Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Storage effect wikipedia , lookup
Theoretical ecology wikipedia , lookup
Unified neutral theory of biodiversity wikipedia , lookup
Habitat conservation wikipedia , lookup
Ecological fitting wikipedia , lookup
Biodiversity action plan wikipedia , lookup
Occupancy–abundance relationship wikipedia , lookup
Fauna of Africa wikipedia , lookup
Introduced species wikipedia , lookup
Island restoration wikipedia , lookup
Latitudinal gradients in species diversity wikipedia , lookup
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