variation in cliff top vegetation (46, 47).
While doubt remains about the ecological meaning of the mathematical
models used in ordination, exploratory ordination methods cannot provide
unequivocal evidence for the continuum concept. Two approaches may be
adopted to improve the relevance of ordination methods to vegetation theory in
general, and to the continuum concept in particular. Methods could be de-
veloped that are robust to differences in the underlying models (e. g. 64), or new
vegetation models could be tested and explicit methods based on the models
developed. Either environmental ordination or experimental studies of plant
communities on environmental gradients can be used to test new vegetation
models.
Environmental Ordination
Whittaker (114, 115, 116, 120, 122) pioneered the direct gradient analysis
(DGA) approach to environmental ordination using elevation and topography.
Austin et al (15) discuss the methodology, indicating the close relationship with
Jenny's soil-factor equation (63). Greig-Smith (54: Ch. 9) reviews many of the
possible methods of vegetation/environment correlation. Because of the restric-
tive mathematical assumptions of multivariate statistical methods (PCA, CVA,
CCA) as they may relate to the ecological behavior of organisms, these methods
will not be considered further here (4, 96, 125).
DIRECT GRADIENT ANALYSIS This method is a simple graphical technique

for displaying patterns of species distribution (environmental niche) in relation
to environmental factors presumed to be important. Direct gradient analysis has
remained substantially unchanged since the original work (37, 66, 90, 91, 114,
115, 116, 122). In theory any two or more environmental factors may be used;
in practice DGA involves the sampling of vegetation along a topographic
moisture gradient in a series of altitudinal belts. The position on the topographi-
cal moisture gradient is calculated by arranging samples in each of nine
topographic positions from ravines to exposed dry slopes, assigning weights to
species with modes in different topographic positions, and using the weights to
derive indirect indices of sample position along the gradient. Elements of
circular reasoning show up here: those species are weighted that show clear
patterns, and then the species are used to claim that the topographic moisture
gradient is a major determinant of the vegetation variation. The topographic
moisture gradient is thus an indirect (compositional) rather than a direct (en-
vironmental) gradient (8, 120).
Three aspects of DGA give rise to concern: (a) the lack of an explicit
50 AUSTIN
sampling strategy; (b) the rejection rate of samples (see Figure 2 in 120); and (c)
the absence of statistical analysis. Recent improvements have been made. Peet
(90, 91) tested the correlation between the weighted average gradient and
independent measures of topographic position (e.g. potential incident solar
radiation) and found it to be very high. In addition, he specified his sampling
frame and intensity and so ensured that at least one sampling unit occurred in
every cell-2 10-of a three-dimensional table of elevation, topographic posi-
tion, and stand age. Such an experimental design is not explicit in the original
studies (115, 116, 122). Comments (115, p. 58) may be found suggesting
perhaps 13% of samples were excluded from the graphical analysis. Recent
graphical DGA studies (e.g. 66) have presented only response isolines without
scatter-plots showing the distribution of data points on which they are based. It
is impossible to evaluate such graphs.
Austin et al (15) make the point that environmental variables or gradients are
of various kinds: (a) those which have no necessary physiological influence on
plant growth (indirect gradients such as elevation)-correlation of these with
vegetation patterns is likely to be location specific; (b) those which have a
physiological influence (direct gradients) but are not a resource for plant growth
(e.g. pH), and (c) -those (resource gradients) where the environmental variable
can be directly used as a resource by plants (e.g. available soil nitrate). Clearly a
spectrum of gradient types exists. In principle, DGA should be based on the
most proximal variables that can be measured or estimated. Numerous studies
have devised environmental scalars of increasing relevance to plant growth (70,
1 1 1, 124, 127). Attempts to analyze niche relationships with simple, separate,
and indirect environmental variables like slope and aspect are unlikely to be
profitable, particularly when confounded with differences in location (e.g.
126). More attention to environmental processes for estimating environmental
gradients is necessary.
NEW STATISTICAL APPROACH Problems of experimental design and analy-

sis need not limit the potential of the approach. Recently, co-workers and I (11,
14, 15) have applied the statistical technique of generalized linear models
(GLM) (82) to DGA. GLM provides a less restrictive form of regression and
allows error distributions of the dependent variable other than the normal to be
used. Categorical factors may be easily combined with continuous variates as
predictors in the analysis. We examined the relative importance of four en-
vironmental variables (gradients) in predicting the presence (qualitative niche;
14, 15) or stem density (quantitative niche; 1 1) of eucalypt species. Altitude,
mean annual rainfall, radiation index (measure of aspect and slope), geology,
and the interactions of these were, using GLM, shown to be important, to
different degrees, for individual eucalypt species in southeastern Australia
using GLM (14). We (15) then used continuous variables to examine whether
(a) Mean annual rainfall (mm)

500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000
4 - Radiation index 0-70 - Radiation index 10 - Radiation index 1.12
6-
8-
10 -
12-
16- E rossil E -rossii 0 r oss
4- Radiation index 0r70 R adiation index ( s 1 0 t Radiation index r 112
6-
C 0-05~~
c 12-
c 14-
2 16 ~~E dairympleana -Edairympleana E dairympleana
Figure 2 The qualitative environmental niche (expressed

Eucalyptus species in relation to three environmental factor
mean annual rainfall and radiation index (measure of topog
on sedimentary geologies. The response surface was calculated using GLM. The outline indicates
the distribution of the 1286 samples used in the analysis (reproduced by permission of Dr J. Junk
from Austin et al. (15)).
the qualitative niche of species approximated a bell-shaped response curve.

Quadratic relationships were fitted simultaneously for three continuous en-
vironmental variables-mean annual temperature, mean annual rainfall, and
radiation index-plus a categorical factor, geology. Six eucalypt species were
tested. All showed significant bell-shaped responses to mean annual tempera-
ture (Figure 2). Two species had a quadratic (bell-shaped) response for rainfall,
two had linear responses, and two showed no significant response. Only one
species showed a quadratic response to radiation index, while five species were
significantly related to geology. Another recent application is that of Minchin
(80). A. 0. Nicholls, C. R. Margules, and I (in preparation) have used GLM to
investigate whether species response curves depart from a symmetric, bell-
shaped ideal for direct environmental gradients (distal factors) such as tempera-
ture, rainfall, and radiation. For those species tested, the degree of interaction
between environmental gradients and the degree of skewness of response varied
with species.
Rigorous analysis of the realized niche of species is now possible with GLM,
provided we use our understanding of environmental processes to develop
scalars that are increasingly more relevant and proximal to the causal factors
of vegetation variation (7, 15). Many of the ideas discussed by Whittaker in
his pioneering studies of the vegetation continuum should now be put to the
test.
52 AUSTIN
EXPERIMENTAL APPROACHES TO CONTINUUM

CONCEPT
Experimental studies of continua would provide much stronger tests o

propositions put forward by Gauch & Whittaker (39) than statistical ana
field observations. Continuum experiments require multispecies mixtur
numerous treatment levels to simulate a gradient (cf 24).
The early and much neglected work of Ellenberg (32, 33, see also 3-
translation by Aukland) showed that striking results could be obtained with
simple gradient experiments. Ellenberg examined species response in mono-
culture and multispecies mixture along an environmental gradient of depth to
permanent water table for six grasses. The physiological optimal (monoculture)
yields did not correspond with the ecological optimal (multispecies) mixture
yields, in terms of dry matter. All species had similar physiological optima in
terms of water table depth but the majority of ecological optima were displaced
to either a shallower or a deeper water table. Evidence for bimodal response
curves was obtained (32, 33; see also a reanalysis by Austin, 7). The ecological
optima form a sequence along the gradient and constitute an experimental
demonstration of a vegetational continuum measured as dry matter yield over
one season. Ernst (34) questioned whether competition actually displaced the
species optima. The disagreement depends on the type of standardization of the
yield totals used (9). If the observed carrying capacity at each position along the
gradient is taken into account, then it can be shown that the optima of relative
physiological performance correspond to the relative ecological optima (9), an
interpretation different from that of both Ellenberg and Ernst. Water-table
depth is a complex-factor gradient in Whittaker's terms (120); it can be
regarded as a distal factor that determines the availability of nitrogen, soil
aeration, and water supply, all of which vary along the gradient and act as
proximal factors determining the relative performance of the species (33).
Difficulties in interpretation and design of such experiments exist (7, 9, 10,
32, 33, 34), but these do not explain the relative neglect of such work over the
last 30 years (8). Recent studies (10, 16) have examined the technical problems
associated with multispecies gradient experiments. For example, Ellenberg
(32, 33) and Austin & Austin (10) had no replication but did have 10 and 16
treatment levels respectively. Subsequent work (16) with six thistle species,
using three replicates in each of 12 treatment levels, has confirmed the feasibil-
ity of the experimental design.
Mueller-Dombois & Ellenberg (81) put forward a graphical statement of the
expected types of species response curves based on their experiments and field
observations. Physiological optima and response curves (fundamental niches)
are assumed to be the same (bell-shaped) for different species, while ecological
response curves may vary from bell-shaped, to skewed, to symmetric, or
markedly asymmetric bimodal curves, due to competition. Though compatible

with experimental results, this graphical model differs radically from the usual
assumption of niche theory-that species' fundamental niches, though over-
lapping, usually differ in their optima, and competition tends to restrict the
realized niche to optimum physiological conditions.
Community parameters, dominance, diversity, and total yield also show
distinct response patterns to experimental gradients (8, 10). Ecological per-
formance (realized niche) can be predicted from physiological performance
(fundamental niche) for one five-species and two ten-species communities (9).
The predictive equations showed consistent trends along the nutrient concentra-
tion gradient used. Similar results were obtained with thistles (16); however, it
was necessary to use two species properties determined from monoculture-
vegetative shoot yield and shoot/root ratio-in order to achieve reasonable
prediction of yield in mixture. The generality of these results requires further
testing.
Few similar plant experiments have been done (9, 58), though see Pickett &
Bazzaz (95) and McCormick et al (77). Colwell & Fuentes (24) review a
number of experimental studies of niche with a wide variety of organisms and
gradients. The potential now exists for combining these experimental
approaches to provide a powerful series of tools for the study of vegetation
along environmental gradients.
CONTINUUM STUDIES AND NICHE THEORY
The development of the niche concept has been primarily based on birds (22,
65, 103) and lizards (94, 103), while the continuum concept is based on higher
plants. There has been a common methodology (e.g. use of PCA and CCA),
though few studies recognize the commonality (see though 78, 79, 104).
Zoologists continue to accept PCA. Numerous examples could be cited of
workers who interpret PCA axes as having physical meaning and who display
little appreciation of the problems of curvilinear distortion and independence of
components. As Meents et al (78) point out, few ecological reasons lead us to
expect only linear relationships between organisms and features of their en-
vironment. Nonlinear relationships may also occur between habitat measures.
Relationships between organisms may show very different forms of correlation
depending on their overlap along gradients (54).
One interesting example of the problems is provided by Rotenberry & Wiens
(102), who use PCA to examine the niche distribution of bird species in relation
to habitat at a continental scale in steppe vegetation. They apply PCA to a
correlation matrix between various variables measuring the coverage and the
vertical and horizontal structure of vegetation. The relationships between
variables are likely to be at least monotonic, if not linear. However, variables
54 AUSTIN
such as the percentage cover of grass, herbs, and litter are necessarily logically
correlated, and the number of variables in each of such correlated groups will
determine the relative magnitude of the components. The choice of relative
numbers of attributes will determine also the relative orientation of the com-
ponents (4, 54). Rotenberry & Wiens (102) argue that the use of continuously
distributed standardized environmental variables largely circumvents
nonlinearity problems. This does not mean that PCA provides ecologically
independent components to which bird species may respond (54).
The authors use linear correlation to examine the relationship between bird
species abundance and the vegetation habitat components. Although some
show strong correlations with the first component, many do not, but examina-
tion of a figure in Rotenberry & Wiens (102: Figure 5, p.2261) shows a striking
sequence of bird distributions along the first vegetation component, a sequence
which Rotenberry & Wiens do not discuss. The sequence of species is very
clear from Dickcissel, with a negative linear response, through truncated
bell-shaped responses, to clear symmetric bell-shaped responses (e.g. Lark
bunting and western meadow-lark occupying central positions along the com-
ponent), to Sage sparrow showing a positive response. In this case, PCA has
proved robust, but the use of the linear correlation coefficient appears to have
limited the interpretation. On the basis of Rotenberry & Wiens' figure, the
hypothesis can be advanced that plots of species abundance in the three major
dimensions of the PCA will show considerable niche partitioning. In this
example, it is possible to make suggestions of alternative interpretations of the
results; in many uncited examples it is not. Where the possibility of curvilinear-
ity is recognized, it may be ascribed to less important components of the
problem. For example, Miles & Ricklefs (79) suggest that curvilinearity is due
to being constrained to sum to a constant, while Dunn & Everitt (31) suggest it
is due to use of presence/absence data for species abundance (cf 54, 78, 96).
In fauna studies, workers are beginning to recognize the role of MDS (96',
113), though the claim that NM-MDS had not yet been applied in quantitative
ecology in 1980 (113) highlights the lack of communication between plant and
animal ecologists. The absence of any mechanistic model for the shape of the
utilization function curve for habitat gradients argues strongly that as few
assumptions as possible should be incorporated into any mathematical analysis
of niche.
Plant ecologists use environmental factor complexes or compositional di-
mensions as their gradients, while animal ecologists usually choose habitat or
food. Plant ecologists tend to ignore functional differences between species
(e.g. tree, shrub, and herbs), and hence often adopt Gleason's individualistic
hypothesis; many animal ecologists, by concentrating on guilds, emphasize
resource partitioning and competition. In many continental scale comparisons
of niche structure (22, 94, 102), the changes in habitat gradients (e.g. shrub
heights) are confounded with climatic changes. Temperature has a physiolog-

ical impact on animals as much as on plants, and such distal factors need to be
incorporated into faunal niche studies.
In the current debate on research methods in ecology (see the special issue of
American Naturalist, 1983, Vol. 122, No. 5), possible lessons from the
continuum/community controversy are not considered. No discussion occurs
about the meaning of the terms "community" or "species assemblage" (68).
Strong (107) does give a definition of an ecological community as "groups of
species living closely enough together for the potential of local interaction." An
even more general definition is that of MacArthur (71) "any set of organisms
currently living near each other and about which it is interesting to talk." It is
questionable how useful such definitions are. If a continuum exists, then any
studies of such communities will contain variation that will be correlated with
changes in environment. Hence, the estimation of statistics will be biased
depending on the range of environment encompassed by the definition of
community.
Research on vegetation can contribute alternative ideas to these discussions.
For plants, the hypothesis that species are randomly distributed with respect to
environmental gradients constitutes a null model. The hypothesis that species
are nonrandomly distributed along an environmental gradient can be falsified
by observational analysis of the kind reviewed here. The hypothesis that the
ecological response of a species to a gradient is determined solely by the
physiological properties of that species can be tested by experiments of the
Ellenberg type (32, 33), that can define both the physiological and the ecologi-
cal responses. However, the observation that grasses show similar physiolog-
ical optima (10, 32) raises an important possible distinction between plants and
animals. Many of the resources used by plants are continuous. A gradient of
nitrate-nitrogen supply cannot be equated with a food-particle size gradient.
Plants may adapt to different nitrogen sources but responses to a single source
could be expected to show a limiting factor response rather than a "normal
curve," i.e. the shape of the fundamental niche for plants may be different from
that of animals.
Experiments indicate that extremes for species may be governed by physi-
ological processes, and limits towards more "favorable" environments may be
due to competition or predation (25). The role of competition in extreme as
opposed to mesic environments has been a source of controversy in vegetation
studies. Grime (55, 56) maintains competition is more important in mesic as
compared with "stressed" environments; Newman (83) holds the opposite
view. Patterns of species control factors (e.g. predation or competition) may be
determined by environmental gradients of favorableness and predictability (9,
56). Defining environmental position may indicate the relative importance of
competition, predation, and proximal environmental factors. Students of niche
56 AUSTIN
theory need to incorporate results from plant community ecology, concerning

both methodology and species behavior; otherwise their theories will be in-
complete.
FUTURE PROSPECTS
In a summary of current controversy regarding experimental methods in co

munity ecology, Salt (105) suggests that a number of steps are necessary in
research: observation, modelling (theory), experimentation, field-testing, and
judgment. After a long preoccupation with descriptive observation, plant
community ecology seems poised to address many of the larger issues and to
take further steps along this research sequence.
Observational Analysis
Explicit hypotheses regarding the patterns and shapes of species response to
distal environmental gradients can be derived from the propositions of Gauch &
Whittaker (39) regarding the continuum concept. These can now be tested
statistically with the use of DGA and new statistical procedures such as GLM
(15). The possibility exists that a major environmental variable has been
missed, however. Exploratory techniques, and ordination in particular, will
then be needed to examine whether major axes of variation correspond to DGA
gradients used. Two alternative developments for exploratory ordination exist
at present: to develop (a) methods robust to differences in the underlying
models or (b) methods based on an explicit model of vegetation response.
Controversy exists over whether either DCA or NM-MDS is a sufficiently
robust solution to the first option at present; the second option will not be viable
until we have more conclusive information about species responses.
Theoretical Developments
Vegetation science has no theoretical basis at present, and a satisfactory
synthesis of continuum ideas and niche theory is urgently needed, as are
mechanistic models of how plant species behave along environmental gra-
dients. These models should define a species' fundamental niche and show how
species physiology determines the shape of this niche. Only then will it be
possible to discriminate between models of niche overlap and packing based on
competition, predation, or stochasticity. A possible null model for species is
that the fundamental niche remains unchanged in the presence of other species.
If the controlling influence of competition varies along environmental gra-
dients, more elaborate theories and null models will be required.
Experimental Analysis
In principle, Ellenberg (32, 33) demonstrated an experimental approach to
community ecology incorporating environmental gradients. In practice, the
equation of vegetative biomass with physiological performance and the abili

to reproduce and persist is likely to prove inadequate. Problems of experime
design also remain (9, 16).
Niche Theory
Those plant ecologists whose interests are in community analysis and the
hypotheses associated with the as yet unproven continuum concept are studying
topics very similar to those of animal ecologists interested in niche theory.
Communication and methodological problems are not the only difficulties
inhibiting a synthesis of ideas. Plant ecologists have expressed their hypotheti-
cal patterns in terms of distal factors (e.g. mean annual rainfall) that determine
the broad properties and dynamics of an ecosystem. Niche ecologists have been
more concerned with proximal factors of food, competition, and predation.
Integration of these differing levels of abstraction will be necessary before any
theoretical communication will be possible.
CONCLUSION
In a review of statistical plant ecology, Goodall (50) hoped that explorat

methods would give way to hypothesis-testing; Green (53) expressed s
sentiments when reviewing ecological similarity. The present situation
summarized by Salt (105): "[T]he currently popular response is to subj
data pile to . .. multivariate analysis in the hope that . .. something of
may be sifted out." Goodall (50) does not emphasize the absence of suita
vegetation models but does suggest that "the subject is likely to mark time
its biological aspects resume their rightful place i.e. as master rather tha
to mathematical and statistical considerations". The current preoccupatio
ordination methodology, as opposed to hypothesis-testing of niche or c
tinuum propositions for plants, is not encouraging; neither is the lack
communication between plant and animal ecologists over suitable metho
gy. I have tried to show in this review, however, that the ingredients of
observational technique, theory, and experimental design now exist and make
possible a suitable synthesis.
ACKNOWLEDGMENTS
I thank P. R. Minchin, N. J. McKenzie, D. Potts, I. R. Noble, D. P. Faith, C.

R. Margules, B. Wellington, and M. Strasser for comments on the manuscript
and particularly P. R. Minchin for permission to quote from his unpublished
thesis. I gratefully acknowledge the efforts of N. Parnell and M. L. Lowe in
preparing the manuscript for publication.
58 AUSTIN
Literature Cited
1. Allen, T. F. H., Shugart, H. H. 1983. mental scalars and statistical curve-fitting

Ordination of simulated complex forest procedures. Vegetatio 55:11-27
succession: a new test of ordination 16. Austin, M. P., Groves, R. H., Fresco, L.
methods. Vegetatio 51:141-55 F. M., Kaye, P. E. 1985. Relative
2. Anderson, A. J. B. 1971. Ordination growth of six thistles along a nutrient
methods in ecology. J. Ecol. 59:713-26 gradient with multispecies competition.
3. Aukland, P. 1978. Translation of two of J. Ecol. In press
Heinz Ellenberg's papers. CSIRO Div. 17. Beals, E. W. 1985. Bray-Curtis Ordina-
Land Use Res. Tech. Memo 78/7 tion: an effective strategy for analysis of
4. Austin, M. P. 1968. An ordination study multivariate ecological data. Adv. Ecol.
of a chalk grassland community. J. Ecol. Res. 14:1-55
56:739-57 18. Bray, J. R., Curtis, J. T. 1957. An
5. Austin, M. P. 1976a. On non-linear spe- ordination of the upland forest com-
cies response models in ordination. Vege- munities of southern Wisconsin. Ecol.
tatio 33:33-41 Monogr. 27:325-49
6. Austin, M. P. 1976b. Performance of 19. Brown, R. T., Curtis, J. T. 1952. The
four ordination techniques assuming upland conifer-hardwood forests of
three different non-linear species re- northern Wisconsin. Ecol. Monogr.
sponse models. Vegetatio 33:43-49 22:217-34
7. Austin, M. P. 1979. Current approaches 20. Chardy, P., Glenmarec, M., Laurec, A.
to the non-linearity problems in vegeta- 1976. Application of inertia methods to
tion analysis. In Contemporary Quantita- benthic marine ecology: Practical im-
tive Ecology and Related Ecometrics, ed. plications of the basic options. Estuarine
E. P. Patil, M. Rosenzweig, Stat. Ecol. Coastal Mar. Sci. 4:179-205
No. 12, pp. 197-2 10. Fairland, Md: Int. 21. Clymo, R. S. 1980. Preliminary survey
Co-op. of the peat-bog Hummell Knowe Moss
8. Austin, M. P. 1980. Searching for a mod- using various numerical methods. Vege-
el for use in vegetation analysis. Vegeta- tatio 42:129-48
tio 42:11-21 22. Cody, M. L. 1975. Towards a theory of
9. Austin, M. P. 1982. Use of a relative continental species diversity. See Ref.
physiological performance value in the 23, pp. 214-57
prediction of performance in multispe- 23. Cody, M. L., Diamond, J. M. eds. 1975.
cies mixtures from monoculture perform- Ecology and Evolution of Communities.
ance. J. Ecol. 70:559-70 Cambridge, Mass: Belknap
10. Austin, M. P., Austin, B. 0. 1980. Be- 24. Colwell, R. K., Fuentes, E. R. 1975.
haviour of experimental plant communi- Experimental studies of the niche. Ann.
ties along a nutrient gradient. J. Ecol. Rev. Ecol. Svst. 6:281-310
68:891-918 25. Connell, J. H. 1972. Community in-
11. Austin, M. P., Cunningham, R. B. 1981. teractions on marine rocky intertidal
Observational analysis of environmental shores. Ann. Rev. Ecol. Syst. 3:169-92
gradients. Proc. Ecol. Soc. Aust. 1 1: 26. Curtis, J. T. 1959. The Vegetation of
109-19 Wisconsin: An Ordination of Plant Com-
12. Austin, M. P., Orloci, L. 1966. munities. Madison: Univ. Wis. Press
Geometric models in ecology. II. An 27. Curtis, J. T., McIntosh, R. P. 1951. An
evaluation of some ordination tech- upland forest continuum in the prairie-
niques. J. Ecol. 54:217-27 forest border region of Wisconsin. Ecolo-
13. Austin, M. P., Noy-Meir, I. 1971. The gy 32:476-96
problem of non-linearity in ordination: 28. Dale, M. B. 1975. On objectives of
experiments with two gradient models. J. methods of ordination. Vegetatio 30:15-
Ecol. 59:762-73 32
14. Austin, M. P., Cunningham, R. B., 29. Dansereau, P. 1968. The continuum con-
Good, R. B. 1983. Altitudinal distribu- cept of vegetation: responses. Bot. Rev.
tion in relation to other environmental 34:253-332
factors of several Eucalypt species in 30. Daubenmire, R. 1966. Vegetation:
southern New South Wales. Aust. J. identification of typal communities. Sci-
Ecol. 8:169-80 ence 151:291-398
15. Austin, M. P., Cunningham, R. B., 31. Dunn, G., Everitt, B. S. 1982. An In-
Fleming, P. M. 1984. New approaches to troduction to Mathematical Taxonomy.
direct gradient analysis using environ- Cambridge: Cambridge Univ. Press
CONTINUUM, ORDINATION. AND NICHE 59
32. Ellenberg, H. 1953. Physiologisches und glesey. II. Experimental studies. J. Ecol.
6kologisches Verhalten derselben Pflan- 61:819-30
zenarten. Ber. Dtsch. Bot. Ges. 65:35 1- 48. Goodall, D. W. 1954. Objective methods
62 for the classification of vegetation. III.
33. Ellenberg, H. 1954. Uber einige Fort- An essay in the use of factor analysis.
schritte der kausalen Vegetationskunde. Aust. J. Bot. 2:304-24
Vegetatio 5/6:199-211 49. Goodall, D. W. 1963. The continuum
34. Ernst, W. 1978. Discrepancy between and the individualistic association. Vege-
ecological and physiological optima of tatio 11:297-316
plant species: A re-interpretation. Oecol. 50. Goodall, D. W. 1970. Statistical plant
Plant. 13:175-88 ecology. Ann. Rev. Ecol. Syst. 1:99-
35. Fasham, M. J. R. 1977. A comparison of 124
nonmetric multidimensional scaling, 51. Gower, J. C. 1966. Some distance prop-
principal components and reciprocal erties of latent root and vector methods
averaging for the ordination of simulated used in multivariate analysis. Biometrika
coenoclines and coenoplanes. Ecology 53:325-38
58:551-61 52. Green, R. H. 1979. Sampling Design and
36. Feoli, E., Feoli-Chiapella, L. 1980. Statistical Methods for Environmental
Evaluation of ordination methods Biologists. New York: Wiley
through simulated coenoclines: some 53. Green, R. H. 1980. Multivariate
comments. Vegetatio 42:35-41 approaches in ecology: The assessment
37. Gauch, H. G. Jr. 1982. Multivariate of ecologic similarity. Ann. Rev. Ecol.
Analysis in Community Ecology. Cam- Syst. 11:1-14
bridge: Cambridge Univ. Press 54. Greig-Smith, P. 1983. Quantitative
38. Gauch, H. G., Wentworth, T. R. 1976. Plant Ecology. Oxford: Blackwell. 3rd
Canonical correlation analysis as an ed.
ordination technique. Vegetatio 33:17- 55. Grime, J. P. 1973. Competitive exclu-
22 sion in herbaceous vegetation. Nature
39. Gauch, H. G., Whittaker, R. H. 1972a. 242:344-47
Coenocline simulation. Ecology 53:446- 56. Grime, J. P. 1979. Plant Strategies and
51 Vegetation Processes. Chichester: Wiley
40. Gauch, H. G., Whittaker, R. H. 1972b. 57. Grubb, P. J. 1977. The maintenance of
Comparison of ordination techniques. species-richness in plant communities:
Ecology 53:868-75 the importance of the regeneration niche.
41. Gauch, H. G., Whittaker, R. H., Singer, Biol. Rev. 52:107-45
S. B. 1981. A comparative study of 58. Harper, J. L. 1977. Population Biology
nonmetric ordinations. J. Ecol. 69:135- of Plants. London: Academic
52 59. Hill, M. 0. 1973. Reciprocal averaging:
42. Gauch, H. G., Whittaker, R. H., Went- an eigenvector method of ordination. J.
worth, T. R. 1977. A comparative study Ecol. 61:237-49
of reciprocal averaging and other ordina- 60. Hill, M. 0. 1974. Correspondence analy-
tion techniques. J. Ecol. 65:157-74 sis: A neglected multivariate method. J.
43. Gittins, R. T. 1963. A study of the struc- R. Stat. Soc. Ser. C. 23:340-54
ture of some grassland and heath com- 61. Hill, M. 0. 1979. DECORANA -A FOR-
munities in Anglesey. PhD thesis. Univ. TRAN Program for detrended Corre-
Wales, Bangor spondence Analysis and Reciprocal
44. Gittins, R. 1979. Ecological applications Averaging. Ithaca, NY: Cornell Univ.
of canonical analysis. In Multivariate 62. Hill, M. O., Gauch, H. G. 1980. De-
Methods in Ecological Work, eds. L. trended Correspondence Analysis, an im-
Orloci, C. R. Rao, W. M. Stiteler. Stat. proved ordination technique. Vegetatio
Ecol. No. 7, pp. 309-535. Fairland, Md: 42:47-58
Int. Co-op. 63. Jenny, H. 1980. The Soil Resource: Ori-
45. Gleason, H. A. 1926. The individualistic gin and Behavior. Ecological Studies:
concept of the plant association. Bull. Analysis and Prediction. No. 37. New
Torrey. Bot. Club 53:1-20 York: Springer-Verlag
46. Goldsmith, F. B. 1973a. The vegetation 64. Johnson, R. W., Goodall, D. W. 1982.
of exposed sea cliffs at South Stack, An- Non-linear ordination in several di-
glesey, I. The multivariate approach. J. mensions a maximum likelihood ap-
Ecol. 61:787-818 proach. Vegetatio 48:197-208
47. Goldsmith, F. B. 1973b. The vegetation 65. Karr, J. R., James, F. C. 1975. Ecomor-
of exposed sea cliffs at South Stack, An- phological configurations and con-
60 AUSTIN
vergent evolution. See Ref. 23, pp. 258- Field Plateau, Tasmania. PhD thesis.
291 Univ. Tasmania, Hobart
66. Kessell, S. R. 1979. Gradient Model- 81. Mueller-Dombois, D., Ellenberg, H.
ling. New York: Springer-Verlag 1974. Aims and Methods of Vegetation
67. Kessell, S. R., Whittaker, R. H. 1976. Ecology. New York: Wiley
Comparisons of three ordination tech- 82. Nelder, J. A., Wedderburn, R. W. M.
niques. Vegetatio 32:21-29 1972. Generalized linear models. J. R.
68. Krebs, C. J. 1978. Ecology: the Ex- Stat. Soc. Ser. A. 135:370-84
perimental Analysis of Distribution and 83. Newman, E. I. 1973. Competition and
Abundance. New York: Harper, Row. diversity in herbaceous vegetation. Na-
2nd ed. ture 244:310
69. Levins, R., Lewontin, R. 1980. Di- 84. Nichols, S. 1977. On the interpretation of
alectics and reductionism in ecology. In principal components analysis in ecologi-
Conceptual Issues in Ecology, ed. E. cal contexts. Vegetatio 34:191-97
Saarinen, pp. 107-38. Dordrecht: Reidel 85. Noy-Meir, 1. 1974. Multivariate analysis
70. Loucks, 0. L. 1962. Ordinating forest of semi-arid vegetation in south-eastern
communities by means of environmental Australia. II. Vegetation catenae and en-
scalars and phytosociological indices. vironmental gradients. Aust. J. Bot. 22:
Ecol. Monogr. 32:137-66 115-40
71. MacArthur, R. 1971. Patterns of terres-86. Noy-Meir, I., Austin, M. P. 1970. Prin-
trial bird communities. In Avian Biology, cipal component ordination and sim-
Vol. 1, ed. D. S. Farner, J. R. King, K. ulated vegetational data. Ecology
C. Parkes. New York: Academic 51:551-2
72. Magdych, W. P. 1984. Salinity stress 87. Noy-Meir, I., Whittaker, R. H. 1978.
along a complex river continuum: effects Recent developments in continuous mul-
on mayfly (Ephemeroptera) distribu- tivariate techniques. See Ref. 119, pp.
tions. Ecology 65:1662-72 337-78
73. Mathews, J. A. 1978. An application of 88. Oksanen, J. 1983. Ordination of boreal
non-metric multidimensional scaling to heath-like vegetation with principal com-
the construction of an improved species ponents analysis, correspondence anal-
plexus. J. Ecology 66:157-173 ysis and multidimensional scaling. Vege-
74. Mathews, J. A. 1979. The vegetation of tatio 52:181-89
the Storbreen gletschervorfeld, Jotunhei- 89. Orloci, L. 1966. Geometric models in
men, Norway. II. Approaches involving ecology. 1. The theory and application of
ordination and general conclusions. J. some ordination techniques. J. Ecol.
Biogeogr. 6:133-67 54:193-215
75. May, R. M. 1981. Theoretical Ecology, 90. Peet, R. K. 1978a. Latitudinal variation
Principles and Applications. Oxford: in southern Rocky Mountain forests. J.
Blackwell. 2nd ed. Biogeogr. 5:275-89
76. McIntosh, R. P. 1967. The continuum 91. Peet, R. K. 1978b. Forest vegetation of
concept of vegetation. Bot. Rev. 33:130- the Colorado Front Range: patterns of
87 species diversity. Vegetatio 37:65-78
77. McCormick, J. F., Lugo, A. E., Sharitz, 92. Peet, R. K., Christensen, N. L. 1980.
R. R. 1974. Experimental analysis of Hardwood forest vegetation of the North
ecosystems. In Vegetation and Environ- Carolina piedmont. Veroeff. Geobot.
ment: Handbook of Vegetation Science, Inst. Eidg. Tech. Hochsch. Stift. Ruebel
Part VI, ed. B. R. Strain, J. W. D. Bil- Zuerich 69:14-39
lings, pp. 151-79. The Hague: Junk 93. Peet, R. K., Loucks, D. L. 1977. A gra-
78. Meents, J. K., Rice, J., Anderson, B. dient analysis of southern Wisconsin for-
W., Ohmart, R. D. 1983. Nonlinear rela- ests. Ecology 58:485-99
tionships between birds and vegetation. 94. Pianka, E. R. 1981. Competition and
Ecology 64:1022-27 niche theory. See Ref. 75, pp. 167-196
79. Miles, D. B., Ricklefs, R. E. 1984. The 95. Pickett, S. T. A., Bazzaz, F. A. 1978.
correlation between ecology and Organization of an assemblage of early
morphology in deciduous forest passer- successional species on a soil moisture
ine birds. Ecology 65:1629-40 gradient. Ecology 59:1248-55
80. Minchin, P. R. 1984. A comparative 96. Pimental, R. A. 1981. A comparative
evaluation of techniques for ecological study of data and ordination techniques
ordination using simulated vegetation based on a hybrid swarm of sand ver-
data and an integrated ordination- benas (Abronia Juss.) Syst. Zool.
classification analysis of the alpine and 30:250-67
subalpine plant communities of the Mt.97. Poole, R. W. 1974. An Introduction to
CONTINUUM. ORDINATION. AND NICHE 61
Quantitative Ecology. Tokyo: McGraw- and Baja California and the community/
Hill continuum debate. Vegetatio 52:3-19
98. Prentice, I. C. 1977. Non-metric ordina- 113. Whitmore, R. C., Harner, E. J. 1980.
tion methods in ecology. J. Ecol. 65:85- Analyses of multivariately determined
94 community matrices using cluster analy-
99. Prentice, I. C. 1980a. Multidimensional sis and multidimensional scaling.
scaling as a research tool in Quaternary Biometr. J. 22:715-23
palynology: a review of theory and 114. Whittaker, R. H. 1951. A criticism of the
methods. Rev. Palaeobot. Palvnol. 31: plant association and climatic climax
71-104 concepts. Northwest Sci. 26:17-31
100. Prentice, I. C. 1980b. Vegetation analy- 115. Whittaker, R. H. 1956. Vegetation of the
sis and order invariant gradient models. Great Smoky Mountains. Ecol. Monogr.
Vegetatio 42:27-34 26:1-80
101. Rachel, F. J. 1984. Factors structuring 116. Whittaker, R. H. 1960. Vegetation of the
fish assemblages along a bog lake suc- Siskiyou Mountains, Oregon and Cali-
cessional gradient. Ecology 65:1276-89 fornia. Ecol. Monogr. 30:279-338
102. Rotenberry, J. T., Wiens, J. A. 1980. 117. Whittaker, R. H. 1967. Gradient analysis
Habitat structure, patchiness, and avian of vegetation. Biol. Rev. 42:207-64
communities in North American steppe 118. Whittaker, R. H. 1975. Communities and
vegetation: a multivariate analysis. Ecol- Ecosystems. New York: MacMillan. 2nd
ogy 61:1228-50 ed.
103. Roughgarden, J. 1983. Competition and 119. Whittaker, R. H., ed. 1978a. Ordination
theory in community ecology. Am. Nat. of Plant Communities, Handbook of
122:583-601 Vegetation Science, No. 5. The Hague:
104. Sabo, S. R. 1980. Niche and habitat rela- Junk. 2nd ed.
tions in subalpine bird communities of 120. Whittaker, R. H. 1978b. Direct gradient
the White Mountains of New Hampshire. analysis. See Ref. 119, pp. 7-50
Ecol. Monogr. 50:241-59 121. Whittaker, R. H., Gauch, H. G. Jr. 1978.
105. Salt, G. W. 1983. Roles: their limits and Evaluation of ordination techniques. See
responsibilities in ecological and evolu- Ref. 119, pp. 277-336
tionary research. Am. Nat. 122:697-705 122. Whittaker, R. H., Niering, W. A. 1965.
106. Schum, M. 1984. Phenetic structure and Vegetation of the Santa Catalina Moun-
species richness in North and Central tains, Arizona: A gradient analysis of the
America bat faunas. Ecology 65:1315-24 south slope. Ecology 46:429-52
107. Strong, D. R. 1983. Natural variability 123. Whittaker, R. H., Levin, S. A., Root, R.
and the manifold mechanisms of ecologi- B. 1973. Niche, habitat and ecotope. Am.
cal communities. Am. Nat. 122:636-60 Nat. 107:321-38
108. Swan, J. M. A. 1970. An examination of 124. Wikum, D. A., Wali, M. K. 1974. Anal-
some ordination problems by use of sim- ysis of a North Dakota gallery forest:
ulated vegetation data. Ecology 51:89- vegetation in relation to topographic and
102 soil gradients. Ecol. Monogr. 44:441-64
109. Tilman, D. 1982. Resource Competition 125. Williams, B. K. 1983. Some observa-
and Community Structure. Princeton, tions on the use of discriminant analysis
NJ: Princeton Univ. Press in ecology. Ecology 64:1283-91
110. Van der Maarel, E. 1976. On the es- 126. Yeaton, R. I., Cody, M. L. 1979. The
tablishment of plant community bound- distribution of cacti along environmental
aries. Ber. Dtsch Bot. Ges. 89:415-43 gradients in the Sonoran and Mohave
111. Waring, R. M., Major, J. 1964. Some deserts. J. Ecol. 67:529-41
vegetation of the California coastal red- 127. Zobel, D. B., McKee, A., Hawk, G. M.
wood region in relation to gradients of 1976. Relationships of environment to
moisture, nutrients, light and tempera- composition, structure and diversity of
ture. Ecol. Monogr. 34:167-215 forest communities of the central western
112. Westman, W. E. 1983. Xeric Mediterra- Cascades of Oregon. Ecol. Monogr.
nean-type shrubland associations of Alta 46:135-56

Annual Reviews Annual Review of Ecology and Systematics

Uploaded by

Copyright:

Available Formats

Annual Reviews Annual Review of Ecology and Systematics

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Annual Reviews Annual Review of Ecology and Systematics

Uploaded by

Copyright:

Available Formats

Continuum Concept, Ordination Methods, and Niche Theory

Department of Environmental Biology, Research School of Biological Sciences, Aus-

Research on niche theory (75) often ignores ordination stud

A statement of the principle of community continuity is found in Whittak

these methods showed extensive overlap of species distributions, with no

Figure I Alternative models for vegetation organization along an environmental gradient. a)

Modes of minor species' abundance are randomly distributed along an environ-

In plant ecology, ordination is the process of arranging vegetation sampl

Second is indirect gradient analysis (1 17, 119), or vegetational ordination (4),

MULTI-DIMENSIONAL SCALING Multi-dimensional scaling (MDS) is an

models can be accommodated by this weak assumption. Gauch (37, p. 162)

COMPARATIVE STUDIES Numerous comparative studies of RA and PCA

hypothesis-testing phase. Two studies have used experiments to test ordination

DIRECT GRADIENT ANALYSIS This method is a simple graphical technique

NEW STATISTICAL APPROACH Problems of experimental design and analy-

(a) Mean annual rainfall (mm)

4 - Radiation index 0-70 - Radiation index 10 - Radiation index 1.12

16- E rossil E -rossii 0 r oss

4- Radiation index 0r70 R adiation index ( s 1 0 t Radiation index r 112

2 16 ~~E dairympleana -Edairympleana E dairympleana

Figure 2 The qualitative environmental niche (expressed

the qualitative niche of species approximated a bell-shaped response curve.

EXPERIMENTAL APPROACHES TO CONTINUUM

Experimental studies of continua would provide much stronger tests o

markedly asymmetric bimodal curves, due to competition. Though compatible

CONTINUUM STUDIES AND NICHE THEORY

heights) are confounded with climatic changes. Temperature has a physiolog-

theory need to incorporate results from plant community ecology, concerning

In a summary of current controversy regarding experimental methods in co

equation of vegetative biomass with physiological performance and the abili

In a review of statistical plant ecology, Goodall (50) hoped that explorat

I thank P. R. Minchin, N. J. McKenzie, D. Potts, I. R. Noble, D. P. Faith, C.

1. Allen, T. F. H., Shugart, H. H. 1983. mental scalars and statistical curve-fitting

You might also like