Download 10/4/06 version

The nature of the plant community: a reductionist view
1
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
Chapter 5: Assembly rules
44
45
1
2
Introduction ................................................................................................................................ 1
What rules are we searching for, and how? ............................................................................... 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
4
Species sorting.......................................................................................................................... 10
4.1
Species associations in succession ................................................................................... 10
4.2
Compositional convergence ............................................................................................. 11
5
Richness ................................................................................................................................... 12
6
Limiting similarity.................................................................................................................... 13
6.1
Limiting similarity in morphological characters .............................................................. 14
6.2
Limiting similarity in phenology...................................................................................... 17
7
Guild proportionality ................................................................................................................ 19
7.1
Concept............................................................................................................................. 19
7.2
Evidence: constancy in space ........................................................................................... 20
7.3
Patch models .................................................................................................................... 23
7.4
Evidence: removal experiments ....................................................................................... 24
7.5
Evidence: successional convergence ................................................................................ 24
7.6
Intrinsic guilds .................................................................................................................. 25
8
Texture convergence ................................................................................................................ 28
9
Time ......................................................................................................................................... 32
10
Abundance ............................................................................................................................ 33
10.1 Biomass constancy ........................................................................................................... 33
10.2 Relative abundance distribution (RAD) ........................................................................... 33
10.3 Sparse species ................................................................................................................... 35
11
Keystone species .................................................................................................................. 36
12
Exotic species as community structure probes ..................................................................... 37
12.1 The nature of exotic species ............................................................................................. 37
12.2 Exotic establishment and community assembly............................................................... 41
13
Conclusions, and the Otago Botany Lawn ........................................................................... 42
1 Introduction
We have outlined the processes that occur in plant communities: interference, subvention
46
litter effects and autogenic disturbance. Many ecologists wish to go no further with plant
47
communities than look at such processes, but we want to make generalisations at the plant
Wilson and Agnew, chapter 5, Assembly rules, page 2 of 50
48
community level. The ‘phytosociologists’ wish to make regional vegetation inventories using the
49
methods originating with Braun-Blanquet (1932), identifying and naming communities. This has
50
value in conservation advocacy and an ecological tourist’s guide, but we have to ask: where are the
51
testable hypotheses? In our approach we look for the rules of engagement in plant associations that
52
would arise from the interspecific interactions that we described in chapter 2. These are the
53
assembly rules, which we define as "restrictions on the observed patterns of species presence or
54
abundance that are based on the presence or abundance of one or other species or groups of species
55
(not simply the response of individual species to the environment)" (J.B. Wilson 1999a). This is
56
close to Hubbell’s (2005) definition of assembly as “which species, having which niche traits, and
57
how many species, co-occur in a given community”. We could argue that this is the true meaning
58
of phytosociology.
59
Some suggest that after a disturbance assembly rules will not be found until the community
60
regains equilibrium (Bartha et al. 1995). Support for this comes from speculation rather than from
61
evidence, and in any case disturbance of the autogenic kind is endemic to plant communities
62
(chapter 2). However, we shall tend we shall tend to concentrate on what seem to be equilibrium
63
communities. Another question comes from Yodzis’ (1978; 1986) distinction between founder
64
control of community composition and dominance/niche control. If the former be operating, the
65
species composition of a community will depend largely on which species arrives first and there
66
will be no further predictability, no rules. Ozinga et al. (2005) addressed this issue using a 20,000-
67
quadrat database. On average among species the first four axes of a CCA ordination constrained
68
by six Ellenberg scores explained only 7.7 % of species occurrences, though the value was 10.3 %
69
for species with long-lived seeds and a mechanism for long-distance dispersal. This implies a rôle
70
for founder control, though this conclusion relies on the completeness of the environmental
71
characterisation, the same problem we encountered with ‘chance’ as a mechanism of coexistence
72
(chap. 4, sect. 9). Another problem is that the species the ecologist sees are not those the
73
taxonomist sees. We have forsworn, in general, consideration of within-species genetic (e.g.
74
ecotypic) differences and plastic responses in this book, but both are important in the world. The
75
studies are often dealing with the realised niche of the species, which may be considerably
76
different from its fundamental niche (Austin and Gaywood 1994), and not easily predicted from it.
77
There is a widespread and commendable scepticism as to whether assembly rules occur at
78
all (e.g. Ulrich 2004). This may not be our conclusion, but our reductionist aim demands that we
79
start with such a null model and that we be especially careful in examining the evidence.
Wilson and Agnew, chapter 5, Assembly rules, page 3 of 50
80
2 What rules are we searching for, and how?
81
2.1 Inductive versus deductive
82
Inductive and deductive approaches both have value in community ecology (Dale 2002;
83
J.B. Wilson 2003) and both will be seen below. An example of the deductive approach is guild
84
proportionality in forest: differences between species in their mature height are well established
85
and we can reason that these represent different niches with the species potentially capable of
86
occupying to each niche constituting a guild. We can reason that a species will invade more readily
87
where few members of its guild are already present. If the null model is disproved, and if we can
88
rule out other explanations such as environmental effects, the existence of the rule has been
89
proved, though not its exact mechanism. On the other hand, a search for intrinsic guilds is
90
inductive in that we are not assuming any structure save that guilds might exist, but so long as the
91
guilds are formed and tested on independent data there is a strong pointer to where to seek the
92
processes that are structuring the community. Finding a repeated pattern is the first step to finding
93
its cause.
94
2.2 Randomisation tests
95
To demonstrate assembly rules, is it essential to compare an observed pattern with that
96
expected under a null model. However, the null model is often difficult to frame. What does a
97
plant community look like when it isn’t there? A prior question is what pattern to seek: what does
98
a plant community look like when it is there?
99
In these comparisons, randomisation tests are often needed, in which a test statistic is
100
calculated on the observed data, then on data randomised under a certain null model, and
101
significance (i.e. the probability that the observed results would occur under the null model) is
102
determined from the proportion of randomised values that are equal to, or more extreme than, the
103
observed one. There are traps here. Any test statistic can validly be chosen, though we should
104
ensure it tests the ecological question asked. Selection of the null model is more crucial; many
105
studies have come unstuck from choosing the wrong one and demonstrating as a result an obvious
106
fact such that species differ in frequency (J.B. Wilson 1995). We use the Tokeshi principle, that
107
the null model must include all the features of the observed data except the one it is intended to
108
test (J.B. Wilson 1999a). Lastly, tails: if it is conceivable that the observed data could differ from
109
the null model in either direction, i.e. results either way will be noticed, a 2-tailed test must be
110
used. This comprises either doubling the p value obtained, or using, say, two 2.5 % tails for a 5 %
111
test.
Wilson and Agnew, chapter 5, Assembly rules, page 4 of 50
112
113
2.3 Ruling out environmental variation
We must examine natural mixtures of species in a way that takes into account gross
114
environmental heterogeneity. It is no surprise that species are adapted top particular places along
115
an environmental gradient. The rules we find have to transcend in their generality ones of the type:
116
“species x occurs at low/high values of environmental factor z” (the “easy task” of community
117
ecology: Warming 1909). We need to search for reasons for species’ relative positions which are
118
not primarily environmental, but are based on species interactions. Therefore, in seeking assembly
119
rules, i.e. the repeated patterns of MacArthur (1972):
120
(a) The rules we seek will not necessarily depend on the identity of particular species. This
121
contrasts with Diamond’s (1975) original assembly rules, but that approach has not proved
122
useful.
123
(b) They will not simply describe the fact that species are correlated with their environment.
124
However, the rules cannot be expected to apply worldwide, in all habitats. For example, rules
125
based on stratification cannot apply to the very few communities that have no stratification, and
126
communities in deserts can be expected to be constructed quite differently from those in
127
rainforests. For character-based, limiting-similarity rules, the characters involved will be different
128
in different habitats, where different resources are limiting.
129
Environmental correlations, Warming’s “easy task” to investigate for their own sake, are
130
actually a huge problem in seeking assembly rules. Environmental variation occurs at all scales in
131
all communities (Goodall 1954). Often, when we are seeking assembly rules, environmental
132
variation acts as noise. Very commonly, the null model against which we are testing the observed
133
pattern assumes no environmental variation, so that if the analysis disproves the null model this
134
could be either because there really is an assembly rule or because environmental variation has
135
mimicked the effect. The latter possibility would not be interesting. Eliminating effects of the
136
environment is not easy.
137
Take the simple case of testing whether variance in richness differs from a null model.
138
Suppose there is environmental variation such that some habitats have few species (just ‘A’ in Fig.
139
5.1), but others have many (‘A B C D’) – the ‘waterhole effect’ of Edith Pielou (1975) (Fig. 5.1) –
140
with no variation of species richness within those habitats. The community structure is in fact
141
determinate, but will appear as greater variation in species richness than expected at random if an
142
overall randomisation – a ‘site’ model – is used.
Wilson and Agnew, chapter 5, Assembly rules, page 5 of 50
143
Environ. 1
Environ. 2
A
A
A
A
A
A
A
A
AB AB AB
C
AB AB AB
C
AB AB AB
C
AB AB AB
C
Environ. 3
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
Fig. 5.1: Four environments containing different species assemblages, consistent within each
144
environment.
145
Suppose the number of species is the same in each quadrat, and they are the same species
146
in each quadrat within each of two environments (Fig. 5.2). Randomisations will include some
147
quadrats with 0, 1, 3 and 4 species, and the observed state will look like constant richness
148
compared to this. The effect is real in that there is the same number of species in each
149
environment. However, this is being tested 20 times in each environment: pseudoreplication. A
150
test over several environments would be valid and interesting, but then one has to include each
151
community only once and one needs many environments.
Environment 1
152
153
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
Fig. 5.2: Two environments containing different species assemblages, but the same richness.
The best answer to these problems is to use a patch model rather than a site model. This
154
comprises making a prediction for each quadrat (the ‘target’ quadrat) on the basis of a limited
155
number of adjacent or otherwise similar quadrats (Fig. 5.3). The patch can be square (Fig. 5.3) or
156
linear, or a grouping of quadrats can be determined a priori as being similar in some other way.
157
Wilson and Agnew, chapter 5, Assembly rules, page 6 of 50
AB B
B
D
AB
C
CD
A
B
B
D
D
158
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
159
species A in the 3×3 patch is 3/9 = 0.333, so in the randomisation species A has a 0.333
160
probability of occurring in the central square.
161
162
2.4 Taxonomic-based limiting similarity
In animal ecology, membership of a genus is commonly used to indicate similarity in alpha
163
niche. In plants, the niche is commonly more independent of taxonomy and phylogeny, and
164
sometimes membership of a genus is more representative of a species’ beta niche (e.g. Salicornia
165
spp. all in saline areas). However, the genus is clearly an ecologically-objective and a priori
166
classification, and if taxonomy is not a good guide to ecology the result will be non-significance,
167
not spurious significance.
168
2.5 Process versus pattern
169
Ecologists often suggest that ‘assembly rule’ should mean the process by which the
170
community is established. Whilst this is a logical thought, Diamond (1975) first used the term for
171
the results of that process. Most later workers have used it in this way and we do so here.
172
3 Zonation
173
3.1 Boundaries in zonation
174
As Robert H. Whittaker (1975a) pointed out, the ideal way to determine whether species
175
are associated into discrete communities is to see whether their boundaries are clustered on an
176
environmental gradient, e.g., to distinguish between the situations in Fig. 5.4 a and b. Answering
177
the question is much more difficult (J.B. Wilson 1994b).
178
Species abundance
Wilson and Agnew, chapter 5, Assembly rules, page 7 of 50
a. Clustered
boundaries
b. Randomlyspaced
boundaries
179
180
Fig. 5.4: Whittaker’s diagram (part of) of different distributions of species along an environmental
181
gradient.
182
Shipley and Keddy (1987) examined the upper and lower species boundaries on 13
183
transects along 200 m of a lake shore in Ontario, Canada, and concluded they were significantly
184
clustered. There are some problems with pseudoreplication (J.B. Wilson (1994b). However, the
185
real problem is that Shipley and Keddy used elevation as the gradient. It is a proxy factor for those
186
actually affecting the plants and probably not linearly related to any of them. However, we do not
187
know the true factors, or on what scale to express them. The study of Auerbach and Shmida (1993)
188
of altitudinal zonation on Mt Hermon, Israel, has the same problems. Bimodality of species
189
distributions would be a mildly interesting feature, but evidence for it (e.g. Whittaker 1960; 1967)
190
is weak (J.B. Wilson et al. 2004).
191
The problem of defining the scale of environmental gradient was solved by Dale (1984) by
192
abolishing it. He took up a previous implication that looking at the sequence of top- and bottom-
193
boundaries up a gradient (an intertidal shore in his case) the top boundary (T) of one species would
194
be immediately followed by the bottom boundary (B) of another (the one replacing the other in the
195
same alpha niche): a TB pair. Therefore, overall there would be an excess of TB pairs compared to
196
expectation. This test is non-parametric, in that it is absolutely unaffected by any monotonic
197
rescaling of the axis. However, the non-null (H1) hypothesis assumes very precise replacement of
198
one species by another, with a small gap, which is hard to envisage in the real world (J.B. Wilson
199
1994b). It is surprising that Dale himself found excesses of TB pairs significantly often. Thomas et
200
al. (1999), using Dale’s method, did not.
201
It seems that since it is impossible to obtain evidence on community structure from overall
202
zonation, valid answers can be obtained only by changing the question. J.B. Wilson and Lee
203
(1994) formed a null model in which the number of species, their frequency patterns and positions
204
along an altitudinal gradient in the Murchison Mountains, southern NZ, and the number of species
205
in each genus were all held as observed. The test statistic was the amount of overlap along the
206
gradient between species in the same genus and in the null model species were assigned to genera
Wilson and Agnew, chapter 5, Assembly rules, page 8 of 50
207
randomly, without replacement. The concept is that members of one genus will tend to be similar
208
in alpha niche. They will compete with each other in either ecological or evolutionary time (the
209
“ghosts of competition past”), and hence be spread out in beta niche (altitude), with less overlap
210
than expected for a random selection of species. The results are complicated because testing
211
several genera separately comprises making multiple significance tests. Some genera are known to
212
have altitudinal biases (Pielou 1978, showed that this was true overall for the distribution of algal
213
congeners along a latitudinal gradient) and others have too few species to give significance.
214
However, taking all this into account Wilson and Lee concluded that there was evidence that the
215
species of a genus were more spread out in altitude than expected at random. However, we have
216
not solved the problem of how to find clustered boundaries over all species;
217
3.2 Fundamental and realised niche
218
Beta niche
219
We know that a species’ realised niche is related to its fundamental one (chap. 1, sect. 4.1),
220
but it is not clear just how. Generally, when two species with largely overlapping fundamental
221
niches meet in the field, their realised niches are different. For example, Kenkel et al. (1991) grew
222
three species, one a facultative halophyte, in a range of rather low salinities in sand culture. In
223
monoculture, they all grew best with no added NaCl, but in mixed pots they sorted themselves into
224
three realised-niche optima along the gradient. In most situations, one species moves further along
225
the gradient than the other. A well-known example is the work of Grace and Wetzel (1981)
226
growing two Typha (cattail) species on a gradient of average water depth. In monoculture, both
227
had the same optimum depth of 50 cm. In mixture they hardly overlapped in the depths at which
228
they grew: Typha latifolia moved its optimum to 15 cm and T. angustifolia to 80 cm. Similarly,
229
Pennings et al. (2005) investigated a southeast USA saltmarsh, where Juncus roemerianus grows
230
higher up on the marsh and Spartina alterniflora grows lower, with a sharp boundary between
231
them. The lower limit of J. roemerianus is set by the physical environment (salt and/or
232
waterlogging), but the upper limit of S. alterniflora is set by interference for without interference
233
from J. roemerianus it grew if anything slightly better in the latter’s normal zone than in its own.
234
Can we generalise? Austin (1982) grew five grass species in a greenhouse sand culture
235
with a range of nutrient solution concentrations, both in monoculture and in a five-species mixture.
236
Performance was calculated as shoot dry mass relative to the highest-yielding species in those
237
conditions. He found that in most concentrations the performance of a species in mixture was
238
correlated with its performance in monoculture, but the shape of the relation depended on the
239
nutrient level and was often markedly non-linear. Thus, a species’ realised niche could generally
240
but not always be predicted from its fundamental niche. Pickett and Bazzaz (1978) grew six
Wilson and Agnew, chapter 5, Assembly rules, page 9 of 50
241
species along an experimental soil moisture gradient in a greenhouse, in monoculture and in a 6-
242
species mixture. The optimum stayed in the same position for four of the species, but for most
243
species the peak of the response was sharper in the mixture. Fascinating results came from S.D.
244
Wilson and Keddy (1985), who examined a field gradient in organic content along a lakeshore.
245
The gradient is probably caused by wave action and it is correlated also with soil mechanical
246
composition, nutrients and water depth. Twelve of the species were also grown in sand : field-
247
organic mixtures, in pots but out-of-doors. The shape of the response to the gradient, field versus
248
experimental, was:
249

Not or hardly related: 5 species.
250

The opposite skewness: 3 species.
251

Related or vaguely related: 4 species. (The response was sharper in the field in one of
252
these, less sharp in another, equal in a third, and the relation was too vague to see in the
253
fourth.)
254
One possibility is that weaker competitors are pushed towards the less favourable end of the
255
gradient. This can be seen in the work of Pickett and Bazzaz (1978), where one of the two species
256
most suppressed by interference, Polygonum pensylvanicum, was pushed in mixed stands to the
257
dry end of the gradient, where overall growth was less. This seems to be the situation for Spartina
258
alterniflora in the work of Pennings et al. (2005), the species being restricted by interference to the
259
lower marsh. But can we generalise? No, not yet anyway.
260
Alpha niche
261
Niche shift (including 'Habitat shift') is a change in mean/modal resource usage by a single
262
species in different areas (Schoener 1986). It is the difference between fundamental and realised
263
alpha niches, or between realised niches with different associates. Such differences have long been
264
recognised (Gleason 1917). Niche expansion is a similar concept, except that the niche width
265
changes, not the mean/mode. There is disagreement in the literature, sometimes even within one
266
paper, as to whether these responses are plastic/behavioural or genetic.
267
Silvertown (1983) investigated whether the depths of species in limestone pavement grykes
268
were more different when they co-occurred (sympatry) than when they were alone (allopatry) – a
269
test for niche shifts. However, he found that the species occurred at more similar depths when in
270
sympatry. Presumably any niche shifts were obscured by differences between grykes, e.g. species
271
can occur deeper in large grykes. Veresoglou and Fitter (1984) suggested that when Holcus
272
lanatus was growing with certain species (their Area III), its nutrient uptake peaked earlier than in
273
other communities. However, this was true for only one of the two nutrients they examined. Even
274
then, Area III could have been different in other ways. Niche shift has been found in rooting depth.
Wilson and Agnew, chapter 5, Assembly rules, page 10 of 50
275
Nobel (1997) found that rooting depths for the three co-dominant species in a site in the Sonoran
276
Desert were 9-10 cm for isolated plants, but roots for interspecific pairs in close proximity
277
averaged 2-3 cm more shallow for Agave deserti and 2-3 cm deeper for the other two species. The
278
results of O'Brien et al. (1967; see chap. 3, sect. 6 above) and Bookman and Mack (1982; see chap.
279
2, sect 2.2 above) are similar.
280
4 Species sorting
281
4.1 Species associations in succession
282
Greig-Smith (1952) suggested that species associations would change through succession,
283
and Gitay and Wilson (1995) synthesised these suggestions with the terms of Watt (1947) to
284
suggest three phases in succession:
285
286
287
288
289
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, 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
290
larger scale, giving stronger associations, with negative ones predominating if different
291
communities have approximately equal species richness.
292
Gitay and Wilson (1995) analysed tussock grasslands with a known time of secondary succession
293
since burning. The expected pattern was seen – association was low and rather negative for the
294
first 10 years, close to zero (negative and positive associations balancing) at 10-20 years, and more
295
negative beyond 20 years. The processes seem likely, but the model was probably subsequent to
296
the data. The model was confirmed in a restoration experiment at Monks Wood, eastern England,
297
where over 13 yr rank consistency (Watkins and Wilson 1994) increased during the pioneer phase,
298
was maximal in the Building phase and then decreased markedly in the Mature phase. An identical
299
but non-significant trend was seen over the 6 yr of a restoration experiment in the midlands of
300
England.
301
Greig-Smith (1952) in Trinidad tropical rain forest found evidence for the Pioneer and
302
Building phase, but there was little indication of non-random distribution in 1.5 × 1.5 m plots.
303
O'Connor and Aarssen (1987) in Ontario sand quarries of various ages, expected to see what we
304
have called the Mature phase developing, but in fact the frequency of negative species associations
305
decreased with time. Malanson (1982) approached this question differently: vegetation patches on
306
canyon walls in Utah should show greater dissimilarities if they were safe from floods and the
307
species had time to assemble into communities, but if anything the opposite was true.
Wilson and Agnew, chapter 5, Assembly rules, page 11 of 50
308
Aarssen and Turkington (1985a) compared three pastures of different age in western
309
Canada. They claimed consistently stronger and more negative associations between grass species
310
in the older pastures, though the relevant information presented shows that the total number of
311
significant associations (positive plus negative) is lower in the oldest pasture. They do give figures
312
to demonstrate that the number of associations were more consistent over seasons and years in
313
both direction and significance in older pastures. Turkington and Mehrhoff (1990) interpret this as
314
“transition from an essentially unorganised assemblage of species to a more organised
315
community”.
316
This approach is potentially interesting. No investigation so far has given firm indication of
317
deterministic structure and several results have been opposite to theoretical expectation. However,
318
there seems to be only weak theoretical support for the concepts in the first place.
319
4.2 Compositional convergence
320
It would be fascinating to see how similar species assembly was in identical conditions. We
321
can never quite do this, but Crawley et al. (1999) approached it by sowing a mixture of 80 forbs
322
into six replicate blocks in an experimental field. After seven years, Tanacetum vulgare (tansy)
323
predominated among the sown species , varying across five of the blocks from 10 % to 72 % of the
324
standing crop (Table 5.x). However, it comprised only 0.1 % in the block 3, where four of the
325
other five sown species present exceeded it. There is no convergence here. Amongst the volunteers
326
the most abundant was Alopecurus pratensis (foxtail) which reached 86 % in one block but was
327
absent in block 6, then Holcus lanatus (Yorkshire fog) with a 64 % maximum but absent from four
328
of the six blocks, and Arrhenatherum elatius (oat grass) varying 0 – 31 %. Again, huge ranges.
329
Recall that these plots had been made as similar as possible. Crawley et al. describe this variation
330
amongst them as a “quite remarkable degree of similarity”, but we would describe it as quite
331
remarkable dissimilarity. Crawley et al. (1999) went further and described the blocks as
332
remarkably similar in species diversity, but in fact variance in species richness between blocks was
333
three times greater than expected at random, and significantly so (for the method, see J.B. Wilson
334
et al. 1987).
335
Table 5.x. Dry mass (g m-2) of selected species in six replicate plots in the experiment of Crawley
336
et al. (1999).
Replicate →
Tanacetum vulgare
Alopecurus pratensis
Holcus lanatus
Arrhenatherum elatius
1
10
86
–
4
2
32
33
–
31
3
<1
61
–
3
4
72
28
–
<1
5
12
46
41
–
6
27
–
64
9
Wilson and Agnew, chapter 5, Assembly rules, page 12 of 50
337
5 Richness
338
A basic question in community ecology is whether there is a limit to the number of species
339
that can be packed locally. Testing for low variance in species richness (J.B. Wilson et al. 1987) is
340
a direct approach to this. If the niches are not primarily defined by the species themselves (chap. 1,
341
sect. 4.1) there will be a limit to the number of niches in a type of community. Since there can be
342
no more species present in a sample (quadrat) than there are niches, the number of species in a
343
quadrat should be limited by the number of niches and be rather constant across quadrats. To be
344
precise, there should be lower variance in the species richness of quadrats than would be expected
345
under a null model in which the number of occurrences of each species is held at that observed, but
346
those occurrences are scattered across the quadrats, independently of other species. It is often
347
difficult to see such an effect because of overlain environmental variation and disturbances, and
348
perhaps the presence of empty niches. Possibly for this reason J.B. Wilson et al. (1987) failed to
349
show variance in two communities at scales of 5 × 5 m and 2 × 2 cm respectively, and J.B. Wilson
350
and Sykes (1988) at 10 × 10 m. However, Watkins and Wilson (1992) found lower variance than
351
expected under the null model at the scale of 13 × 13 mm, and this remained for six of the 12
352
lawns when analysed with a patch model (section 2.3 above). There may be remaining doubts that
353
the limitation at this scale is due to geometric packing of individuals, but variance in richness
354
remains a basic question.
355
In other approaches to niche limitation, Levine (2001) sowed seeds of several native and
356
exotic plant species into tussocks of Carex nudata and found even the most diverse tussocks were
357
colonised, and concluded that they had not been completely saturated with species. However,
358
some of these species might not have persisted in the long term. E.O. Wilson (1961) concluded
359
that most of the ant faunas of the Moluccas-Melanesian are saturated, using as evidence a close
360
correlation between the size of the fauna and the area of the island. Cornell and Lawton (1992)
361
suggested that it would be possible to identify niche saturation from the relation between regional
362
and local richness. If there be niche saturation, then as the regional species pool increases, local
363
richness will increase proportionally at first, but level off to a maximum. If there be no saturation,
364
the relation will continue to be linear. It is easy to show in models of community assembly that
365
such saturation will occur (Fukami 2004), but will it in the real world? Although it is easy to
366
determine richness at the site level, the estimation of regional species pools involves too many
367
arbitrary and subjective decisions. There is also a problem of circularity: local richness is affected
368
by the regional richness, as Cornell and Lawton reasoned, but regional richness is a combination of
369
local (alpha) richness and beta richness and so is not independent of it. J.B. Wilson and Anderson
370
(2001) concluded that comparisons between habitats are not appropriate for such tests because of
Wilson and Agnew, chapter 5, Assembly rules, page 13 of 50
371
overlap of the species pools and because ecological processes differ between habitats. Only
372
comparisons between equivalent habitats on different continents are valid, and they cannot be
373
made because there are too few continents for a statistical analysis. A wooden light bulb is
374
beautiful and interesting, but of little use (J.B. Wilson and Anderson 2001); likewise saturation
375
from the species pool concept is stimulating, but it is probably operationally impossible to test.
376
There have also been simple comparisons between different continents in florule size and
377
quadrat species richness. As Orians and Paine (1983) say, “Implicit in community convergence in
378
species richness patterns is the notion that assemblages eventually reach some saturation level”.
379
However, such comparisons have generally found the areas compared to differ in richness at both
380
area and quadrat scales, e.g. annual grassland in California and Chile by Gulmon (1977), the floras
381
of California and Israel by Shmida (1981) and in the brown intertidal algae in various points
382
around the Atlantic, Pacific and Southern oceans by Orians and Paine (1983). Richness
383
convergence would have implied niche saturation; divergence does not disprove saturation,
384
because the habitats may not be as similar as we hope, or there might be niche straddling/splitting,
385
it is impossible to tell. Robert H. Whittaker travelled the world recording species diversity in a
386
standard way and in plots whose exact location was carefully selected (JBW pers. obs.), attempting
387
to find patterns and thus predictability. In Whittaker (1977) he had reached the conclusion, which
388
he put in a more straightforward way in seminars, “We once thought species diversity was the one
389
fixed, predictable feature of plant communities. But it isn’t”.
390
6 Limiting similarity
391
Abrams (1990) assumed that if two species were too similar in resource-use patterns one
392
would be excluded. This is a present-day reassertion of the Principle of Gause (1934), that species
393
that are too similar will tend not to occur together. The concept has also been referred to as
394
‘community-wide character displacement’ or ‘ecological character displacement’ (Strong et al.
395
1979). Hutchinson (1959) instigated this topic, as he instigated so much in ecology, by observing
396
that in some mammals and birds of Britain, Iran and the Galapagos Islands, the morphological size
397
ratio between each species and the next larger one was about 1 : 1.3 for a linear measure. He
398
actually reported a range of 1:1.1 to 1:1.4, but this is usually forgotten. Hutchinson implied that
399
this was partly due to within-species character displacement and there is some evidence for that in
400
his data. So far as we know this has not been applied to plants. MacArthur and Levins (1967) put
401
the idea of a limiting similarity between the niches of co-existing species on a solid mathematical
402
foundation, be it with some assumptions.
403
404
The quantitative predictions of the MacArthur and Levins theory have not been tested, but
even qualitative testing has been fraught. It is even difficult to know what test statistic to use – e.g.
Wilson and Agnew, chapter 5, Assembly rules, page 14 of 50
405
minimum distance, even distances, greater range – or which characters are appropriate (Stubbs and
406
Wilson 2004). It is usually unclear what we are trying to test: plastic responses, exclusion by
407
interference between species, character displacement or the co-evolution of species. Hubbell
408
(2005) concluded: “The empirical evidence, in general, has not borne out these [MacArthur and
409
Levins, etc.] predictions …, particularly in plant communities”, and further, “Does a limiting niche
410
similarity for species in functional groups exist? … I believe the answer to [this] question is no (at
411
least in plants)” (op. cit.). We wish to look further, and with plants at that.
412
Terminology has been a problem. When co-occurring species are closer in character space
413
(i.e. more similar) than expected, the terms used have included ‘clumped’ and ‘aggregated’; when
414
they are less similar terms have been ‘evenly-spread’, ‘evenly-spaced’, ‘spaced-out’, ‘staggered’
415
and ‘regular’. These terms are self-explanatory. ‘Overdispersed’ and ‘underdispersed’ and have
416
also been used. This is unfortunate because overdispersed is the mathematical term for clumped
417
and underdispersed for evenly-spread (Greig-Smith 1983). For obvious reasons undergraduates
418
often use them in the opposite, incorrect, senses, and this can be found even in the literature (e.g.
419
Weiher et al. 1998). They are therefore ambiguous in usage, and are best avoided.
420
As elsewhere, species have usually been used as units, ignoring polyploids, other within-
421
species variation, within-plant somatic variation and generally dioecy.
422
6.1 Limiting similarity in morphological characters
423
Cody (1986) reported a number of pieces of evidence for limiting similarity amongst
424
woody plants of desert and South African fynbos. In the Granite Mountains, Mojave Desert,
425
California, he demonstrated that the Opuntia species which are shallow-rooted, were negatively
426
associated, but Yucca schidigera, which is somewhat deeper-rooted, was positively associated with
427
all the Opuntia spp. For four fynbos sites, he showed spreading-out of species of the major
428
proteaceous shrubs in a morphology space of leaf shape and leaf length, with little overlap
429
between species. Positions in morphological space were occupied by different species in different
430
sites and the position of some species changed between sites, both making the spread that was
431
observed even more notable. However, no probabilistic test against a null model was made and a
432
null model would probably not be easy to frame, but the patterns are compelling. The one
433
exception to the morphological sorting was between Protea eximia and P. nitida, and they
434
occurred in different aspect micro-habitats. Most remarkably, in some species, notably
435
Leucadendron salignum, plants of the two sexes overlapped considerably on each of the axes, yet
436
were largely separate in the 2-dimensional morphological space. For Leucadendron, Cody offers
437
evidence that species pairs that are more similar in the 2-D space co-occur less often than expected
438
at random. He also found indication that the 80 species of Leucadendron in Cape Province, South
Wilson and Agnew, chapter 5, Assembly rules, page 15 of 50
439
Africa, were more spread in morphological space than expected by chance, but with only 20
440
randomisations the probability cannot be accurately determined, and details of the null model are
441
not clear, especially the treatment of the edges of morphological space. Cody’s work is fascinating
442
and it would be wonderful for some of these leads to be followed up in more detail.
443
In careful work, Armbruster (1986) examined the association of Dalechampia species at 12
444
sites in Central and northern South America with unique combinations of Dalechampia species
445
(reduced from 26 populations observed in the field). In the ecological sorting (“pure assemblage”)
446
null model, the Dalechampia species richness of each site was fixed at that observed and the
447
species frequencies, whilst not so fixed, were taken as probabilities of occurrence. As with most
448
assembly rule work, environmental differences between sites are potentially confounding, no less
449
and probably no more so than in work on a micro scale. Armbruster coped with this by using five
450
different species pools taking into account climatic and geographical ranges. In effect this is a
451
patch model on a grand scale. The test statistic was the number of cases where two species that
452
were similar in pollinator usage co-occurred (within 50 m) at a site, pollination vectors being
453
determined by observation and flower morphology. After this careful work p was 0.16, not
454
significant. Twelve sites are really too few for a good test. Another model, with character
455
displacement, does not strictly concern us here since we are limiting ourselves to ecological
456
assembly eschewing ecotypic differentiation, but the results were significant, though only using a
457
1-tailed test which is debatable. A decade later at 25 sites in Western Australia, Armbruster et al.
458
(1994) performed a similar study on Stylidium species: another genus with complicated floral
459
organs. The test statistic was overlap in the morphological similarity in the flowers of species co-
460
occurring at a site, and again there was a large-scale patch model based on habitat and geography.
461
Only one site with overlap was observed, compared to an average of 4.38 expected under the null
462
model, but this result was not significant (p = 0.055, but perhaps we should double this to 0.11 for
463
a 2-tailed test). Again there was significant character displacement.
464
Weiher et al. (1998) tested for limiting similarity in herbaceous riverside vegetation, with
465
quadrats placed to deliberately give a range in environment (soil fertility and disturbance) and
466
vegetation (“from cattail marshes, to wet sedge meadows to sandy beaches”), measuring 11
467
vegetative characters. They found a significant tendency for the minimum nearest-neighbour
468
distance in 11-character space to be greater than expected under their null model, though other test
469
statistics did not give significance. Four of the individual characters showed even spreading. They
470
concluded that there are morphological assembly rules that constrain wetland plant community
471
composition. The main problem with this work is that there was no attempt to avoid environmental
472
heterogeneity, or to allow for such heterogeneity in the analysis by a patch model or the like, so
473
the null model they used combined species from several species pools. This means that the
Wilson and Agnew, chapter 5, Assembly rules, page 16 of 50
474
departures of the observed data from their null model are likely to reflect species habitat
475
preferences, rather than community structure resulting from limiting similarity, as discussed
476
above. To put it another way, there was pseudo-replication of the habitat differences, as there
477
would be if we analysed the data of Fig. 5.2 by randomising across habitats. It is as if we saw one
478
person’s garden with two species, one a dicot and one a monocot, and a second person’s garden
479
also with one dicot and one monocot. Is this limiting similarity in the plants that gardeners choose?
480
Perhaps, but we would never get significance from four species in two gardens. Suppose we
481
sampled each garden 100 times, with of course identical results, and analysed the whole dataset.
482
We would get significance, but it would be spurious because we had pseudoreplicated. That is in
483
effect the trap into which the brave attempt of Weiher et al. fell, an illustration of the traps that
484
await those who are less careful than Armbruster was.
485
Stubbs and Wilson (2004) attempted to avoid previous traps when they tested for limiting
486
similarity in a New Zealand sand-dune community. Twenty three functional characters were
487
measured on each of the species, covering the morphology of the shoot and root systems and
488
nutrient status and intended to represent modes of resource acquisition. Since it is not clear at what
489
scale limiting similarity would occur, sampling was at four spatial scales, from a single point up to
490
a scale of 50 m2. These multiple scales allowed patch models to be used. A carefully-selected
491
range of test statistics was used, for example excluding any that were affected by the range of
492
character values. A test over all characters found that the mean dissimilarity between nearest-
493
neighbour species in functional space and the minimum dissimilarity were greater than expected
494
under the null model at the 0.5 × 0.5 m scale, supporting the MacArthur and Levins (1967)
495
limiting similarity concept. However, the actual community did not follow the theory to the extent
496
of showing an even spread of species in functional space. Limiting similarity effects were seen
497
even more consistently in separate root and leaf characters when within-species variation was
498
taken into account to calculate measures of overlap – the test most closely aligned to MacArthur
499
and Levins’ original theory. The characters showing limiting similarity were mainly those related
500
to rooting patterns and leaf water control and thus probably reflected the acquisition of nutrients
501
and / or water. The implication that competition for water and nutrients limit coexistence seems
502
reasonable for a sand dune. The main problem with this work is the number of tests made – four
503
spatial scales, 23 characters and different test statistics. This seems inevitable when analysis of
504
limiting similarity in plant communities is in its early stages and it is not yet known at what scales,
505
in what characters and how it will operate. However, the overall results are convincing.
506
Armbruster (1995) suggested that limiting similarity due to ecological sorting would
507
operate more readily in vegetative characters than in reproductive ones, and comparison of his own
508
ecological-sorting results with the results of Cody and of Stubbs and Wilson supports this. Hubbell
Wilson and Agnew, chapter 5, Assembly rules, page 17 of 50
509
(2005) was too dismissive. Limiting similarity exists in plant communities and can be
510
demonstrated.
511
6.2 Limiting similarity in phenology
512
The simplicity of time as a niche axis has led to several attempts to ask the question – are
513
the flowering times of the species in a community evenly-spread? That is, is there a constraint on
514
the phenology of species which can co-occur? In such work, either the position of species
515
flowering peaks can be compared, or the time span of flowering, or quantitative measures such as
516
the number of flowers open at any time. The ecological and evolutionary selective pressures
517
against species that are too similar in flowering time would come from several interactions
518
discussed in chapter 2, such as competition for pollinators/dispersers, pollen wastage, interference
519
on the stigma and mal-adapted hybrids. On the other hand, aggregation of reproductive events
520
could be an adaptation to attract pollinators/dispersers, to combat predators, or a response to
521
pollinator/disperser availability (Thompson and Willson 1979).
522
Investigation was sparked when Stiles (1977) claimed to find evenly-spread flowering for
523
hummingbird-pollinated plants in a Costa Rican tropical forest. Statistical analysis of this dataset,
524
and of such datasets in general, has proved difficult and controversial; an excellent summary is
525
given by Gotelli and Graves (1996). In general the more recent studies use appropriate
526
randomisation tests and are valid. Similar tests have been made for an even-spread of fruiting.
527
Ashton et al. (1988), examining the six species of Shorea section Mutica in tropical rain
528
forest in Malaya, found even spread “at the 4.6 % confidence level”, but it is not clear whether this
529
was a 2-tailed test. Wright and Calderon (1995) tested separately 59 genera from Barro Colorado
530
Island. Flowering times were aggregated in some genera, but evenly-spread in six genera, so far as
531
one can tell, converting the two 1-tailed tests into a 2-tailed one and within the limited number of
532
randomisations used. Thies and Kalko (2004) found that eight forest Piper species flowered within
533
a short period and at random within that, but fruiting was evenly-spread. The p values were not
534
adjusted to give a 2-tailed test, though the results may have been significant anyway, again with
535
few randomisations. Burns (2005), in 10 woody angiosperms common below the canopy of conifer
536
forest in an area of British Columbia, Canada, found no evidence for a significantly even spread of
537
fruiting times. Poulin et al. (1999) examined fruiting phenology in central America. Data for the
538
fruiting times of Miconia (Melastomataceae) species from Barro Colorado Island were not
539
significantly different from a null model, but those from the genus in Trinidad and Colombia
540
showed significantly-even fruiting times, though again with few randomisations. In Psychotria
541
(Rubiaceae), fruiting times were aggregated. Overall conclusions are difficult, especially with the
Wilson and Agnew, chapter 5, Assembly rules, page 18 of 50
542
danger that non-significant results or aggregation are under-reported, but it seems that even
543
spreading of reproductive phenology sometimes occurs.
544
Not all niche differences in pollination are via phenology and interesting conclusions can
545
be made bringing in other information. Pleasants (1980) calculated from flowering-time overlap
546
and flower densities the potential for competition for pollinators between bumblebee-pollinated
547
species in some Rocky Mountain Meadow species; he found that such competition was negatively
548
correlated with presence/absence association between the species.
549
There are major problems with all such studies:
550
a. It is difficult to know whether to compare overlap between the most similar neighbours, or
551
between all possible pairs of species (Pleasants 1990). Probably species are affected by the
552
cumulative competitive pressure from several, but not all, species.
553
b. Flowering times are usually aggregated on a seasonal scale. In temperate areas, few species
554
flower in winter, but there is normally aggregation in the tropics too, corresponding to
555
wet/dry seasons (Stiles 1979; Wright and van Schaik 1994). There can be up to three peaks
556
per year (Parrish and Bazzaz 1979). It is very difficult to demonstrate even spread when it is
557
laid over aggregation.
558
c.
Even within the flowering season (or within a clump), there is usually variation, with fewer
559
species flowing at the beginning and end. Although it would be possible to estimate this
560
variation from the data, incorporation of it in a null model starts to involve circular
561
reasoning. This problem is probably insoluble.
562
d. There will probably also be variation in pollinator availability, so pollination competition
563
will be more intense at the two ends of the season with few insects (e.g.) around. This will
564
actually tend to mitigate problem ‘c’ above.
565
e. The patterns in flowering/fruiting could be caused by any of four processes: (1) ecological
566
assembly by exclusion by interference between pre-adapted species or ecotypes (i.e.
567
ecological sorting), (2) coevolution of species, (3) evolution of co-adapted ecotypes within
568
species (i.e. character displacement), or (4) plastic responses (i.e. niche shift). Rarely is it
569
clear which process particular workers have been intending to test. Most recent studies have
570
been based on in-situ observations of phenology. Although this sounds commendable, it
571
would actually be preferable to use data on the species generally, even from deliberately
572
outside the area, in order to exclude ‘3’ and ‘4’ and narrow the possible explanations. Co-
573
evolution of species (‘2’) seems unlikely here because most species occur in several
574
different communities, with different neighbouring species, and could not adapt their
575
flowering times to each community. Ecotypic differentiation (‘3’) would be difficult when
576
species associations are constantly changing. Plasticity, (‘4’), at sight unlikely, is possible
Wilson and Agnew, chapter 5, Assembly rules, page 19 of 50
577
since fruit removal from a plant often causes its flowering period to be extended. Analysis
578
with multiple null models (as performed by Armbruster 1986; Armbruster et al. 1994) would
579
be needed to distinguish between these possibilities.
580
581
f. Relative flowering time may not be consistent from year to year, because species are
responding to different signals (Rathcke and Lacey 1985).
582
Vegetative phenology might also constrain the coexistence of species. For example, Parrish and
583
Bazzaz (1976) commented that among the six oldfield species they examined only one pair was
584
similar in the time of peak root growth. Comparison with a null model would have been useful.
585
Veresoglou and Fitter (1984) found differences in vegetative phenology (growth and nutrient
586
uptake) between co-occurring grasses, suggesting that this helped permit coexistence between
587
them, but again they compared with no null model. Rogers (1983) examined sorting of species by
588
vegetative phenology amongst the vernal guild of herbs in North American deciduous forest.
589
Effects of environment producing negative correlations were potentially removed by excluding
590
species pairs with negative correlations at a larger scale (50 × 100 cm), though in fact none were
591
found, an approach conceptually related to the method of Dale (1985). Associations between
592
species in the same guild (ephemeroid, summergreen, annual) were no more or less frequent than
593
between species in different guilds.
594
This is an interesting approach to community structure. It is mainly restricted by
595
difficulties in specifying a null model in which the test focuses on possible assembly rules. Some
596
evidence for such rules has emerged.
597
Cody and Prigge (2003) made the curious observation that individual shrubs of Quercus
598
cornelius-mulleri affect each others' phenology of leaf replacement. Late and early timing
599
alternated annually within individuals and between large or close individuals in space. People are
600
asking: “How do the plants decide which is to go early and which later?”. The authors proposed
601
that these phenomena could be due to resource depletion or the cost of early bud break. Cody and
602
Prigge do not suggest how staggering of leaf replacement affects fitness. This is an interesting case
603
which could be considered as either interference or subvention. It seems a sort of assembly rule,
604
but it is difficult to know how to characterise it.
605
7 Guild proportionality
606
7.1 Concept
607
608
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:
609
1. Species arrive at a point and some establish (cf. chapt 1, sects. 2.3-2.5: the challenge).
610
2. A further species arrives:
Wilson and Agnew, chapter 5, Assembly rules, page 20 of 50
611
2a. The species may fail to establish. Failure is more likely if the new species is similar in
612
resource use to the majority of the species already present, i.e. it is a member of the
613
same alpha guild (Fig. 5.8), or
614
2b. If the new species does establish, and species previously present are excluded, the
615
excluded species are more likely to be from the same alpha guild as the newly-
616
established species.
617
Fig. 5.8: In Patch 1 there is only one tree species, so further tree species will find it easier to
618
invade; in Patch 2 there is only one understorey species, so further understorey species will
619
find it easier to invade. This will even up the guild proportions between the two patches.
620
We have to be careful here, because in the simple scenario above invasion will be determined by
621
the total abundance of each guild, not the number of species in it, so we have to suppose within-
622
guild differences, micro-niches. Note that the mechanism is assumed to operate at a small enough
623
scale to allow the constant possibility of movement of disseminules and challenge. The result
624
would be a tendency towards a relative constancy in the proportion of species from each of the
625
guilds - 'guild proportionality' (J.B. Wilson 1989b). Not exact constancy in the real world, but less
626
variation than in a null model and the appropriate null model here is one that holds both quadrat
627
richnesses and species frequencies equal to those observed. The finding of guild proportionality
628
would mean: (1) there is constraint on species presence and, (2) it is at least partially related to the
629
characters used in the guild classification. These must be alpha guilds since, to quote Pianka
630
(1980), they refer to niche in the “narrow sense of resource utilization”.
631
7.2 Evidence: constancy in space
632
The first application of the concept to a plant community was by J.B. Wilson et al. (1989b) in
633
a New Zealand rainforest, sampled with quadrats 2 m in diameter. The guilds were synusiae (strata,
634
lianes and epiphytes). The ground and herb strata showed significant guild proportionality when
635
coastal broadleaved forest and Nothofagus forest were combined, which is not ideal, and a site model
Wilson and Agnew, chapter 5, Assembly rules, page 21 of 50
636
was used, which casts doubt on the results. Bycroft et al. (1993) found significant guild
637
proportionality at the scale of 1 × 1 m in the herb stratum of a New Zealand Nothofagus forest, but
638
only with a site model, not with a patch model, reinforcing the doubts on the J.B. Wilson (1989b)
639
analyses. J.B. Wilson and Watkins (1994), sampling eleven lawns at a scale of c. 13 × 13 mm and
640
using a 3 × 3 quadrat patch model, found significant guild proportionality between graminoids and
641
forbs in three of the lawns, but only in the more species-rich quadrats as if the limitation did not
642
operate whilst there were empty niches. This was promising. J.B. Wilson and Roxburgh (1994)
643
sought guild proportionality in one of those three lawns, the University of Otago Botany Lawn,
644
using point quadrats. Again there was a significant guild proportionality using graminoid versus
645
forb guilds. We shall synthesise the Botany Lawn data in section 13.
646
Elsewhere, Klimeš et al. (1995) recorded for five years 30 × 30cm permanent quadrats in
647
two meadow communities, that differed in fertilisation and mowing regimes. There were many
648
cases of guild proportionality using a wide variety of guild classifications and fewer cases of
649
variance excess. Yet, to be frank, plant community structure is often so elusive that we should be
650
cautious when it is found. Using a site model, there could possibly be problems with
651
environmental heterogeneity even within the 1.5 × 1.5 m area, but more worrying is that many of
652
the guilds that showed significance were in characters typically of beta-niche differentiation, not
653
characters that represent differences in resource use at one spot (i.e. alpha). Light response could
654
relate to stratification in the community, but how could there be alpha niche differentiation, i.e. at
655
one point, in pH and soil nitrogen? The winter-green guild is more convincing, suggesting
656
phenological guilds, and with that guild there were significant differences in the fertilised meadow
657
in 4 years out of the 5 recorded.
658
Weiher et al. (1998) analysed their rivershore data (see above) for guild proportionality.
659
They reported significant guild proportionality for three guilds, but discounted them after
660
Bonferroni correction. The use of Bonferroni is problematic here since the tests include
661
complementary guilds and are thus far from independent. However, the much greater problem is
662
the deliberate combining of different habitats. J.B. Wilson and Whittaker (1995) found highly
663
significant guild proportionality on a saltmarsh for two, though related, a priori guild
664
classifications: narrow versus broad leaves and monocots versus dicots; they analysed with a patch
665
model. J.B. Wilson and Gitay (1999) found significant guild structure at 10 × 10 cm scale in the
666
inter-tussock vegetation of 21 sites of a New Zealand grassland. Kikvidze et al. (2005) analysed
667
subalpine meadows in Georgia (Caucasus), using 4× 4 cm quadrats. Index of guild proportionality
668
RVgp for the proportions of monocots and dicots was 0.64, impressively below the null-model
669
value of 1.0 and highly significant. A site model was used, but the reality of the result was
670
reinforced by an interference experiment, where the yield of a monocot+dicot mixture was greater
Wilson and Agnew, chapter 5, Assembly rules, page 22 of 50
671
than for either monocots or dicots alone. Bossuyt et al. (2005) analysed 52 1 ×1 m quadrats, each
672
in a different dune slack in western Belgium and northern France, using forb versus graminoid
673
versus shrub guilds. They found highly significant guild proportionality with forbs. The sampling
674
of 52 slacks differing in age from 5 to 45 years makes us worry about environmental artefacts.
675
Using C, S, R they found significant guild proportionality with ruderals. This is difficult to
676
understand. There could well be disturbed patches for ruderals within each 1 × 1 m quadrat, but a
677
proportion more constant than expected at random? How would this arise?
678
Great care is necessary with evidence for guild proportionality, partly because community
679
structure is so elusive, and partly because it is so easy to obtain artefacts from habitat variation.
680
The danger is that with habitat variation the null model may be inappropriate. In the case of guild
681
proportionality, if A and C in Fig. 5.2 are in one guild and B and D in another, each observed
682
quadrat has guild proportions of 0.5:0.5, with zero variance in this. If occurrences could be
683
randomised (i.e. with somewhat different quadrat and species totals), absolutely constant guild
684
proportions would be seen, compared to considerable variation in the null model. This would be
685
guild proportionality that was highly significant but spurious, being not from species interactions
686
but from environmental control. It is a real result that each environment has one species from each
687
guild, but we are multiplying the difference between the environments 20 times –
688
pseudoreplication.
689
The concept of guild proportionality can be seen at a biogeographic scale in the conclusion
690
of Gentry (1988) that the familial composition of tropical rain forests is remarkably constant. For
691
example, members of Fabaceae virtually always dominate neotropical and African "lowland
692
primary forests"; the plant families represented are "almost entirely" the same in the New World as
693
the Old. He saw similarity at the generic level too, for example between the New World and
694
Madagascar. These are fascinating observations. Gentry comments that it "can hardly be due to
695
chance", but he made no comparison with a null model. The finding is relevant to guild
696
proportionality only if families occupy particular niches, Gentry's "familial-specific niches", but
697
how else could the result arise? As with taxonomic guilds in general, non-significant results would
698
be unsurprising, but significant ones are valid.
699
Mohler (1990) made a comparison at the subgeneric level, within Quercus (oak) at various
700
sites across the USA. For 12 of the 14 regions that he examined (apparently with a variety of
701
quadrat sizes) there was a significant tendency for the two most abundant oak species to be from
702
different subgenera. This was not related to consistent pairing of particular species. His null
703
hypothesis was a 0.5 chance of each subgenus, which assumes they are equal in size, but this
704
would bias the test against the situation he found. The data were collected in various ways, but his
705
consistent result is in spite of this. It was apparently an a posteriori test (i.e., he thought he saw an
Wilson and Agnew, chapter 5, Assembly rules, page 23 of 50
706
interesting effect so he tested it), but the consistency of the effect over several regions largely
707
overcomes this problem. Mohler examined various explanations: disease/pest pressure, niche
708
differences in fruiting phenology through mast fruiting, dispersal differences, etc., but could not
709
find any clear single explanation. This approach was considerably extended in careful work by
710
Cavender-Bares et al. (2004). They examined several Quercus spp. in three reserves in central
711
Florida, USA. Characters that tended to be similar in more frequently co-occurring species
712
included bark thickness, radial growth rate, seedling absolute growth rate and rhizome resprouting.
713
These are characters that probably adapt to water stress, fire tolerance and soil fertility. Habitat
714
preferences were more scattered across the phylogeny than expected at random, suggesting that the
715
three sub-genera occupied different alpha niches and within those had evolved to cover the beta-
716
niche range, mainly in moisture availability. In reconstructions of phylogeny from ribosomal
717
DNA, the characters indicated as changing less within a clade included acorn maturation time,
718
embolism due to freezing, wood density and second-year vessel diameters. Seedling leaf lifespan
719
and perhaps SLW tended non-significantly in that direction. Characters that tended to be dissimilar
720
in co-occurring species, indicative of different alpha niches, were acorn maturation time, embolism
721
due to freezing, leaf life span and first-year vessel diameters and, non-significantly, SLW and
722
perhaps seedling leaf lifespan. Because of the tendency for species from far parts of the phylogeny
723
to co-occur, this should be a similar list to the list of conservative characters and it is almost
724
identical. These should be characters that are related to alpha niche and it is less easy to see how
725
they are. Cavander-Bares et al. suggest that acorn maturation time might be related to phenological
726
niche differentiation in masting and seedling regeneration, they imply that frost tolerance might be
727
related to year-to-year weather variation and leaf lifespan to timing of nutrient uptake. The crucial
728
correlation (p < 0.034) is that species that co-occur more often are more distant on their
729
‘phylogenetic tree’. However, this is essentially a test between habitats and therefore their 74 plots
730
were not all independent. Again we see the ugly head of pseudoreplication via what we might call
731
environmental autocorrelation.
732
7.3 Patch models
733
We have referred repeatedly to the problem of spurious ‘guild proportionality’ due to
734
environmental differences and consequent pseudoreplication. The solution, as mentioned above, is
735
not to randomise over all the quadrats. J.B. Wilson and Roxburgh (1994) made some attempt by
736
having their points arranged in ten 24 × 24 cm plots, randomising occurrences only within each
737
plot, and accumulating the departures from the null models over the ten plots. J.B. Wilson and
738
Gitay (1999) used a similar technique creating separate null models for each of their 21 sites and
739
then combining the results to give an overall test, and J.B. Wilson and Whittaker (1995) did the
Wilson and Agnew, chapter 5, Assembly rules, page 24 of 50
740
same over six sampling lines. An even better technique is to form a separate null model for each
741
quadrat, randomising over a few quadrats adjacent to it: the ‘patch model’ technique described
742
above (Fig. 5.3). Bycroft et al. (1993) did this by using a linear 7-quadrat patch based on the target
743
quadrat, with the result that the proportionality that had been seen with a site model was reduced in
744
size and no longer significant. Although the loss of significance could be due to the reduced power
745
of patch model, the effect size was less too – only half. This was in vegetation selected to be
746
uniform, and warns us to be careful about any study that does not use some kind of patch model.
747
J.B. Wilson and Watkins (1994) used a patch of 9-quadrats centred contiguously on the grid. This
748
is probably the ideal and in their work some significant guild proportionality was seen.
749
7.4 Evidence: removal experiments
750
It should be possible to see equivalent guild effects in perturbation experiments. If
751
member(s) of one guild are removed, the species that increase should be from the same guild.
752
Indeed, when Herben et al. (2003) removed the dominant grass species, Festuca rubra, from a
753
mountain grassland, it was grass biomass that increased more than that of dicotyledons. However,
754
the species responding differed depending on the year in which the removals started. Symstad
755
(2000) removed three guilds – forbs, C3 graminoids and C4 graminoids – from existing Cedar
756
Creek grassland. After three years of growth, seeds of 16 native prairie species were added:
757
legumes, nonleguminous forbs, C3 graminoids and C4 graminoids. There was only weak evidence
758
that resident species repelled functionally similar invaders. Fargione et al. (2003) used plots at
759
Cedar Creek that had been planted with 1-24 species in 1994. Then in 1997, 27 species were added
760
that occurred in the area but had not been planted in 1994. Multiple regression of the 1999 guessed
761
cover of four invader guilds on the resident guilds indicated that each guild as a resident had a
762
greater inhibitory effect on invasion by its own guild, though all invader guilds were inhibited
763
most by C4 grasses. Von Holle and Simberloff (2004) marked out field plots on a floodplain, and
764
weeded particular subjective guilds from some. They then planted in 10 species commonly found
765
in those floodplains. There was no tendency for species to survive better or grow more when
766
planted into a plot from which their guild had been removed. In summary, these removal
767
experiments gave little evidence for guild-based assembly rules. However, such removal
768
experiments are prone to high experimental error.
769
7.5 Evidence: successional convergence
770
Fukami et al. (2005) reported an experiment in which outdoor plots were sown to a mixture
771
of 15 species, or to five different combinations of four species out of those 15. Unfortunately,
772
cover was guessed (in six categories, which only discards information). One year after
Wilson and Agnew, chapter 5, Assembly rules, page 25 of 50
773
establishment, the species composition of the 15-species plots was very similar between five
774
replicates, as was that of five plots that started with bare soil. However, the (unreplicated) 5-
(a) Species composition
(b) Guild composition
Fig. 5.9: Time trends in the species composition and guild composition of plots planted with
different mixtures of species in the experiment of Fukami et al. (2005).
775
species mixtures showed considerable differences that year and those differences remained eight
776
years after sowing with no sign of convergence (Fig. 5.9). The authors called this priority effects,
777
which might imply a switch, but there might be an effect of inertia due to competitive abilities
778
being rather similar. But in spite of the persistent differences in species composition, the different
779
5-species mixtures converged in terms of composition of 14 guilds (a typical guild being
780
“Autumn-germinating annuals, typically tall with semi-rosette form and wind-dispersed seeds”.
781
There is a danger that the clearer trend with guilds was because they averaged over several
782
species, but the authors disproved this with a randomisation test.
783
7.6 Intrinsic guilds
784
The majority of guild investigations have used extrinsic guilds, designated by a priori
785
criteria (J.B. Wilson 1999b). Sometimes, the guilds have been pre-determined (e.g. MacNally
786
2000). Sometimes several characters have been chosen and multivariate methods have been used
787
to classify species into guilds (e.g. Landres and MacMahon 1980; Willby et al. 2000), but this begs
788
the question of whether the characters measured are the appropriate ones and whether they have
789
been weighted correctly. Tests for the reality of such guilds using field associations (e.g. Hallett
790
1982) or perturbation experiments (e.g. Hairston 1981) can indicate that some guild structure has
791
been found, but not that it is the true guild structure of the community. Wiens (1989) summarised
792
the problem:
793
"There is an arbitrariness to guild classification and the determination of guild
794
membership, which is especially evident in subjective a priori classifications. This raises
795
the prospect that the guild 'patterns' that emerge from studies based on such classifications
Wilson and Agnew, chapter 5, Assembly rules, page 26 of 50
796
are consequences of imposing an arbitrary arrangement on a community that is actually
797
structured ecologically in some other way altogether (or is not structured at all). Using
798
multivariate statistical procedures does not grant immunity from this problem."
799
A solution to Wiens' dilemma is to “interview the plants”, to select an index of guild structure and
800
to find the guild classification that maximises this index. This classification is the intrinsic guild
801
structure. J.B. Wilson and Roxburgh (1994) introduced this concept: determining the guilds
802
according to the ways the species actually behave, asking the plants what guilds they are working
803
by.
804
Distributional data
805
J.B. Wilson and Roxburgh (1994) used distributional information to find intrinsic guilds.
806
To avoid circularity they divided the data in two, optimising the guild classification on one half of
807
the quadrats and testing it on the others. With field data it is impossible to examine every possible
808
guild classification, the number is generally astronomical (2(number of species - 1) - 1), so they took their
809
a priori graminoid versus forb+bryophyte classification, and swapped species iteratively to reduce
810
guild proportionality index RVgp. This showed that some forbs were better assigned to the
811
'graminoid' guild, perhaps because of the role of their laminae in the upper canopy, and vice versa.
812
After many iterations the process converged to intrinsic guilds that gave an even stronger tendency
813
towards guild proportionality, not only in the optimisation subset, but also in the independent test
814
subset that had not been used in the optimisation process. Searches for intrinsic guilds starting
815
from two random initial configurations resulted in classifications quite similar to the optimised
816
'Graminoid' versus 'forb+bryophyte' guilds, and with further optimisation using the whole dataset
817
the three optimised classifications converged to become identical. It is important to remember that
818
these intrinsic guilds are alpha guilds, not beta ones. That is, there is a tendency for the species of
819
one guild not to occur together. Presumably the reason is that they are too similar in resource use,
820
and exclusion by interference occurs. Rather, at a 2-species point, e.g., there will tend to be one
821
species from one guild and one from the other.
822
J.B. Wilson and Whittaker (1995) used the method on their Welsh saltmarsh data. Three
823
searches produced very similar guild classifications, which converged to become identical after
824
further whole-dataset optimisations, indicating that real guilds were occurring in the saltmarsh.
825
Intrinsic guild membership could subsequently be correlated with leaf morphology; all the
826
monocots were in one guild together with other narrow-leaved species as in a lawn previously
827
examined by J.B. Wilson and Roxburgh (1994). This suggests that canopy interactions may be
828
important in controlling species coexistence.
Wilson and Agnew, chapter 5, Assembly rules, page 27 of 50
829
J.B. Wilson and Gitay (1999) performed 100 random-start searches (computer processing
830
power had increased in the interim) on the tussock-grassland data. A guild classification that
831
showed significant guild proportionality in the test subset was found in a significantly greater
832
number of searches than expected by chance (28 out of 100) and the ten classifications that gave
833
the lowest RVgp comprised three groups. Further optimisation of representatives of these groups
834
using the whole dataset confirmed that the community contained at least two genuinely
835
independent, alternative guild classifications. It seems that two or more guild classifications can
836
exist within the same set of species in a community, orthogonal in the sense that they are unrelated
837
to each other and operate simultaneously. This is not surprising; the true guild relations are
838
probably quite complex. The intrinsic guilds showed some relation to growth form/height.
839
The general impression from these results is that guild membership in these grasslands
840
depends on canopy relations, especially vertical stratification as affected by leaf morphology.
841
However, this may be partly due to the characters considered and other characters, correlated with
842
them, may be the real determinants.
843
Interference experimental data
844
J.B. Wilson and Roxburgh (2001) used an interference experiment to seek intrinsic guilds.
845
Seven species from the Otago Botany Lawn had been grown in boxes in all possible 2-species
846
mixtures. They argued that when a species from one alpha guild was grown with a species from
847
another alpha guild, by definition differing in resource use, then by the Jack Spratti effect the yield
848
of the mixture should be considerably greater than the mean of the two monocultures, as measured
849
with index RYM (Relative Yield of the Mixture, J.B. Wilson 1988c). With only seven species it
850
was possible to test all possible 2-guild classifications to find the one that maximised the mean
851
RYM of mixtures and this resulted in guilds very similar to those obtained from distributional
852
data.
853
Experimental removals data
854
Clements et al. (1929) had experimented with removing species from communities and
855
Fowler (1981) took this approach by removing single species from a North Carolina grassland. For
856
all removals, at least one other species was affected significantly. Often several species were
857
affected. Usually removal effects between a pair of species were not reciprocal. There was no sign
858
of guilds of species that especially affected each other and it was hard to predict which species
859
would be affected when one was removed. A few negative effects were seen, in which removal of
860
a species decreased the yield of another; if these effects were real, they could have been due to
861
subventions or to indirect interactions via a third species. The conclusion is that species
862
interactions in that grassland were complicated, often indirect, and diffuse. Intrinsic guilds were
Wilson and Agnew, chapter 5, Assembly rules, page 28 of 50
863
not present. Similar experiments, with similar conclusions, were performed by E.B. Allen and
864
Forman (1976) on a New Jersey oldfield, Abul-Fatih and Bazzaz (1979) on an Illinois oldfield,
865
Silander and Antonovics (1982) on North Carolina dune, slack and saltmarsh, del Moral (1983) in
866
Washington subalpine meadows and Gurevitch and Unnasch (1989) on a New York oldfield.
867
These results exclude a simple model of community structure, e.g. with distinct guilds. It implies,
868
but does not prove, stochastic structure. There is a necessary compromise in this work, in that the
869
community is disturbed by the perturbation, probably repeatedly, and the removals tend to lower
870
plant density. A greater problem is that the intensity of work required limits replication, so the
871
statistical errors are usually large and many of the interesting effects are not significant.
872
Conclusion on intrinsic guilds
873
A major advantage of the intrinsic guild approach is that it can fail. Approaches such as
874
multivariate classification of characters must give guilds, whether any exist or not. In contrast, a
875
search for intrinsic guilds by minimising RVgp, maximising RYM or examining the pattern of
876
response to removals can result in all the species being in one guild, or in a guild structure that is
877
non-significant, as it did for J.B. Wilson et al. (2000a), and in a more informal way for Fowler
878
(1981) when she failed to see clear groups in removal results. That is, if there is no guild structure,
879
the intrinsic guild approach can indicate this.
880
Although functional-character relations between species are often expressed in a
881
classification, ordinations have also been used to see trends and continuous variation. It would be
882
good to have an intrinsic equivalent to ordination, placing the species on guild gradients according
883
to their distributions or their responses in experiments.
884
8 Texture convergence
885
Vegetation texture was defined by Jan Barkman (1979) as: "the qualitative and quantitative
886
composition of the vegetation as to different morphological elements, regardless of their
887
arrangement". Ecologists would these days extend it beyond morphology into physiological
888
characters, and use the term ‘functional characters’, but the aim remains to describe communities
889
not by the names of the species, but by plant characters, assuming that similar characters indicate
890
similar function. As an assembly rule, the concept of texture convergence is that in comparable
891
habitats in different areas, whilst the actual species present may be different, the texture may be
892
the same (Fig. 5.10). Constraints of the physical environment and of species interactions will cause
893
convergence to the same texture. The concept is similar to that of guild proportionality, except that
894
instead of dividing the variation into groups (guilds), it looks at the whole distribution. The
895
abundance of species with different characters can be taken into account (Wilson and Smith 2001).
Wilson and Agnew, chapter 5, Assembly rules, page 29 of 50
896
It is possible that the mean texture might converge, but not the distribution of characters (Fig.
897
5.11a), or the distribution could converge, but not the mean (Fig. 5.11b), or of course both or
898
neither.
899
There has long been interest in the idea of convergence between the plants and animals of
900
areas on different continents, with a similar environment (mainly climate) but taxonomically
901
different biotas. Work with plant communities has been almost entirely on mediterranean-climate
902
areas, such as in California, Chile, the Cape, SW Australia and the Mediterranean itself. Mooney
903
and Dunn (1970) suggested that the mediterranean environment in particular imposes several
904
limitations on plant growth, with only a limited number of strategies possible, the evergreen
905
sclerophyll strategy being one.
906
907
Fig. 5.10. The concept of texture convergence. A similar range of characters is present on the two
Smallest
Leaf width
Mean
Largest
Smallest
Leaf width
Largest
909
Largest
Smallest
Leaf width
Largest
Mean
Leaf width
Mean
Leaf width
continents, even though the species involved are different.
Leaf width
908
Largest
Smallest
(a) Convergence
in mean
Smallest
(b) Convergence
in distribution
Largest
Fig. 5.11. Texture convergence can be in: (a) mean or (b) shape.
Smallest
Wilson and Agnew, chapter 5, Assembly rules, page 30 of 50
910
A few of these studies have measured texture and looked for convergence. Parsons (1976)
911
compared scrub communities California and Chile (‘chaparral’ and ‘matorral’) under very similar
912
climates, recording 24 plant characters: growth form, many leaf characters, reproductive
913
characters, etc. Some plant characters were present in species with comparable abundance in
914
similar altitude/aspect habitats, e.g. lobed leaves and winter-deciduousness in high-altitude ravines
915
and large leaves in low-mid altitude ravines. Others, such as summer-deciduousness, were present
916
in both areas but in somewhat different environments. However, small leaved plants were
917
prevalent on low-altitude ridges in California, but absent in Chile, where spiny-leaved species
918
were present instead. Parsons attributed some of the differences to land-use history. Cowling and
919
Witkowski (1994) compared sclerophyllous shrubland in mediterranean Western Australia and in
920
South Africa, and found similar texture between the continents in terms of growth form
921
(shrub/graminoid/forb), leaf consistency (sclerophylly and succulence) and SLW, but divergence
922
in spininess. Canopy-storage of seed diverged, but dispersal type (wind/vertebrate/ant/other)
923
generally converged. However, convergence here is being judged from non-significance of
924
difference and no conclusions can be drawn from lack of significance.
925
Schluter (1990) introduced the concept of “species-for-species matching”, where there are
926
species in the same positions in niche space in different areas. However, that would not necessarily
927
be expected. Nor should the same number of species be expected, since a niche filled by one
928
species in area 1 could be split between three species in area 2. The requirement is only that the
929
same niche space is occupied, and that it is fully occupied in both communities (Fig. 5.12).
Species abundance
Area 1, with 7 species, A to G
C D
E
X
Y
F
B
Z
B
W
G
A
930
931
Area 2, with 4 species, W to Z
Character value
Character value
Fig. 5.12. Site 1 has the same texture as Site 2 with respect to the character, even though they
932
differ in the number and abundances of species.
933
J.B. Wilson et al. (1994), compared convergence between two carr (i.e. wooded fen)
934
communities in Britain and two in New Zealand, in five functional characters related to light
935
capture, such as SLW and PSU support fraction (PSU = photosynthetic unit). In the null model,
Wilson and Agnew, chapter 5, Assembly rules, page 31 of 50
936
the species present were swopped at random between sites with no constraint on the co-occurrence
937
of species similar in morphology. The test is one for co-evolutionary convergence and co-
938
ecological sorting, not for similarity of adaptation to the environment. In fact, the texture of the
939
four carrs diverged when weighting species equally. However, weighting the species by their
940
photosynthetic biomass, convergence was seen for PSU width and possibly for PSU area. Note
941
that this does not represent adaptation to the overall environment, because the comparison was
942
with random draws of the species present at the site, not with an exterior species pool. It means
943
that each community has representation from the range of functional characters present in those
944
carr communities, strong evidence that species are being sorted by their characters, evolutionarily
945
or ecologically, for their entry into the community. There are niches for particular types of species
946
that are filled by immigration or by evolution.
947
The first studies compared continents, but comparisons can be made between nearby sites,
948
or between patches within sites with slight adjustment to the null model (Watkins and Wilson
949
2003). This is close to the intuitive question looking at different patches within an area of
950
vegetation: is the texture similar, do similar species trade off against each other? One just has to
951
realise that it is ecological, not evolutionary, convergence. However, any evolutionary
952
convergence is just a genetic fixation of ecological convergence (Smith and Wilson 2002). Smith
953
et al. (1994) investigated sites in conifer/broadleaved forest in southern New Zealand, recording
954
similar characters to those used by J.B. Wilson et al. (1994) and found convergence in all
955
characters, but as in the Wilson et al. study only when characters of the species were weighted by
956
the abundance of the species. Matsui et al. (2002) conducted the same type of investigation but
957
more locally, within three sites, and evidence of convergence was found for a subalpine grassland:
958
each patch (quadrat) tended to comprise a mixture of small-leaved species and large-leaved
959
species, a more constant mixture than expected if the species were being swapped between
960
quadrats at random. Watkins and Wilson (2003) took this approach further by examining replicate
961
quadrats within twelve herbaceous communities, measuring eleven characters that were intended
962
to reflect the functional above-ground niche of the species and laboriously obtaining the biomass
963
of each species in each quadrat. Biomass weighting allows for true characterisation of the texture
964
of the quadrat. Convergence was seen in chlorophyll content, indicating a significant tendency for
965
each patch in a community to comprise a rather constant mixture of species types in terms of their
966
different chlorophyll contents, though other results were non-significant or showed divergence. In
967
these local convergence studies it is explicit that the question is of ecological assortment.
968
As so often, environmental differences act as noise. As Schluter (1990) wrote: “recall that
969
we are seeking communities more similar than would be expected on the basis of random sampling
970
from the same underlying probability distribution of possible species values. Any factor that
Wilson and Agnew, chapter 5, Assembly rules, page 32 of 50
971
causes the underlying distributions to differ will quickly decrease their chance that a too-small
972
difference between communities will arise”. This is matching.
973
9 Time
974
Time has done natural experiments for us. When the climate has changed, e.g. in the c.
975
15,000 yr since the last glaciation, species have moved around. But have they been restricted to
976
reassembly into the same communities, or as the species reassembled as they pleased? Clements
977
(1936) wrote that “climaxes have evolved, migrated and disappeared under the compulsion of
978
great climatic changes from the Paleozoic onwards, but [the student of past vegetation] is also
979
insistent that they persist through millions of years in the absence of such changes”. He continued:
980
“The prairie climax has been in existence for several millions of years at least, and with most of
981
the dominant species of today”. Clearly his concept of the community as a complex organism led
982
to a conclusion that there were only a limited number of combinations in which species could
983
assemble. Sure, in the very long term new communities could “evolve” and some disappear, but
984
the changes in climate since the last glaciation would result largely in the migration of existing
985
combinations. However, several palaeoecologists have suggested that many of the communities, as
986
seen in pollen assemblages, that were extant earlier in the Holocene are not found anywhere on
987
Earth today: they are ‘no-analogue’ communities. This challenges Clements’ concept of the plant
988
community.
989
There are actually many possible explanations and Jackson and Williams (2004) evaluate
990
them carefully. They discuss the problem of how different, and by what criterion, a ‘no-analogue’
991
community has to be. They reject, as major causes of no-analogues artefacts, such as differential
992
pollen preservation, mixing of sediments, different pollen production by some species in the [CO2]
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 realisable niches of Species X and Y overlap in an area of
environmental hyperspace that exists. At Time 2, the combination of environmental
conditions where they overlap does not occur. Inspired by Jackson and Williams
(2004).
Wilson and Agnew, chapter 5, Assembly rules, page 33 of 50
993
obtaining then and a different juxtaposition of communities over the landscape. It is remarkably
994
difficult to find exact matches between any two current climates, and this is probably even more
995
true for the past, and differences in [CO2] will be present too, making matching impossible.
996
Jackson and Williams suggest that the most likely explanation for no-analogue communities is that
997
whilst similar ranges of climatic variates occurred, often the combinations that are around today
998
did not (Fig. 5.13).
999
This interpretation is supported by comparing the degree of mismatch between
1000
reconstructed past plant communities and the closest modern analogues with the degree of
1001
mismatch between reconstructed past climates (from general circulation models) and the best
1002
modern fits. Community misfits (no-analogues) tend to occur in the same place/time as climate
1003
misfits (Williams et al. 2001). This evidence is at variance with Clements’ interpretation of
1004
constant communities moving around the landscape. However, it does not distinguish between
1005
species reacting individualistically to the climate, as suggested by Gleason in some of his writings
1006
(chap. 6, sect. 3) from a model in which the occurrence of a species is determined by the identity
1007
of other species present, a view attributed with some truth to Clements (chap. 6, sect. 2).
1008
10 Abundance
1009
10.1 Biomass constancy
1010
The constancy of biomass per unit area, compared to null models in which species
1011
abundances are random, has been used as an assembly rule (J.B. Wilson and Gitay 1995a). This is
1012
not a deep rule, but it is a demonstration from the field that interference is occurring and causing
1013
community structure. It has the ability to distinguish between communities (J.B. Wilson et al.
1014
2000a).
1015
10.2 Relative abundance distribution (RAD)
1016
Various models of community construction give predictions for the relative abundance
1017
distribution between species (RAD; J.B. Wilson 1991; Fig. 5.14). The Niche-preemption
1018
(Geometric) model is based on competition and the Zipf-Mandelbrot can be interpreted as
1019
succession/facilitation. The Broken stick and the Sequential Breakage (General Lognormal)
1020
models are alternative models of the random assignment of resources (i.e. alpha niche widths)
1021
between species. Several other subtly different models of this type can be constructed (Tokeshi
1022
1996). All of these are null models: alternative models of what is going on when nothing is going
1023
on. This means that we are liable to end up testing between null models, not against them. It is also
1024
a problem that some of the distributions, notably the General lognormal, can be derived from
Wilson and Agnew, chapter 5, Assembly rules, page 34 of 50
1025
alternative assumptions. With so many different models, and with sampling variations, we might
1026
worry that it would be impossible to discriminate between them. However, in a 15-species
1027
community, for example, one can identify the correct model with reasonable correctness given 10
1028
or more quadrats; it depends on the model and the number of species (Mouillot and Wilson 2002).
1029
Most types of evidence for community structure involve comparisons in time or space; RADs are
1030
one of the very few types of evidence available for one point in space and time.
1031
In MacArthur’s (1957) "broken-stick" model abundances reflect the partitioning of resources
1032
among competing species by random divisions along a one-dimensional gradient. This ecological
1033
model can be tested by comparing its RAD predictions with those observed. However, the concept
1034
of a one-dimensional resource gradient applies uneasily to partitioning of most plant resources.
1035
Other ecological models can give the same distribution, including models with no restrictions on
1036
niche overlap (Cohen 1968).
1037
Preston (1948) proposed the use of a lognormal distribution for empirical reasons, though it
1038
might express community structure in two ways. Plant growth will be affected by several
1039
environmental factors. By the Central Limit Theorem, this will give a near-normal distribution.
1040
Since plants have intrinsic logarithmic growth, the distribution will be lognormal (May 1975).
1041
Alternatively, MacArthur's Broken Stick model, but with the breaks sequential and breakage
1042
probability independent of length, gives a lognormal distribution. This can be seen as the
1043
occupation and subsequent division of niches by species (Pielou 1975). Preston (1962) proposed
1044
further that the distribution was a reduced-parameter subset of lognormal distributions that he
1045
called 'Canonical lognormal', defined by the mode of the individuals curve coinciding with the last
1046
point on the species curve (i.e. gamma = 1). The hypothesis was empirical; there is no ecological
1047
basis for it (Caswell 1976). Whether it is a mathematical artefact is controversial (May 1975;
1048
Connor and McCoy 1979; Sugihara 1980; Connor et al. 1983).
1049
The geometric model (Whittaker 1965) suggests that the 'most successful species' takes
1050
fraction k of the resources and therefore forms approximately k of the abundance. The second most
1051
successful species takes k of the remainder (i.e. a total of k(1-k)), etc. This model can be tested.
1052
However, ‘most successful’ can be taken either ecologically, as the first species to arrive when
1053
interference is cumulative (chap. 2, sect. 2.3), or evolutionarily, as the species that has the highest
1054
intrinsic interference ability (Watkins and Wilson 1994), and a test is available between those
1055
possibilities.
1056
After many years of parallel presentation of theoretical curves of theory and data, Wilson
1057
(1991) showed how the two could be compared. The results have been frustrating. J.B. Wilson and
1058
Gitay (1995b) found that in four dune slacks in west Wales the best fit was given by either
1059
Geometric or General Lognormal models, but there was no consistency between two subsites
Wilson and Agnew, chapter 5, Assembly rules, page 35 of 50
1060
within each slack as to which gave the better fit. J.B. Wilson et al. (1996a) fitted RAD models to
1061
plots from three experiments; basically there were no trends except those reflecting the higher
1062
evenness in plots to which P had been applied, and it is hard to see much ecological meaning even
1063
in this. Watkins and Wilson (1994) sought a relation between the level of vertical complexity in a
1064
community in which RAD model fitted, but could find none. The model fitting best can be
1065
dependent on the scale of sampling (J.B. Wilson et al. 1998).
Fig. 5.14: A RAD plot for biomass in a Spanish hay meadow. From J.B. Wilson (1991).
1066
1067
Species diversity can be split into richness and evenness, and the latter represents in one value
1068
some of the information in dominance-diversity curves. Caswell (1976) compared evenness to that
1069
expected from a null model. He found that tropical rain forests tended to be less even than
1070
predicted from the null model; temperate deciduous forests of eastern North America were
1071
significantly more even than the null model. The contrast was the opposite of what he expected
1072
from previous theories. Other attempts to obtain evidence on community assembly from evenness
1073
have not been fruitful.
1074
We conclude that the information analysed here is potentially useful. Fits to a model based
1075
on ecological theory would be most interesting, though usually ambiguous, and no conclusions of
1076
real ecological value have emerged yet. Any regularity, such as adherence to Preston's Canonical
1077
hypothesis, would be that the structure was deterministic.
1078
10.3 Sparse species
1079
Species that are sparse (or ‘rare’) within the community, are a puzzle. Firstly: are they
1080
filling special niches that exist for sparse species? Zobel et al. (1994) investigated this in a wooded
1081
meadow in Estonia by removing 10-17 species from certain plots, all with a cover of 1 % or less (a
1082
different list for each plot), repeating the removals for 5 years. There were no visible gaps and they
1083
say very little biomass was removed, but species richness was reduced by 25-33 %. Species did
Wilson and Agnew, chapter 5, Assembly rules, page 36 of 50
1084
not immigrate to fill the gaps: the number of immigrants was no higher than in control (i.e. no-
1085
removal) plots, actually non-significantly lower. There seemed to be no special niches for the
1086
sparse species.
1087
Then do the sparse species have a distinct effect on the major species? Lyons and Schwartz
1088
(2001) in a meadow in the mountains of California manipulated the species richness by removing
1089
either: (a) all plants of the least abundant species, thus reducing species richness to between two
1090
and seven species, or (b) an equivalent biomass of the most common species (to control for
1091
possible disturbance by the removals in treatment ‘a’. The exotic grass, Lolium temulentum
1092
(darnel) was then introduced. Its establishment was higher when more sparse species were
1093
removed, indicating a rôle for the sparse species in invasion resistance. It is not immediately
1094
obvious how this result squares with that of Zobel et al. We are far from generalisations in these
1095
questions.
1096
11 Keystone species
1097
A valuable concept in describing communities has been that of ‘keystone species’, defined
1098
by Paine (1969) as a single native species high in the food web that, whilst perhaps unimportant as
1099
an energy transformer, is vital for the maintenance of the community. This cannot be applied
1100
literally to plants, but others have seen a keystone species as being the one in a community with
1101
the greatest effect on others, or the greatest effect relative to its biomass (Jordán et al. 1999). Since
1102
plants dominate the biomass and carbon capture of their systems, one could almost see all green
1103
plants as keystone species. They affect lower (decomposer) and higher trophic levels – usually
1104
more than one higher level. Their effect is often via herbivory of their vegetative parts, but the
1105
contribution of Ficus spp. to frugivores has led to their being called keystone species (Patel 1997;
1106
Nason et al. 1998). The term has also been applied to plants with intransigent litter (Empetrum
1107
hermaphroditum; Mallik 2003) and here it seems to be a switch maintaining the current state via
1108
litter that produces polyphenol-rich humus with low pH (chap. 3, sect. 5.4.E above). The
1109
contribution of plants as furniture for birds has been seen as keystone (arborescent succulents by
1110
Midgley et al. 1997) and this may also operate as a switch. Hurlbert (1997) says, "the metaphor
1111
'keystone species' was appealing and harmless" but "has come to mean little more than 'important
1112
for something'". And why not? As Bond (1993) says, "If loss of a species results in a large effect
1113
on some functional property of the ecosystem, that species may be called a keystone". In fact, a
1114
species with a strong reaction on the environment will either change the current state, in which
1115
case it would not be called a keystone, or it will reinforce the current state, in which case it is a
1116
keystone because it operates a switch. Top predators can be keystones because of cascade effects,
1117
and plants can be keystones when they operate switches.
Wilson and Agnew, chapter 5, Assembly rules, page 37 of 50
1118
1119
12 Exotic species as community structure probes
In some parts of the world, exotic species have displaced much of the native cover (e.g. the
1120
Seychelles, Hawaii, New Zealand: McDonald and Cooper 1995). It is not always easy to define
1121
what an exotic species is, but most cases are clear. Exotic species are opportunities for the
1122
theoretical community ecologist, natural experiments.
1123
12.1 The nature of exotic species
1124
In one way invasion by exotic species is surprising: the native species have presumably
1125
evolved to meet the local environment, physical and biotic. Moreover, exotic species cannot be
1126
intrinsically different because all species are native somewhere (except species of garden origin
1127
and a few species of recent origin such as Spartina anglica). The concept ‘exotic species’ is
1128
deficient in logic. Leger and Rice (2003) found the alien (Chilean) ecotype of Eschscholzia
1129
californica to be more vigorous in California than the native genotype. Would the Californian
1130
genotype, as an alien, be more vigorous in Chile than the native one? How would that situation
1131
arise? It is far from clear that exotics are consistently different. Kissel et al. (1987) found no
1132
consistent difference in water relations between the three major native woody species and four exotic
1133
ones of the most semi-arid area of New Zealand. King and Wilson (in press) found no difference in
1134
experimental water stress tolerance or nutrient response, though the exotic species did have a greater
1135
RGRmax. We suspect that often generalisations are made from special cases, especially ones with a
1136
practical impact.
1137
Exotics have been implicated in destroying the structure of the whole community. Hubbard
1138
and Wilson, surveying the semi-arid Upper Clutha catchment, NZ, where massive exotic invasion
1139
has occurred, found very weak community structure as seen in an inability of an ordination to predict
1140
species presence/absence. J.B. Wilson (1989a) attributed this to conflicting structure in the native
1141
and exotic guilds. Sanders et al. (2003) studied invasion by Linepithema humile (the Argentine ant)
1142
in California. They examined chequerboarding – the tendency of species to be mutually exclusive so
1143
that a site/species table looks like a chess board – by calculating index C for the ground-foraging ant
1144
community. Positive values of C indicate segregation, i.e. less species co-occurrence than expected
1145
under a null model, more mutual exclusions, a predominance of negative associations. Negative
1146
values of C indicate aggregation, i.e. more species co-occurrence than expected under a null model.
1147
It is difficult to interpret an uncontrolled natural experiment, but Sanders et al. took the best
1148
approach possible, comparing: (a) quadrats sampled in one year that had not been invaded versus
1149
those that had, and (b) particular plots in the year before and the year after invasion. They found that
1150
before invasion C was generally positive and significant; after invasion it was never significantly
1151
positive, and sometimes significantly negative. If we can take chequerboarding as evidence of
Wilson and Agnew, chapter 5, Assembly rules, page 38 of 50
1152
community structure, the exotic ant had destroyed it. A similar study with plants would be
1153
fascinating if one could find a situation in which to do it.
1154
There are many examples of invaders successfully entering natural, allogenically
1155
undisturbed communities: in Britain Rhododendron ponticum can invade forest, in northeastern
1156
USA Ligustrum spp. (privets) can invade forest, in New Zealand Berberis darwinii (barberry) and
1157
Mycelis muralis can invade forest and Juncus gerardii (a rush) saltmarsh.
1158
This whole approach has been questioned, as to whether in invasions the exotics are the
1159
cause of the change – the ‘drivers’ – or whether they just take advantage of a disturbance – the
1160
‘passengers’. Corbin and D'Antonio (2004) addressed this for the grasslands of California, which
1161
200 years before had been dominated by native perennial grasses with associated annual and
1162
perennial dicot species. These were almost completely displaced by European and Asian species.
1163
Under the ‘passenger’ hypothesis the change came about due to tilling for agriculture, introduction
1164
of livestock and a severe drought in the 19th Century, leaving disturbed conditions. Corbin and
1165
D’Antonio experimentally removed the vegetation, then sowed plots with three native perennial
1166
grass species, with three exotic annual grass species, or with both. Over time, the native grasses
1167
reduced the productivity of the exotic annuals, whilst the impact of the latter on the native
1168
perennials was minor and decreasing. The ‘passenger’ concept was supported. Further south in
1169
California, Stylinski and Allen (1999) compared almost undisturbed sites of chaparral and sage
1170
shrublands with nearby areas disturbed by vehicles, excavation or agriculture. Percent cover of
1171
shrubs was measured by canopy intercept, but of that herbs and seedlings only guessed. The
1172
vegetation of the disturbed areas comprised mainly exotic annuals (60 %), whilst the undisturbed
1173
areas had 68 % cover of native shrubs. This situation remained essentially unchanged in a site
1174
disturbed 71 years earlier, and the authors concluded that after invasion by exotics the vegetation
1175
reached an alternative stable state. Presumably a switch was operating, so that the passengers took
1176
over driving the vehicle, but we do not know through what factor the switch was operating.
1177
Five major explanations have been given for the ability of exotics to invade:
1178
(a) depauperate floras, (b) weak competitors, (c) the invaders are r species ,(d) escape from natural
1179
enemies and, (e) co-evolution.
1180
Islands are often given as examples of where exotics are invading areas with depauperate
1181
floras (e.g. NZ: Dulloo et al. 2002). The depauperisation can be in the number of species leaving
1182
empty niches, or whole guilds (functional types) can be missing. Shimizu and Tabata (1985)
1183
explained the invasion of Pinus lutchensis into the shrublands of the Ogasawara Islands, Japan, by
1184
postulating that there had been an empty niche for an emergent tree. Ricciardi and Atkinson (2004)
1185
examined in a literature survey whether aquatic invaders amongst fish, invertebrates, algae and
1186
vascular plants were more likely to have a high impact in terms of local extirpation / severe decline
Wilson and Agnew, chapter 5, Assembly rules, page 39 of 50
1187
of a native species if there were no congeners in the native biota. For four of seven systems,
1188
including the NZ coast, this hypothesis was supported. This implies that species could invade more
1189
readily when there were missing guilds (many of the comparisons were with animals, for which
1190
genera are often reasonable guild substitutes). Similarly, Cappuccino and Carpenter (2005)
1191
comparing invasive and non-invasive exotic plant species in natural areas in Ontario, New York
1192
and Massachusetts, found that invasive plants were more taxonomically isolated than non-invasive
1193
plants, belonging to families with 75 % fewer native North American genera, and Strauss et al.
1194
(2006) found the same with grasses of California, this time using a reconstructed phylogeny rather
1195
than taxonomy. There does seem to be some evidence for the empty niche / missing guild idea.
1196
The second explanation for the success of exotics is that the native species might not be
1197
vigorous enough. Macdonald and Cooper (1995) said “an individual island’s biota is based on too
1198
small a sub-sample of the global gene pool to have generated robust competitors for every
1199
available niche. … Insular species are frequently outcompeted by species that have been honed in
1200
much more exacting biotic communities of the mainland. … [suggesting] superior competitive
1201
ability of mainland species”. For New Zealand, Dansereau (1964) wrote: of “apparently vacant
1202
space” , occupied only by “weaker” species. Is this really true? Perhaps super-species, once limited
1203
by dispersal (e.g. to the Old or New, Northern or Southern, hemispheres), are now able to spread
1204
everywhere. In that case, homogenisation of the flora is set to change the world (which it is). Still,
1205
these super-species don’t seem to have been that super in their original hemisphere. In Britain,
1206
when one meets a yellow composite herb with rosette leaves one has to key it out between a
1207
number of quite likely possibilities. In New Zealand Hypochaeris radicata (cat’s ear) is present in
1208
a huge range of environments and often quite frequent with them, so for those environments the
1209
answer 95 % of the time is ‘Hypochaeris radicata’. An exception may be Ammophila spp. It has
1210
been suggested, with some truth, that when high coastal dunes are built it is always by species of
1211
Ammophila. It seems to operate a switch, trapping sand and tolerating burial.
1212
The third possibility is that the exotics could invade because they are r species, short-lived
1213
and rapidly reproducing in ephemeral habitats. These are the R species of Grime (2001): fast-
1214
growing in open conditions, with quick and extensive seed reproduction. Why should there be
1215
more r species amongst exotics? Probably disturbed habitats are much more common and
1216
extensive than before humans changed the landscape. This has been an explanation for the origin
1217
of arable weeds: that once they were only in local disturbed areas such as riverbanks and with
1218
cultivation they expanded their geographical range into arable fields. In some floras the number of
1219
r species may have been very small, for example Allan (1937) gives 6 % of the flora of NZ as
1220
being annual, certainly an over-estimate, and a similar situation may have been true of many areas
1221
before humans appeared.
Wilson and Agnew, chapter 5, Assembly rules, page 40 of 50
1222
A fashionable explanation for the high fitness of exotic invaders is that having escaped
1223
from their natural specific enemies they have been able to evolutionarily discard their defences to
1224
those enemies and the resources involved have been used in growth and reproduction instead.
1225
Presumably the enemies will catch up in dispersal time, as has happened with the invasion of
1226
Lupinus arboreus in New Zealand, now largely suppressed by the lupin anthracnose fungus
1227
Colletotrichum gloeosporioides (Molloy et al. 1991), or in evolutionary time. The general pattern,
1228
whether the pests are insects, crustaceans, fungi or viruses, is indeed a lesser impact on
1229
populations in the exotic range of a species, presumably because the pests specific to the species
1230
are missing (Vilà et al. 2005; Bossdorf et al. 2005; C.E. Mitchell and Power 2003). An increase in
1231
growth apparently resulting from the loss of defences has been seen comparing ecotypes from the
1232
native and exotic ranges grown in a common garden (e.g. Blair and Wolfe 2004), though in other
1233
studies the effect has been absent (e.g. Bossdorf et al. 2005). Thébaud and Simberloff (2001) used
1234
maximum heights given in floras to compare species between the U.S.A. and Europe: invaders in
1235
both directions. In some comparisons populations were no different, and in some species
1236
populations were taller in their native range, the opposite of the effect expected under the enemy-
1237
release hypothesis. This study has the advantage of surveying many species and avoiding possible
1238
bias of choosing problem weeds, but it is not clear from where the flora writers obtained this
1239
information, nor how maximum height was defined. A complication has been illustrated for
1240
Senecio jacobaea, native to Europe but invasive in North America, Australia, NZ and elsewhere,
1241
that defence against specialist herbivores Tyria jacobaeae (cinnabar moth) and Longitarsus
1242
jacobaeae (ragwort flea beetle) has been lost, but some of the resources saved seem to have been
1243
put into increased protection against generalist lepidopteran herbivores via pyrrolizidine alkaloids
1244
(Joshi and Vrieling 2005; Stastny et al. 2005).
1245
A further possibility is that the resident plant species in a community have been able to
1246
coevolve resistance to each others’ allelochemical toxins. Callaway and Aschehoug (2000)
1247
suggested this when they found in a greenhouse experiment that Centaurea diffusa, exotic in
1248
Montana (USA) had greater interference effect on Montana grasses than on related species from
1249
Georgia (Caucasas) and the difference was removed by adding active carbon.
1250
Tropical rain forests in the tropics are an interesting case, since they are generally less
1251
invaded by exotic species. It would be helpful to conservationists to ascribe the lack of exotics to
1252
the saturation of available niche space through high diversity of species or guilds, but some
1253
species-poor types of tropical forest also have no invaders (Gilbertiodendron dewevrei: Richards
1254
1996). A more likely explanation is that most of the “exotic species that are transported to tropical
1255
countries lack specific the life history traits, most importantly shade tolerance, that are necessary
1256
for successful invasion of undisturbed tropical forests” (Fine 2002). Rejmánek (1996) suggested
Wilson and Agnew, chapter 5, Assembly rules, page 41 of 50
1257
that the paucity of invaders was because fast growth in that environment resulted in rapid canopy
1258
closure after disturbances.
1259
12.2 Exotic establishment and community assembly
1260
The most fascinating way to use exotics as probes into community structure is to ask how
1261
they assemble when they reach new territory. J.B. Wilson (1989a) examined the native and exotic
1262
plant origin guilds of the Upper Clutha catchment, NZ. The two guilds produced classifications of
1263
the quadrats that were no more different than those using random groups of species, suggesting that
1264
the two guilds follow the same vegetational boundaries. However, there was some evidence that the
1265
guilds differ in which environmental factors controlled their distribution.
Fig. 5.15: A minimum spanning tree for the species composition of British and New
Zealand roadside vegetation.
1266
1267
The roadsides of New Zealand generally comprise exotic species that have reassembled
1268
into communities there. J.B. Wilson et al. (2000b) examined an area of southern NZ containing
1269
152 exotic species, mainly from Britain for environmental and cultural reasons. Quadrats from
1270
these NZ roadsides were fitted to the British National Vegetation Classification (NVC). After
1271
excluding species that are not present in New Zealand, the fit was 61 %. Randomising the
1272
species/quadrats occurrences of the NZ data gave on average a 59 % fit to the NVC, so the fit of
1273
the real quadrats was only slightly, though significantly (p < 0.001) better than the random ones.
1274
British roadside communities were also compared to the NVC, as a control; they gave a 66 % fit.
1275
Thus, the New Zealand communities bear little relation to NVC communities in Britain (though
1276
the British communities did not fit brilliantly either). Comparing the NZ and British quadrats
1277
directly using a minimum spanning tree to connect similar quadrats, the two formed two almost
Wilson and Agnew, chapter 5, Assembly rules, page 42 of 50
1278
distinct groups (Fig. 5.15). The conclusion must be that the British species have re-assembled into
1279
communities in NZ most of which are new, i.e. distinct from those that occur in the native range of
1280
the species in Britain. The evidence points to community assembly by pre-adaptation.
1281
Wilson et al. (1988) reached similar conclusions comparing an area in southern England
1282
with one in southern New Zealand, which had considerable overlap in florules but no similarity in
1283
species associations.
1284
Lord et al. (2000) studied in a similar way the re-assembly of species introduced from
1285
Britain in NZ calcareous-soil grasslands (4-24 % CaCO3) that were largely composed of such
1286
species. Analysed as with roadsides, the fits for six sites ranged 48-77 %. Two of the six sites
1287
fitted British calcareous grassland communities. These two sites are on thinner soil (< 10 cm
1288
depth), under lower rainfall, more likely to be influenced by the base rock, and for these sites the
1289
environment of the community in Britain matched very well that of the N.Z. site.
1290
Together, these three reassembly studies suggests that only strong environmental filtering
1291
is able to reassemble communities. Even though the roadside dataset spanned a wide and very
1292
comparable environmental range in the two countries (e.g. rainfall 345 – 3460 mm and mean
1293
temperature in the warmest month 12-17 °C in New Zealand versus 485-1777 mm and 14-17 °C in
1294
Britain), it appears that environmental filters and assembly rules were not strong enough to
1295
reassemble the same communities. Instead, alternative states have been reached. We could not tell
1296
whether they are stable, and if so what switch is responsible, but the consistent separation Fig. 5.15
1297
is remarkable.
1298
13 Conclusions, and the Otago Botany Lawn
1299
It is difficult to draw conclusions on assembly rules. Plants interact (chapter 2) and plant
1300
species differ (chapter 1) so there must be limitations to coexistence. However, the difficulty of
1301
finding assembly rules, and the difficulty of ensuring that tests for them are valid, combine to
1302
make it difficult to confirm that this is so in the real world.
Wilson and Agnew, chapter 5, Assembly rules, page 43 of 50
1303
In most studies just a few assembly rules have been tested on a site, and these might not
1304
have been the ones operating. The Botany Lawn of the University of Otago (Fig. 5.16) has surely
1305
been more intensively studied in this way than any other community and offers a case study. It has
1306
also yielded the best evidence that such rules exist. The lawn was established c. 1965 with the
1307
sowing of Agrostis capillaris (bent) and Festuca rubra. The former is still prominent, but the bulk
1308
of the 36 species present within the current community have arrived through natural dispersal, the
1309
commonest being the grass Holcus lanatus (Yorkshire fog), forbs Trifolium repens (white clover)
1310
and Hydrocotyle heteromeria (a New Zealand native) and mosses Eurhynchium praelongum and
Fig. 5.16: Profile through a part of the Botany Lawn.
1311
Acrocladium cuspidatum. Since its establishment, the lawn has been maintained under a consistent
1312
regime of cutting to a height of c. 2.7 cm, fortnightly in the growing season and monthly in winter.
1313
There has been no application of fertilizer, herbicide or irrigation (the average annual rainfall is
1314
784 m yr-1). This constant management, together with the short lifespan of individual ramets in the
1315
lawn, has created the opportunity for the community to come to equilibrium, and indeed the
1316
species composition of the lawn is quite constant over time. There are seasonal changes on the
1317
lawn, but there is little evidence of directional change between years, and the abundance ranks of
1318
species have remained almost constant (Roxburgh and Wilson 2000b).
1319
There is considerable stratification of species in the lawn (Figs. 5.16). Even when the
1320
sward is only 2.7 cm high after cutting there is significant evidence for three strata (Fig. 5.17a),
1321
and when the species have regrown 14 days later there are many more significant vertical relations
1322
between species, with evidence for four strata (Fig. 5.17b).
Wilson and Agnew, chapter 5, Assembly rules, page 44 of 50
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
species are omitted.
1323
1324
The variance in species richness across the lawn has been demonstrated to be lower than
1325
expected in a null model. This is seen at the scale of 13 × 13 mm (Watkins and Wilson 1992) and
1326
the effect at that scale does not seem to be an artefact of environmental variation since its
1327
significance remains using a patch model. In fact, it was one of three out of the 12 lawns in that
1328
investigation to show a deficit of variance significant and greater than 20 %. A similar deficit in
1329
variance richness can be seen at the scale of a point (J.B. Wilson et al. 1992b). The possibility has
1330
been raised that the effect is due to a physical limitation in packing plant modules at that scale.
1331
However, up to five species can be found at a point in this lawn, and on average only 1.45 species
1332
are, so space does not seem to be a limitation. Plants do not compete for space (Chiarucci et al.
1333
2002), and the profile diagram (Fig. 5.16, drawn from life) confirms that the canopy is largely
1334
empty.
1335
The restrictions on species coexistence can probably seen better by analyzing guild
1336
proportionality. This removes us from questions of the number of modules that can be physically
1337
packed, by using a null model in which the numbers of species in each quadrat do not differ from
1338
those observed, and indicating restrictions in terms of types of species. J.B. Wilson and Watkins
1339
(1994) analysed thus at the 13 × 13 mm scale. Testing over all richness categories there was no
1340
significant (p = 0.074) guild proportionality for graminoid versus forb guilds, but examining 4-
1341
species quadrats alone there was (p = 0.005). This was true for one other NZ lawn and one Fiji lawn.
1342
Likewise, grass versus legume guild proportionality was significant in the Botany Lawn in 3-species
1343
quadrats. J.B. Wilson and Roxburgh (1994) found significant guild proportionality at a point using
Wilson and Agnew, chapter 5, Assembly rules, page 45 of 50
1344
graminoid versus forb guilds, and whether or not the two bryophyte species were included with the
1345
forbs. There was no evidence that the rule was based on grass/legume interactions. There was also
1346
guild proportionality using as guilds the species that tended to be in the upper stratum of the lawn
1347
versus those that were basal, but only if the stratum assignments were based on species' positions
1348
at the end of the 14-day mowing/regrowth cycle. The constancy of the graminoid versus forb
1349
proportions increased as the number of species at a point did. All these results indicate that when
1350
there are few species present at a point there is less constraint on which types, but as the species start
1351
to pack in, their ability to enter the community depends on their characters.
1352
The a priori guilds that were used are not necessarily the true ones. At the scale of 13 × 13
1353
mm, although two of the three grass-grass associations negative as one would expect, so were
1354
those between Plantago lanceolata and two of the grasses (J.B. Wilson and Watkins 1994). We
1355
can determine the guilds as perceived by the plants using the intrinsic guild approach. With
1356
distributional data (minimising guild proportionality index RVgp) the intrinsic guilds that resulted
1357
from the optimisation process generally confirmed both the particular rôle of graminoids and the
1358
importance of leaf position in the canopy (Table 5.4; J.B. Wilson and Roxburgh 1994). For
1359
example, Trifolium repens (white clover) with its horizontal laminae is often in the canopy fighting
1360
with the grasses (Fig. 5.16), and it appeared in the same intrinsic guild as four of the five grasses
1361
(Table 5 .4). Some other forbs were better assigned to the 'graminoid' guild too, again apparently
1362
because of their rôle in the upper canopy. All this suggests that there is one niche for species that
1363
occupy the upper canopy towards the end of the mowing/regrowth cycle, based on the interaction
1364
of lamina shape and position, and another for the basal species that may absorb the light that
1365
reaches further down just after mowing. Strong, almost surprising, support came from the intrinsic
1366
guilds obtained from the interference experiment by maximising the RYT, relative yield total, i.e.
1367
tendency towards overyield (J.B. Wilson and Roxburgh 2001). The guilds formed gave, for the
1368
seven species included in the experiment, perfect agreement with those obtained from the
1369
distributional data (Table 5.4). These intrinsic guilds are real community ecology, because we are
1370
allowing the species to tell us what is happening in the community. This is inductive science, and
1371
made deductive for the distributional data by testing the guilds on independent data and for
1372
experimental data by confirming the results from the distributional data.
Wilson and Agnew, chapter 5, Assembly rules, page 46 of 50
1373
Table 5.4. Intrinsic guild classifications of species of a lawn obtained from: (a) distributional data
1374
(J.B. Wilson and Roxburgh 1994) and (b) the interference-experiment data of Roxburgh
1375
and Wilson (2001).
Species
Characteristics
Guild from
distributional
data
Agrostis capillaris
Anthoxanthum odoratum
Bellis perennis
Holcus lanatus
Hydrocotyle moschata
Linum catharticum
Poa pratensis
Ranunculus repens
Trifolium dubium
Trifolium repens
Acrocladium cuspidatum
Cerastium fontanum
Cerastium glomeratum
Eurhynchium praelongum
Festuca rubra
Hydrocotyle heteromeria
Hypochaeris radicata
Prunella vulgaris
Ranunculus repens
Sagina procumbens
Grass
Grass
Dicot, rosette
Grass
Dicot, horizontal lamina
Dicot, upright
Grass
Dicot
Legume, horiz. lamina
Legume, horiz. lamina
Moss
Dicot, erect
Dicot, erect
Moss
Grass
Dicot, horizontal lamina
Dicot, rosette
Dicot, creeping
Dicot, creeping
Dicot, creeping
A
A
A
A
A
A
A
A
A
A
B
B
B
B
B
B
B
B
B
B
Guild from
interference
experiment
data
A
A
A
A
B
B
B
1376
Mason and Wilson (2006) examined the traits of the seven most common species in
1377
intrinsic each guild. The intrinsic guild approach does not make any assumptions about the
1378
characters that determine coexistence, but the two guilds differed in Mowing Removal Index
1379
(MRI), calculated as the proportion of a species’ mass typically removed during mowing (Fig.
1380
5.18), though not in other characters related to light capture, such as specific leaf area (leaf area
1381
per mass), leaf area ratio (the leafiness of a plant) and six photosynthetic pigment characters. This
1382
confirms the importance of canopy interactions, but sheds doubt on whether they involve light
1383
capture.
Wilson and Agnew, chapter 5, Assembly rules, page 47 of 50
1.0
0.8
Mowing Removal Index (MRI)
p = 0.008
p = 0.022
p = 0.030
p = 0.033
p = 0.016
0.6
Guild A
Guild B
0.4
0.2
0.0
4
8
12
16
20
Time since mowing (days)
Fig. 5.18: Mean Mowing Removal Index (MRI) of each guild at each sampling date. The Pvalues are from t-tests for differences between guilds in mean Mowing Removal Index.
1384
1385
Mason and Wilson (2006) also used the approach of Stubbs and Wilson (2004) on new
1386
point-quadrat data (separate from those used by Wilson and Roxburgh), testing the limiting-
1387
similarity concept directly by examining the characters of the species co-occurring at a point.
1388
Greater variance among those characters than expected at random would indicate limiting
1389
similarity: a tendency for species that were alike not to co-occur. MRI (Fig. 5.19) and leaf length
1390
showed significant limiting similarity at all five times since mowing, as did two correlated
1391
characters, leaf area and length:width ratio. However, none of the other characters gave more than
1392
sporadic indication of limiting similarity. PSU length:width ratio showed significant limiting
1393
similarity for three of the dates, but it is related to MRI. Anthocyanin / dry mass demonstrated
1394
limiting similarity for in the first two samples after mowing and marginally (p = 0.072) after 20
1395
days. None of PSU width, PSU thickness, PSU dry mass, SLW, ratio of lamina area or mass to
1396
shoot mass, chlorophylls a or b per dry mass, chlorophyll a:b ratio or UV pigment content were
Wilson and Agnew, chapter 5, Assembly rules, page 48 of 50
significant for more than one period out of the five times.
Observed / exoected variance in MRI
1397
1.20
1.10
1.00
0
5
10
15
Days after mowing
20
Fig. 5.19: Variation in the Mowing Removal Index (MRI) of species co-occurring at a point in
1398
1399
the Botany Lawn.
How can the restrictions on coexistence be due to canopy interactions yet not be related to
1400
light capture? One possibility, by analogy with the apparent importance of NPK and water
1401
resources in the results of Stubbs and Wilson (2004) is that although the guilds are canopy-related
1402
the basic effect is below ground. After defoliation there is generally ‘root growth stoppage’.
1403
Species with a high MRI would be affected by this because more leaf is removed. The temporary
1404
cessation of root growth would affect P uptake, which is rather dependent on exploration of the
1405
soil by new roots. Species with a low MRI could carry on growing, not only absorbing light
1406
temporarily available by canopy removal, but with a continuing P supply. However, some support
1407
for the rôle of light comes from the local texture convergence study of Watkins and Wilson (2003).
1408
They found overall convergence between quadrats in chlorophyll, mainly due to strong
1409
convergence in two of the 12 sites, one of which was the Botany Lawn. It is simplistic to expect
1410
one process to be limiting coexistence.
1411
Why is the evidence for assembly rules stronger in the Botany Lawn than anywhere else?
1412
Firstly, it has been more intensively studied than any other community. The short stature probably
1413
contributes to the ease of finding assembly rules. The canopy is in some ways like a forest canopy
1414
in miniature, but the relations are easier to see: in a forest it is hard to determine just which part of
1415
the canopy a ground herb is influenced by. However, the major factor is probably not that it is
1416
easier to find rules but that they have shaped the lawn community to a greater extent because it has
1417
reached equilibrium. It has been undisturbed for 30-40 years, with a constant mowing regime and
1418
no fertilisation or weedkilling. The lifespan of a ramet in the lawn is probably about a year, giving
1419
30-40 generations of turnover. For forest trees, with lifespans of say 300 years, the equivalent
1420
would be 9000-12000 years. In temperate areas the forests have not been around that long since
1421
the glaciation, and in tropical areas there would almost certainly have been major disturbance.
Wilson and Agnew, chapter 5, Assembly rules, page 49 of 50
1422
There is possibly no plant community anywhere closer to its equilibrium than the Botany Lawn. If
1423
the community is close to equilibrium, we can ask about its stability, and as we discussed in
1424
Chapter 3 the Botany Lawn community has been analysed for stability more intensively than any
1425
other community (Roxburgh and Wilson 2000a), and found to be on the borderline of stability, a
1426
conclusion confirmed by its response to perturbation (Roxburgh and Wilson 2000b). This stability
1427
is probably both the cause and the result of the assembly rules demonstrated.
1428
TABLES and ILLUSTRATIONS
1429
Table 5.1. Which species has the higher interference ability? The starting biomass for both species
1430
was 1.00.
1431
Table 5.2. Competitive hierarchy from Mouquet et al. (2004), strong competitors at the top
1432
Fig. 5.1: Four environments containing different species assemblages, consistent within each
1433
environment.
1434
Fig. 5.2: Two environments containing different species assemblages, but the same richness.
1435
Fig. 5.3. A patch randomisation model based on a grid of contiguous quadrats. The frequency of
1436
species A in the 3×3 patch is 3/9 = 0.333, so in the randomisation species A has a 0.333
1437
probability of occurring in the central square.
1438
1439
1440
1441
1442
1443
Fig. 5.4: Whittaker’s diagram (part of) of different distributions of species along an environmental
gradient.
Fig. 5.9: Time trends in the species composition and guild composition of plots planted with
different mixtures of species in the experiment of Fukami (2005).
Fig. 5.10. The concept of texture convergence. A similar range of characters is present on the two
continents, even though the species involved are different.
1444
Fig. 5.11. Texture convergence can be in: (a) mean or (b) shape.
1445
Fig. 5.12. Site 1 has the same texture as Site 2 with respect to the character, even though they
1446
differ in the number and abundances of species.
1447
Fig. 5.13. At Time 1, the realisable niches of Species X and Y overlap in an area of environmental
1448
hyperspace that exists. At Time 2, the combination of environmental conditions where they
1449
overlap does not occur. Inspired by Jackson and Williams (2004).
1450
Fig. 5.14. A RAD plot for biomass in a Spanish hay meadow. From J.B. Wilson (1991).
1451
Fig. 5.15: A minimum spanning tree for the species composition of British and New Zealand
1452
1453
roadside vegetation.
Fig. 5.16: Profile through a part of the Botany Lawn.
Wilson and Agnew, chapter 5, Assembly rules, page 50 of 50
1454
Fig. 5.17: Stratification in the Botany Lawn, (a) just after cutting to 2.7 cm and (b) after 14 days
1455
regrowth. Lines connect species pairs that are significantly different in vertical position.
1456
Rare species are omitted.
1457
Fig. 5.18: Mean Mowing Removal Index (MRI) of each guild at each sampling date. The P-values
1458
1459
are from t-tests for differences between guilds in mean Mowing Removal Index.
Fig. 5.19: Variation in the Mowing Removal Index (MRI) of species co-occurring at a point in the
1460
Botany Lawn.
“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.”
i