WinTWINS教程

WinTWINS version 2.3

Mark O. Hill

Petr Šmilauer

2005

1 Preface

TWINSPAN, based partly on an earlier program called ‘Indicator Species Analysis’ (Hill et al. 1975), was written in 1979, five years before the Apple Macintosh revolutionized personal computing, and two years before MS DOS was launched. The first version of Windows did not appear till November 1985. In 1979, a personal computer was an expensive luxury possessed only by a few geeks; all serious calculations were made on mainframes. At Cornell University, newly-available computer terminals had made programming much easier than in earlier years. Programs and problems could be submitted electronically (to another room in the Cornell’s Langmuir Lab), though the output was still normally on paper. The time was therefore ripe for development of numerical methods in ecology to the point where they could become routine tools rather than interesting prospects for development.

The early proponents of numerical methods, notably Goodall (1953a, 1953b), had seen themselves as champions of objectivity. They were uncomfortable about the Zürich-Montpellier tradition of continental Europe, which had sought to construct a comprehensive system of knowledge. In the eyes of many British and American ecologists, the Z-M system was subjective and therefore intellectually dubious, because field workers sampled in a way that allowed them to prove what they wanted to find out in the first place. However, not everybody in Britain and America was convinced that objectivity and the Z-M system were in opposition. R.H. Whittaker (1962) urged ecologists to be pragmatic. He visited Tüxen in Germany, and with Tüxen’s blessing edited the monumental Ordination and classification of communities (1973) in which the various protagonists set out their points of view. The publication of a good English-language manual by the German ecologists Mueller-Dombois and Ellenberg (1974), included an exposition of ‘Tablework’, a nearly-algorithmic method of sorting two-way tables. This narrowed the gap still further.

Thus it was natural that Mark Hill, visiting Whittaker in Cornell, should seek to develop an algorithm whose purpose was to sort tables in an objective way but in the spirit of Z-M methodology. There remained a difficulty, namely that existing numerical methods almost all classified either the samples (the so-called Q methods) or the species (R methods) but did not seek to arrange both together. Underlying this difficulty was the fact that many mathematicians were committed to the metric paradigm, according to which a classification should reflect as faithfully as possible a metric compositional distances between samples (for Q methods) or between species (for R methods). By 1979, the metric paradigm for ordination had already been subjected to severe criticism (Austin 1976), and its applicability to classification was therefore also open to question.

This was the background to the writing of TWINSPAN. A practical problem that had already been solved for Indicator Species Analysis was to keep magnitude of the calculation so that it rose only linearly with the size of the dataset. (Clearly any explicit calculation of distance matrices would increase the problem to magnitude m2 if m is the number of samples, or n2 where n is the number of species.) An algorithm linear in the size of the dataset was achieved by, figuratively speaking, sending signals through the data matrix in search of resonances in which the species and samples sounded together, and then dividing the data accordingly. When the samples and subsequently the species were repeatedly divided, TWINSPAN resulted.

TWINSPAN was originally written in FORTRAN 4, a language well suited to mathematical calculations but with poor handling of alphabetical data. Some improvements