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