Classification and Ordination: Symposium on advances in vegetation science, Nijmegen, The Netherlands, May 1979

184 Pages·1980·6.455 MB·English

by  P. Greig-Smith (auth.), Eddy van der Maarel (eds.)

CLASSIFICATION AND ORDINATION
Advances in vegetation science 2
Edited by
EDDY VAN DER MAAREL
Dr. W. JUNK bv PUBLISHERS THE HAGUE - BOSTON - LONDON 1980
Classification and ordination
Symposium on advances
in vegetation science,
Nijmegen, The Netherlands, May 1979
Edited by
EDDY VAN DER MAAREL
Dr. W. JUNK bv PUBLISHERS THE HAGUE - BOSTON - LONDON 1980
Distributors:
for the United States and Canada
Kluwer Boston, Inc.
190 Old Derby Street
Hingham, MA 02043
USA
for all other countries
Kluwer Academic Publishers Group
Distribution Center
P.O. Box 322
3300 AH Dordrecht
The Netherlands
Library of Congress Cataloging in Publication Data CIP
Symposium on Advances in Vegetation Sciences, Nijmegen,
The Netherlands, 1979.
Classification and ordination.
(Advances in vegetation science; v. 2)
'Reprinted from Vegetatio, vol. 42,1980.'
I. Vegetation classification - Congresses.
1. Maarel, E. van der. II. Vegetatio. III. Title.
IV. Series.
QK91 1. S94 1979 580'. 12 80-23092
1SBN-13: 978-94-009-9\99-6 e-\SBN-13: 978-94-009-9\97-2
DOl: \0.\007/978-94-009-9\97-2
Reprinted from Vegetatio, Vol. 42, 1980
Copyright© 1980 Dr. W. Junk bv Publishers, The Hague.
Softcover reprint of the hardcover 1st edition 1980
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form
or by any means, mechanical, photocopying, recording, or otherwise, without the prior written permission of the publishers.
Dr. W. Junk bv Publishers, P.O. Box 13713, 2501 ES The Hague, The Netherlands.
CONTENTS
E. van der Maarel, Introduction . . . . . . . . . . . . . . . . . . . vn
L. Orl6ci, Preface to 'Classification and ordination' . . . . . . . . . . . IX
P. Greig-Smith, The development of numerical classification and ordination. 1
M. P. Austin, Searching for a model for use in vegetation analysis . . II
L. Orl6ci, An algorithm for predictive ordination . . . . . . . . . . . . 23
I. C. Prentice, Vegetation analysis and order invariant gradient models. . . 27
E. Feoli & L. Feoli Chiapella, Evaluation of ordination methods through simulated coenoclines: Some comments 35
E. van der Maarel, On the interpretability of ordination diagrams . . . . . . . . . . . . . . . . . . . .. 43
M. O. Hill & H. G. Gauch, Detrended correspondence analysis: An improved ordination technique . . . . .. 47
T. J. Carleton, Non-centred component analysis of vegetation data: A comparison of orthogonal and oblique rotation 59
G. Wiegleb, Some applications of principal components analysis in vegetation: Ecological research of aquatic com-
munities. . . . . . . . . . . . . . . . . . . • . . . . . . . . . . . . . . . . . . . . . 67
R. del Moral, On selecting indirect ordination methods. . . . . . . . . . . . . . . . . . . . . . 75
J. Lepart & M. Debussche, Information efficiency and regional constellation of environmental variables. 85
M. B. Dale, A syntactic basis of classification. . . . . . . . . . . . . . . . . 93
R. C. Jancey, The minimisation of random events in the search for group structure. . . 99
H. G. Gauch, Rapid initial clustering of large data sets. . . . . . . . . . . . . . . 103
E. Feoli & D. Lausi, Hierarchical levels in syntaxonomy based on information functions 113
S. Jensen & E. van der Maarel, Numerical approaches to lake classification with special reference to macrophyte
communities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
R. S. Clymo, Preliminary survey of the peat-bog Hummell Knowe Moss using various numerical methods. . . . . 129
V. Komarkova, Classification and ordination in the Indian Peaks area, Colorado Rocky Mountains . . . . . . . 149
P. van Heeke, I. Impens, R. Goossens & F. Hebrant, Multivariate analysis of multispectral remote sensing data on
grasslands from different soil types. . . . . . . . . . . . . . . . . . . . . . 165
R. K. Peet, Ordinations as a tool for analyzing complex data sets . . . . . . . . . . . . . . . . . . 171
O. Wildi, Management and multivariate analysis of large data sets in vegetation research . . . . . . . . 175
S. Pignatti, Reflections on the phytosociological approach and the epistemological basis of vegetation science . 181
List of participants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
INTRODUCTION
Eddy V AN DER MAAREL
This volume is the first of two volumes covering the Sym computer programmes for the rapid clustering and ordina
posium 'Advances in vegetation science', which was held at tion of very large sets of reI eves and for (subsequent) table
Nijmegen, The Netherlands, from 15-19 May 1979. This rearrangement (this volume as well as the book Data
symposium was organized on behalf of the Working Group Processing in Phytosociology contain various new pro
for Data-Processing of the International Society for Vege grams). What we do not have is a manual in which the
tation Science. After this group held its final meeting two apparently successful methods are compared and applied
years earlier it decided to continue its activities, but within a to some data-sets. H. Lieth, editor-in-chief of a new Junk
wider scope. Most members of the Group felt that the series 'Tasks for vegetation science' already suggested to
original aim, i.e. the introduction of data-processing and produce such a manual in this series.
multivariate methods for use in the systematic description The present volume contains the texts of the lectures and
of plant communities, was more or less fulfilled. The book most of the poster demonstrations of the first three sessions
Data -Processing in Phytosociology, largely based on papers of the Symposium, dealing with classification and ordina
in Vegetatio, edited by E. van der Maarel, L. Orloci & S. tion, as well as some manuscripts which were submitted by
Pignatti, and to be published by Dr. W. Junk Publishers, the time of the Symposium and fitted very well in the
may be considered as a comprehensive report on the Work Volume. (Such a rather effective procedure can easily be
ing Group's activities. developed if Symposium Proceedings are published as
The wider scope I mentioned can be described as 'theore special volumes of Vegetatio !).
tical vegetation ecology' or 'descriptive and experimental The first session presided by L. Orloci, started with a
analysis of vegetation'. The further development of multi broad introductory survey by P. Greig-Smith and was
variate methods is certainly a matter of interest for the continued by M.P. Austin and I.e. Prentice. Session 2,
extended Working Group, but those methods should presided by P. Greig-Smith included lectures by L. Orloci,
also be applied to other than synsystematical purposes, R. del Moral, M.B. Dale, R.e. Jancey and D. Lausi.
for instance to site - time series. In relation to this develop Session 3 largely contained examples of applications and
ment general models for vegetation analysis and for comparisons of methods, by M.A. Ayyad, S. Jensen, V.
vegetation succession should be elaborated. Besides Komarkova, O. Wildi, E. Feoli, P. van Hecke, J. Lepart
general aspects of diversity and structural - functional and R.S. Clymo. The contribution by Ayyad and an
properties and population dynamics are worth while study inserted lecture by H. Lieth will be published in current
ing. issues ofVegetatio. (Herewith we have a second advantage
Indeed, this all would imply a new organisation ofvege of incorporating Symposium Proceedings in Vegetatio:
tation ecologists. I hope the Nijmegen Symposium will there are always contributions which do not fit in the
contribute to this new development. Symposium theme. By exluding them from the proper
As I indicated already multivariate methods will remain proceedings those become more homogeneous and by
important for the period to come. Unlike some colleagues publishing them in the same way we do justice to the
who are satisfied with the present state of the art and now contributors !).
wish to achieve results of large scale data treatments Poster demonstrations were given by most of the speakers
I believe that we have not reached that stage. To mention mentioned above, and in this way we would compensate
one example: we do have a rather complete series of for the short speaking time (15-20 min.) most of them got.
VII
Besides a poster demonstration by G. Wiegleb and a Looking back at the Symposium and reading through
discussion contribution by E. van der Maarel were in the manuscripts I came to the conclusion that this volume
corporated. reasonably reflects the variety and quality we can find in the
The organisation of the volume is as follows: theories and methods of numerical vegetation science.
A. General: Orloci's opening remarks and the lectures by I am sure that most of the participants feel the same way.
Greig-Smith and Austin. I hope many readers will join this conclusion! What the
B. Ordination, theory and methods: Symposium contribu reader cannot detect is the cooperative and stimulating
tions by Orloci, Prentice, Feoli & Feoli Chiapella, van der atmosphere during the many discussions, both plenary
Maarel, del Moral and Wiegleb, and inserted manuscripts and private in the poster hall. We can only hope, that more
by Hill & Gauch and Carleton. The contribution by Lepart of such symposia, with short but representative lectures and
& Debussche on ecological variables fits in this section. intensive personal contact will follow!
C. Classification, theory and methods: Symposium con Finally I wish to remark that this symposium was
tributions by Dale, Jancey, Feoli & Lausi, Jensen & van sponsored by the Faculty of Science of the University of
der Maarel and an inserted manuscript by Gauch. Nijmegen. I thank the Director of the Faculty, Dr. C.J.M.
D. Classification and ordination,' examples and comparisons: Aarts, for his support and cooperation and I am especially
Symposium contributions by Clymo, Komarkova, van grateful for the considerable help which was given by Peter
Hecke et aI. and Peet. Wildi's paper (dealing with a new Toll, L.L.D. of the Faculty's Directorate, both in the
package of programs) joins this series. organization of the symposium and in taking care of the
E. Epilogue: Pignatti's closing contribution on the episto symposium guests! From our Department Rita Dubbers,
mological basis of vegetation science forms a logical end of Henk Butteling and Dr. Marinus Werger deserve being
our Symposium. acknowledged for their help.
VIII
PREFACE TO 'CLASSIFICATION AND ORDINATION'
Laszlo H. ORLOcI
The Symposium comes at a time when our outlook about axiomatic formulations, stress on generality, arbitrary
the nature of the relevant statistical universe in plant confinement of the practice to the simple case of linear
ecology is changing, and at a time when we have a strongly data structures, and all these in the interest of elegant,
felt need for a new methodology to deal with the new axiomatic developments.
realities in data analysis. 3. The incorporation of the idea of probability in the
As I see them, these realities may present problems on methodologies to render them statistical is highly desired.
three separate counts: Yet, not much has been done to develop a new statistics
I. The assumption of a non-linear species response, and which can handle non-linear covariation.
non-linear covariation among species is pivotal in current My feeling is that at this stage in the evolution of data
thinking. Yet, the methodologies in such important analysis, ecologists should not feel bound by the orthodox
fields as the multivariate analysis of variance, discriminant statistical tradition in which data analysis degenerated into
analysis, canonical analysis, component analysis, etc., a rigid exercise constrained by formality and hosts of
cannot handle anything but linear covariation. restrictive assumptions. They should rather view and
2. The heuristic approaches in data analysis are closely in practise data analysis as an opportunity in experimen
line with current preferences for flexibility in method tation, to see ideas and techniques evolve, and results
selection and maximum local relevance of the results. Yet emerge which can illuminate as many aspects of the
what is being offered by way of the prevailing formal information in the data as reasonable possible.
methodologies is characterized by adherence to rigid
THE DEVELOPMENT OF NUMERICAL CLASSIFICATION AND ORDINATION
P. GREIG-SMITH
School of Plant Biology, University College of North Wales, Bangor LL57 2UW, United Kingdom
The invitation to open this session set me thinking about association with the continuum ooncept of vegetation, in
the development of numerical procedures of classification fluenced its reception.
and ordination. Their technical development has been re The attitude of most ecologists to mathematics at that
viewed by various authors from various viewpoints, e.g. time can fairly be described as one of suspicion. This is
Cormack (1971), Orl6ci (1975, 1978), Dale (1975), Goodall curious because there was at the same time considerable
(1970), Greig-Smith (1954, 1964, 1980), Whittaker (1967, respect for quantification, so that much time was some
1973). I do not intend to discuss this in more than broad times devoted to obtaining quantitative data in the field,
terms, but there is another aspect that has received less data largely wasted because no further analysis was made.
attention. This concerns the influences and constraints It is interesting that Tansley could write in 1923 'in pro
which have affected the development of numerical methods portion to the advance of a branch of science its methods
and their acceptance by phytosociologists. Their accep become more quantitative. This is true of biology in general
tance is particularly important; numerical methods are and of ecology in particular as of other branches of science.'
tools and unless they are used in the investigation of real With characteristic percipience he went on to warn against
ecological problems we are wasting our time in developing gathering quantitative data for their own sake, but this
them. warning was not infrequently ignored. The first edition of
What I have to say represents a personal view, but will, I Fisher's Statistical Methods for Research Workers, which
hope, be of some interest. My excuse for attempting a broad was to have such a profound impact on biolcgy in general,
survey of this kind is that I have been closely involved with appeared in 1925. One wonders whether this may paradox
numerical classification and ordination throughout their ically have delayed the development of numerical methods
development. This has given me the opportunity to look of phytosociology. For most biologists the kind of sta
back at the misconceptions and the failures to recognise tistical analysis developed by Fisher, with its emphasis on
what now seems obvious, fhat occur in all scientific fit to hypothesis, probability and tests of significance,
development, but which are not so often talked about. became the only kind of mathematics that was relevant;
Numerical analysis of plant communities as we now it did not prove helpful in dealing with plant communities.
understand it originated approximately 30 years ago, The early advocates of ordination techniques were all
though interest in certain community attributes e.g. spe supporters ofthe interpretation of variation in vegetational
cies/area relationships and in numerical approaches to the composition as a continuum e.g. Ramensky, Curtis, Good
distribution of individual species had developed earlier and all, Whittaker. Acceptance of the continuum view was un
was sufficient to justify review articles in 1936 and 1948 doubtedly a powerful stimulus to the development of tech
(Ashby 1936, 1948). At this time classification was a long niques of ordination, but it is now generally accepted that
established approach, though there was controversy about the choice between classification and ordination depends
the most appropriate system to use and the importance of on the objective of data analysis and the structure of the
classification to a broader understanding of vegetation. data set being examined, rather than on preconceptions
Although Ramensky had developed a technique of ordi about the nature of vegetation. Presentation of ordination
nation (see Sobolev & Utekhin 1973) this was little known techniques in the context of continuum undoubtedly led
outside Russia and it came as a new approach to most to their being ignored by many ecologists. Misunderstan
phytosociologists. Ordination was thus linked with numeri ding was increased by the chance that the first practical
cal approaches from the start and this, together with its technique (Curtis & McIntosh 1951) was illustrated by a
Vegetatio vol. 42: 1-9, 1980
set of data in which the first and only axis extracted was a feel for a numerical approach, provided they were prepared
successional one and the method was regarded as aiming to accept the context of continuum in which they were
at elucidating succession rather than of more general ap presented. On the other hand, the emphasis on single
plication. recognisable environmental gradients has had an influence
[he initial techniques were mostly crude and some could on later developments which has not been wholly bene
scarcely be regarded as numerical. There was little to at ficial.
tract the interest of mathematicians. Later, as techniques At this stage the limitations of available techniques were
developed, mathematicians were to look at them critically that indirect gradient analysis allowed for the derivation
and helpfully, as in Cormack's (1971) review of classifi of a single axis only and was difficult to apply unless a
cation, but in the early stages numerical methods were single gradient was overriding, as in a successional situa
regarded as irrelevant by most ecologists and as unworthy tion, and direct gradient analysis depended on prior recog
of notice by mathematicians. nition of the most important environmental gradients; it
Because of the influence of ordination techniques on was no accident that Whittaker's method was developed
classificatory techniques, it is convenient to consider ordi in a region of marked topographic variation. It was not
nation first. Dale (1975) has pointed out that there were surprising that attention was then concentrated on indirect
three major sources of methods of ordination. One, direct methods that allowed more flexible derivation of several
gradient analysis, was dependent on recognition of envi axes.
ronmental gradients and sought to relate vegetational data There were two independent approaches to multiaxis
to them (Whittaker, 1952, 1956). The other two both sought ordination, both dimension-reducing. Goodall (1954) ap
patterns in the vegetational data and only after these had plied a form of principal component analysis and Bray &
been identified was their relation to environmental gra Curtis (1957) used an essentially geometric approach to
dients established. The first of these accepted, though not reduce the dimensionality of a matrix of interstand dissimi
always explicitly, that a set of data could be considered larities or digtances. There is an interesting contrast in the
in relation to as many independent axes as there were spe subsequent developments from these two pioneer papers.
cies present and argued that the dimensionality of the set Principal component analysis was not followed up as a
could be reduced without serious loss of information if practical tool until considerably later, though Dagnelie
species occurrences are correlated (Goodall 1954, Bray & (1960) discussed the use of both it and factor analysis.
Curtis 1957). The second, path-seeking or more recently Bray & Curtis' technique was relatively soon being used
(Noy-Meir 1974) 'catenation', aimed to order stands so and appears as an important tool in Curtis' The Vegetation
that species had a unimodal distribution along the axis of Wisconsin (Curtis 1959). Although principal components
(Curtis & McIntosh 1951). Reduction in dimensionality analysis is more difficult to understand than the very simple
emphasises the overall relations between stands, catenation Bray & Curtis technique, the delay in following up Good
emphasises the relations between a stand and those most all's suggestion resulted primarily from the computational
similar to it. Though the distinction between these ap load involved. Not until electronic computers of sufficient
proaches is evident in retrospect, it was not, I think, speed and capacity had been developed did it become
generally realised at the time. Even those most interested feasible to use principal component analysis for any
regarded Bray & Curtis' method at the time as a promising considerable. body of data. This is not the only case where
extension of that of Curtis & McIntosh to more than one the application of analytical techniques has been delayed by
axis, rather than a fundamentally different approach. the unavailability of adequate computing facilities, rapid
The first approaches were in terms of a single axis only. though the development of the latter has been. By contrast,
It is true that Whittaker (1956, 1960), by examining mois Bray & Curtis' technique is feasible 'by hand' i.e. with only
ture gradients at different altitudes, produced what was in a simple desk calculator; Ashton (1964) analysed a very
effect a two-dimensional ordination and, by comparing considerable body of data from a species-rich tropical
moisture and altitudinal effects on two contrasting soil forest by hand. That it was developed in the very active
types, was even able to include two different levels on a group led by the late J.T. Curtis at the University of
third axis, but the technique was essentially one of examina Wisconsin also contributed to its rapid adoption.
tion of single environmental gradients. There are two The simple ordination of Bray & Curtis was criticised on
considerations of interest. These very simple ordinations various grounds and modifications were proposed (e.g.
were readily understood, even by those with little or no Orl6ci 1966, 1974, Austin & Orl6ci 1966, Swan, Dix &
2