Download original version of Chapter 5

1
The nature of the plant community: a reductionist view
2
3
J. Bastow Wilson
Botany Department, University of Otago, Box 56, Dunedin, New Zealand.
4
5
Andrew D.Q. Agnew
Institute of Biological Sciences, University of Wales Aberystwyth, SY23 3DA, U.K.
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
Chapter 5: Assembly rules
1
2
Introduction ................................................................................................................................ 2
What rules are we searching for? ............................................................................................... 3
2.1
Inductive versus deductive ................................................................................................. 3
2.2
Randomisation tests............................................................................................................ 3
2.3
Ruling out environmental variation.................................................................................... 4
2.4
Taxonomic-based limiting similarity ................................................................................. 6
2.5
Process versus pattern ........................................................................................................ 6
3
Zonation ..................................................................................................................................... 7
3.1
Boundaries in zonation ....................................................................................................... 7
3.2
Fundamental and realised niche ......................................................................................... 8
Beta niche ................................................................................................................................... 8
4
Species sorting.......................................................................................................................... 10
4.1
Species associations in succession ................................................................................... 10
4.2
Compositional convergence ............................................................................................. 11
4.3
Transitivity in interference networks ............................................................................... 12
5
Richness ................................................................................................................................... 17
6
Limiting similarity.................................................................................................................... 18
6.1
Limiting similarity in morphological characters .............................................................. 19
6.2
Limiting similarity in phenology...................................................................................... 22
7
Guild proportionality ................................................................................................................ 24
7.1
Concept............................................................................................................................. 24
7.2
Evidence: constancy in space ........................................................................................... 25
7.3
Patch models .................................................................................................................... 28
7.4
Evidence: removal experiments ....................................................................................... 29
7.5
Evidence: successional convergence ................................................................................ 29
7.6
Intrinsic guilds .................................................................................................................. 30
8
Texture convergence ................................................................................................................ 33
9
Time ......................................................................................................................................... 36
10
Abundance ............................................................................................................................ 38
10.1 Biomass constancy ........................................................................................................... 38
10.2 Relative abundance distribution (RAD) ........................................................................... 38
10.3 Sparse species ................................................................................................................... 40
11
Keystone species .................................................................................................................. 41
12
Exotic species as community structure probes ..................................................................... 41
12.1 The nature of exotic species ............................................................................................. 42
12.2 Exotic establishment and community assembly............................................................... 46
13
Conclusions, and the Otago Botany Lawn ........................................................................... 47
Wilson and Agnew, chapter 5, Assembly rules, page 2 of 53
46
47
1 Introduction
We have outlined the processes that occur in plant communities: interference and
48
subvention. Many ecologists wish to go no further with plant communities than look at such
49
processes, but we want to make generalisations at the plant community level.
50
Some ecologists, the ‘phytosociologists’, wish to make worldwide vegetation inventories
51
using the methods originating with Braun-Blanquet (1932) to identify and name communities. In
52
the English-speaking world, the methods and indeed the very aims of such phytosociology were
53
once questioned (Poore 1955), to the extent that the approach is now largely ignored. This is
54
regretted by many European-continental ecologists, even by a few English-speaking ones. We
55
personally appreciate efforts to classify vegetation into numbered/named communities when they
56
are based on objective methods: the British National Vegetation Classification stands out for its
57
better methodology (Rodwell 1991-2000). However, the value of such exercises is mainly as a tool
58
in conservation advocacy, and as an ecological tourist’s guide if the generalisations are accurate.
59
Still, we have to ask, is this science? Where are the testable hypotheses? Where are the tests? It is
60
important now to look for a classification system that would be more substantive. I bet we won’t
61
find one.
62
Our approach is different, in that we look for the rules of engagement in plant associations,
63
based on the processes of species interaction that we have described. These are the assembly
64
rules, which we define as "restrictions on the observed patterns of species presence or abundance
65
that are based on the presence or abundance of one or other species or groups of species (not
66
simply the response of individual species to the environment)" (Wilson 1999 %chapter). This is
67
close to Hubbell’s (2005 %166) definition of assembly as “which species, having which niche
68
traits, and how many species, co-occur in a given community”. We could argue that this is the true
69
meaning of the term phytosociology.
70
Our enquiry must attempt to establish whether assembly rules exist, and deal with the
71
possibility that there are no such rules. There is a widespread and commendable scepticism as to
72
whether assembly rules occur at all (e.g. Ulrich 2004). This may not be our conclusion, but our
73
reductionist aim demands that we start with such a null model, and that we be especially careful in
74
examining the evidence. One suggestion is that they will not be found after disturbance, not until
75
the community regains equilibrium (Bartha et al. 1995). Without accepting this, for its support
76
comes from speculation rather than from evidence, and in any case we have described in Chapter 2
77
how prevalent autodisturbance is, we shall tend to concentrate on what seem to be equilibrium
78
communities. Another question comes from Yodzis’ (1986; 1978) distinction between founder
79
control of community composition and dominance/niche control. If the former be operating, the
Wilson and Agnew, chapter 5, Assembly rules, page 3 of 53
80
species composition of a community will depend largely on which species arrives first, and there
81
will be no further predictability, no rules. Ozinga et al. (2005) addressed this issue using a 20,000-
82
quadrat database. On average among species the first four axes of a CCA ordination constrained
83
by six Ellenberg scores explained only 7.7 % of species occurrences, though the value was 10.3 %
84
for species with long-lived seeds and a mechanism for long-distance dispersal. This implies a rôle
85
for founder control, though the conclusion relies on the completeness of the environmental
86
characterization. We shall discuss in chapter 6 ideas on the rôle of chance. Another problem is that
87
the species the ecologist sees are not those the taxonomist sees. We have forsworn, in general,
88
consideration of within-species genetic (e.g. ecotypic) differences and plastic responses in this
89
book, but both are important in the world. We are often dealing with the realised niche of the
90
species, which may be considerably different from its fundamental niche (Austin and Gaywood
91
1994), and not easily predicted from it.
92
There is a widespread and commendable skepticism as to whether assembly rules occur at
93
all (e.g. Ulrich 2004). This may not be our conclusion, but our reductionist aim demands that we
94
start with such a null model, and that we be especially careful in interpreting departures from it.
95
Said before
96
2 What rules are we searching for, and how?
97
2.1 Inductive versus deductive
98
99
Inductive and deductive approaches both have their value in community ecology (Dale
2002 %191; Wilson 2003 %216), and both will be seen below. An example of the deductive
100
approach is guild proportionality in forest: the differences between species in their mature height
101
are well established, we can reason that these represent different niches, with the species
102
potentially capable of occupying to each niche constituting a guild. We can reason that a species
103
will invade more readily where few members of its guild are already present. If the null model is
104
disproved, and if we can rule out other explanations such as environmental effects, the existence of
105
the rule has been proved, though not its exact mechanism. On the other hand, a search for intrinsic
106
guilds is inductive in that we are not assuming any structure save that guilds might exist, but so
107
long as the guilds are formed and tested on independent data we then have a strong pointer to
108
where to seek the processes that are structuring the community. Finding a repeated pattern is the
109
first step to finding its cause.
110
2.2 Randomisation tests
111
112
To demonstrate assembly rules, we have to compare an observed pattern with that expected
under a null model. These are often difficult to frame. What does a plant community look like
Wilson and Agnew, chapter 5, Assembly rules, page 4 of 53
113
when it isn’t there? A prior question is what pattern to seek: what does a plant community look
114
like when it is there?
115
In these comparisons, randomisation tests are often needed, in which a test statistic is
116
calculated on the observed data, then on data randomised under a certain null model, and
117
significance (probability that the observed results would occur under the null model) is determined
118
from the proportion of randomised values that are equal to, or more extreme than, the observed
119
one. There are traps here. Any test statistic can validly be chosen, though one should ensure it tests
120
the ecological question asked. Selection of the null model is more crucial; many studies have come
121
unstuck from choosing the wrong one and demonstrating as a result an obvious fact such that
122
species differ in frequency (Wilson 1995 % 543). We use the Tokeshi principle, that the null
123
model must include all the features of the observed data except the one it is intended to test
124
(Wilson 1999 %130). Lastly, tails: if it is conceivable that the observed data could differ from the
125
null model in either direction, i.e. results either way will be noticed, a 2-tailed test must be used.
126
This comprises either doubling the p value obtained, or using say two 2.5% tails for a 5% test.
127
2.3 Ruling out environmental variation
128
We must examine natural mixtures of species in a way that takes into account gross
129
environmental heterogeneity. It is no surprise that species are adapted top particular places along
130
an environmental gradient. The rules we find have to transcend in their generality ones of the type:
131
“species x occurs at low/high values of environmental factor z”, described as the “easy task” of
132
community ecology by Warming (1909). We need to search for reasons for species’ relative
133
positions which are not primarily environmental. Therefore, in seeking assembly rules, i.e. the
134
repeated patterns of MacArthur (1972):
135
(a) The rules we seek will not necessarily depend on the identity of particular species. This
136
contrasts with Diamond’s (1975) original assembly rules, but that approach has not proved
137
useful.
138
(b) They will not simply describe the fact that species are correlated with their environment.
139
However, we cannot expect that the rules will apply worldwide, in all habitats. For example, rules
140
based on stratification cannot apply to the very few communities that have no stratification, and
141
we might expect that communities in deserts would be constructed quite differently from those in
142
rain forests. For character-based, limiting-similarity rules, the characters involved will be different
143
in different habitats, where different resources are limiting.
144
Environmental correlations, Warming’s “easy task” to investigate for their own sake, are
145
actually a huge problem in seeking assembly rules. Environmental variation occurs at all scales in
146
all communities (Goodall 1954). We have to seek assembly rules against a background of this
Wilson and Agnew, chapter 5, Assembly rules, page 5 of 53
147
local environmental variation, which is difficult to define. Often, when we are seeking assembly
148
rules, environmental variation acts as noise. Very commonly, the null model against which we are
149
testing the observed pattern assumes no environmental variation, so that if we succeed in
150
disproving the null model we do not know whether this is because there really is an assembly rule,
151
or because environmental variation has mimicked the effect. The latter possibility would not be
152
interesting. Eliminating effects of the environment is not easy.
153
Take the simple case of testing whether variance in richness differs from a null model.
154
Suppose there is environmental variation such that some habitats have few species (just ‘A’ in Fig.
155
5.1), but others have many (‘A B C D’) – the ‘waterhole effect’ of Edith Pielou (1975) (Fig. 5.1),
156
but with no variation of species richness within those habitats. The pattern is in fact determinate,
157
but will appear as greater variation in species richness than expected at random if an overall
158
randomisation – a ‘site’ model – is used.
Environ. 1
Environ. 2
Environ. 3
A
A
A
A
A
A
A
A
AB AB AB
C
AB AB AB
C
AB AB AB
C
AB AB AB
C
AB
C
AB
C
AB
C
AB
C
Environ. 4
AB
CD
AB
CD
AB
CD
AB
CD
AB
CD
AB
CD
AB
CD
AB
CD
159
Fig. 5.1: Four environments containing different species assemblages, consistent within each
160
environment.
161
Suppose the number of species is the same in each quadrat, and they are the same species
162
in each quadrat within each of two environment (Fig. 5.2). Randomisations will include some
163
quadrats with 0, 1, 3 and 4 species, and the observed state will look like constant richness
164
compared to this. The effect is real, in that there is the same number of species in each
165
environment. However, this is being tested 20 times in each environment: pseudoreplication. A
166
test over several environments would be valid and interesting, but then one has to include each
167
community only once and one needs many environments.
Environment 1
Environment 2
AB
AB
AB AB CD CD CD CD
AB
AB
AB AB CD CD CD CD
AB
AB
AB AB CD CD CD CD
AB
AB
AB AB CD CD CD CD
Wilson and Agnew, chapter 5, Assembly rules, page 6 of 53
AB
168
AB
AB AB CD CD CD CD
Fig. 5.2: Two environments containing different species assemblages, but the same richness.
169
The best answer to these problems is to use a patch model rather than a site model. This
170
comprises making a prediction for each quadrat (the ‘target’ quadrat) on the basis of a limited
171
number of adjacent or otherwise similar quadrats (Fig. 5.3). The patch can be square, as in Fig.
172
5.3, or linear, or a grouping of quadrats can be determined a priori as being similar in some other
173
way.
174
AB B
B
D
AB
C
CD
A
B
B
D
D
175
Occurrence of
species A in the
target quadrat
is based on the number
of occurrences of A in
a patch of nine
quadrats centered on it
Fig. 5.3. A patch randomisation model based on a grid of contiguous quadrats. The frequency of
176
species A in the 3×3 patch is 3/9 = 0.333, so in the randomisation species A has a 0.333
177
probability of occurring in the central square.
178
179
2.4 Taxonomic-based limiting similarity
In animal ecology, membership of a genus is commonly used to indicate similarity in alpha
180
niche. In plants, the niche is commonly more independent of taxonomy, and sometimes
181
membership of a genus is more representative of a species’ beta niche (e.g. Salicornia spp. all in
182
saline areas). However, the genus is clearly an ecologically-objective and a priori classification,
183
and if taxonomy is not a good guide to ecology the result will be non-significance, not spurious
184
significance.
185
2.5 Process versus pattern
186
Ecologists often suggest that ‘assembly rule’ should mean the process by which the
187
community is established. Whilst this is a logical thought, Diamond (1975) first used the term for
188
the results of that process. Most later workers have used it in this way, and we do so here.
Wilson and Agnew, chapter 5, Assembly rules, page 7 of 53
189
3 Zonation
190
3.1 Boundaries in zonation
191
As Robert H. Whittaker pointed out, the ideal way to determine whether species are
192
associated into discrete communities is to see whether their boundaries are clustered on an
193
environmental gradient, e.g. to distinguish between the situations in Fig. 5.4 a and b. Answering
194
the question is much more difficult (Wilson 1994 %275).
195
196
197
Fig. 5.4: Whittaker’s diagram.
Shipley and Keddy (1987) examined the upper and lower species boundaries on 13
198
transects along 200 m of a lake shore in Ontario, Canada, and concluded they were significantly
199
clustered. There are some problems with pseudoreplication (Wilson (1994 %275). However, the
200
real problem is that Shipley and Keddy used elevation as the gradient. It is a proxy factor for those
201
actually affecting the plants, and probably not linearly related to any of them. How does it matter?
202
We know this way that species guilds/clusters happen because of some difference in a combination
203
of factors. That’s a start! We do not really know the true factors, or on what scale to express them.
204
The study of Auerbach and Shmida (1993) of altitudinal zonation on Mt Hermon, Israel, has the
205
same problems. Bimodality of species distributions would be a mildly interesting feature, but
206
evidence for it (e.g. Whittaker 1960; 1967) is weak (Wilson et al. 2004 %254).
207
The problem of defining the scale environmental gradient was solved by Dale (1984) by
208
abolishing it. He took up a previous implication that looking at the sequence of top- and bottom-
209
boundaries up a gradient (an intertidal shore in his case) the top boundary (T) of one species would
210
be immediately followed by the bottom boundary (B) of another (the one replacing the other in the
211
same alpha niche): a TB pair. Therefore, overall there would be an excess of TB pairs compared to
212
expectation. This test is non-parametric, in that it is absolutely unaffected by any monotonic
213
rescaling of the axis. However, the non-null (H1) hypothesis assumes very precise replacement of
Wilson and Agnew, chapter 5, Assembly rules, page 8 of 53
214
one species by another, with a small gap, which is hard to envisage in the real world (Wilson 1994
215
%275). It is surprising that Dale himself found excesses of TB pairs significantly often. Thomas et
216
al. (1999), using Dale’s method, did not.
217
It seems that since it is impossible to obtain evidence on community structure from overall
218
zonation, valid answers can be obtained only by changing the question. Wilson and Lee (1994)
219
formed a null model in which the number of species, their frequency patterns and positions along
220
an altitudinal gradient in the Murchison Mountains, southern NZ and the number of species in
221
each genus were all held as observed. The test statistic was the amount of overlap along the
222
gradient between species in the same genus and in the null model species were assigned to genus
223
randomly. The concept is that members of one genus will tend to be similar in alpha niche. They
224
will compete with each other in either ecological or evolutionary time (the “ghosts of competition
225
past”), and hence be spread out in beta niche (altitude), with less overlap than expected from a
226
random selection of species. The results are complicated because testing several genera separately
227
comprises making multiple significance tests. Some genera are known to have altitudinal biases
228
(Pielou 1978, showed that this was true overall for the distribution of algal congeners along a
229
latitudinal gradient), and others have too few species to give significance. However, taking all this
230
into account Wilson and Lee concluded that there was evidence that the species of a genus were
231
more spread out in altitude than expected at random. However, we have not solved the problem of
232
how to find clustered boundaries over all species, have done changed the question.
233
3.2 Fundamental and realised niche
234
Beta niche
235
We know that a species’ realised niche is related to its fundamental one (this vol., chapt. 1,
236
sect. 4.1), but it is not clear just how. Generally, when two species with largely overlapping
237
fundamental niches meet in the field, their realised niches are different. For example, Kenkel et al.
238
(1991 %2497) grew three species, one a facultative halophyte, in a range of rather low salinities. In
239
monoculture, they all grew best with no added NaCl, but in mixed sand culture pots they sorted
240
themselves into three realised-niche optima along the gradient. In most situations, one species
241
moves further along the gradient than the other. A well-known example is the work of Grace and
242
Wetzel (1981) growing two Typha (cattail) species on a gradient of water depth. In monoculture
243
both had same optimum of 50 cm average water depth. In mixture they hardly overlapped in the
244
depths at which they grew: Typha latifolia moved its optimum to 15 cm, T. angustifolia to 80 cm.
245
Similarly, Pennings et al. (2005) investigated a SE USA saltmarsh, where Juncus roemerianus
246
grows higher up on the marsh and Spartina alterniflora grows lower, with a sharp boundary
247
between them. The lower limit of J. roemerianus is set by the physical environment (salt and/or
Wilson and Agnew, chapter 5, Assembly rules, page 9 of 53
248
waterlogging) but the upper limit of S. alterniflora is set by competition, for without competition
249
from J. roemerianus it grew if anything slightly better in the latter’s normal zone than in its own.
250
Can we generalise? Austin (1982 %559) found that for several grasses the growth of a
251
species in mixture is generally correlated with its performance in monoculture, but the relation
252
depends on the nutrient level, is often markedly non-linear, and both the relative performance in
253
monoculture and the difference from that in mixture can change with nutrient level. Pickett and
254
Bazzaz (1978) grew six species in along an experimental soil moisture gradient in a greenhouse, in
255
monoculture and in a 6-species mixture. The optimum stayed at the same state out of six for four
256
of the species, but it was for most species sharper in the mixture. Fascinating results came from
257
Wilson and Keddy (1985 %851), who examined a field gradient in organic content along a
258
lakeshore. The gradient is probably caused by wave action, and is correlated also with mechanical
259
composition, nutrients and water depth. Twelve of the species were also grown in sand : field-
260
organic mixtures, in pots but out-of-doors. The shape of the response to the gradient, field versus
261
experimental, was:
262

Not or hardly related: 5 species.
263

The opposite: 3 species.
264

Related or vaguely related: 4 species. (The response was sharper in the field in one of
265
these, less sharp in another, equal in a third, and the relation was too vague to see in the
266
fourth.)
267
One possibility is that weaker competitors are pushed towards the less favourable end of the
268
gradient. This can be seen in the work of Pickett and Bazzaz (1978), where one of the two species
269
most suppressed by competition, Polygonum pensylvanicum, is pushed in mixed stands to the dry
270
end of the gradient, where overall growth is less. Doesn’t that make the species a stress-
271
competitor? This seems to be the situation for Spartina alterniflora in the work of Pennings et al.
272
(2004), the species being restricted by competition to the lower marsh. But can we generalise? No,
273
not yet anyway.
274
Alpha niche
275
Niche shift (including 'Habitat shift') is a change in mean/modal resource usage by a single
276
species in different areas (Schoener 1986). Niche shifts are the difference between fundamental
277
and realised alpha niches, or between realised niches with different associates. Such differences
278
have long been recognised (Gleason 1917). Niche expansion is a similar concept, except that the
279
niche width changes, not the mean/mode. There is disagreement in the literature, sometimes even
280
within one paper, as to whether these responses are plastic/behavioural or genetic.
Wilson and Agnew, chapter 5, Assembly rules, page 10 of 53
281
Silvertown (1983) investigated whether the depths of species in limestone pavement grykes
282
were more different when they co-occurred (sympatry) than when they were alone (allopatry) - a
283
test for niche shifts. However, he found the species occurred at more similar depths when in
284
sympatry. Presumably any niche shifts were obscured by differences between grykes, e.g. species
285
can occur deeper in large grykes. Veresoglou and Fitter (1984) suggested that when Holcus
286
lanatus was growing with certain species (their Area III), its nutrient uptake peaked earlier than in
287
other communities. However, this was true for only one of the two nutrients they examined. Even
288
then, that Area III could have been different in other ways. Niche shift has been found in rooting
289
depth. Nobel (1997) found that rooting depths for the three co-dominant species in a site in the
290
Sonoran Desert were 9-10 cm for isolated plants, but roots for interspecific pairs in close
291
proximity averaged 2-3 cm more shallow for Agave deserti and 2-3 cm deeper for the other two
292
species. And O'Brien et al. (1967) found that the depth from which three grasses took up
293
phosphate depended on their mixture with each other, and this can be related to the ability of those
294
mixtures to overyield (Whittington and O'Brien 1968).
295
4 Species sorting
296
4.1 Species associations in succession
297
Greig-Smith (1952) suggested that species associations would change through succession,
298
and Wilson and Gitay (1995 %775) synthesised these suggestions with the terms of Watt (1947) to
299
suggest three phases in succession:
300
301
302
303
304
1. Pioneer: Initially colonisation will be essentially at random, with weak associations between
species, those tending to be negative.
2. Building: As dispersal removes the effects of chance dispersal, some positive and negative
association will appear due to micro-habitat sorting.
3. Mature: Species will sort themselves by micro-habitat and assembly rules, especially at a
305
larger scale, giving stronger associations, with negative ones predominating if different
306
communities have approximately equal species richness.
307
Wilson and Gitay (1995 %775) analysed tussock grasslands with a known time of secondary
308
succession since burning. The expected pattern was seen – association was low and rather negative
309
for the first 10 years, close to zero (negative and positive associations balancing) at 10-20 years,
310
and more negative beyond 20 years. The processes seem likely, but the model was probably
311
subsequent to the data. The model was confirmed in a restoratation experiment at Monks Wood,
312
England, where over 13 yr rank consistency (Watkins and Wilson 1994 %91) increased during the
313
pioneer phase, was maximal in the Building phase and then decreased markedly in the Mature
Wilson and Agnew, chapter 5, Assembly rules, page 11 of 53
314
phase. An identical but non-significant trend was seen over the 6 yr of a restoration experiment
315
elsewhere in England.
316
Greig-Smith (1952) in Trinidad tropical rain forest found evidence for the Pioneer and
317
Building phase, but there was little indication of non-random distribution in 1.5 × 1.5 m plots.
318
O'Connor and Aarssen (1987), in Ontario sand quarries of various ages, expected to see what we
319
have called the Mature phase developing, but in fact the frequency of negative species associations
320
decreased with time. Malanson (1983) approached this question differently: vegetation patches on
321
canyon walls in Utah should show greater dissimilarities if they were safe from floods, and the
322
species had time to assemble into communities, but if anything the opposite was true.
323
Aarssen and Turkington (1985 JE, p 585) comparing three pastures of different age in
324
western Canada. claimed consistently stronger and more negative associations between grass
325
species in the older pastures, though the relevant information presented shows that the total
326
number of significant associations (positive plus negative) is lower in the oldest pasture. They do
327
give figures to demonstrate that the number of associations were more consistent over seasons and
328
years in both direction and significance in older pastures. Turkington and Mehrhoff (1990)
329
interpret this as “transition from an essentially unorganised assemblage of species to a more
330
organised community”.
331
This approach is potentially interesting. No investigation so far has given any indication of
332
deterministic structure; most results have been opposite to theoretical expectation. However, there
333
seems to be only weak theoretical support for the concepts in the first place.
334
4.2 Compositional convergence
335
It is fascinating to see how similar species assembly was in identical conditions. We can
336
never do this, but Crawley et al. (1999) approached this situation by sowing a mixture of 80 forbs
337
into six replicate blocks in an experimental field. After seven years, Tanacetum vulgare (tansy*)
338
predominated among the sown species, varying across five of the blocks from 9.7 % to 72.1 % of
339
the standing crop – a wide range – with no other species reaching 0.1 %. In the sixth block, T.
340
vulgare comprised only 0.1 % and other species exceeded it, with Achillea millefolium (yarrow)
341
2.5% and Cirsium vulgare (spear thistle) 1.0%. There is no convergence here. Amongst the
342
volunteers the most abundant was Alopecurus pratensis (foxtail) varying 0 – 86 %, then Holcus
343
lanatus (Yorkshire fog) with a 64 % maximum but absent from four of the six blocks and
344
Arrhenatherum elatius (oat grass) varying 0 – 31 %. Again, huge ranges were found. Crawley et al
345
describe this as “quite remarkable degree of similarity”, but we would describe it as remarkable
346
dissimilarity amongst plots that had been made as similar as possible. Crawley et al. (1999)
Wilson and Agnew, chapter 5, Assembly rules, page 12 of 53
347
describe the blocks as being remarkably similar in species diversity, but in fact species richness is
348
three times greater than expected at random, and significantly so.
349
4.3 Transitivity in interference networks
350
The terms ‘transitivity’ and ‘interference network’ hide a rather simple question: is there a
351
‘pecking order’ so that it is possible to arrange a set of species (or genotypes) so that one higher in
352
the order can always competitively exclude one lower down? If a pecking order exists, the
353
interactions between species is said to be transitive (this vol., chapt. 5, sect. 4.3).
354
Simple questions do not always lead to simple answers. First, we note that the question can
355
be asked only in one environment, for competitive abilities will change with the environment.
356
There have been doubts whether relative competitive abilities between species change with
357
environment, but it does (Keddy et al. 2000 %413; Fynn et al. 2005), and clearly it must. That is
358
the main reason we get different vegetation in different places. More basically, if we talk merely of
359
competitive ability, the answer can depend on the proportions of the two species, and they will
360
change as competition proceeds. The trouble is that the species that has the higher relative growth
361
rate will be the one that dominates the mixture, but the relative RGRs of the two species will
362
change as the proportions of the two species in the mixture changes. Therefore, since we want to
363
know the eventual result, we have to talk in terms of competitive exclusion. Yet we know that for
364
a variety of reasons (Chapter 4) competitive exclusion does not always occur. For these cases, the
365
question of transitivity cannot be asked.
366
But is this the way to determine competitive ability? Several studies have used
367
comparisons of species’ performance in mixture with those in monoculture, but Connolly (1997)
368
pointed logical flaws. Correction can be made for the “size-bias”, but the basic error has been
369
comparison with a monoculture. Many studies conclude that if species A grows more slowly in
370
mixture than in its monoculture whilst species B grows faster in mixture than in its monoculture, B
371
has the higher competitive ability. Yet Connolly’s table (5.1), over the undefined period of his
372
artificial data and assuming a starting biomass of 1, gives an example where A does worse in
373
mixture than in monoculture, and B does better in mixture than in monoculture. Still, A has the
374
faster growth rate in mixture (loge 2.77 – loge 1 = 1.02) than Species B (loge 2.71 – loge 1 = 1.00),
375
and will come to exclude its competitor from the mixture (subject to the conditions mentioned
376
above). If B goes extinct it can hardly be said to have the higher competitive ability. (We use the
377
term ‘competitive ability’ since it is usual in the literature, but since we rarely know the
378
mechanism the more general ‘interference ability’ should strictly be used.)
Wilson and Agnew, chapter 5, Assembly rules, page 13 of 53
Table 5.1. Which species has the higher competitive ability? The
starting biomass for both species was 1.00
Species
379
Biomass in
Biomass
monoculture
in mixture
A
3.00
decrease
2.77 Winner in mixture
B
2.64
increase
2.71
It turns out that what is essential in designing such an experiment is not the monocultures,
380
as many people had thought, but two harvests so RGR can be calculated. This invalidates almost
381
all the studies of transitivity done so far. So all we have to do is to wait, perhaps for close to
382
infinite time, and see which species has the higher growth rate as the mixture approaches
383
competitive exclusion. This is coming to be one off those community ecology questions that are
384
impossible to answer.
385
At the moment, it is interesting to look at the imperfect evidence we do have. Buss and
386
Jackson (1979) claimed several competitive cycles for coral reef sedentary organisms, as seen in
387
static evidence for overtopping. Likewise, Russ (1982) claimed non-transitive relations between
388
species in the overgrowth of sedentary marine organisms observed colonising experimental plastic
389
sheets in the sea in Australia, though no cycle can be made out of his results.
390
Turning to pure plant work, Mouquet et al. (2004 %77) grew eight meadow herbs species
391
in replacement competition in all possible pairs. Using relative yield (biomass in mixture / biomass
392
in monoculture), if the species form a transitive hierarchy it should be possible to arrange them so
393
if species i is further up the hierarchy than species j, and RYi,j is the relative yield of species i
394
when growing with species j, RYi,j-RYj,i is always positive. In his experiment, at both low and
395
high density it almost is, with a very similar order (Table 5.2).
396
Table 5.1. Competitive hierarchy from Mouquet et al. (2004), strong competitors at the top
High_density
Holcus_lanatus
Rumex_acetosella
Cerastium_glomeratum
Anthoxanthum_odoratum
Festuca_rubra
Arabidopsis_thaliana
Lamium_pupureum
Veronica_arvensis
397
Low_density
Holcus_lanatus
Rumex_acetosella
Cerastium_glomeratum
Anthoxanthum_odoratum
Festuca_rubra
Lamium_pupureum
Arabidopsis_thaliana
Veronica_arvensis
Wilson and Agnew, chapter 5, Assembly rules, page 14 of 53
398
At each density, there is one negative RY1,2-RY2,1 indicting a conflict with the hierarchy, and it is
399
between species not contiguous in the hierarchy, but of size -0.05 or -0.06 which is clearly within
400
the experimental error.
401
A study that returned a clear answer to the question of transitivity is that of Roxburgh and
402
Wilson (2000 %395). It relates to a particular community, since the seven species used in the
403
competition experiment were taken from that community, the University of Otago Botany Lawn,
404
grown in lawn soil in boxes placed near the lawn. The use of 10 replicates in careful experimental
405
conditions allowed significance tests. The seven species could be arranged in a hierarchy to which
406
all significant competitive relations conformed, i.e. if species X is higher in the hierarchy and
407
species Y lower, then the suppressive effect of X on Y is greater than that of Y on X. In fact,
408
relations between all pairs of species, significant or not, were compatible with the hierarchy.
409
410
411
Fig. 5.5: Competitive relations in seven species from the University of Otago Botany Lawn
The experimental design of Keddy et al. (1998 %777) comprised planting a number of
412
‘wetland’ species into a number of swards of wetland species. They report results for 18 species
413
planted into five swards. The 18 species tended to respond similarly to different swards, e.g.
414
Kendal’s coefficient of concordance took a rank of 0.7 (1.0 = complete agreement as to which
415
target suffered more/less), highly significant. Some of the variation in invader/sward combinations
416
could be due to experimental error (no replication was possible), but some results are impressive,
417
e.g. the rank of Carex crinita varied from 14 to 17 across the 5 swards (18=suppressed most), and
418
Lythrum salicaria varied from 4 to 7 (1= suppressed least).
Wilson and Agnew, chapter 5, Assembly rules, page 15 of 53
419
In a different approach, Silvertown et al. (1992) used data from an experiment where
420
several species had been planted in adjacent hexagons, and invasion between hexagons recorded.
421
Examining the difference between the invasion of Species A into Species B and that of Species B
422
into Species A, replacement rates could be calculated, and the interference hierarchy.
Agrostis stolonifera
Holcus lanatus
A
Poa trivialis
D
B
C
Lolium perenne
Cynosurus cristatus
423
424
Key:
Strong (> 0.2) difference in invasion rates
425
Weak-moderate difference in invasion rates
426
Invasion rates equal (i.e. no net invasion)
427
Fig. 5.5. The competitive hierarchy from invasion rates in data of Silvertown et al. (1992).
428
A pecking order can be formed from these results, with no discrepancies (though L.
429
perenne and C. cristatus could equally well exchange positions)—yes, for a very specific set of
430
conditions… if the conditions are changed regularly, perhaps then the factors that contribute to the
431
formation of the pecking order may be easier to determine.. There are qualitative discrepancies,
432
e.g. since H. lanatus can invade P. trivialis (A) and P. trivialis can strongly invade L. perenne (B),
433
we would expect that H. lanatus would be able to invade L. perenne even more strongly, but in
434
fact their invasion rates are exactly balanced (C). Moreover, although the species A. stolonifera at
435
the top of the order can invade C. cristatus at the bottom, the rate of replacement is less than for
436
other pairs (D).
437
In a similar experiment Silvertown et al. (1994) used only four species, so there was less
438
opportunity for intransitivity, but in any case there was none in any of the four grazing treatments
439
(Table 5.3).
440
Table 5.3: Competitive hierarchy of four species in four treatments in Silvertown et al. (1994).
Summer sward
grazing height
Winter and
spring
Invasion ability: greater → lesser
3 cm
Grazed
Lolium perenne → Festuca rubra → Schedonorus phoenix → Poa pratensis
3 cm
Ungrazed
Festuca rubra → Lolium perenne → Poa pratensis → Schedonorus phoenix
Wilson and Agnew, chapter 5, Assembly rules, page 16 of 53
9 cm
Grazed
Festuca rubra → Lolium perenne → Schedonorus phoenix → Poa pratensis
9 cm
Ungrazed
Lolium perenne → Festuca rubra → Poa pratensis → Schedonorus phoenix
441
442
It’s interesting to wonder what ecological processes would give rise to intransitivity (Fig. 6).
(1)
A is taller
than B and
shades it
out
C produces an
allelopathic
chemical,
toxic to A
C
(3)
C is shadetolerant, and
scavenges
nitrogen
C
443
444
(2)
A
A
o
Tree A is taller
than shrub B
and
shades it
out
grass C lowers the
temperature, and
suppresses
seedlings of A
B
B is taller
than C
A
B
C
shrub B shades out grass
C, and is not affected by
lower temperature
(4)
A is taller
than B and
shades it
out
B is taller
than C and
fixes N
B
A
C with A is taller
than it, and
shades it out
C
A is taller
than B and
shades it
out
B with C is taller
than it, and shades
it out
B
Fig. 5.6. Possible causes of intransitivity between three species: A, B and C.
445
In scenario ‘1’, we use an allelopathic chemical produced only by C and toxic only to A.
446
This works, but species-specific allelopathy is rather like Getafix’s magic potions in the Asterix
447
books: it can perform/explain any wonder. Scenario ‘2’ is similar, except that the third factor is
448
lower temperature (Ball et al. 2002) rather than a toxin. In ‘3’, we have to ask why C can suppress
449
A; presumably the shade-tolerance of C minimises the competition for light, so competition for N
450
becomes important, and C has the lower Tilman R*. Why cannot C suppress B? Perhaps because it
451
is shorter and so cannot compete for light, and its low R* for N does not help because B can fix N.
452
Does this work? Probably. In all three cases, not all pairs are interfering using the same
453
resource/factor. Could we envisage a 3-species solution using competition for light (‘4’)? How can
454
we have heights of A>B, B>C and C>A? We can have such magic by means of differential
455
plasticity: in this case probably by red:far-red effects (this vol., chapt. 2, sect. 1.9). However, we
456
are again introducing a second factor: light spectrum in addition to light intensity. All this it is
457
rather convoluted, which suggests that intransitivity will not be the norm.
Wilson and Agnew, chapter 5, Assembly rules, page 17 of 53
458
We conclude that intransitivity could occur, but the evidence from competition and
459
invasion experiments is that it is uncommon and has not been observed in plants. In retrospect we
460
should have expected that, because we had not thought what mechanisms would cause it.
461
5 Richness
462
A basic question in community ecology is whether there is a limit to the number of species
463
that can be packed locally. Testing for low variance in species richness (Wilson et al. 1997) is a
464
direct approach to this. If the niches are not primarily defined by the species themselves (this vol.,
465
chapt. 1, sect. 4.1) there will be a limit to the number of niches in a type of community. Since there
466
can be no more species present in a sample (quadrat) than there are niches, the number of species
467
in a quadrat should be limited by the number of niches, and be rather constant across quadrats. To
468
be precise, there should be lower variance in the species richness of quadrats than would be
469
expected under a null model in which the number of occurrences of each species is held at that
470
observed, but those occurrences are scattered across the quadrats, independently of other species.
471
It is often difficult to see such an effect because of overlain environmental variation, and perhaps
472
disturbances and the presence of empty niches. Possibly for this reason Wilson et al. (1987 %391)
473
failed to show variance in two communities at scales of 5 × 5 m and 2 × 2 cm respectively, and
474
Wilson and Sykes (1988) at 10 × 10 m. However, Watkins and Wilson (1992) found lower
475
variance than expected under the null model at the scale of 13 × 13 mm, and this remained for six
476
of the 12 lawns when analysed with a patch model. There may be remaining doubts that the
477
limitation at this scale is due to geometric packing of individuals, but this remains a basic question.
478
In other approaches to niche limitation, Levine (2001 %397), by sowing seeds of a variety
479
of native and exotic plant species into tussocks of Carex nudata found even the most diverse
480
tussocks were colonized, and concluded that they had not been completely saturated with species,
481
but then some species might not have persisted in the long term, and Wilson (1961) concluded that
482
most or the ant faunas of the Moluccas-Melanesian are saturated, using as evidence a close
483
correlation between the size of the fauna and the area of the island. Cornell and Lawton (1992)
484
suggested that it would be possible to identify niche saturation from the relation between local and
485
regional richness. If there be niche saturation, then as the regional species pool increases, local
486
richness will increase proportionally at first, but level off to a maximum. If there is no saturation,
487
the relation will continue to be linear. It is easy to show in models of community assembly that
488
saturation will occur (Fukami 2004 %137). But will it in the real world? Although it is easy to
489
determine richness at the site level, the estimation of regional species pools involves too many
490
arbitrary and subjective decisions. There is also a problem that local richness is affected by the
491
regional richness, as Cornell and Lawton reasoned, but regional richness is a combination of local
Wilson and Agnew, chapter 5, Assembly rules, page 18 of 53
492
(alpha) richness and beta richness, and so not independent of it. Wilson and Anderson (2001)
493
concluded that comparisons between habitats are not appropriate because of non-independence due
494
to overlap of species and because of ecological non-comparability between habitats. Only
495
comparisons between equivalent habitats on different continents are valid, and they cannot be
496
made because there are too few continents for a statistical analysis. A wooden light bulb is
497
beautiful and interesting but of little use (Wilson and Anderson 2001); likewise the species pool
498
concept is stimulating but it is probably operationally impossible to test.
499
There have also been simple comparisons between different continents in the florule size
500
and quadrat species richness of. As Orians and Paine (1983) say: “Implicit in community
501
convergence in species richness patterns is the notion that assemblages eventually reach some
502
saturation level”. However, such comparisons have generally found the areas compared to differ in
503
richness at both area and quadrat scales, e.g. annual grassland California and Chile by Gulmon
504
(1977), in the brown intertidal algae in various points around the Atlantic, Pacific and Southern
505
oceans by Orians and Paine (1983), California and Israel by Shmida (1981). Richness convergence
506
would have implied niche saturation; divergence does not disprove saturation, because the habitats
507
may not be as similar as we hope, or there might be niche straddling/splitting, we cannot tell.
508
Robert H. Whittaker travelled the world recording species diversity in a standard way, and in plots
509
whose exact location was carefully selected (JBW, pers. obs.), attempting to find patterns and thus
510
predictability. In Whittaker (1977 %1) he had reached the conclusion, which he put in a more
511
straightforward way in seminars: “We once thought species diversity was the one fixed,
512
predictable feature of plant communities. But it isn’t”.
513
6 Limiting similarity
514
Abrams (1990) assumed that if two species were too similar in resource-use patterns one
515
would be excluded. This is a present-day reassertion of the Principle of Gause (1936), that species
516
that are too similar will tend not to occur together. The concept has also been referred to as
517
‘community-wide character displacement’ or ‘ecological character displacement’ (Strong et al.
518
1979). Hutchinson (1959) instigated this topic, as he instigated so much in ecology, by observing
519
that in some mammals and birds of Britain, Iran and the Galapagos Islands the morphological size
520
ratio between each species and the next larger one was about 1 : 1.3 for a linear measure. He
521
actually reported a range of 1:1.1 to 1:1.4, but this has usually been forgotten. Hutchinson implied
522
that this is partly due to within-species character displacement, and there is some evidence for that
523
in his data. So far as we know this has not been applied to plants. MacArthur and Levins (1967)
524
put this idea that there should be a limiting similarity between the niches of co-existing species on
525
a solid mathematical foundation, be it with some assumptions.
Wilson and Agnew, chapter 5, Assembly rules, page 19 of 53
526
The quantitative predictions of the MacArthur and Levins theory has not been tested, but
527
even qualitative testing has been difficult. It is even difficult to know what the test statistic to use –
528
e.g. minimum distance, even distances, greater range – or which characters are appropriate (Stubbs
529
and Wilson 2005). It is usually unclear what we are trying to test: plastic responses, competitive
530
exclusion between species, character displacement or the co-evolution of species. Hubbell (2005
531
% 166) concluded: “The empirical evidence, in general, has not borne out these [MacArthur and
532
Levins, etc.] predictions …, particularly in plant communities”, and further “Does a limiting niche
533
similarity for species in functional groups exist? … I believe the answer to [this] question is no (at
534
least in plants)” (op. cit.). We wish to look further, and with plants at that.
535
Terminology has been a problem. When co-occurring species are closer in character space
536
(i.e. more similar) than expected, the terms used have included ‘clumped’ and ‘aggregated’; when
537
they are less similar terms have been ‘evenly-spread’, ‘evenly-spaced’, ‘spaced-out’, ‘staggered’
538
and ‘regular’. These terms are self-explanatory. ‘Overdispersed’ and ‘underdispersed’ and have
539
also been used. This is unfortunate because overdispersed is the mathematical term for clumped
540
and underdispersed for evenly-spread (Greig-Smith 1983). For obvious reasons undergraduates
541
often use them in the opposite, incorrect, senses, and this can be found even in the literature (e.g.
542
Weiher et al. 1988). They are therefore ambiguous in usage, and are best avoided.
543
As elsewhere, we generally have to take species as units, ignoring polyploids, other within-
544
species variation, within-plant somatic variation and generally dioecy.
545
6.1 Limiting similarity in morphological characters
546
Cody (1986) reported a number of pieces of evidence for limiting similarity amongst
547
woody plants of desert and South African fynbos. In the Granite Mountains, Mojave Desert,
548
California, he demonstrated that the Opuntia species, which are shallow-rooted, are negatively
549
associated, but Yucca schidigera, which is somewhat deeper-rooted, was positively associated with
550
all the Opuntia spp. For four fynbos sites, he showed spreading-out of species of the major
551
Proteaceae shrubs in morphology space of leaf shape and leaf length, with little overlap between
552
species. Positions in morphological space were occupied by different species in different sites, and
553
the position of some species changed between sites, both making the spread that was observed
554
even more notable. However, no probabilistic test against a null model was made and a null model
555
would probably not be easy to frame, but the patterns are compelling. The one exception to the
556
morphological sorting was between Protea eximia and P. nitida and they occurred in different
557
aspect micro-habitats—Are we sure that all other variation isn’t also due to differences in micro-
558
habitats, instead of competitive exclusion due to similarities in traits?. Most remarkably, in some
559
species, notably Leucadendron salignum, plants of the two sexes overlapped considerably on each
Wilson and Agnew, chapter 5, Assembly rules, page 20 of 53
560
of the axes, yet were largely separate in the 2-dimensional morphological space. For
561
Leucadendron, Cody offers evidence that species pairs that are more similar in the 2-D space co-
562
occur less often than expected at random. He also found indication that the 80 species of
563
Leucadendron in Cape Province, South Africa, were more spread in morphological space than
564
expected by chance, but with only 20 randomisations the probability cannot be accurately
565
determined, and details of the null model are not clear, especially the treatment of the edges of
566
morphological space. Cody’s work is fascinating, and it would be wonderful for some of these
567
leads to be followed up in more detail.
568
In careful work, Armbruster (1986) examined the association of Dalechampia species at 12
569
sites in Central and northern South America with unique combinations of Dalechampia species
570
(reduced from 26 populations observed in the field). In the ecological sorting (“pure assemblage”)
571
null model, the Dalechampia species richness of each site was fixed at that observed, and the
572
species frequencies, whilst not so fixed, were taken as probabilities of occurrence. As with most
573
assembly rule work, environmental differences between sites are potentially confounding, no less
574
and probably no more so than in work on a micro scale. Armbruster coped with this by using five
575
different species pools taking into account climatic and geographical ranges. In effect this is a
576
patch model on a grand scale. The test statistic was the number of cases where two species that
577
were similar in pollinator usage co-occurred (within 50 m) at a site, pollination vectors being
578
determined by observation and flower morphology. After this careful work p was 0.16, not
579
significant. Twelve sites are really too few for a good test. Another model, with character
580
displacement, does not strictly concern us here since we are limiting ourselves to ecological
581
assembly, eschewing ecotypic differentiation, but the results were significant, though only using a
582
1-tailed test which is debatable. A decade later, Armbruster et al. (1994) performed a similar study
583
on Stylidium species at 25 sites in Western Australia: another genus with complicated floral
584
organs. The test statistic was overlap in the morphological similarity in the flowers of species co-
585
occurring at a site, and again there was a large-scale patch model based on habitat and geography.
586
Only one site overlap was observed, compared to an average of 4.38 expected under the null
587
model, but this result was not significant (p = 0.055, but perhaps we should double this to 0.11 for
588
a 2-tailed test). Again there was significant character displacement.
589
Weiher et al. (1988) tested for limiting similarity in herbaceous riverside vegetation, with
590
quadrats placed to deliberately give a range in environment (soil fertility and disturbance) and
591
vegetation (“from cattail marshes, to wet sedge meadows to sandy beaches”), measuring 11
592
vegetative characters. They found a significant tendency for the minimum nearest-neighbour
593
distance in 11-character space to be greater than expected under their null model, though other test
594
statistics did not give significance. Four of the individual characters showed even spreading. They
Wilson and Agnew, chapter 5, Assembly rules, page 21 of 53
595
concluded that there are morphological assembly rules that constrain wetland plant community
596
composition. The main problem with this work is that there was no attempt to sample under
597
environmental heterogeneity, or to allow for such heterogeneity in the analysis by a patch model or
598
the like, so the null model they used combined species from several species pools. This means that
599
the departures of the observed data from their null model are likely to reflect species habitat
600
preferences, rather than community structure resulting from limiting similarity, as discussed
601
above. To put it another way, there was pseudo-replication of the habitat differences. The study
602
was a brave attempt, but one that illustrates the traps that await those who are less careful than
603
Armbruster was.
604
Stubbs and Wilson (2005) attempted to avoid previous traps when they tested for limiting
605
similarity in a New Zealand sand-dune community. Twenty three functional characters were
606
measured on each of the species, covering the morphology of the shoot and root systems and
607
nutrient status, and intended to represent modes of resource acquisition. Since it is not clear at
608
what scale limiting similarity would occur, sampling was at four spatial scales, from a single point
609
up to a scale of 50 m2. These multiple scales allowed patch models to be used. A carefully-selected
610
range of test statistics was used, for example excluding any that were affected by the range of
611
character values. A test over all characters found that the mean dissimilarity between nearest-
612
neighbour species in functional space, and the minimum dissimilarity, were greater than expected
613
under the null model at the 0.5 × 0.5 m scale, supporting the MacArthur and Levins limiting
614
similarity concept. However, the actual community did not follow the theory to the extent of
615
showing an even spread of species in functional space. Limiting similarity effects were seen even
616
more consistently in separate root and leaf characters when within-species variation was taken into
617
account to calculate measures of overlap – the test most closely aligned to MacArthur and Levins’
618
original theory. The characters showing limiting similarity were mainly those related to rooting
619
patterns and leaf water control, and thus probably reflected the acquisition of nutrients and / or
620
water. The implication that competition for water and nutrients limit coexistence seems reasonable
621
for a sand-dune. The main problem with this work is the number of tests made – four spatial
622
scales, 23 characters and different test statistics. This seems inevitable when analysis of limiting
623
similarity in plant communities is in its early stages and we do not know at what scales, in what
624
characters and how it will operate, but the overall results are convincing.
625
Armbruster (1995) suggested that limiting similarity due to ecological sorting would
626
operate more readily in vegetative characters than in reproductive ones, and comparison of his own
627
ecological-sorting results with the results of Cody and of Stubbs and Wilson supports this. Hubbell
628
(2005 %166) was too dismissive. Limiting similarity exists in plant communities and can be
629
demonstrated.
Wilson and Agnew, chapter 5, Assembly rules, page 22 of 53
630
6.2 Limiting similarity in phenology
631
The simplicity of time as a niche axis has led to several attempts to ask the question – are
632
the flowering times of the species in a community evenly-spread? That is, is there a constraint on
633
the phenology of species which can co-occur? In such work, either the position of species
634
flowering peaks can be compared, or the time span of flowering, or quantitative measures such as
635
the number of flowers open at any time. The selective pressures against species too similar in
636
flowering time would come from several interactions discussed in chapter 2, such as competition
637
for pollinators/dispersers, pollen wastage, interference on the stigma and mal-adapted hybrids. On
638
the other hand, aggregation could be an adaptation to attract pollinators/dispersers, to combat
639
predators, or a response to pollinator/disperser availability (Thompson and Willson 1979).
640
Investigation was sparked when Stiles (1977) claimed to find evenly-spread flowering for
641
hummingbird-pollinated plants in a Costa Rican tropical forest. Statistical analysis of this dataset,
642
and of such datasets in general, has proved difficult and controversial; an excellent summary is
643
given by Gotelli and Graves (1996). In general the more recent studies use appropriate
644
randomisation tests, and are valid. Similar tests have been made for an even-spread of fruiting.
645
Ashton et al. (1988), examining the six species of Shorea section Mutica in tropical rain
646
forest in Malaya, found even spread “at the 4.6% confidence level”, but it is not clear whether this
647
was a 2-tailed test. Wright and Calderon (1995) tested separately 59 genera from Barro Colorado
648
Island. Flowering times were aggregated in some genera, but evenly-spread in six genera, so far as
649
one can tell converting the two 1-tailed tests into a 2-tailed one and within the limited number of
650
randomisations used. Thies and Kalko (2004) found that eight forest Piper species flowered within
651
a short period and at random within that, but fruiting was evenly-spread. The p values were not
652
adjusted to give a 2-tailed test, though the results may have been significant anyway, again with
653
few randomisations. Burns (2005), in 10 woody angiosperms common below the canopy of conifer
654
forest in an area of British Columbia, Canada, found no evidence for significantly even spread of
655
fruiting times. Poulin et al. (1999) examined fruiting phenology in central America. Data for the
656
fruiting times of Miconia (Melastomataceae) species from Barro Colorado Island were not
657
significantly different from a null model, but those from the genus in Trinidad and Columbia
658
showed significantly even fruiting times, though again with few randomisations. In Psychotria
659
(Rubiaceae), fruiting times were aggregated. Overall conclusions are difficult, especially with the
660
danger that non-significant results or aggregation are under-reported, but it seems that even
661
spreading sometimes occurs.
662
663
Not all niche differences in pollination are via phenology, and interesting conclusions can
be made bringing in other information. Pleasants (1980) calculated from flowering-time overlap
Wilson and Agnew, chapter 5, Assembly rules, page 23 of 53
664
and flower densities the potential for competition for pollinators between bumblebee pollinated
665
species in some Rocky Mountain Meadow species; he found that such competition was negatively
666
correlated with presence/absence association between the species.
667
668
There are major problems with all such studies:
a. It is difficult to know whether to compare overlap between the most similar neighbours, or
669
overlap between all possible pairs of species (Pleasants 1990). Probably species are affected
670
by the cumulative competitive pressure from several, but not all, species.
671
b. Flowering times are usually aggregated on a seasonal scale. In temperate areas, few species
672
flower in winter, but there is normally aggregation in the tropics too, corresponding to
673
wet/dry seasons (Stiles 1979; Wright and van Schaik 1994). There can be up to three peaks
674
per year (Parrish and Bazzaz 1979). It is very difficult to demonstrate even spread when it is
675
laid over aggregation.
676
c.
Even within the flowering season (or within a clump), there is usually variation, with fewer
677
species flowering at the beginning and end. Although it would be possible to estimate this
678
variation from the data, incorporation of it in a null model would involve circular reasoning.
679
This problem is probably insoluble.
680
d. There will probably also be variation in pollinator availability, so pollination competition
681
will be more intense at the two ends of the season with few insects (e.g.) around. This will
682
actually tend to mitigate problem ‘c’ above.
683
e. The patterns in flowering/fruiting could be caused by any of four processes: (1) ecological
684
assembly by competitive exclusion between pre-adapted species (i.e. ecological sorting), (2)
685
coevolution of species, (3) evolution of co-adapted ecotypes within species (i.e. character
686
displacement), or (4) plastic responses (i.e. niche shift). Rarely is it clear which process
687
workers have been intending to test. Most recent studies have been based on in-situ
688
observations of phenology. Although this sounds commendable, it would actually be
689
preferable to use data on the species generally, even from deliberately outside the area, in
690
order to exclude ‘3’ and ‘4’, and narrow the possible explanations. Co-evolution of species
691
(‘2’) seems unlikely here because most species occur in several different communities, with
692
different neighbouring species, and could not adapt their flowering times to each
693
community. Ecotypic differentiation (‘3’) would be difficult when species associations are
694
constantly changing. Plasticity, (‘4’), at sight unlikely, is possible since fruit removal from a
695
plant often causes its flowering period to be extended. Armbruster-type (1986; Armbruster
696
et al. 1994) analysis with multiple null models would be needed to distinguish between these
697
possibilities.
Wilson and Agnew, chapter 5, Assembly rules, page 24 of 53
698
699
f. Relative flowering time may not be consistent from year to year, because species are
responding to different signals (Rathcke and Lacey 1985).
700
Vegetative phenology might also constrain the coexistence of species. For example, Parrish and
701
Bazzaz (1976) commented that among the six oldfield species they examined only one pair was
702
similar in the time of peak root growth. Comparison with a null model would have been useful.
703
Veresoglou and Fitter (1984) found differences in vegetative phenology (growth and nutrient
704
uptake) between co-occurring grasses, suggesting that this helped permit coexistence between
705
them, but again they compared with no null model. Rogers (1983) examined sorting of species by
706
vegetative phenology amongst the vernal guild of herbs in North American deciduous forest.
707
Effects of environment producing negative correlations were potentially removed by excluding
708
species pairs with negative correlations at a larger scale (50 × 100 cm), though in fact none were
709
found, an approach conceptually related to the method of Dale (1985). Associations between
710
species in the same guild (ephemeroid, summergreen, annual) were no more or less frequent than
711
between species in different guilds.
712
This is an interesting approach to community structure. It is mainly restricted by the
713
difficulties in specifying a null model in which the test focuses on possible assembly rules. Some
714
evidence for such rules has emerged.
715
Cody and Prigge (2003) made the curious observation that individual shrubs of Quercus
716
cornelius-mulleri affect each others' phenology of leaf replacement. Late and early timing
717
alternated annually within individuals and between large or close individuals in space. The authors
718
proposed that these phenomena could be due to resource depletion or the cost of early bud break.
719
Cody and Prigge do not suggest how the fitness of individuals may be affected. This is an
720
interesting case which could be considered as either facilitation or interference, but is undoubtedly
721
an environmental disturbance of the individual shrub. It seems a sort of assembly rule, but it is
722
difficult to know how to characterize it.
723
7 Guild proportionality
724
7.1 Concept
725
726
Guild proportionality is based on the concept of Pianka (1980): species that are in the same
alpha guild will tend to exclude each other. The process would be:
727
1. Species arrive at a point and some establish (cf. chapt 1, sects. 2.3-2.5: the challenge).
728
2. A further species arrives:
729
2a. The species may fail to establish. Failure is more likely if the new species is similar in
730
resource use to the majority of the species already present, i.e. it is a member of the
731
same alpha guild (Fig. 5.7), or
Wilson and Agnew, chapter 5, Assembly rules, page 25 of 53
732
2b. If the new species does establish, and species previously present are excluded, the
733
excluded species are more likely to be from the same alpha guild as the newly-
734
established species.
735
Fig. 5.7:
736
Note that dimensions are necessarily small here to allow constant possibility of challenge. This is
737
not island biogeography, but micromovement within a sward, heath or forest patch.
738
The result would be a tendency towards a relative constancy in the proportion of species
739
from each of the guilds - 'guild proportionality' (Wilson 1989). Not exact constancy in the real
740
world, but less variation than in a null model, and the appropriate null model here is one that holds
741
both quadrat richnesses and species frequencies equal to those observed. The finding of guild
742
proportionality would mean: (1) there is constraint on species presence, and (2) it is at least
743
partially related to the characters used in the guild classification. These must be alpha guilds since,
744
to quote Pianka (1980), they refer to niche in the “narrow sense of resource utilization”.
745
7.2 Evidence: constancy in space
746
The first application of the concept to a plant community was by Wilson et al. (1989 %263)
747
in a New Zealand rainforest, sampled with quadrats 2 m in diameter. The guilds were synusiae
748
(strata, lianes and epiphytes). The ground and herb strata showed significant guild proportionality
749
when coastal broadleaved forest and Nothofagus forest were combined, which is not ideal and the
750
use of a site model casts doubt on the results. Bycroft et al. (1993) found significant guild
751
proportionality at the scale of 1 × 1 m in the herb stratum of an NZ Nothofagus forest, but only with
752
a site model, not with a patch model. Wilson and Watkins (1994 %591), sampling eleven lawns at a
753
scale of c. 13 × 13 mm and using a 3 × 3 quadrat patch model, found significant guild
754
proportionality between graminoids and forbs in three of the lawns, but only in the more species-rich
755
quadrats as if the limitation did not operate whilst there were empty niches. Wilson and Roxburgh
756
(1994 %267) sought guild proportionality in one of those three lawns, the University of Otago
Wilson and Agnew, chapter 5, Assembly rules, page 26 of 53
757
Botany Lawn, using point quadrats. Again there was a significant guild proportionality using
758
graminoid versus forb guilds. We shall synthesise the Botany Lawn data later.
759
Elsewhere, Klimeš et al. (1995) recorded 30 × 30cm permanent quadrats for five years in
760
two meadow communities, that differed in fertilisation and mowing regimes. There were many
761
cases of guild proportionality using a wide variety of guild classifications and fewer cases of
762
variance excess. Yet, to be frank, plant community structure is often so elusive that we should be
763
cautious when it is found. Using a site model, there could possibly be problems with
764
environmental heterogeneity even within the 1.5 × 1.5 m area, but more worrying is that many of
765
Klimeš et al.’s guilds showing significance were in characters typically of beta-niche
766
differentiation, not characters that represent differences in resource use at one spot (i.e. alpha).
767
Light response could relate to stratification in the community, but how could there be alpha niche
768
differentiation, i.e. at one point, in pH and soil nitrogen? The winter-green guild is more
769
convincing, suggesting phenological guilds, and with that guild there were significant differences
770
in the fertilised meadow in 4 years out of the 5 recorded.
771
Weiher et al. (1998) analysed their rivershore data (see above) for guild proportionality.
772
They reported significant guild proportionality for three guilds, but discounted them after
773
Bonferroni correction. The use of Bonferroni is problematic here, since the tests include
774
complementary guilds, and are thus far from independent. However, the much greater problem is
775
the deliberate combining of different habitats (see above). Wilson and Whittaker (1995) found
776
highly significant guild proportionality for two, though related, a priori guild classifications:
777
narrow versus broad leaves and monocots versus dicots. Wilson and Gitay (1999 %566) found
778
significant guild structure at 10 × 10 cm scale in the inter-tussock vegetation of 21 sites of a New
779
Zealand grassland. Kikvidze et al. (2005) analysed subalpine meadows in Georgia (Caucasus),
780
using 4× 4 cm quadrats. RVgp for the proportion of was 0.64, impressively below the null-model
781
value of 1.0 and highly significant. A site model was used, but the reality of the result was
782
reinforced by a competition experiment, where the yield of a monocot+dicot mixture was greater
783
than for either monocots or dicots alone. Bossuyt et al. (2005) analysed 52 1 ×1 m quadrats, each
784
in a different dune slack in W Belgium and N France, using forb versus graminoid versus shrub
785
guilds. They found highly significant guild proportionality with forbs. The sampling of 52 slacks
786
differing in age from 5 to 45 years makes us worry about environmental artefacts. Using C, S, R
787
they found significant guild proportionality with ruderals. This is difficult to understand. There
788
could well be disturbed patches for ruderals within each 1 × 1 m quadrat, but a proportion more
789
constant than expected at random? How would this arise?
790
We have to be very careful with evidence for guild proportionality, partly because
791
community structure is so elusive, and partly because it is so easy to obtain artefacts from habitat
Wilson and Agnew, chapter 5, Assembly rules, page 27 of 53
792
variation. The danger is that with habitat variation the null model may be inappropriate. In the case
793
of guild proportionality, if A and C in Fig. 5.2 are in one guild and B and D in another, each
794
observed quadrat has guild proportions of 0.5:0.5, with zero variance. If occurrences could be
795
randomised (i.e. with somewhat different quadrat and species totals), we would see highly
796
significant but spurious ‘guild proportionality’, not from species interactions but from
797
environmental control. It is a real result that each environment has one species from each guild,
798
but we are multiplying that fact several times – pseudoreplication. This makes sense—but perhaps
799
a numerical example of how it would be pseudoreplication would help?
800
We can see the concept of guild proportionality at a biogeographic scale in the conclusion
801
of Gentry (1988a) that the familial composition of tropical rain forests is remarkably constant. For
802
example, members of Fabaceae virtually always dominate neotropical and African "lowland
803
primary forests"; the plant families represented are "almost entirely" the same in the New World as
804
the Old. He saw similarity at the generic level too, for example between the New World and
805
Madagascar. These are fascinating observations. Gentry comments that it "can hardly be due to
806
chance", but he made no comparison with a null model. The finding is relevant to guild
807
proportionality only if families occupy particular niches, Gentry's "familial-specific niches", but
808
how else could the result arise? As with taxonomic guilds in general, we cannot be surprised if the
809
results are non-significant, but they are valid if significant.
810
Mohler (1990) made a comparison at the subgeneric level, within Quercus (oak) at various
811
sites across the USA. For 12 of the 14 regions that he examined (apparently with a variety of
812
quadrat sizes) there was a significant tendency for the two most abundant oak species to be from
813
different subgenera. This was not related to consistent pairing of particular species. His null
814
hypothesis was a 0.5 chance of each subgenus, which assumes they are equal in size, but this
815
would bias the test against the situation he found. The data were collected in various ways, but his
816
consistent result is in spite of this. It was apparently an a posteriori test (i.e., he thought he saw an
817
interesting effect, and tested it), but the consistency of the effect over several regions largely
818
overcomes this problem. Mohler examined various explanations: disease/pest pressure, niche
819
differences in fruiting phenology through mast fruiting, dispersal differences, etc., but could not
820
find any clear single explanation. This approach was considerably extended in careful work by
821
Cavender-Bares et al. (2004). They examined the associations, habitat correlations, characters for
822
several Quercus spp. in three reserves in central Florida, USA. Characters that tended to be similar
823
in more frequently co-occurring species included bark thickness, radial growth rate, seedling
824
absolute growth rate (AGR) and rhizome resprouting. These are characters that probably adapt to
825
water stress, fire tolerance and soil fertility. Habitat preferences were more scattered across the
826
phylogeny than expected at random, suggesting that the three sub-genera occupied different alpha
Wilson and Agnew, chapter 5, Assembly rules, page 28 of 53
827
niches, and within those had evolved to cover the range, mainly in moisture availability. In
828
reconstructions of phylogeny from ribosomal DNA, the characters indicated as changing less
829
within a clade included acorn maturation time, embolism due to freezing, woody density and
830
second-year vessel diameters. Seedling leaf lifespan and perhaps SLW tended non-significantly in
831
that direction. [SLW = specific leaf weight, the weight of unit area of leaf. It is the reciprocal of
832
SLA.] Characters that tended to be dissimilar in co-occurring species, indicative of different alpha
833
niches: acorn maturation time, embolism due to freezing, leaf life span and first-year vessel
834
diameters, and non-significantly SLW and perhaps seedling leaf lifespan. Because of the tendency
835
for species from far parts of the phylogeny to co-occur, this should be a similar list to the list of
836
conservative characters, and it is almost identical. These should be characters that are related to
837
alpha niche, and it is less easy to how they are. Cavander-Bares et al. suggest that acorn maturation
838
time might be related to phenological niche differentiation in masting and seedling regeneration,
839
they imply that frost tolerance might be related to year-to-year weather variation, and leaf lifespan
840
to timing of nutrient uptake. The crucial correlation is that species that co-occur more often are
841
more distant on their ‘phylogenetic tree’ (p < 0.034). However, we have to bear in mind that this is
842
essentially a test between habitats, and therefore their 74 plots were not all independent. Again we
843
see the ugly head of pseudoreplication via what we might call environmental autocorrelation.
844
7.3 Patch models
845
We have referred repeatedly to the problem of spurious ‘guild proportionality’ due to
846
environmental differences and consequent pseudoreplication. The solution, as mentioned above, is
847
not to randomise over all the quadrats. Wilson and Roxburgh (1994 %267) made some attempt by
848
having their points arranged in ten 24 × 24 cm plots, randomising occurrences only within each
849
plot, and accumulating the departures from the null models over the ten plots. Wilson and Gitay
850
(1999 %566) used a similar technique, creating separate null models for each of their 21 sites. An
851
even better technique is to form a separate null model for each quadrat, randomising over a few
852
quadrats adjacent to it: the ‘patch model’ technique described above (Fig. 5.3). Bycroft et al.
853
(1993) did this by using a linear 7-quadrat patch based on the target quadrat; the proportionality in
854
the what? that had been seen with a site model was reduced in size and no longer significant.
855
Although the loss of significance could be due to the reduced power of patch model, the effect of
856
size was less too – only half. This was in vegetation selected to be uniform, and warns us to be
857
careful about any study that does not use some kind of patch model. Watkins and Wilson (1994
858
%591) used a patch of 9-quadrats centred contiguously on the grid. This is probably the ideal, and
859
in their work some significant guild proportionality was seen with such a model.
Wilson and Agnew, chapter 5, Assembly rules, page 29 of 53
860
861
7.4 Evidence: removal experiments
It should be possible to see equivalent guild effects in perturbation experiments. If
862
member(s) of one guild are removed, the species that increase should be from the same guild.
863
Indeed, when Herben et al. (2003) removed the dominant grass species, Festuca rubra, from a
864
mountain grassland, it was grass biomass that increased more than that of dicotyledons. However,
865
the species responding differed depending on the year in which the removals started. Symstad
866
(2000 %99) removed three guilds – forbs, C3 graminoids, and C4 graminoids – from existing
867
Cedar Creek grassland. After three years of growth, seeds of 16 native prairie species were added:
868
legumes, nonleguminous forbs, C3 graminoids and C4 graminoids. There was only weak evidence
869
that resident species repelled functionally similar invaders. Such removal experiments are prone to
870
high experimental error. Fargione et al. (2003 %8916) used plots at Cedar Creek that had been
871
planted with 1-24 species in 1994, and 27 species that occurred in the area but had not been
872
planted were added in 1997. Multiple regression of the 1999 guessed cover of four invader guilds
873
on the resident guilds indicated that each guild as a resident had a greater inhibitory effect on
874
invasion by its own guild, though all invader guilds were inhibited most by C4 grasses. Von Holle
875
and Simberloff (2004) marked out field plots on a floodplain, and weeded particular subjective
876
guilds from some. They then planted in 10 species commonly found in those floodplains. There
877
was no tendency for species to survive better or grow more when planted into a plot from which
878
their guild had been removed. In summary, these removal experiments gave little evidence for
879
guild-based assembly rules.
880
7.5 Evidence: successional convergence
881
Fukami et al. (2005) reported an experiment in which outdoor plots were sown to a mixture
882
of 15 species, or to five different combinations of four species out of those 15. Unfortunately,
883
cover was guessed (in six categories, which only discards information). One year after
884
establishment, the species composition of the 15-species plots was very similar between five
885
replicates, as was that of five plots that started with bare soil. However, the (unreplicated) 5-
886
species mixtures showed considerable differences that year, and those differences remained eight
887
years after sowing with no sign of convergence (Fig. 8a). The authors called this priority effects,
Wilson and Agnew, chapter 5, Assembly rules, page 30 of 53
888
which might imply a switch, but there might be an effect of inertia due to competitive abilities
889
being rather similar. But in spite of the persistent differences in species composition, the different
890
5-species mixtures converged in terms of composition of 14 functional types (a typical type being
891
“Autumn-germinating annuals, typically tall with semi-rosette form and wind-dispersed seeds”.
892
Fig. 5.9: Fukami.
893
There is a danger that this was just due to averaging, but the authors disproved this with a
894
randomisation test.
895
7.6 Intrinsic guilds
896
The majority of guild investigations have used extrinsic guilds, designated by a priori
897
criteria (Wilson 1999). Sometimes, the guilds have been pre-determined (e.g. MacNally 2000).
898
Sometimes several characters have been chosen and multivariate methods have been used to
899
classify species into guilds (e.g. Landres and MacMahon 1980; Willby et al. 2000), but this begs
900
the question of whether the characters measured are the appropriate ones, and whether they have
901
the correct weighting. Tests for the reality of such guilds using field associations (e.g. Hallett
902
1982) or perturbation experiments (e.g. Hairston 1981) can indicate that some guild structure has
903
been found, but not that it is the true guild structure of the community. Wiens (1989) summarised
904
the problem:
905
"There is an arbitrariness to guild classification and the determination of guild
906
membership, which is especially evident in subjective a priori classifications. This raises
907
the prospect that the guild 'patterns' that emerge from studies based on such classifications
908
are consequences of imposing an arbitrary arrangement on a community that is actually
909
structured ecologically in some other way altogether (or is not structured at all). Using
910
multivariate statistical procedures does not grant immunity from this problem."
911
A solution to Wiens' dilemma is to “interview the plants”, to select an index of guild structure and
912
to find the guild classification that maximises this index. This classification is the intrinsic guild
913
structure. Wilson and Roxburgh (1994) introduced this concept: determining the guilds according
914
to the ways the species actually behave, asking the plants what guilds they are working by.
915
Distributional data
916
Wilson and Roxburgh (1994 %267) used distributional information to find intrinsic guilds.
917
To avoid circularity they divided the data in two, optimising the guild classification on one half
918
and testing it on the other. With field data it is impossible to examine every possible guild
919
classification, the number is generally astronomical (2(number of species - 1) - 1), so they took their a
920
priori graminoid versus forb+bryophyte classification, and swapped species iteratively to reduce
Wilson and Agnew, chapter 5, Assembly rules, page 31 of 53
921
guild proportionality index RVgp. This showed that some forbs were better assigned to the
922
'graminoid' guild, perhaps because of the role of their laminae in the upper canopy, and vice-versa.
923
After many iterations the process converged to intrinsic guilds that gave an even stronger tendency
924
towards guild proportionality, not only on the optimisation subset but also on the independent test
925
subset that had not been used in the optimisation procedure. Searches for intrinsic guilds starting
926
from two random initial configurations resulted in classifications quite similar to the optimised
927
'Graminoid' versus 'Forb+bryophyte' guilds, and with further optimisation using the whole dataset
928
the three optimised classifications converged to become identical. It is important to remember that
929
these intrinsic guilds are alpha guilds, not beta ones. That is, there is a tendency for the species of
930
one guild not to occur together. Presumably the reason is that they are too similar in resource use,
931
and competitive exclusion occurs. Rather, say we are dealing with a 2-species point, there will
932
tend to be one species from one guild and one from the other.
933
Wilson and Whittaker (1995) used the method on their Welsh saltmarsh data. Three
934
searches produced very similar guild classifications, which converged to become identical after
935
further whole-dataset optimizations, indicting that real guilds occur in the saltmarsh. Intrinsic guild
936
membership could subsequently be correlated with leaf morphology, all the monocots were in one
937
guild together with other narrow-leaved species as in a lawn previously examined by Wilson and
938
Roxburgh (1994 %267). This suggests that canopy interactions may be important in controlling
939
species coexistence.
940
Wilson and Gitay (1999 %566) performed 100 random-start searches (computer processing
941
power had increased in the interim) on the tussock-grassland data. A guild classification that
942
showed significant guild proportionality in the test subset was found in a significantly greater
943
number of searches than expected by chance (28 out of 100), and the ten classifications that gave
944
the lowest RVgp comprised three groups. Further optimisation of representatives of these groups
945
using the whole dataset confirmed that the community contained at least two genuinely
946
independent, alternative guild classifications. It seems that two or more guild classifications can
947
exist within the same set of species in a community, orthogonal in the sense that they are unrelated
948
to each other and operate simultaneously. We should not be surprised at this, the true guild
949
relations are probably quite complex. These intrinsic guilds showed some relation to growth
950
form/height.
951
The general impression from these results is that guild membership in these grasslands
952
depends on canopy relations, especially vertical stratification as affected by leaf morphology.
953
However, this may be partly due to the characters considered, and other characters, correlated with
954
them, may be the real determinants.
Wilson and Agnew, chapter 5, Assembly rules, page 32 of 53
955
Competition experimental data
956
Wilson and Roxburgh (2000 %189) used a competition experiment to seek intrinsic guilds.
957
Seven species from the Otago Botany Lawn had been grown in boxes in all possible 2-species
958
mixtures. They argued that when a species from one alpha guild was grown with a species from
959
another alpha guild, by definition differing in resource use, then by the Jack Spratt1 effect the yield
960
of the mixture should be considerably greater than the mean of the two monocultures, as measured
961
with index RYM (Wilson 1988%279). With only seven species it was possible to test all possible
962
2-guild classifications to find the one that maximised the mean RYM of mixtures, and this resulted
963
in guilds very similar to those obtained from distributional data.
964
Experimental removals data
965
Clements et al. (1929) had experimented with removing species from communities, and
966
Fowler (1981) took this approach by removing single species from a North Carolina grassland. For
967
all removals, at least one other species was affected significantly. Often several species were
968
affected. Usually removal effects between a pair of species were not reciprocal. There was no sign
969
of guilds that affected each other and it was hard to predict which species would be affected. A
970
few negative effects were seen, in which removal of a species decreased the yield of another; if
971
these effects were real, they could have been due to mutualisms or to indirect interactions via a
972
third species. The conclusion is that species interactions in that grassland were complicated, often
973
indirect, and diffuse. Intrinsic guilds were not present. Similar experiments, with similar
974
conclusions, were performed by Allen and Forman (1976) on a New Jersey oldfield, Abul-Fatih
975
and Bazzaz (1979) on an Illinois oldfield, Silander and Antonovics (1982) on North Carolina dune,
976
slack and saltmarsh, del Moral (1983) in Washington subalpine meadows, and Gurevitch and
977
Unnasch (1989) on a New York oldfield. These results exclude a simple model of community
978
structure, e.g. with distinct guilds. It implies but does not prove stochastic structure. There is a
979
necessary compromise in this work, in that the community is disturbed by the perturbation,
980
probably repeatedly, and the removals tend to lower plant density. A greater problem is that the
981
intensity of work required limits replication, so the statistical errors are usually large, and many of
982
the interesting effects are not significant.
983
Conclusion on intrinsic guilds
984
A major advantage of the intrinsic guild approach is that it can fail. Approaches such as
985
multivariate classification of characters must give guilds, whether any exist or not. In contrast, a
986
search for intrinsic guilds by minimizing RVgp, maximizing RYM or examining the pattern of
“Jack Sprat could eat no fat and his wife could eat no lean, and so between the two of them they wiped
the platter clean.”
1
Wilson and Agnew, chapter 5, Assembly rules, page 33 of 53
987
response to removals can result in all the species being in one guild, or in a guild structure that is
988
non-significant, as it did for Wilson et al. (2000 %749), and in a more informal way for Fowler
989
(1981) when she failed to see clear groups in removal results. That is, if there is no guild structure,
990
the intrinsic guild approach can indicate this.
991
Although functional-character relations between species are often expressed in a
992
classification, ordinations have also been used to see trends and continuous variation. It would be
993
good to have an intrinsic equivalent to ordination, placing the species on guild gradients according
994
to their distributions or their responses in experiments.
995
8 Texture convergence
996
Vegetation texture was defined by Jan Barkman (1979) as: "the qualitative and quantitative
997
composition of the vegetation as to different morphological elements, regardless of their
998
arrangement". We would these days extend it beyond morphology into physiological characters,
999
and use the term ‘functional characters’, but the aim remains to describe communities not by the
1000
names of the species but by plant characters, assuming that similar characters indicate similar
1001
function. As an assembly rule, the concept of texture convergence is that in comparable habitats in
1002
different areas, e.g. on different continents, whilst the actual species present are probably different,
1003
but the texture may be the same (Fig. 5.10). Constraints of the physical environment and of species
1004
interactions will cause convergence to the same texture. Why doesn’t he just call it convergent
1005
evolution? The concept is similar to that of guild proportionality, except that instead of dividing
1006
the variation into groups (guilds), it looks at the whole distribution. It is possible that the mean
1007
texture might converge but not the distribution of characters (Fig. 5.11a), or the distribution could
1008
converge but not the mean (Fig. 5.11b), or of course both or neither.
1009
There has long been interest in the idea of convergence between the plants and animals of
1010
areas on different continents, with a similar environment (mainly climate) but taxonomically
1011
different biotas. Work with plant communities has been almost entirely on mediterranean-climate
1012
areas, such as in California, Chile, the Cape, SW Australia and the Mediterranean itself. Mooney
1013
and Dunn (1970) suggested that the mediterranean environment in particular imposes several
1014
limitations on plant growth, with only a limited number of strategies possible, the evergreen
1015
sclerophyll strategy being one.
Wilson and Agnew, chapter 5, Assembly rules, page 34 of 53
Smallest
Leaf width
Mean
Largest
Smallest
Leaf width
Largest
1018
Largest
Smallest
Leaf width
Largest
Mean
Convergence
in mean
Leaf width
Mean
Leaf width
Fig. 5.10. The concept of texture convergence.
Leaf width
1016
1017
Largest
Smallest
Smallest
Convergence
in distribution
Largest
Smallest
Fig. 5.11. convergence can be in mean or in shape.
1019
A few of these studies have measured texture and looked for convergence. Parsons (1976)
1020
compared scrub communities California and Chile (‘chaparral’ and ‘matorral’) under very similar
1021
climates, recording 24 plant characters: growth form, many leaf characters, reproductive
1022
characters, etc. Some plant characters were present in similar altitude/aspect habitats in
1023
comparable abundance among species, e.g. lobed leaves and winter-deciduousness in high-altitude
1024
ravines, and large leaves in low-mid altitude ravines. Others, such as summer-deciduousness, were
1025
present in both areas but in somewhat different environments. However, small leaved plants were
1026
prevalent on low-altitude ridges in California but absent in Chile, where spiny-leaved species were
1027
present instead. Parsons attributed some of the differences to land-use history. Cowling and
1028
Witkowski (1994) compared sclerophyllous shrubland in mediterranean Western Australia and in
1029
South Africa, and found similar texture between the continents in terms of growth form
1030
(shrub/graminoid/forb), and leaf consistency (sclerophylly and succulence) and SLW, but
Wilson and Agnew, chapter 5, Assembly rules, page 35 of 53
1031
divergence in spininess. Canopy-storage of seed diverged, but dispersal type
1032
(wind/vertebrate/ant/other) generally converged. However, convergence here is being judged from
1033
non-significance of difference, and no conclusions can be drawn from lack of significance.
1034
Schluter (1990) introduced the concept of “species-for-species matching”, where there are
1035
species in the same positions in niche space in different areas. However, we would not necessarily
1036
expect that. Nor should we necessarily expect the same number of species, since a niche filled by
1037
one species in area 1 could be split between three species in area 2. All that we need to require is
1038
that the same niche space is occupied, and that it is fully occupied in both communities (Fig. 5.12).
Area 1, with 7 species, A to G
Area 2, with 4 species, W to
Species abundance
Z
C D
E
B
Y
Z
B
W
G
A
1039
1040
X
F
Character value
Character value
Fig. 5.12. Site 1 has the same texture as Site 2 with respect to the character, even though they
1041
differ in the number and abundances of species.
1042
Wilson et al. (1994 %109), compared convergence between two carr (i.e. wooded fen)
1043
communities in Britain and two in New Zealand, in five functional characters related to light
1044
capture, such as SLW and support fraction. In the null model the species present were swopped at
1045
random between sites with no constraint on the cooccurrence of species similar in morphology.
1046
The test is one for coevolutionary convergence and coecological sorting, not for similarity of
1047
adaptation to the environment. Rather the convergence in texture of the four carrs there was
1048
divergence when weighting species equally. However, weighting the species by their
1049
photosynthetic biomass, convergence was seen for PSU width and possibly for PSU area. [PSU =
1050
photosynetetic unit, i.e. leaflet, leaf, cladode, unit of green stem, etc.] Note that this does not
1051
represent adaptation to the overall environment, because the comparison was with random draws
1052
of the species present at the site, not with an exterior species pool. It means that each community
1053
has representation from each of the functional types present in those carr communities, strong
1054
evidence that species are being sorted by their characters, evolutionarily or ecologically, for their
1055
entry into the community.
Wilson and Agnew, chapter 5, Assembly rules, page 36 of 53
1056
The first studies compared continents, but comparisons can be made between nearby sites,
1057
or between patches within sites. This is close to the intuitive question looking at different patches
1058
within an area of vegetation: do similar species trade off against each other? One just has to realise
1059
that it is ecological, not evolutionary, convergence,. However, any evolutionary convergence is
1060
just a genetic fixation of ecological convergence (Smith and Wilson 2003). Smith et al. (1994
1061
%041) investigated sites in conifer/broadleaved forest in southern New Zealand, recording similar
1062
characters to those used by Wilson et al. (1994 %109) and found convergence in all characters, but
1063
as in the Wilson et al. study only when characters of the species were weighted by the abundance
1064
of the species. Matsui et al. (2002 %015) conducted the same type of investigation but locally,
1065
within three sites, and evidence of convergence was found for a subalpine grassland: each patch
1066
(quadrat) tended to comprise a mixture of small-leaved species and large-leaved species, a more
1067
constant mixture than expected if the species were being swapped between quadrats at random,
1068
irrespective of their characters. Watkins and Wilson (2003 %525) took this approach further by
1069
examining replicate quadrats within twelve herbaceous communities, measuring eleven characters
1070
that were intended to reflect the functional above-ground niche of the species, and meticulously
1071
obtaining the biomass of each species in each quadrat. Biomass weighting allows for true
1072
characterisation of the texture of the quadrat. Convergence was seen in chlorophyll content,
1073
indicating a significant tendency for each patch in a community to comprise a rather constant
1074
mixture of species types in terms of their different chlorophyll contents, though other results were
1075
non-significant or showed divergence. In these local convergence studies it is explicit that the
1076
question is of ecological assortment.
1077
As so often, environmental differences act as noise. As Schluter (1990) wrote: “recall that
1078
we are seeking communities more similar than would be expected on the basis of random sampling
1079
from the same underlying probability distribution of possible species values. Any factor that
1080
causes the underlying distributions to differ will quickly decrease their chance that a too-small
1081
difference between communities will arise”. This is matching.
1082
9 Time
1083
Time has done natural experiments for us. When the climate has changed, e.g. in the c.
1084
15,000 yr since the last glaciation, species have moved around. But have they reassembled into the
1085
same communities, or as they pleased? Clements (1936 %252) wrote that “climaxes have evolved,
1086
migrated and disappeared under the compulsion of great climatic changes from the Paleozoic
1087
onwards, but [the student of past vegetation] is also insistent that they persist through millions of
1088
years in the absence of such changes”. He continued: “The prairie climax has been in existence for
1089
several millions of years at least, and with most of the dominant species of today”. Clearly his
Wilson and Agnew, chapter 5, Assembly rules, page 37 of 53
1090
concept of the community as a complex organism led to a conclusion that there were only a limited
1091
number of combinations in which species could occur. Sure, in the very long term new
1092
communities could ‘evolve’ and some disappear, but the changes in climate since the last
1093
glaciation would result largely in the migration of existing combinations. However, several
1094
palaeoecologists have suggested that many of the communities, as seen in pollen assemblages, that
1095
were extant earlier in the Holocene are not found anywhere on Earth today: they are ‘no-analogue’
1096
communities. This challenges Clements’ very concept of the plant community.
1097
There are actually many possible explanations, and Jackson and Williams (2004) evaluate
1098
them carefully. They discusses the problem of how different, and by what criterion, a ‘no-
1099
analogue’ community has to be. They reject as major causes of no-analogues artefacts such as
1100
differential pollen preservation, mixing of sediments, different pollen production by some species
1101
in the [CO2] obtaining then and a different juxtaposition of communities over the landscape. It is
1102
remarkably difficult to find exact matches between any two current climates, and this is probably
1103
even more true for the past, and [CO2] will be present too, and Jackson and Williams suggest that
1104
the most likely explanation for no-analogue communities is that whilst similar ranges of climatic
1105
variates occurred, often the combinations that we see today did not (Fig. 5.13). So then
1106
communities are like species—they can go extinct. Perhaps this has some implications in our
1107
classification and conservation of communities? Just like a species’ niche may no longer exist due
1108
to changes in climate, geology, etc;, perhaps a community’s niche space can be destroyed. That
1109
doesn’t make the classification of communities based on niche space any less legitimate, does it?
1110
This interpretation is supported by comparing the degree of mismatch between
1111
reconstructed past plant communities and the best modern fits with the degree of mismatch
1112
between reconstructed past climates (from GCMs, general circulation models) and the best modern
1113
fits. Community misfits (no-analogues) tend to occur in the same place/time as climate misfits
1114
(Williams et al. 2001). This evidence is at variance with Clements’ interpretation of constant
Existing
environments
Niche of
Species X
Species Y
Environmental factor A
Time 2
Environmental factor B
Environmental factor B
Time 1
Existing
environments
Environmental factor A
Fig. 5.13. At Time 1, the area in which the realisable niche of Species X and Y overlap in
environmental space that exists. At Time 2, the combination of environmental factors in
which they could co-occur does not exist. Inspired by Jackson and Williams (2004).
Wilson and Agnew, chapter 5, Assembly rules, page 38 of 53
1115
communities moving around the landscape. However, it does not distinguish between species
1116
reacting individualistically to the climate, as suggested by Gleason in some of his writings (this
1117
vol., chapt. 6, sect. 3) from a model in which the occurrence of a species is determined by the
1118
identity of other species present, a view attributed with some truth to Clements (this vol., chapt. 6,
1119
sect. 2).
1120
10 Abundance
1121
10.1 Biomass constancy
1122
The constancy of biomass per unit area, compared to null models in which species
1123
abundances are random, has been used as an assembly rule (Wilson and Gitay 1995 %369). This is
1124
not a deep assembly rule, but it is a demonstration from the field that competition is occurring and
1125
causing community structure. It has the ability to distinguish between communities (Wilson et al.
1126
2000 %749).
1127
10.2 Relative abundance distribution (RAD)
1128
Concepts
1129
Various models of community construction give predictions for the relative abundance
1130
distribution between species (RAD; Wilson 1991): the Niche-preemption (Geometric) model is
1131
based on competition, the Zipf-Mandelbrot can be interpreted as succession/facilitation. The
1132
Broken stick and the Sequential Breakage (General Lognormal) models are alternative models of
1133
the random assignment of resources (alpha niche widths) between species. Several others of this
1134
type can be constructed (Tokeshi 1996 %35). Note that several are null models: alternative models
1135
of what is going on when nothing is going on. This means that we are liable to end up testing
1136
between null models, not against them. It is also a problem that some of the distributions, notably
1137
the General lognormal, can be derived from alternative assumptions, and on the other hand how
1138
many distributions can be found from subtly different models (Tokeshi 1996 %35). With so many
1139
different models, and with sampling variations, one might worry that it would be impossible to
1140
discriminate between them. However, in a 15-species community, for example, one can identify
1141
the correct model with reasonable correctness given 10 or more quadrats; it depends on the model
1142
and the number of species (Mouillot and Wilson 2002).
1143
The RAD for a community is potentially useful for several reasons. Most types of evidence
1144
for community structure involve comparisons in time or space; RADs are one of the very few types
1145
of such evidence (evenness is one aspect of the RAD).
Wilson and Agnew, chapter 5, Assembly rules, page 39 of 53
1146
MacArthur (1957) proposed a "broken-stick" model, in which abundances reflected the
1147
partitioning of resources among competing species, by random divisions along a one-dimensional
1148
gradient. This ecological model can be tested by comparing its dominance/diversity predictions with
1149
those observed. However, the concept of a one-dimensional resource gradient applies uneasily to
1150
partitioning of most plant resources. Other ecological models can give the same distribution,
1151
including models with no restrictions on niche overlap (Cohen 1968).
1152
The geometric model (Whittaker 1965) suggests that the 'most successful species'
1153
(presumably the one with the highest competitive ability) takes fraction 'k' of the resources, and
1154
therefore forms approximately (Whittaker 1965) k of the abundance. The second most successful
1155
species takes k of the remainder (i.e. a total of k(1-k)), etc. Again, this ecological model can be
1156
tested.
1157
1158
1159
Preston (1948) proposed the use of a lognormal distribution for empirical reasons, though it
might express community structure:
a. Plant growth will be affected by several environmental factors. By the Central Limit
1160
Theorem, this will give a near-normal distribution. Since plants have intrinsic logarithmic
1161
growth, the distribution will be lognormal (May 1975), or
1162
b. MacArthur's Broken Stick model, but with the breaks sequential and breakage probability
1163
independent of length, gives a lognormal distribution. This can be seen as the occupation and
1164
subsequent division of niches by species (Pielou 1975).
1165
Preston (1962) proposed further that the distribution was a reduced-parameter subset of
1166
lognormal distributions that he called 'Canonical lognormal', defined by the mode of the
1167
individuals curve coinciding with the last point on the species curve (i.e. gamma = 1). The
1168
hypothesis was empirical; there is no ecological basis for it (Caswell 1976). Whether it is a
1169
mathematical artefact is controversial (May 1975; Connor and McCoy 1979; Sugihara 1980;
1170
Connor et al. 1983). Why does he describe so many theories that no one seems to subscribe to? It’s
1171
like a history book.
1172
Evidence
1173
Almost all comparisons of actual dominance-diversity curves with theoretical ones has
1174
been by comparison of shapes (e.g Whittaker 1965). For valid comparison, a best fit needs to be
1175
calculated for each model (Wilson, in prep).
1176
Species diversity can be split into richness and evenness. Evenness represents in one value
1177
some of the information in dominance-diversity curves. Caswell (1976) examined evenness,
1178
compared to that expected from a null model. He found that tropical rain forests tended to be less
1179
even than predicted from the null model; temperate deciduous forests of eastern North America
Wilson and Agnew, chapter 5, Assembly rules, page 40 of 53
1180
were significantly more even than the null model. The contrast was the opposite of what he
1181
expected from previous theories.
1182

1183
1184
Wilson (1991), Watkins and Wilson (1994), Wilson and Gitay (1995), Wilson et al. (1996
Compton), Wilson et al. (1998 King).

Wilson et al. (1996 %527) fitted RAD models to plots from three experiments; basically
1185
there were no trends except those reflecting the higher evenness in plots to which P had
1186
been applied. Watkins and Wilson (1994 %91) sought a relation between the level of
1187
vertical complexity in a community in which RAD model fitted, but could find none.
1188
The model fitting best can be dependent on the scale of sampling (Wilson et al. 1998
1189
1190
1191
1192
%213).
Fig. 5.14.
Conclusion
The information analysed here is potentially useful. Fits to a model based on ecological
1193
theory would be most interesting, though usually ambiguous. Any regularity, such as adherence to
1194
Preston's Canonical hypothesis, would be that the structure was Deterministic. Tests of the
1195
Canonical hypothesis have therefore fascinated ecologists.
1196
10.3 Sparse species
1197
Species that are sparse (or ‘rare’) within the community, are a puzzle. The first question is:
1198
are they filling special niches that exist for rare species? Zobel et al. (1994) investigated this in a
1199
wooded meadow in Estonia by removing 10-17 species from certain plots, all with a cover of 1%
1200
or less (a different list for each plot), repeating the removals for 5 years. There were no visible
1201
gaps and they say very little biomass was removed, but species richness was reduced by 25-33%.
1202
Species did not immigrate to fill the gaps: the number of immigrants was no higher than in control
1203
(i.e. no-removal) plots, actually non-significantly lower. There seemed to be no special niches for
1204
the sparse species.
1205
Another question is whether sparse species have a distinct effect on the major species.
1206
Lyons and Schwartz (2001) in a meadow in the mountains of California manipulated the species
1207
richness by removing either: (a) all plants of the least abundant species, thus reducing species
1208
richness to between two and seven species, and (b) an equivalent biomass of the most common
1209
species (to control for possible disturbance by the removals in treatment ‘a’. The exotic grass,
1210
Lolium temulentum (darnel) was then introduced. Its establishment was higher when more rare
1211
species were removed, indicating a rôle for the sparse species in invasion resistance. It seems one
1212
would have to conduct this experiment with many other exotic species before one could conclude
1213
that rare species help with resistance to invasion; it could be that the rare species’ niche happened
Wilson and Agnew, chapter 5, Assembly rules, page 41 of 53
1214
to overlap with that of the invader in this case. This may just be a case for the resilience of
1215
biodiversity in general. We cannot tell why this conclusion differs from that of Zobel et al., and we
1216
are far from a generalisation.
1217
11 Keystone species
1218
A valuable concept in describing communities in recent years has been that of ‘Keystone
1219
species’; defined by Paine (1969 %91) as a single native species high in the food web that, whilst
1220
perhaps unimportant as an energy transformer, is vital for the maintenance of the community. This
1221
cannot be applied literally to plants, but others have seen a keystone species as being the one in a
1222
community with the greatest effect on others, or the greatest effect relative to its biomass (Jordán
1223
et al. 1999). Interesting attempt to quantify the concept—but one would have to measure not only
1224
‘biomass’ but ‘effect’ objectively. Since plants dominate the biomass and carbon capture of their
1225
systems, one could almost see all green plants as keystone species. They affect lower
1226
(decomposer) and higher trophic levels – usually more than one higher level. Their effect is often
1227
via herbivory of their vegetative parts, but the contribution of Ficus spp. to frugivores has led to
1228
their being called keystone species (Patel 1997; Nason et al. 1998). The term has also been applied
1229
to plants with intransigent litter (Empetrum hermaphroditum; Mallik 2003) and here it seems to be
1230
a switch maintaining the current state (this vol., chapt. 3, sect. 5.4.E), via litter that produces
1231
polyphenol-rich humus with low pH. The contribution of plants as furniture for birds has been
1232
seen as keystone (arborescent succulents by Midgley et al. 1997), and this may be a switch too.
1233
Hurlbert (1997) says, "the metaphor 'keystone species' was appealing and harmless" but "has come
1234
to mean little more than 'important for something'". And why not? As Bond (1993) says "If loss of
1235
a species results in a large effect on some functional property of the ecosystem, that species may
1236
be called a keystone". In fact, a species with a strong reaction on the environment will either
1237
change the current state, in which case it would not be called a keystone, or it will reinforce the
1238
current state, in which case it is a keystone because it operates a switch. Top predators can be
1239
keystones because of cascade effects, and plants can be keystones when they operate switches. So
1240
what? Does every community have a keystone? Can communities be described based on a
1241
keystone? What’s the connection?
1242
12 Exotic species as community structure probes
1243
In some parts of the world, exotic species have displaced much of the native cover (e.g. the
1244
Seychelles, Hawaii, New Zealand (McDonald and Cooper 1995). It is not always easy to define
1245
what an exotic species is, but most cases are clear. Exotic species are an opportunity for the
1246
theoretical community ecologist.
Wilson and Agnew, chapter 5, Assembly rules, page 42 of 53
1247
12.1 The nature of exotic species
1248
In one way invasion by exotic species is surprising: the native species have presumably
1249
evolved to meet the local environment, physical and biotic. Moreover, exotic species cannot be
1250
intrinsically different because all species are native somewhere (except species of garden origin
1251
and a few species of recent origin such as Spartina anglica). The concept is deficient in logic.
1252
Leger and Rice (2003) found the alien (Chilean) ecotype of Eschscholzia californica to be more
1253
vigorous in California than the native genotype. Would the Californian genotype, as an alien, be
1254
more vigorous in Chile than the native one? How would that situation arise? It is far from clear
1255
that exotics are consistently different. Kissel et al. (1987) found no consistent difference in water
1256
relations between the three major native woody species and four exotic ones of the most semi-arid
1257
area of NZ. King and Wilson (in press) found no difference in experimental water stress tolerance or
1258
nutrient response, though the exotic species did have a greater RGRmax. We suspect that often
1259
generalisations are being made from special cases.
1260
Exotics have been implicated in destroying the structure of the whole community. Wilson
1261
and Hubbard (1988), surveying the semi-arid Upper Clutha catchment, New Zealand, where massive
1262
exotic invasion has occurred, found very weak community structure as seen in an inability of an
1263
ordination to predict species presence/absence. Wilson (1989) attributed this to conflicting structure
1264
in the native and exotic guilds. Sanders et al. (2003 %2474) examined invasion by Linepithema
1265
humile (the Argentine ant) in California. They examined chequerboarding – the tendency of species
1266
to be mutually exclusive so that a site/species table looks like a chess board – by calculating index C
1267
for the ground-foraging ant community. Positive values of C indicate segregation, i.e. less species
1268
co-occurrence than expected under a null model, more mutual exclusions, a predominance of
1269
negative associations. Negative values indicate aggregation, i.e. more species co-occurrence than
1270
expected under a null model. It is difficult to see the effect of an uncontrolled natural experiment, but
1271
Sanders et al. took the best approach possible, comparing quadrats sampled in one year that had not
1272
been invaded versus those that had, and comparing particular plots the year before and after
1273
invasion. They found that before invasion C was generally positive and significant; after invasion it
1274
was never significantly positive, and sometimes significantly negative. If we can take
1275
chequerboarding as evidence of community structure, the exotic ant had destroyed it.
1276
There are many examples of invaders successfully entering natural, allogenically
1277
undisturbed communities: in Britain Acer pseudoplatanus (sycamone), Rhododendron ponticum
1278
and Reynoutria japonica (Japanese knotweed), in New Zealand Berberis darwinii and Mycelis
1279
muralis can invade forest and Juncus gerardii saltmarsh.
Wilson and Agnew, chapter 5, Assembly rules, page 43 of 53
1280
This whole approach has been questioned, as to whether in invasions the exotics are the
1281
cause of the change – the ‘drivers’ – or whether they just take advantage of a disturbance– the
1282
‘passengers’. Corbin and D'Antonio (2004 %1273) addressed this for the grasslands of California,
1283
which 200 years ago before had been dominated by native perennial grasses with associated
1284
annual and perennial dicot species. These were almost completely displaced by European and
1285
Asian species. Under the ‘passenger’ hypothesis the change came about due to tilling for
1286
agriculture, introduction of livestock and a severe drought in the 19th Century, leaving disturbed
1287
conditions. Corbin and D’Antonio experimentally removed the vegetation, then sowed plots with
1288
three native perennial grass species, with three exotic annual grass species, or with both. Over
1289
time, the native grasses reduced the productivity of the exotic annuals, whilst the impact of the
1290
latter on the native perennials was minor and decreasing. The ‘passenger’ concept was supported.
1291
Further south in California, Stylinski and Allen (1999) compared almost undisturbed sites of
1292
chaparral and sage shrublands with nearby areas disturbed by vehicles, excavation or agriculture.
1293
Percent cover of shrubs was measured by canopy intercept, but of that herbs and seedlings only
1294
guessed. The vegetation of the disturbed areas comprised mainly exotic annuals (60 %), whilst the
1295
undisturbed areas had 68 % cover of native shrubs. This situation remained essentially unchanged
1296
in a site disturbed 71 years, and the authors concluded that after invasion by exotics the vegetation
1297
reached an alternative stable state. Presumably a switch was operating, so that the passengers took
1298
over driving the vehicle, but we do not know through what factor the switch was operating.
1299
Five major explanations have been given for the ability of exotics to invade: (a)
1300
depauperate floras, (b) weak competitors, (c) the invaders are r species and (d) escape from natural
1301
enemies, (e) coevol.
1302
The ‘depauperate flora’ concept is that exotics are invading areas with depauperate floras.
1303
Islands are often given as examples of depauperate floras (e.g. NZ: Dulloo et al. 2002). The
1304
depauperisation can be in the number of species, leaving empty niches, or guilds (functional types)
1305
can be missing. Shimizu and Tabata (1985) explained the invasion of Pinus lutchensis into the
1306
shrublands of the Ogasawara Islands, Japan, by postulating that there had been an empty niche for
1307
an emergent tree. Ricciardi and Atkinson (2004) examined in a literature survey whether aquatic
1308
invaders amongst fish, invertebrates, algae and vascular plants, were more likely to have a high
1309
impact in terms of local extirpation / severe decline of a native species if there no congeners in the
1310
native biota. For four of seven systems, including the NZ coast, they were. This implies that
1311
species could invade more readily when there were empty niches (many of the comparisons were
1312
with animals, for which genera are often reasonable guild substitutes). If there were no congeners,
1313
surely there would be no species to push out. Similarly, Cappuccino and Carpenter (2005 %435)
1314
comparing invasive and non-invasive exotic plant species in natural areas in Ontario, New York
Wilson and Agnew, chapter 5, Assembly rules, page 44 of 53
1315
and Massachusetts, found that invasive plants were more taxonomically isolated than non-invasive
1316
plants, belonging to families with 75% fewer native North American genera, and Strauss et al.
1317
(2006) found the same with grasses of California, this time using a reconstructed phylogeny rather
1318
than taxonomy. There does seem to be some evidence for the empty niche / missing guild idea.
1319
The second explanation for the success of exotics is that the native species might not be
1320
vigorous enough. MacDonald and Cooper (1995) said “an individual island’s biota is based on too
1321
small a sub-sample of the global gene pool to have generated robust competitors for every
1322
available niche. … Insular species are frequently outcompeted by species that have been honed in
1323
much more exacting biotic communities of the mainland. … [suggesting] superior competitive
1324
ability of mainland species”. For New Zealand, Dansereau (1964) wrote: of “apparently vacant
1325
space” , occupied only by “weaker” species. Is this really true? Perhaps super-species, once limited
1326
by dispersal (e.g. to the old or new, Northern or Southern, hemispheres), are now able to spread
1327
everywhere. In that case, homogenisation of the flora is set to change the world (which it is). Still,
1328
these super-species don’t seem to have been that super in their original hemisphere. In Britain,
1329
when one meets a yellow composite herb with rosette leaves one has to key it out between a
1330
number of quite likely possibilities. In New Zealand Hypochaeris radicata (cat’s ear) is present in
1331
a huge range of environments and often quite frequent with them, so the answer 95 % of the time
1332
is ‘Hypochaeris radicata’. This all supports the above hypothesis—that a plant species faces
1333
greater competition on the main land, and have been forced by evolution into a competition-based
1334
niche that an isolated community may be unable to defend against due to a lack of similar forces of
1335
selection. An example of an exception may be Ammophila spp. It has been suggested, with some
1336
truth, that when high coastal dunes are built it is always by species of Ammophila. It seems to
1337
operate a switch, trapping sand and tolerating burial.
1338
The third possibility is that the exotics could invade because they are r species, short-lived
1339
and rapidly reproducing in ephemeral habitats. These are the R species of Grime (2001): fast-
1340
growing in open conditions, with quick and extensive seed reproduction. Why should there be
1341
more r species amongst exotics? Probably disturbed habitats are much more common and
1342
extensive than before humans changed the landscape. This has been an explanation for the origin
1343
of arable weeds: that once they were only in local disturbed areas such as riverbanks, and with
1344
cultivation they expanded their geographical range into arable fields. In some floras the number of
1345
r species may have been very small, for example Allan (1937) gives 6 % of the flora of NZ as
1346
being annual, and this is almost certainly an over-estimate, and a similar situation may have been
1347
true of many areas before humans appeared.
1348
1349
A fashionable explanation for the invasion of exotics and their apparently higher fitness
than the natives is that they have escaped from their natural specific enemies, they have therefore
Wilson and Agnew, chapter 5, Assembly rules, page 45 of 53
1350
evolved to discard their defences to these enemies and the resources involved have been used
1351
instead in growth and reproduction. Presumably the enemies will catch up in dispersal time (as has
1352
happened with the invasion of Lupinus arboreus in NZ, now largely suppressed by the lupin
1353
anthracnose fungus Colletotrichum gloeosporioides in N.Z.: Molloy et al. 1991) or in evolutionary
1354
time. The general pattern, whether the pests are insects, crustaceans, fungi or viruses, is a lesser
1355
impact on populations in the exotic range of a species (Vila et al. 2005; Bossdorf et al. 2005 %1;
1356
Mitchell and Power 2003 %625), presumably because the pests specific to the species are missing.
1357
The effect has been found comparing ecotypes from the native and exotic range grown in a
1358
common garden (e.g. Blair and Wolfe 2004). Sometimes, the increase in growth and reproduction
1359
in exotic populations of a species due to release from natural enemies has not been seen (Bossdorf
1360
et al. 2005 %1). Thébaud and Simberloff (2001) used maximum heights given in floras to compare
1361
species between the U.S.A. and Europe: invaders in both directions. In some comparisons
1362
populations were no different, and in others taller in their native range, the opposite of the effect
1363
expected under the enemy-release hypothesis. This study has the advantage of surveying many
1364
species, and in avoiding possible bias of choosing problem weeds, but it is not clear from where
1365
the flora writers obtain this information, nor how maximum height is defined. A complication has
1366
been illustrated for Senecio jacobaea, native to Europe but invasive in North America, Australia,
1367
New Zealand and elsewhere, that defence against specialist herbivores Tyria jacobaeae (cinnabar
1368
moth) and Longitarsus jacobaeae (ragwort flea beetle) has been lost, but some of the resources
1369
saved seem to have been put into increased protection against generalist lepidopteran herbivores
1370
via pyrrolizidine alkaloids (Joshi and Vrieling 2005 %704; Stastny et al. 2005 % 27).
1371
A further possibility is that the resident species in a community have been able to coevolve
1372
resistance to each others’ toxins. Thus, when Callaway and Aschehoug (2000 %521) found in a
1373
greenhouse experiment that Centaurea diffusa, exotic in US (Montana) had greater interference
1374
effect on Montana grasses than on related species from Georgia (Caucasas), and the difference was
1375
removed by adding active carbon, suggesting it was an allelochemical effect.
1376
Tropical rain forests in the tropics are an interesting case, since they are generally less
1377
invaded by exotic species. It would be helpful to conservationists to ascribe the lack of exotics to
1378
the saturation of available niche space through high diversity of species or functional types, but
1379
some species-poor types of tropical forest also have no invaders (Gilbertiodendron dewevrei:
1380
Richards 1996). A more likely explanation is that most of the “exotic species that are transported
1381
to tropical countries lack the specific life history traits, most importantly shade tolerance, that are
1382
necessary for successful invasion of undisturbed tropical forests” (Fine 2002). Rejmánek (1996
1383
%153) suggested that this was because of fast growth in that environment resulted in rapid canopy
1384
closure after disturbances.
Wilson and Agnew, chapter 5, Assembly rules, page 46 of 53
1385
1386
12.2 Exotic establishment and community assembly
The most fascinating way to use exotics as probes into community structure is to ask how
1387
they assemble when they reach new territory. Wilson (1989 %223) examined the native and exotic
1388
plant origin guilds of the Upper Clutha catchment, New Zealand. The two guilds produced
1389
classifications of the quadrats that were no more different than those using random groups of
1390
species, suggesting that the two guilds follow the same vegetational boundaries. However, there was
1391
some evidence that the guilds differ in the environmental factors with which they are correlated.
1392
Fig. 5.15: A minimum spanning tree for the species composition of British and New
Zealand roadside vegetation.
1393
1394
The roadsides of New Zealand generally comprise exotic species that have reassembled
1395
into communities there. Wilson et al. (2000 %757) examined an area of southern NZ containing
1396
152 exotic species, mainly from Britain for environment and cultural reasons. Quadrats from these
1397
NZ roadsides were fitted to the British National Vegetation Classification (NVC). After excluding
1398
species that are not present in New Zealand, the fit was 61%. Randomising the species/quadrats
1399
occurrences of the NZ data gave on average a 59% fit to the NVC. The fit of the real quadrats was
1400
only slightly, though significantly (p < 0.001) better than the random ones. British roadside
1401
communities were also compared to the NVC, as a control; they gave a 66% fit. We see that the
1402
New Zealand communities bear little relation to NVC communities in Britain. Comparing the NZ
1403
and British quadrats directly using a minimum spanning tree to connect similar quadrats, similarity
1404
was low, the two forming two almost distinct groups (Fig. 5.15). We have to conclude that the
1405
British species have re-assembled into communities in N.Z. most of which are new, i.e. distinct
Wilson and Agnew, chapter 5, Assembly rules, page 47 of 53
1406
from those that occur in the native range of the species in Britain. The evidence points to
1407
community assembly by pre-adaptation.
1408
Lord et al. (2000 %213) studied in a similar way the re-assembly of species introduced
1409
from Britain in NZ calcareous soil grasslands (4-24% CaCO3) that were largely composed of such
1410
species. Fitting as with roadsides, the fits for six sites ranged 48-77%. Two of the six sites fitted
1411
British calcareous grassland communities. These two sites are on thinner soil (< 10 cm depth),
1412
under lower rainfall, more likely to be influenced by the base rock, and for these sites the
1413
environment of the community in Britain matched very well that of the N.Z. site.
1414
Comparison of these two reassembly studies suggests that strong environmental filtering is
1415
able to reassemble communities. Even though the roadside dataset spanned a wide and very
1416
comparable environmental range in the two countries (e.g. rainfall 345 – 3460 mm and mean
1417
temperature in the warmest month 12-17 °C in New Zealand versus 485-1777 mm and 14-17 °C in
1418
Britain), it appears that environmental filters were not strong enough to reassemble the same
1419
communities. Assembly rules were not strong enough to do so. Instead, alternative states have
1420
been reached. We cannot tell whether they are stable, and if so what switch is responsible, but the
1421
consistent separation in the MST is remarkable.
1422
13 Conclusions, and the Otago Botany Lawn
1423
It is difficult to draw conclusions on assembly rules. We know that plants interact (this
1424
vol., chapt 2), we know that plant species differ (this vol., chapt. 1), and this must mean that there
1425
are limitations to coexistence. However, the difficulty of finding assembly rules, and the difficulty
1426
of ensuring that tests for them are valid, combine to make it difficult to confirm that this is so in
1427
the real world.
1428
The Botany Lawn of the University of Otago (Fig. 5.16) has surely been more intensively
1429
studied for assembly rules than any other community and offers a case study. It has also yielded
1430
the best evidence that such rules exist. The lawn was established c. 1965 with the sowing of a
Fig. 5.16: Profile through a part of the Botany Lawn.
Wilson and Agnew, chapter 5, Assembly rules, page 48 of 53
1431
Agrostis capillaris / Festuca rubra mix. The bulk of the 36 species present within the current
1432
community have arrived through natural dispersal, the commonest being the grasses Holcus
1433
lanatus (Yorkshire fog) and Agrostis capillaris (bent), forbs Trifolium repens (white clover) and
1434
Hydrocotyle heteromeria (a New Zealand native) and mosses Eurhynchium praelongum and
1435
Acrocladium cuspidatum. Since its establishment, the lawn has been maintained under a consistent
1436
regime of cutting to a height of c. 2.7 cm fortnightly in the growing season and monthly in winter.
1437
There has been no application of fertilizer, herbicide or irrigation (the average annual rainfall is
1438
784 m yr-1) That’s a lot of rain. This constant management, together with the short lifespan of
1439
individual ramets in the lawn, has created the opportunity for the community to come to
1440
equilibrium, and indeed the species composition of the lawn is quite constant over time. There
1441
were seasonal changes on the lawn, but there was little evidence of directional change between
1442
years, and the abundance ranks of species remained almost constant (Roxburgh and Wilson 2000).
1443
There is considerable stratification of species in the lawn (Figs. 5.16). Even when the
1444
sward is only 2.7 cm high after cutting there is significant evidence for three strata (Fig. 5.17a),
1445
and when the species have regrown 14 days later there are many more significant vertical relations
1446
between species, with evidence for four strata (Fig. 5.xb).
Fig. 5.17: Stratification in the Botany Lawn, (a) just after cutting to 2.7 cm and (b) after 14 days
regrowth. Lines connect species pairs that are significantly different in vertical position. Rare
1447
1448
species are omitted.
The variance in species richness across the lawn has been demonstrated to be lower than
1449
expected in a null model. This is seen at the scale of 13 × 13 mm (Watkins and Wilson 1992 %15),
1450
and the effect at that scale does not seem to be an artefact of environmental variation since its
1451
significance remains using a patch model. In fact, it was one of three out of the 12 lawns in that
Wilson and Agnew, chapter 5, Assembly rules, page 49 of 53
1452
investigatin to show a deficit of variance significant and greater than 20%. A similar deficit in
1453
variance richness can be seen at the scale of a point (Wilson et al. (1992 %711). The possibility
1454
has been raised that the effect is due to a physical limitation in packing plant modules at that scale.
1455
However, up to five species can be found at a point in this lawn, and on average only 1.45 species
1456
are, so space does not seem to be a limitation. We submitted earlier that plants do not compete for
1457
space (Chiarucci et al. 2002 %333), and the profile diagram (Fig. 5.16, drawn from life) confirms
1458
that the canopy is largely empty.
1459
The restrictions on species coexistence can probably seen better by analyzing guild
1460
proportionality. This removes us from questions of the number of modules that can be physically
1461
packed by using a null model in which the number of species in each quadrat does not differ between
1462
that observed and that in the null model, and it indicates restrictions in terms of species. Wilson and
1463
Watkins (1994 %591) analysed thus at the 13 × 13 mm scale. Testing over all richness categories
1464
there was no significant (p = 0.074) guild proportionality for graminoid versus forb guilds, but
1465
examining 4-species quadrats alone there was (p = 0.005). This was true for one other NZ lawn and
1466
one Fiji lawn. Likewise, grass versus legume guild proportionality was significant in the Botany
1467
Lawn in 3-species quadrats. Wilson and Roxburgh (1994 %267) found significant guild
1468
proportionality at a point using graminoid versus forb guilds, and whether or not the two
1469
bryophyte species were included with the forbs. There was no evidence that the rule was based on
1470
grass/legume interactions. There was also guild proportionality using as guilds the species that
1471
tended to be in the upper stratum of the lawn versus those that were basal, but only if the stratum
1472
assignments were based on species' positions at the end of the 14-day mowing/regrowth cycle. The
1473
constancy of the graminoid/forb proportion increased as the number of species at a point did. This
1474
suggests that when there are few species present at a point there is less constraint on which ones, but
1475
as the species start to pack in their ability to enter the community depends on their characters.
1476
The a priori guilds that we formed are not necessarily the true ones. At the scale of 13 × 13
1477
mm, although two of the three grass-grass associations negative as one would expeCt, so were
1478
those between Plantago lanceolata and two of the grasses (Watkins and Wilson 1994%591). We
1479
can determine the guilds as perceived by the plants using the intrinsic guild approach. With
1480
distributional data (minimizing guild proportionality index RVgp) the intrinsic guilds generally
1481
confirmed both the particular rôle of graminoids and the importance of the position in the canopy
1482
(Table 5.x; Wilson and Roxburgh 1994 %267). For example, Trifolium repens (white clover) with
1483
its horizontal laminae is often in the canopy fighting with the grasses (Fig. 5.16), and it appeared
1484
in the same intrinsic guild as four of the five grasses. Some other forbs were better assigned to the
1485
'graminoid' guild too, apparently because of the rôle of their laminae in the upper canopy. All this
1486
suggests that there is one niche for species that occupy the upper canopy towards the end of the
Wilson and Agnew, chapter 5, Assembly rules, page 50 of 53
1487
mowing/regrowth cycle, based on the interaction of lamina shape and position, and another for the
1488
basal species. Strong, almost surprising, support came from the intrinsic guilds obtained from the
1489
competition experiment (Wilson and Roxburgh 2001 %189). The guilds formed by maximizing
1490
the RYT (relative yield total, i.e. tendency towards overyield) gave, for the seven species included
1491
in the experiment, perfect agreement with those obtained from the distributional data (Table 5.x).
1492
These intrinsic guilds are real community ecology, because we allow the species to tell us what is
1493
happening in the community. This is inductive science, and made deductive for the distributional
1494
data by testing the guilds on independent data and for experimental data by confirming the results
1495
from the distributional data.
1496
Table 2. Intrinsic guild classifications of species of a lawn obtained from: (a) distributional data
1497
(Wilson and Roxburgh 1994 %267) and (b) the competition-experiment data of Roxburgh
1498
and Wilson (2000 %189).
Species
Characteristics
Guild from
distributional
data
Agrostis capillaries
Grass
A
Anthoxanthum odoratum
Grass
A
Bellis perennis
Dicot, rosette
A
Holcus lanatus
Grass
A
Hydrocotyle moschata
Dicot, horizontal lamina A
Linum catharticum
Dicot, upright
A
Poa pratensis
Grass
A
Ranunculus repens
Dicot
A
Trifolium dubium
Legume, horiz. lamina
A
Trifolium repens
Legume, horiz. lamina
A
Acrocladium cuspidatum
Moss
B
Cerastium fontanum
Dicot, erect
B
Cerastium glomeratum
Dicot, erect
B
Guild from
competition
experiment
data
A
A
A
A
Eurhynchium praelongum Moss
B
Festuca rubra
Grass
B
B
Hydrocotyle heteromeria
Dicot, horizontal lamina B
B
Hypochaeris radicata
Dicot, rosette
B
Prunella vulgaris
Dicot, creeping
B
Ranunculus repens
Dicot, creeping
B
Sagina procumbens
Dicot, creeping
B
B
Wilson and Agnew, chapter 5, Assembly rules, page 51 of 53
1499
This approach does not make any assumptions about the characters that determine
1500
coexistence. Mason and Wilson (2006) examined the traits of seven most common species in each
1501
guild. The two guilds differed in Mowing Removal Index (MRI), calculated as the proportion of a
1502
species’ mass typically removed during mowing (Fig. 5.18), but not in other characters related to
1503
light capture, such as specific leaf area (leaf area per mass), leaf area ratio (the leafiness of a plant)
1504
and six photosynthetic pigment characters. This confirms the importance of canopy interactions,
1505
but sheds light on whether they involve light capture.
1506
Mason and Wilson (2006) also used the approach of Stubbs and Wilson (2004), using
1507
point-quadrat data (new, and thus independent of that used by Wilson and Roxburgh) to test the
1508
limiting-similarity concept directly by examining whether the characters of the species co-
1509
occurring at a point were more different than expected at random. Greater variance among those
1510
characters would indicate limiting similarity: a tendency for species that were alike not to co-
1511
occur. MRI (Fig. 5.19) and leaf length showed significant limiting similarity at all five times since
1512
mowing analysed, as did two correlated characters, leaf area and length:width ratio. However,
1513
none of the other characters gave more than sporadic indication of limiting similarity. PSU
1514
length:width ratio showed significant limiting similarity for three of the dates, but it is related to
1515
MRI. Anthocyanin / dry mass demonstrated limiting similarity for in the first two samples after
1516
mowing, and marginally (p = 0.072) after 20 days. None of PSU width, PSU thickness, PSU dry
1517
mass, SLW, ratio of lamina area or mass to shoot mass, chlorophylls a or b per dry mass,
1518
chlorophyll a:b ratio or UV pigment content were significant for more than one period out of the
1519
five.
1520
How can the restrictions on coexistence be due to canopy interactions yet not be related to
light capture? One possibility, by analogy with the apparent importance of NPK and water
1522
resources in the results of Stubbs and Wilson (2005) is that although the guilds are canopy-related
1523
the basic effect is below ground. After defoliation there is generally ‘root growth stoppage’.
1524
Species with a high MRI would be affected by this because more leaf is removed. The temporary
1525
cessation of root growth would affect P uptake, which is rather dependent on exploration of the
1526
soil by new roots. Species with a low MRI could carry on growing, not only absorbing light
Observed / exoected variance in MRI
1521
1.20
1.10
1.00
0
5
10
15
Days after mowing
20
Wilson and Agnew, chapter 5, Assembly rules, page 52 of 53
1527
temporarily available by canopy removal, but with a continuing P supply. However, some support
1528
for the rôle of light comes from the local texture convergence study of Watkins and Wilson (2003
1529
%525). The result of convergence between quadrats in chlorophyll was mainly due to strong
1530
convergence in two of the 12 sites, one of which was the Botany Lawn. It is simplistic to expect
1531
one process to be limiting coexistence.
1532
Fig. 5.19:
p = 0.008
p = 0.022
p = 0.030
p = 0.033
p = 0.016
0
0.8
Mowing Removal Index (MRI)
Wilson and Agnew, chapter 5, Assembly rules, page 53 of 53
0.6
Guild A
Guild B
0.4
0.2
0.0
4
8
12
16
14
Time since mowing (days)
1533
1534
1535
Fig. 5.18: Mean Mowing Removal Index (MRI) of each guild at each sampling date. The P-values
are from t-tests for differences between guilds in mean Mowing Removal Index.
1536
1537
Why is the evidence for assembly rules stronger in the Botany Lawn than anywhere else?
1538
Firstly, it has been more intensively studied than any other community. The short stature probably
1539
contributes to the ease of finding assembly rules. The canopy is in some ways like a forest canopy
1540
in miniature, but the relations are easier to see: in a forest it is hard to determine just which part of
1541
the canopy a ground herb is influenced by. However, the major factor is probably not that it is
1542
easier to find rules but that they have shaped the lawn community to a greater extent because it has
1543
reached equilibrium. It has been undisturbed for 30-40 years, with a constant mowing regime and
1544
no fertilisation or weedkilling. The lifespan of a ramet in the lawn is probably about a year, giving
1545
30-40 generations of turnover. For forest trees, with lifespans of say 300 years the equivalent
1546
would be 9000-12000 years. In temperate areas, the forests have not been around that long since
1547
the glaciation, and in tropical areas there would almost certainly have been major disturbance.
1548
There is possibly no plant community anywhere closer to its equilibrium than the Botany Lawn. If
1549
the community is close to equilibrium is this sort of equilibrium capable of occurring in nature? If
1550
it isn’t, is ‘equilbrium’ a useful concept?, we can ask about its stability, and as we discussed in
1551
Chapter 3 the Botany Lawn community has been analysed for stability more intensively than any
1552
other community (Roxburgh and Wilson 2002 %395), and found to be on the borderline of
1553
stability, a conclusion confirmed by its response to perturbation (Roxburgh and Wilson 2002
1554
%409). This stability is probably both the cause and the result of the assembly rules demonstrated.
1555