Download 1 - Biology Department | UNC Chapel Hill

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