ECOLOGICAL ASSEMBLY RULES: Perspectives, Advances, Retreats

Home , Assembly rules

1 1 2 2 3 3 This page intentionally left blank 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 1 2 2 3 ECOLOGICAL ASSEMBLY RULES 3 4 Perspectives, advances, retreats 4 5 5 6 It has been more than 20 years since Jared Diamond focused attention on the 6 7 possible existence of assembly rules for communities. Since then, there has 7 8 been a proliferation of studies trying to promote, refute or test the idea that 8 9 there are sets of constraints (rules) on community formation and maintenance 9 10 (assembly). This timely volume brings together carefully selected contributions 10 11 which examine the question of the existence and nature of assembly rules 11 12 with some rigor and in some detail, using both theoretical and empirical 12 13 approaches in a variety of systems. The result is a balanced treatment, which 13 14 encompasses a wide range of topics within ecology, including competition and 14 15 coexistence, conservation and biodiversity, niche theory, and biogeography. 15 16 As such, it provides much to interest a broad audience of ecologists, while 16 17 also making an important contribution to the study of community ecology in 17 18 particular. 18 19 19 20 evan weiher is an Assistant Professor in the Biological Sciences Department, 20 21 University of Winsconsin, Eau Claire, Winsconsin, USA. 21 22 22 23 paul keddy holds the Edward G. Schlieder Endowed Chair for Environmental 23 24 Studies at Southeastern Louisiana University, Hammond, Louisiana, USA. 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 1 2 2 3 ECOLOGICAL ASSEMBLY RULES 3 4 4 5 Perspectives, advances, retreats 5 6 6 7 7 8 8 9 9 10 10 11 11 12 Edited by 12 13 13 14 14 15 EVAN WEIHER AND PAUL KEDDY 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40           The Pitt Building, Trumpington Street, Cambridge, United Kingdom    The Edinburgh Building, Cambridge CB2 2RU, UK 40 West 20th Street, New York, NY 10011-4211, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia Ruiz de Alarcón 13, 28014 Madrid, Spain Dock House, The Waterfront, Cape Town 8001, South Africa http://www.cambridge.org

First published in printed format 1999

ISBN 0-511-03299-4 eBook (Adobe Reader) ISBN 0-521-65235-9 hardback ISBN 0-521-65533-1 paperback 1 1 2 2 3 Contents 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 List of contributorsvii 13 14 14 15 Introduction: The scope and goals of research on assembly rules 15 16 Paul Keddy and Evan Weiher1 16 17 17 18 Part I: The search for meaningful patterns in species assemblages 18 19 1 The genesis and development of guild assembly rules 19 20 Barry J. Fox23 20 21 2 Ruling out a community assembly rule: the method of 21 22 favored states Daniel Simberloff, Lewi Stone, and 22 23 Tamar Dayan58 23 24 3 Community structure and assembly rules: confronting 24 25 conceptual and statistical issues with data on desert rodents 25 26 Douglas A. Kelt and James H. Brown75 26 27 4 Introduced avifaunas as natural experiments in community 27 28 assembly Julie L. Lockwood, Michael P. Moulton, and 28 29 Karla L. Balent108 29 30 5Assembly rules in plant communitiesJ. Bastow Wilson130 30 31 6 Assembly rules at different scales in plant and bird 31 32 communitiesMartin L. Cody165 32 33 7 Impact of language, history and choice of system on the study 33 34 of assembly rulesBarbara D. Booth and Douglas W. Larson206 34 35 35 36 Part II: Other perspectives on community assembly 36 37 8 On the nature of the assembly trajectory James A. Drake, 37 38 Craig R. Zimmerman, Tom Purucker, and Carmen Rojo233 38 39 9 Assembly rules as general constraints on community 39 40 compositionEvan Weiher and Paul Keddy251 40

v vi Contents 1 10 A species-based, hierarchical model of island biogeography 1 2 Mark V. Lomolino272 2 3 11 Interaction of physical and biological processes in the assembly 3 4 of stream ﬁsh communities Elizabeth M. Strange and 4 5 Theodore C. Foin311 5 6 12 Functional implications of trait–environment linkages in 6 7 plant communities Sandra Díaz, Marcelo Cabido, and 7 8 Fernando Casanoves338 8 9 13 When does restoration succeed? Julie L. Lockwood and 9 10 Stuart L. Pimm363 10 11 14 Epilogue: From global exploration to community assembly 11 12 Paul Keddy393 12 13 13 14 Index403 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 1 2 2 3 Contributors 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 Karla L. Balent 13 14 Department of Biological Sciences 14 15 SUNY College at Brockport 15 16 NY 14420 16 17 USA 17 18 18 19 Barbara D. Booth 19 20 Department of Botany 20 21 University of Guelph 21 22 Guelph 22 23 Ontario N1G 2W1 23 24 Canada 24 25 25 26 James H. Brown 26 27 Department of Biology 27 28 University of New Mexico 28 29 Albuquerque 29 30 NM 87131 30 31 USA 31 32 32 33 Marcelo Cabido 33 34 Instituto Multidisciplinario de Biologia Vegetal 34 35 (CONICET – Universidad Nacional de Córdoba) 35 36 Casilla de Correo 495 36 37 5000 Córdoba 37 38 Argentina 38 39 39 40 40

vii viii Contributors 1 Fernando Casanoves 1 2 Grupo de Estadistica y Biometria 2 3 Departmento de Desarrollo Rural 3 4 Facultad de Ciencias Agropecuarias 4 5 Unioversidad de Córdoba 5 6 Casilla de Correo 509 6 7 5000 Córdoba 7 8 Argentina 8 9 9 10 Martin L. Cody 10 11 Department of Biology 11 12 University of California 12 13 Los Angeles 13 14 CA 90095–1606 14 15 USA 15 16 16 17 Tamar Dayan 17 18 Department of Zoology 18 19 Tel Aviv University 19 20 Ramat Aviv 20 21 69978 Tel Aviv 21 22 Israel 22 23 23 24 Sandra Díaz 24 25 Instituto Multidisciplinario de Biologia Vegetal 25 26 (CONICET – Universidad Nacional de Córdoba) 26 27 Casilla de Correo 495 27 28 5000 Córdoba 28 29 Argentina 29 30 30 31 James A. Drake 31 32 Complex Systems Group 32 33 Ecology and Evolutionary Biology 33 34 University of Tennessee 34 35 Knoxville 35 36 TN 37996 36 37 USA 37 38 38 39 39 40 40 Contributors ix 1 Theodore C. Foin 1 2 Division of Environmental Studies 2 3 University of California 3 4 Davis 4 5 CA 95616 5 6 USA 6 7 7 8 Barry J. Fox 8 9 School of Biological Science 9 10 University of New South Wales 10 11 Sydney NSW 2052 11 12 Australia 12 13 13 14 Paul Keddy 14 15 University of Ottawa 15 16 30 Marie Curie Street 16 17 PO Box 450 Stn. A 17 18 Ottawa 18 19 Ontario K1N 6N5 19 20 Canada 20 21 21 22 Douglas A. Kelt 22 23 Department of Wildlife, Fish & Conservation Biology 23 24 University of California 24 25 Davis 25 26 CA 95616 26 27 USA 27 28 28 29 Douglas W. Larson 29 30 Department of Botany 30 31 University of Guelph 31 32 Ontario N1G 2W1 32 33 Canada 33 34 34 35 Julie L. Lockwood 35 36 Department of Ecology and Evolutionary Biology 36 37 The University of Tennessee 37 38 Knoxville 38 39 TN 37996 39 40 USA 40 x Contributors 1 Mark V. Lomolino 1 2 Department of Zoology and Oklahoma Natural Heritage Inventory 2 3 Oklahoma Biological Survey 3 4 University of Oklahoma 4 5 Norman 5 6 OK 73019 6 7 USA 7 8 8 9 Michael P. Moulton 9 10 Department of Wildlife Ecology and Conservation 10 11 The University of Florida 11 12 PO Box 110430 12 13 Gainesville 13 14 FL 32611 14 15 USA 15 16 16 17 Stuart L. Pimm 17 18 Department of Ecology and Evolutionary Biology 18 19 The University of Tennessee 19 20 Knoxville 20 21 TN 37996 21 22 USA 22 23 23 24 Tom Purucker 24 25 Complex Systems Group 25 26 Ecology and Evolutionary Biology 26 27 University of Tennessee 27 28 Knoxville 28 29 TN 37996 29 30 USA 30 31 31 32 Carmen Rojo 32 33 Complex Systems Group 33 34 Ecology and Evolutionary Biology 34 35 University of Tennessee 35 36 Knoxville 36 37 TN 37996 37 38 USA 38 39 39 40 40 Contributors xi 1 Daniel Simberloff 1 2 Department of Ecology and Evolutionary Biology 2 3 University of Tennessee 3 4 Knoxville 4 5 TN 37996 5 6 USA 6 7 7 8 Lewi Stone 8 9 Department of Zoology 9 10 Tel Aviv University 10 11 Ramat Aviv 11 12 69978 Tel Aviv 12 13 Israel 13 14 14 15 Elizabeth M. Strange 15 16 Department of Fishery and Wildlife Biology 16 17 Colorado State University 17 18 Fort Collins 18 19 CO 80523 19 20 USA 20 21 21 22 Evan Weiher 22 23 Department of Biological Sciences 23 24 PO Drawer GY 24 25 Mississippi State University 25 26 MS 39762 26 27 USA 27 28 28 29 J. Bastow Wilson 29 30 Botany Department 30 31 University of Otago 31 32 PO Box 56 32 33 Dunedin 33 34 New Zealand 34 35 35 36 36 37 37 38 38 39 39 40 40 xii Contributors 1 Craig R. Zimmerman 1 2 Complex Systems Group 2 3 Ecology and Evolutionary Biology 3 4 University of Tennessee 4 5 Knoxville 5 6 TN 37996 6 7 USA 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 1 2 2 3 Introduction: The scope and goals 3 4 of research on assembly rules 4 5 5 6 6 7 Paul Keddy and Evan Weiher 7 8 8 9 9 10 10 11 11 12 12 13 Why should assembly rules be studied, why should a symposium be organized, 13 14 why should a volume on the topic be published, and why should anyone 14 15 bother to read it? These are all perfectly reasonable questions, and it is our 15 16 task to briefly address them by way of introducing this volume. 16 17 For some time, ‘community ecology’ has been the name loosely applied 17 18 to a collection of studies and methods that apply to more than one organism, 18 19 but that apply at scales below the landscape. Many books on community 19 20 ecology appear to offer little more than a disparate hodgepodge of studies 20 21 that are unified solely by the above vague restrictions. This may seem a harsh 21 22 criticism, and a peculiar way to open a book written for community ecolo- 22 23 gists. But, these sort of criticisms form the ground of this volume (some might 23 24 suggest charnel ground), and it is not an original observation by any means. 24 25 Indeed, Pianka (1992), a prominent member of our discipline, has felt obliged 25 26 to apologize on our behalf in a paper titled ‘The state of the art in commu- 26 27 nity ecology’, observing therein that ‘community ecology has for too long 27 28 been perceived as repugnant and intractably complex’. He apologizes to a 28 29 world symposium that ‘the discipline has been neglected and now lags far 29 30 behind the rest of ecology’. 30 31 He is not alone in his survey, opinion and prognosis. Nearly two decades 31 32 earlier Lewontin (1974) wrote of the ‘agony’ of community ecology. More 32 33 recently, the late Rob Peters (1991) annoyed many when he decried the lack 33 34 of apparent progress in our discipline. Lawton (1992) then attacked Peters; 34 35 Keddy (1992) criticized his criticisms. Scheiner (1993) then criticized Keddy 35 36 for criticizing Lawton, and Keddy replied (Keddy, 1993), and then Scheiner 36 37 replied to him (Scheiner, 1994). Lacking tangible progress, people turn upon 37 38 one another. If there is agony in community ecology, as Lewontin suggested, 38 39 much of it appears to be self-inflicted. Meanwhile, thick compendia under the 39 40 name of community ecology arise with frustrating regularity and repetitive 40

1 2 P. Keddy and E. Weiher 1 content. This situation is what led Keddy (1993) to observe that, without far 1 2 more emphasis upon measurable properties, critical tests and rational deci- 2 3 sion-making, community ecologists run the risk of becoming more like the 3 4 humanities than the sciences, prone to political and emotional conflicts rather 4 5 than debates using rational criteria. In Camille Paglia’s (1992) essay ‘Junk 5 6 bonds and corporate raiders: academe in the hour of the wolf’ one can read 6 7 her view that ‘the self-made inferno of the academic junk bond era is the 7 8 conferences, where the din of ambition is as deafening as on the floor of 8 9 the stock exchange. The huge, post-1960s’ proliferation of conferences ... 9 10 produced a diversion of professional energy away from study and toward per- 10 11 formance, networking, advertisement, cruising, hustling, glad-handing, back- 11 12 scratching, chitchat, groupthink’. Mercifully, community ecologists (and this 12 13 volume) are completely immune to this risk because we are doing science. 13 14 Why a symposium on assembly rules, and why a book you may still be 14 15 asking? Plans for the symposium were rooted in the above circumstances, 15 16 combined with two common-sense observations. These were: 16 17 17 (a) if there are not some common goals for community ecology, then they 18 18 are unlikely to be achieved; 19 19 (b) a prominent theme in the discipline is the attempt to predict the com- 20 20 position of ecological communities from species pools. 21 21 22 The first observation appears self-evident, but apparently in some circles there 22 23 is still a suspicion of research ‘agendas’. There appears to be a naive belief 23 24 that the way to build a spaceship and land a human on the moon is to trust 24 25 that, if everyone indulges themselves in an idiosyncratic and self-indulgent 25 26 pastime at the taxpayer’s expense, the outcome will be positive. Like 26 27 Voltaire’s Professor Pangloss, there is the insistence that this must be the best 27 28 of all possible worlds, however inefficient or painful it may appear on the 28 29 surface. If progress is forgotten about entirely, and our discipline is seen as 29 30 a mere pastime for the tenured and well to do, there is no particular need to 30 31 be concerned about goals, progress and social contribution. As long as ecolo- 31 32 gists have jobs and the chance at a large grant, why worry? Volumes with 32 33 this kind of philosophy are too frequent as it is, and given that it has been said 33 34 in print (Keddy, 1991), as editors we were careful to insist upon a common 34 35 purpose. Within this common purpose, a diversity of views about the details 35 36 and the strategies and tactics for achieving it was welcomed. 36 37 The second observation may be less evident, but if the many topics that 37 38 have arisen in community ecology over the years are considered, the com- 38 39 mon thread may not be in level of organization, methodology or number of 39 40 species, but in the underlying problem that is being addressed. Moreover, 40 Introduction 3 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 Fig. 1. Assembly rules address a central theme in community ecology: how are com- 14 15 munities assembled from species pools. (Evolutionary ecology, by contrast, deals with 15 the formation of the pool.) 16 16 17 17 18 adopting this goal clarifies a common source of confusion among ecologists 18 19 themselves: community ecology is different from evolutionary ecology. 19 20 Figure 1 shows a possible framework for community ecology. Community 20 21 ecologists are concerned with the question mark: how does one get from the 21 22 pool to the community? Evolutionary ecologists are concerned more with 22 23 the top box and arrows: how do speciation and extinction produce the pool? 23 24 From the perspective of community ecology, the pool is just the source of 24 25 raw materials, and the process that creates the pool, while of interest, gener- 25 26 ally occurs at longer time scales than are normally considered relevant. 26 27 A new name could be invented to describe the study of how communities 27 28 are built from pools, but if the literature is looked back on, there is a good 28 29 deal of terminology that can already be borrowed. There is always a risk in 29 30 using pre-existing terminology, because it all comes with baggage. The bag- 30 31 gage includes assumptions about the kinds of organisms worthy of study, past 31 32 controversies that are actually tangential to the issues at hand, and past con- 32 33 fusions that entangled ecologists. It is for this reason that the Roman armies 33 34 called baggage impedimenta. Ecologists do not need more impediments. But, 34 35 neither does there seem to be any point in inventing new terms when per- 35 36 fectly good ones are already there. To do so would be to throw out the wis- 36 37 dom of past work because the baggage is feared. Thus the term ‘assembly 37 38 rules’ has been adopted to describe the problem of assembly communities from 38 39 pools; this accords rather well with Diamond’s original (1975) usage of the 39 40 word. There may be doubts about using birds as a model system, about 40 4 P. Keddy and E. Weiher 1 descriptive as opposed to experimental studies, about inferences about mech- 1 2 anisms that may not be justified, about controversies that have generated more 2 3 heat than light, and about habitual ways of trying to solve these problems 3 4 that appear self-defeating. All of these objections (and more,) were raised 4 5 either by participants in the symposium, or by other practising ecologists. The 5 6 term assembly rules, however, still captures the essence of the problem in 6 7 Fig. 1. Moreover, it nicely fits in with Pirsig’s (1974) observation that assem- 7 8 bling a rotisserie is not unlike fixing a motorcycle, that the challenge of putting 8 9 something together properly from the pieces (assembling it) is a challenge 9 10 with worthy mechanical and philosophical dimensions. 10 11 And so, the symposium was called ‘assembly rules’, and researchers were 11 12 sought out who were studying how communities were assembled out of pools. 12 13 In spite of ourselves, the perceptive reader will discern certain biases. For 13 14 example, Fig. 2 gives one perspective upon the composition of ecological 14 15 communities upon Earth’s surface; more recent calculations would expand 15 16 the invertebrate and fungal component. This would seem to be a common- 16 17 sense starting point in designing the discipline of community ecology. In 17 18 spite of ourselves, we have ended up with a disproportionate representation 18 19 of vertebrate examples. Our defence is that, while trying to collect a repre- 19 20 sentative set of studies on community assembly, a highly biased and arti- 20 21 ficial pool from which to make the draw was being dealt with, and so the 21 22 distortions of our literature have been included. Our only plea is that we 22 23 consciously tried to avoid the worst distortions. Further, to the extent that 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 Fig. 2. The importance of different life forms in the biosphere, measured according 39 40 to biomass (left) and number of species (right). (From Keddy, 1989.) 40 Introduction 5 1 they have been reproduced, others can only be encouraged to rectify the 1 2 situation. 2 3 3 4 4 5 Two existing paradigms 5 6 Within the literature, and within this volume there are at least two developing 6 7 paradigms for the assembly of communities (Fig. 3). The first we call the 7 8 island paradigm because it deals with mainlands, islands, immigration, and 8 9 coexistence. The second we call the trait–environment paradigm because it 9 10 begins with pools, habitats as filters, and convergence. This is not to suggest 10 11 that there are only two ways forward, or that either of these is the best. These 11 12 simply happen to be two themes which will be evident in this concert. The 12 13 challenge for a musician is to build upon a theme in an entertaining way with- 13 14 out being repetitive. 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 Fig. 3. The two most common paradigms for community assembly are the island 32 paradigm (top) and the trait–environment paradigm (bottom). 32 33 33 34 34 35 35 36 Type 1: Island models 36 37 Many studies are built upon the raw data lists of species on islands. In 37 38 terms of Fig. 1, the pool is the adjoining mainland, and the list of species 38 39 from the island is considered to be the community. The basic series of steps 39 40 is as follows: 40 6 P. Keddy and E. Weiher 1 (a) make lists of organisms in each habitat; 1 2 (b) create one or more null models for possible patterns; 2 3 (c) test for patterns in these lists; 3 4 (d) offer explanations for these patterns; 4 5 (e) state the explicit rules for community assembly. 5 6 6 A good example of this sort of study comes from Diamond’s (1975) work 7 7 on the avifauna of New Guinea (Fig. 4). There is now a large literature on 8 8 this topic, and a growing body of literature on null models, but a good deal 9 9 of controversy about the costs and benefits of null models and the kinds of 10 10 data appropriate to them (Gotelli & Graves, 1996). Moreover, while there 11 11 have been many searches for evidence of pattern, few brave souls have 12 12 reached step (e). 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 Fig. 4. The island paradigm for assembly is built upon studies of bird distribution 34 on offshore islands. Are there patterns, do they differ from those predicted by null 34 35 models, and are there rules that predict them? This example shows the distribution of 35 36 two species of Ptilinopus fruit pigeons, where split circles are co-occurrences and dots 36 are co-absences. (From Keddy, 1989 after Diamond, 1975.) 37 37 38 38 39 39 40 40 Introduction 7 1 1 2 Type 2: Trait–environment models 2 3 One can also approach community assembly not by using lists of organisms, 3 4 but by focusing upon their traits. The environmental factors are then viewed 4 5 as filters acting upon these traits. In this case the procedure is as follows: 5 6 6 (a) determine the key traits the organisms possess; 7 7 (b) relate the traits to key environmental factors; 8 8 (c) specify how trait composition will change with specific changes in envi- 9 9 ronment; 10 10 (d) relate this back to the particular organisms possessing those traits. 11 11 12 We are not interested in general properties of the traits themselves, but in the 12 13 relative abundances of the organisms that possess them. A good example of 13 14 this sort of study is the work on prairie potholes by van der Valk (1981). The 14 15 water level in the pothole acts as a filter determining the kinds of plant species 15 16 that will occur there; the two key states are drained v. flooded (Fig. 5). 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 Fig. 5. The trait environment paradigm for community assembly focuses attention 31 32 upon the pool of species, the traits they possess, and the environmental filters oper- 32 ating in a particular situation. The example shows that the composition of vegetation 33 in a pothole wetland depends upon whether the pool of buried seeds is exposed to 33 34 flooded or drained conditions. (From van der Valk, 1981.) 34 35 35 36 36 37 37 38 What is an assembly rule? 38 39 What would an assembly rule look like if one were found? A goal cannot be 39 40 attainable unless some criteria are set up to tell when it has been achieved. An 40 8 P. Keddy and E. Weiher 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 Fig. 6. A first step in the search for assembly rules is the search for pattern. Island 14 15 modes have tended to search for overdispersion (right) whereas trait–environment 15 models have tended to focus upon trait underdispersion (left). A larger view suggest that 16 both are possible, depending upon the scale of enquiry. (From Weiher & Keddy, 1995.) 16 17 17 18 18 19 assembly rule specifies the values and domain of factors that either structure 19 20 or constrain the properties of ecological assemblages. 20 21 Overall, there are four parts in the procedure of finding assembly rules: 21 22 22 (a) defining and measuring a property of assemblages; 23 23 (b) describing patterns in this property; 24 24 (c) explicitly stating the rules that govern the expression of the property; and 25 25 (d) determining the mechanism that causes the patterns. 26 26 27 Contrary to common practice, merely documenting a pattern is not the study 27 28 of community assembly. Plant ecologists have described plant patterns for more 28 29 than century, and appear ready to continue doing so for yet another; simply 29 30 describing patterns is not the study of assembly rules. Nor is an improve- 30 31 ment, the added demonstration that pattern exists against a null model, suffi- 31 32 cient to qualify. Asking if there is pattern in nature is akin to asking if bears 32 33 shit in the woods. Null models provide a valuable and more rigorous way of 33 34 demonstrating pattern, but they still do not specify assembly rules. Within this 34 35 realm of pattern (step 2), one, of course, needs to ask what kinds of patterns 35 36 might occur and at what scales (Fig. 6). But actual assembly rules, step 3, 36 37 require further effort yet. They might include statements such as the following: 37 38 ‘If an assemblage of plants is flooded, the subset of species that survives 38 39 will all have aerenchyma.’ 39 40 ‘In the absence of predators, a pond in the temperate zone can be expected 40 Introduction 9 1 to have between 5 and 10 amphibian species. If a predatory fish species is 1 2 introduced to the pond, this will fall to between 0 and 2 species.’ 2 3 ‘The ratio of insectivorous to granivorous birds in deciduous forests is 3 4 between 0.25 and 0.33, whereas in boreal forests the ratio falls between 0.45 4 5 and 0.55.’ 5 6 ‘If a herbaceous plant community with biomass of 500 g m−2 is fertilized 6 7 with NPK fertilizer, the mean number of species per m2 will decline by 10% 7 8 with each x g m−2 of fertilizer.’ 8 9 ‘There is a linear relationship between the number of beetles in deciduous 9 10 forests and the volume of coarse woody debris. The equation relating the two 10 11 is as follows ...’ 11 12 These statements all are expressed in terms of measurable properties and 12 13 their range or variation in relation to another factor. The following statements 13 14 would not qualify as assembly rules. 14 15 ‘Similar organisms tend not to coexist.’ 15 16 ‘Competition controls the distribution of birds on islands.’ 16 17 ‘Copepod communities in ponds are not randomly assembled from a species 17 18 pool.’ 18 19 ‘Tree species diversity increases with decreasing latitude.’ 19 20 ‘The distribution of lizards upon islands deviates significantly from null 20 21 models.’ 21 22 Such statements certainly belong within community ecology, and may have 22 23 within them concepts or models that increase our understanding of certain 23 24 assemblages. But they are not assembly rules. Rules themselves must be 24 25 explicit and quantitative if they are to qualify. The other statements are, 25 26 perhaps, steps on the way to the goal. There seems to be some current con- 26 27 fusion on this matter. For example, the existing literature suggest that merely 27 28 finding a pattern in properties is an assembly rule. Confusing a step on the 28 29 path with the attainment of a goal only generates confusion. 29 30 30 31 31 32 Obstacles on the path 32 33 Good generals study the failures of other generals so that they do not repeat 33 34 their mistakes. One of the consequences of accepting a goal is the recogni- 34 35 tion and study of obstacles and past errors. This could be considered annoy- 35 36 ing, something to be avoided at all costs, perhaps because no one wants to 36 37 admit to having fallen prey to an obstacle. Perhaps our early days in Sunday 37 38 school are remembered, being told of our committing a sin. But, the delight- 38 39 ful side of this is that if obstacles cannot be recognized, they cannot be 39 40 avoided. That is why road signs warning ‘detour ahead’ are so useful; without 40 10 P. Keddy and E. Weiher 1 them we could damage our car. Returning to generals, in his chapter Doctrine 1 2 of Command, General Montgomery (1958) wrote ‘I hold the view that the 2 3 leader must know what he himself wants. He must see his objective clearly 3 4 and then strive to attain it; he must let everyone else know what he wants 4 5 and what are the basic fundamentals of his policy.’ (p. 81) 5 6 The frequent reluctance to admit that there are both goals and obstacles to 6 7 our work is certainly unmilitary, and perhaps unprofessional. It may reflect 7 8 a desire to remain child-like, innocent and naive, with no responsibility for 8 9 one’s actions. Once, like General Patton, our intention to be in Berlin next 9 10 year is announced, everyone will know if it is not achieved. It takes some 10 11 bravery to announce our goals, and to suggest that society should care whether 11 12 Berlin or Paris is achieved. Are there some obstacles that have interfered with 12 13 past campaigns in community ecology. What are some of these errors that 13 14 might have led to Pianka’s despair? At the symposium the participants were 14 15 specifically warned against some pitfalls. For those who were not in attendance, 15 16 they are briefly listed below. 16 17 Before the list is presented, one more clarification is necessary. The list 17 18 offers styles of research which are obstacles to advancement. Elements of 18 19 such styles are contained in everyone, but in different relative proportions. In 19 20 an exactly analogous way, all humans have anger, negativity, arrogance, envy, 20 21 ignorance, greed, suspicion in their psychological make-ups. One of the 21 22 reasons humans have lists such as the seven deadly sins, or the three poisons, 22 23 is to be warned to watch out for these states as they arise within own minds. 23 24 Tradition has taught that these states create confusion for ourselves, and prob- 24 25 lems for our human communities. In the same way, the following list of styles 25 26 in intended to illustrate approaches that can be slid into if one remains unaware 26 27 of one’s own behavior. The intention is not to list obstacles in order to imply 27 28 that these flaws are found in only a few bad people (see the exchange between 28 29 Scheiner, 1993 and Keddy, 1993), nor so readers can try to guess who falls 29 30 into which category, but rather to acknowledge that all humans are subject 30 31 to such tendencies. Such a list, then, provides gauges for an instrument panel 31 32 that will warn if one wanders too far into unproductive terrain. Five obstacles 32 33 are considered. 33 34 34 35 (a) Bigger is better (‘Mine is bigger than yours’) 35 36 Sometimes it is thought that if someone does not know where they are going, 36 37 then at least the neighbours can be impressed by seeing them drive a bigger 37 38 car while they try to get there. This is common in the animal and plant king- 38 39 doms. Large insects tend to be dominant over smaller ones (Lawton & Hassell, 39 40 1981) just as large plants tend to be dominant over smaller ones (Gaudet & 40 Introduction 11 1 Keddy, 1988). Display of wealth or economic power is a standard technique 1 2 humans use to maintain authority over neighbours (Kautsky, 1982), the cul- 2 3 mination of which is perhaps seen in the arms races during and after the 3 4 Second World War (Keegan, 1989). There is no need here to impress the 4 5 audience with a Cray 2 supercomputer when only a pocket calculator will make 5 6 the point. Similarly, there is no need to show the helicopter or float plane 6 7 that it took to reach the study site. 7 8 8 9 (b) Complexity is impressive (‘I know more natural history than you’) 9 10 Many people know a great deal about nature. Keddy could, for example, lec- 10 11 ture for hours (and write for pages) upon the plants that grow in wetlands or 11 12 in Lanark County where he lives. If he did so in a sonorous and authorita- 12 13 tive voice, he might even convince you that your time listening was well 13 14 spent. But he would mislead if he tried to pass off his delightful knowledge 14 15 of natural history under the term assembly rules. Merely knowing where 15 16 organisms are found, and describing it in exquisite detail, is not the study of 16 17 assembly rules. Otherwise, Tansley and Adamson (1925) would have to be 17 18 credited with assembly rules for British grassland. In their study of grazing, 18 19 they reported on factors controlling composition in three patches along a gra- 19 20 dient of ‘progressive increase in height and density of vegetation, with an 20 21 increasing number of species’. In one paragraph they observe: 21 22 22 ‘Of the mosses, Barbula cylindrica, B. unguiculata and Bryum capillare 23 23 decrease and disappear with complete closure and increasing depth of the 24 24 turf; Brachythecium purum, one of the most ubiquitous of chalk grassland 25 25 mosses, though not a ‘calcicole’ species, appears in (b) and increases in (c); 26 26 while B. rutabulum, Mnium undulatum and Thuidium tamariscinum first 27 27 appear in the damper conditions of (c)’ 28 28 29 These sorts of descriptions are possible for all manner of ecological systems; 29 30 the real problem is to discover the generalities that underlie the remarkable 30 31 diversity of life (Mayr, 1982). General Montgomery (1958) says ‘It is 31 32 absolutely vital that a senior commander should keep himself from becoming 32 33 immersed in details ... If he gets involved in details ... he will lose sight of 33 34 the essentials which really matter; he will then be led off on side issues which 34 35 will have little influence on the battle ...’ (p. 86). 35 36 In general, simple hypotheses are preferable to complex ones (Aune, 1970). 36 37 We need to remember the late Rob Peters’ warning (1980a,b) that there is an 37 38 important difference between natural history and ecology. It might be recalled 38 39 that even an omniscient and omnipotent deity felt it necessary to give Moses 39 40 only ten rules to guide all the complexities of human conduct. 40 12 P. Keddy and E. Weiher 1 (c) Sycophancy (‘My friends are more important than yours’) 1 2 Humans are primates, and the primate mind appears to have evolved for 2 3 survival in social/tribal settings (e.g., Leakey & Lewin, 1992). It is therefore 3 4 entirely natural for authority figures to be created, like the old silver back 4 5 males in the Gorilla tribe, and then deferred to, whether people live in the 5 6 world’s most powerful democracy (Dye & Zeigler, 1987) or a Stalinist prison 6 7 camp in Siberia (Shalamov, 1982). 7 8 There have been sycophants as long as there have been authority figures. 8 9 Socrates is often presented as a courageous man of science who died rather 9 10 than appease the Athenian rulers. But, in a compelling re-evaluation of the 10 11 historical records, I.F. Stone (1989) reached quite a different conclusion. At 11 12 the time, he notes, Athens was repeatedly threatened by tyrants; in both 411 12 13 and in 404, the conduct of the aristocratic dictators was ‘cruel, rapacious and 13 14 bloody’. Socrates was strangely silent while all this happened. Stone looks 14 15 in vain for Socrates to show compassion for the poor, or opposition to tyranny, 15 16 and notes that the famous Republic, later written by his student Plato, advo- 16 17 cates a tyranny remarkably close to the modern Communist state, complete 17 18 with internal security police (‘guardians’). Socrates, concludes Stone, was a 18 19 sycophant for the tyrants. He did not plead for freedom of speech because it 19 20 was a principle he did not believe in, and he could not bring himself to argue 20 21 before the democrats using their own principles. If Stone’s interpretation 21 22 is correct, using Socrates as an example for scientists may have far darker 22 23 connotations than has been realized, and comes rather close to Saul’s (1992) 23 24 contemporary view that intellect is too often used to reinforce authority rather 24 25 than challenge it. 25 26 Apparently it is natural for humans to divide themselves into tribes and 26 27 attack one another (Ignatieff, 1993). It further seems natural to kick and bite 27 28 (or at least ostracize) members of our tribe who do not seem to want to groom 28 29 the same dominant that the others do (Browning, 1992). They may even favor 29 30 their own gender (Gurevitch, 1988). The fact that this was natural behavior 30 31 for primates on the African plains does not mean that it is defensible or 31 32 useful behavior for contemporary scientists. Indeed, in other settings, such as 32 33 religion, people frequently react with annoyance when they see blind 33 34 obedience and the exercise of authority. Ape instincts such as sycophancy 34 35 seem obvious when they creep out of the tribe and into our of politics, but 35 36 perhaps less so when then appear in the sciences (Keddy, 1989). Appeals to 36 37 authority, the selective citation of friend’s work, and division into tribal units 37 38 do not contribute to the advancement of science. All are all to be avoided 38 39 here. 39 40 40 Introduction 13 1 (d) Self-indulgence (‘Playing with yourself is harmless fun’) 1 2 It may be true that masturbation does not cause blindness, but neither is it 2 3 necessarily a healthy substitute for a relationship with another human being. 3 4 Similarly, self-indulgent work aimed at building up egos may not cause 4 5 irreparable harm to science, but it certainly is no substitute for thoughtful 5 6 goals, collegiality, care, respect and social responsibility. The motivation that 6 7 is brought to our work will necessarily influence the way in which it devel- 7 8 ops. It is therefore necessary to think about where we are going and how our 8 9 discoveries might be of use to society. The alternative is unpleasant for 9 10 everyone. The Roman historian Plutarch describes how one Roman emperor 10 11 after another ‘lavished away the treasures of his people in the wildest extrav- 11 12 agance’. One was killed by a tribune amongst his own guards, another was 12 13 strangled, another killed himself, but the waste of resources caused by this 13 14 self-indulgence caused immense human suffering and the diminution of 14 15 the resources of the Roman empire. Are precious scientific resources being 15 16 squandered in a similar manner today? Is Ecology just National Geographic 16 17 without the color plates? 17 18 Hofstadter (1962) suggests that the current American environment of anti- 18 19 intellectualism is responsible for isolation of scientists from their society: 19 20 ‘Being used to rejection, and having over the years forged a strong traditional 20 21 response to society based upon the expectation that rejection would continue, 21 22 many of them have come to feel that alienation is the only appropriate 22 23 and honorable stance for them to take.’ (p. 393) Further, he continues, it is 23 24 easy to take this to the next logical step and slip into the assumption that 24 25 alienation has some inherent value itself. As an example, a science reporter 25 26 tried to interview one of our participants by asking ‘What is the practical 26 27 importance of this work on assembly rules?’. The answer ‘None, I hope’ 27 28 perhaps illustrates Hofstadter’s point. 28 29 Guarding against this self-indulgence is necessary not just because it is 29 30 other humans who pay our salaries. The attempt to explain ourselves to oth- 30 31 ers and to solve real world problems actually forces us to do better science. 31 32 It may be harmful to indulge in simple natural history, but it is equally dan- 32 33 gerous to take the other extreme and retreat entirely to a cosy world of abstrac- 33 34 tion. When an engineer builds a bad bridge, it falls down. This is a simple 34 35 example of the pragmatic method, which James (1907) described as ‘primarily 35 36 a method of settling metaphysical disputes that might otherwise be inter- 36 37 minable’ (p. 10) In contrast, when an ecologist builds a pointless model or 37 38 publishes a bad paper, it is possible for it to persist and cause unnecessary 38 39 debates for decades. 39 40 40 14 P. Keddy and E. Weiher 1 (e) Lack of historical context (‘My ideas are new and unprecedented’) 1 2 Life is impermanent, and one way to try escape our fears is self-aggrandization. 2 3 The more history is ignored, the more important it can seem to be. It has 3 4 been argued that since the 1960s, ecologists have tended to inflate our self- 4 5 importance by ignoring the roots of our discipline (Gorham, 1953; Jackson, 5 6 1981). Consider a recent example. In 1970 Walker tested whether wetland 6 7 vegetation along lakes showed a progressive and predictable series of devel- 7 8 opmental stages using 159 observed transitions between vegetation types 8 9 extracted from 20 published pollen diagrams (Fig. 7). One dominant course 9 10 was observed (darker lines). Reed swamp (5) was an essential stage through 10 11 which all successions must pass. Some reversals occur (17% of all transi- 11 12 tions), ‘but are usually short-lived and might frequently be due to small 12 13 changes in water level, temperature or trophic status of the lake water ...’ 13 14 Do such observations, perhaps slightly reworded, not qualify as assembly 14 15 rules? The scale in time and space, and the replication is impressive relative 15 16 to many other published studies on assembly rules. So, why is work such as 16 17 Walker so consistently overlooked? See if you can find it cited in the assembly 17 18 rules literature. 18 19 A more recent example of historical revisionism is provided by Gotelli and 19 20 Graves (1996) in their book on null models. The theme of whether plant com- 20 21 munities are discrete communities or random assemblages can be traced back 21 22 through writings by Tansley, Clements, Gleason, Ellenberg and Whittaker 22 23 throughout the period from 1900 to 1970; indeed, the study of species dis- 23 24 tributions along gradients has probably been the defining feature of research 24 25 in plant ecology. In the early 1970s, E.C. Pielou developed a number of null 25 26 models for plant distributions along gradients, which she summarized in her 26 27 1975 book. There have been only five main papers that report the applica- 27 28 tion of her null models in the field (Pielou & Routledge (1976) in salt marshes, 28 29 Keddy (1983) on lake shores, Dale (1984) on marine rocky shores, Shipley 29 30 and Keddy (1987) in marshes, and Hoagland and Collins (1997) in wet 30 31 prairies), and in each case the null model has been rejected. Yet Gotelli and 31 32 Graves do not have a chapter on gradient models, nor does their one para- 32 33 graph on Pielou and Routledge (1976) explain the significance of their work 33 34 – that the rejection of Pielou’s null model constitutes the first demonstration 34 35 that communities occur in discrete clusters rather than random (individualistic) 35 36 associations. On page 1, Gotelli and Graves (two male American zoologists) 36 37 even opine that the term null models for communities originated with two 37 38 male American Zoologists at a Florida symposium in 1981! Chris Pielou was 38 39 a woman, she was Canadian, and she was a plant ecologist: to which of these 39 40 should her erasure from the ecological record be attributed? 40 Introduction 15 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 Fig. 7. Frequencies of transition between 12 vegetation stages (ranging from open 33 water (1) through reed swamp (5) to bog (11) to mixed marsh (12)) in 20 pollen cores 33 34 from a range of wetlands including small lakes, valley bottoms, and coastal lagoons 34 35 in the British Isles. Left, tabulated frequencies, right, transition diagram (after Walker 35 36 1970). 36 37 37 38 In the broader sense, one could argue that our entire culture is becoming 38 39 focused upon the ephemera of the present where everyone will have their 39 40 15 minutes of fame; a consequence is a loss of context or significance for 40 16 P. Keddy and E. Weiher 1 properly evaluating current events (O’Brien, 1994; Saul, 1995)). Booth and 1 2 Larson (this volume) suggest that most of the apparent problems with assem- 2 3 bly rules arise out of a studied ignorance of ecological principles prevalent 3 4 in the early part of this century. 4 5 5 6 6 7 Moving forward 7 8 Ending with a list of obstacles could be seen as unnecessarily defeatist or 8 9 negative. The only reason to study obstacles is to avoid them. It can be hoped 9 10 that, by consciously avoiding these tendencies, everyone will be able to avoid 10 11 repeating patterns of behavior that limit our personal contribution to the 11 12 progress of ecology, or spread discord among our colleagues. By defining 12 13 warning regions, regions which are positive and valuable are equally defined 13 14 (Fig. 8). It is not, then, that a single narrow goal is being advocated, nor a 14 15 single path forward. 15 16 Rather than insist that there is only one goal for assembly rules, it is asked 16 17 that each participant explicitly state their goal; then and only then will it be 17 18 possible to understand their tactics and judge their degree of success in attain- 18 19 ing their goal. To return to our travel analogy, some may be aiming for Paris 19 20 and others for London; so long as they justify their destination, their trip can 20 21 be judged only against the stated goal. Of course, if someone asserts that they 21 22 intend to take us to the Louvre, but then drives us towards London instead, 22 23 it may be gently suggested from the back seat that they review their travel 23 24 plans, consult the map, or else choose another destination. 24 25 Nor would it be desirable to imply that there is room for only a single 25 26 style of science. Within our community it can be accepted that different 26 27 practitioners have different strengths and weaknesses; progress in ecology 27 28 depends upon exploitation of, and respect for, these differences. So while 28 29 Fig. 8 defines some regions as obstacles, it should be apparent that there is 29 30 still plenty of room remaining for a diversity of approaches. Some of the rel- 30 31 atively sterile arguments in ecology may arise out of simple lack of tolerance 31 32 for different styles. It is therefore necessary to be clear in our discussions 32 33 whether disagreement has arisen because of (a) different views on goals, 33 34 (b) different views on tactics and style, (c) different interpretations of data, 34 35 or (d) indulgence in one of the above obstacles. On one hand, it is necessary 35 36 to be open minded to avoid pointless debates that really hinge on differences 36 37 in personality or style. Equally however, being open minded does not mean 37 38 that our brains must be allowed to fall out and outright self-indulgence be 38 39 tolerated. 39 40 Having explicitly considered the need for clearly stated goals, some obstacles 40 Introduction 17 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 Fig. 8. A variety of psychological obstacles can hamper progress in ecology. By explic- 33 itly describing these obstacles, a region is simultaneously defined within which 34 progress is possible; within this region, differences in style enrich collaboration. 34 35 35 36 36 37 to progress, some ways in which researchers may differ in opinion, and 37 38 the merits of tolerating different approaches, let us return to the study of 38 39 assembly rules. The problem remains. How do we get from the pool to the 39 40 community? The introductory talk ended with Fig. 9, so this same figure 40 18 P. Keddy and E. Weiher 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 Fig. 9. Participants in the symposium were each asked to chart out their own course 13 14 though the maze of possibilities that takes us from the pool to the community. 14 15 15 16 16 17 terminates this written introduction. See what the authors have to say in 17 18 response. In the words of Peter Weiss (1965): ‘We ask your kindly indul- 18 19 gence for a cast never on stage before coming to Charenton. But each inmate 19 20 I can assure you will try to pull his weight.’ 20 21 21 22 References 22 23 Aune, B. (1970). Rationalism, Empiricism, and Pragmatism. NY: Random House. 23 24 Browning, C.R. (1992). Ordinary Men: Reserve Police Battalion 101 and the Final 24 25 Solution in Poland. NY: Harper Collins. 25 26 Dale, M.R.T. (1984). The contiguity of upslope and downslope boundaries of 26 species in a zoned community. Oikos 47: 303–308. 27 Diamond, J.M. (1975). Assembly of species communities. In Ecology and 27 28 Evolution of Communities, ed. M.L. Cody & J.M. Diamond, pp. 342–444. 28 29 Cambridge, MA: Belknap Press of Harvard University Press. 29 Dye, T.R. & Zeigler, H. (1987). The Irony of Democracy. 7th edn. Monterey, CA: 30 Brooks/Cole. 30 31 Gaudet, C.L. & Keddy, P.A. (1988). Predicting competitive ability from plant traits: 31 32 a comparative approach. Nature 334: 242–243. 32 Gorham, E. (1953). Some early ideas concerning the nature, origin and 33 development of peat lands. Journal of Ecology 41: 257–274. 33 34 Gotelli, N.J. & Graves, G.R. (1996). Null Models in Ecology. Washington: 34 35 Smithsonian Institution Press. 35 36 Gurevitch, J. (1988). Differences in the proportion of women to men invited to give 36 seminars: is the old boy still kicking? Bulletin of the Ecological Society of 37 America 69: 155–160. 37 38 Hoagland, B.W. & Collins, S.L. (1997). Gradient models, gradient analysis and 38 39 hierarchical structure in plant communities. Oikos 78: 21–30. 39 Hofstadter, R. (1962). Anti-Intellectualism in American Life. NY: Vintage Books. 40 40 Introduction 19

1 Ignatieff, M. (1993). Blood and Belonging. Toronto: Penguin Books Canada Ltd. 1 2 Jackson, J.B.C. (1981). Interspeciﬁc competition and species distributions: the 2 ghosts of theories and data past. American Zoologist 21: 889–901. 3 James, W. (1907). Pragmatism. Reprinted In 1990, Great Books of the Western 3 4 World. ed. M.J. Adler, Vol. 55, pp. 1–89. Chicago, IL: Encyclopedia 4 5 Britannica. 5 6 Kautsky, J.H. (1982). The Politics of Aristocratic Empires Chapel Hill: University 6 of North Carolina Press. 7 Keddy, P.A. (1983). Shoreline vegetation in Axe Lake, Ontario: effects of exposure 7 8 on zonation patterns. Ecology 64: 331–344. 8 9 Keddy, P.A. (1989). Competition. London: Chapman and Hall. 9 Keddy, P.A. (1991). Thoughts on a festschrift: what are we doing with our 10 scientiﬁc lives? Journal of Vegetation Science 2: 419–424. 10 11 Keddy, P.A. (1992). Thoughts on a review of a critique for ecology. Bulletin of the 11 12 Ecological Society of America 73: 231–233. 12 Keddy, P.A. (1993). On the distinction between ad hominid and ad hominem and 13 its relevance to ecological research. A reply to Scheiner. Bulletin of the 13 14 Ecological Society of America 74: 383–385. 14 15 Keegan, J. (1989). The Second World War. NY: Viking Penguin. 15 16 Lawton, J. (1992). Predictable plots. Nature 354: 444. 16 Lawton, J.H. & Hassell, M.P. (1981). Asymmetrical competition in insects. Nature 17 289: 793–795. 17 18 Leakey, R. & Lewin, R. (1992). Origins Reconsidered. In Search of What Makes 18 19 Us Human. New York: Doubleday. 19 Lewontin, R.C. (1974). The Genetic Basis of Evolutionary Change. NY: Columbia 20 University Press. 20 21 Mayr, E. (1982). The Growth of Biological Thought. Diversity, Evolution and 21 22 Inheritance. Cambridge, MA: Belknap Press of Harvard University Press. 22 Montgomery, B. (1958). The Memoirs of Field-Marshall the Viscount Montgomery 23 of Alamein. St James’s Place, London: K.G. Collins. 23 24 O’Brien, C.C. (1994). On the Eve of the Millennium. Ontario: Anansi, Concord. 24 25 Paglia, C. (1992). Sex, Art and American Culture. NY: Random House. 25 26 Peters, R.H. (1980a). Useful concepts for predictive ecology. Synthese 43: 257–269. 26 Peters, R.H. (1980b). From natural history to ecology. Perspectives in Biology and 27 Medicine 23: 191–203. 27 28 Peters, R.H. (1991). A Critique for Ecology. Cambridge: Cambridge University 28 29 Press. 29 Pianka, E.R. (1992). The State of the Art in Community Ecology. Proceedings of 30 the First World Congress of Herpetology. In Herpetology: Current Research 30 31 on the Biology of Amphibians and Reptiles, ed. K. Adler, pp. 141–162. 31 32 Oxford, OH: Society for the Study of Amphibians and Reptiles. 32 Pielou, E.C. (1975). Ecological Diversity. NY: John Wiley. 33 Pielou, E.C. & Routledge, R. (1976). Salt marsh vegetation: latitudinal gradients in 33 34 the zonation patterns. Oecologia 24: 311–321. 34 35 Pirsig, R.M. (1974). Zen and the Art of Motorcycle Maintenance. NY: Morrow. 35 36 Saul, J.R. (1992). Voltaire’s Bastards. Toronto, Ontario: Penguin Books. 36 Saul, J.R. (1995). The Unconscious Civilization. Ontario: Anansi, Concord. 37 Scheiner, S. (1993). Additional thoughts on A Critique for Ecology. Bulletin of the 37 38 Ecological Society of America 74: 179–180. 38 39 Scheiner, S. (1994). Why ecologists should care about philosophy: a reply to 39 Keddy’s reply. Bulletin of the Ecological Society of America 75: 50–52. 40 40 20 P. Keddy and E. Weiher

1 Shalamov, V. (1982). Kolyma Tales. NY: W.W. Norton & Co. (trans. from Russian 1 2 by J. Glad). 2 Shipley, B. & Keddy, P.A. (1987). The individualistic and community-unit concepts 3 as falsifiable hypotheses. Vegetatio 69: 47–55. 3 4 Stone, I.F. (1989). The Trial of Socrates. NY: Anchor Books. 4 5 Tansley, A.G. & Adamson, R.S. (1925). Studies of the vegetation of the English 5 6 chalk. III. The chalk grasslands of the Hampshire–Sussex border. Journal of 6 Ecology XIII: 177–223. 7 van der Valk, A.G. (1981). Succession in wetlands: a Gleasonian approach. 7 8 Ecology 62: 688–696. 8 9 Walker, D. (1970). Direction and rate in some British post-glacial hydroseres. In 9 ed. D. Walker, & R.G. West, Studies in the Vegetational History of the British 10 Isles. pp. 117–139. Cambridge: Cambridge University Press. 10 11 Weiher, E. & Keddy, P.A. (1995). Assembly rules, null models and trait dispersion: 11 12 new questions from old patterns. Oikos 74: 159–164. 12 Weiss, P. (1965). The Persecution and Assassination of Marat as Performed by the 13 Inmates of the Asylum of Charenton under the Direction of the Marquis de 13 14 Sade. (translated by G. Skelton; verse adaptation by A. Mitchell) London: 14 15 Calder & Boyards. 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 Part I: 1 2 2 3 The search for meaningful patterns in 3 4 species assemblages 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 1 1 2 2 3 The genesis and development of 3 4 guild assembly rules 4 5 5 6 6 7 Barry J. Fox 7 8 8 9 9 10 10 11 11 12 12 13 13 Introduction 14 14 15 Diamond’s assembly rule 15 16 The idea that there were rules to govern how communities might be assem- 16 17 bled was first explored by Jared Diamond (1975) in his treatise on ‘assem- 17 18 bly of species communities’. Although there were inklings of what was to 18 19 come in earlier papers (Diamond 1973, 1974) the 1975 paper built on his 19 20 mountain of observational data on the distribution of bird species on the many 20 21 islands surrounding New Guinea to produce ‘incidence functions’. These inci- 21 22 dence functions were then used to deduce or infer the ‘role’ or ‘strategy’ of 22 23 that species, such as the supertramp strategy already described by Diamond 23 24 (1974). Diamond’s ideas were further developed (covering over 100 pages) 24 25 culminating in his assembly rules predicting which species were able to co- 25 26 exist on islands in the New Guinea archipelago, in terms of allowed and 26 27 forbidden combinations. An abbreviated version of his reasoning is shown in 27 28 Fig. 1.1, which matches the resource utilization curves of four species of 28 29 birds (dashed lines) to the availability of resources (resource production 29 30 curves – solid lines) on islands with different levels of resources. By sub- 30 31 tracting individual resource utilization curves from resource production 31 32 curves, it is possible to obtain estimates of the distribution of the remaining 32 33 resources allowing one to see which additional species could survive and 33 34 which species requirements would exceed the resource levels available. In 34 35 this way, Diamond was able to predict allowed and forbidden combinations 35 36 of species. These examples have been used here to illustrate which combi- 36 37 nations of the four species from one guild might be assembled on islands of 37 38 different sizes (one, two or four units of resources, see Fig. 1.1). Diamond 38 39 concluded that one would expect to find only species 3 on small (one unit of 39 40 resource) islands; species 2 and 4 on medium (two units of resource) islands; 40

23 24 B.J. Fox 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 Fig. 1.1. Resource utilization functions are shown for four species (1, 2, 3 and 4) from 17 18 the same guild (a), together with the resource production curves (as a thick line) for 18 the resources used by this guild, for a set of islands of increasing size which produce 19 one unit (b), two units (c), or four units (d) of resources. The right-hand side of each 19 20 part of the figure shows the curve for the resources remaining when the utilization 20 21 curves for species shown on the left-hand side are subtracted from the resource 21 production curve. (Figure adapted from Diamond, 1975: 425.) 22 22 23 23 24 and species 1, 2 and 4 on large (four units of resource) islands (for the 24 25 complete explanation see Diamond 1975: 425). This seminal paper appeared 25 26 in the volume edited by Martin Cody and Jared Diamond (1975) that was 26 27 dedicated to Robert MacArthur to commemorate the bountiful legacy he left 27 28 in so many aspects of ecology, particularly in legitimizing the fledgling 28 29 discipline of ‘community ecology’. The theory of island biogeography that 29 30 MacArthur developed with E.O. Wilson (1967) was clearly the most impor- 30 31 tant contribution to Diamond’s ideas, but the whole concept of niche theory 31 32 and the pre-eminent place afforded to interspecific competition were also 32 33 central. The whole volume (Cody & Diamond, 1975) had an electric effect 33 34 on the development of the discipline and galvanized the careers of many ecol- 34 35 ogists, including myself, and was obligatory reading because of the amount 35 36 of outstanding and interesting research it contained. However, it was the 36 37 assembly rule paper that was the focus of controversy over the next two 37 38 decades. 38 39 39 40 40 Genesis of an assembly rule 25 1 1 2 M’Closkey’s demonstration of the mechanism 2 3 Jared Diamond’s (1975) approach was empirical, the rules were deduced from 3 4 observational data and the analysis of the distributional patterns formed. The 4 5 most important part of these rules was the existence of ‘forbidden’ and 5 6 ‘allowed’ states as they were described by Diamond. The next step in the 6 7 development came not with the the continuing study of birds, but with a study 7 8 of Sonoran desert seed-eating rodents. By accumulating extensive information 8 9 on their habitat niche and food niche, Bob M’Closkey (1978) most elegantly 9 10 demonstrated how his ‘observed’ assemblages were those with minimum mea- 10 11 sured niche separation, which also maximized resource utilization. All the 11 12 other assemblages with greater niche separation he called ‘imaginary’ assem- 12 13 blages, as he did not observe them in any of his field sites. This can be seen 13 14 in Fig. 1.2 which illustrates the mean niche separation of assemblages of 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 Fig. 1.2. Mean niche separation as a function of species richness for assemblages of 32 33 desert rodents. Numbers in brackets for each point (shown as A + B) represent the 33 34 number of sites from which this assemblage was observed from (A) Saguaro National 34 35 Monument, Arizona, and (+ B) Organ Pipe Cactus National Monument. Letters with 35 each point represent the combination of species in the assemblage. Imaginary sites 36 that were not observed are shown as open circles, observed sites shown with filled 36 37 squares, and observed sites with low density are shown with filled diamonds (these 37 −1 38 sites had one species at no more than 1 ha , whereas at all other sites densities were 38 up to 10 or 11 ha−1). The species are Dipodomys merriami (M), Perognathus peni- 39 cillatus (P) {now Chaetodipus penicillatus}, P. amplus (A) and P. baileyi (B) {now 39 40 Chaetodipus baileyi}. (The data were adapted from M’Closkey, 1978.) 40 26 B.J. Fox 1 Sonoran desert seed-eating rodents comprising two, three or four species in 1 2 Saguaro National Monument, Arizona and includes data from a second area 2 3 in Organ Pipe Cactus National Monument. Two main points arise from 3 4 M’Closkey’s work: observed assemblages all have low mean niche separa- 4 5 tion, and the low diversity assemblages are precursors of the higher diversity 5 6 assemblages, in other words the sites are ‘nested’ (see Patterson, 1990; 6 7 Patterson & Brown, 1991). In all cases the greater niche separation of these 7 8 ‘imaginary’ assemblages made them subject to invasion, and the addition of 8 9 another species converted them into a new higher diversity ‘observed’ state. 9 10 For example, in a total of 13 sites in Saguaro National Monument there 10 11 was only one observed combination of two species (Dipodomys merriami (M) 11 12 and Perognathus penicillatus (P) – the latter is now Chaetodipus penicilla- 12 13 tus.). This pair of species had the smallest niche separation of any pair from 13 14 this species pool (see Fig. 1.2 first number = number of sites in Saguaro 14 15 National Monument) and was observed at three sites. No other two-species 15 16 sites were observed in Saguaro National Monument, although there were some 16 17 in Organ Pipe Cactus National Monument (see below). The three-species and 17 18 four-species sites were obtained from the addition of appropriate species, 18 19 which resulted in a marked drop in the mean niche separation. A niche 19 20 separation of over 25 units identified the only ‘imaginary’ pairs of species 20 21 (A–B) (P. amplus (A) and P. baileyi (B) {the latter is now Chaetodipus bai- 21 22 leyi}) that could not be converted to one of the two observed triplets (M–A–P 22 23 and M–B–P) by the addition of a single species. Instead, the addition of 23 24 either of the remaining two species to the A–B pair produced one of the two 24 25 imaginary triplets (A–B–P or M–A–B), both with niche separations over 25 26 10 units, whereas the addition of both produced the four species assemblage 26 27 (M–A–B–P) which was observed (see arrows Fig. 1.2). 27 28 There were 15 sites studied in Organ Pipe Cactus National Monument 28 29 (Fig. 1.2, second number near each point). Two sites with four-species assem- 29 30 blages (M–A–B–P), three sites with triplets (M–A–P) and seven sites with 30 31 pairs (six with M–P and one with M–A). Two other sites had the pair M–B 31 32 and one the pair B–P, but in both cases the density of one of the species was 32 33 1 ha−1 or less, while other species’ densities were up to 10 or 11 ha−1. 33 34 Close analysis of these data clearly reinforces the concepts of both assem- 34 35 bly and invasion, as these communities are assembled by the invasion of 35 36 species to fill the large niche separations found in imaginary assemblages. 36 37 M’Closkey’s contribution, which was confirmed by his later paper (M’Closkey, 37 38 1985), was an important advance as it provided a mechanism to explain how 38 39 the rule proposed by Diamond might work. 39 40 40 Genesis of an assembly rule 27 1 1 2 Connor and Simberloff’s null models 2 3 Controversy arose with the publication of a critique by Ed Connor and Dan 3 4 Simberloff (1979) that challenged the central importance of interspecific com- 4 5 petition and asserted that the patterns observed could equally well be attrib- 5 6 uted to chance events. This controversy continued for over a decade, and 6 7 although at different times the focus has been on methodological differences 7 8 and statistical arguments, ultimately it revolves around whether the patterns 8 9 observed are the result of deterministic or stochastic processes. These argu- 9 10 ments will not be canvassed in detail here as one of the down sides of the 10 11 controversy has been the amounts of time, energy, resources and paper that 11 12 have been consumed. Instead, the positive aspects of the controversy will be 12 13 concentrated on, those that have led to increased knowledge and understanding. 13 14 The first of these that must be recognized was the introduction of null models 14 15 (Connor & Simberloff, 1979). This was an important step, and led to the gen- 15 16 eral acceptance of the need to demonstrate conclusively, using appropriate 16 17 statistical tests, that observed patterns were significantly different from what 17 18 might be expected by chance. Randomization tests and Monte-Carlo simula- 18 19 tions have since became increasingly important and necessary components of 19 20 many ecological studies spreading to much wider fields than assembly rules. 20 21 This approach was used to investigate the relationship between niche para- 21 22 meters and species richness (Fox, 1981). For the small-mammal community 22 23 in patches of heathland habitat from one site on the eastern coast of Australia, 23 24 Monte-Carlo computer simulations were employed to demonstrate that mean 24 25 niche overlap decreased with increasing species richness, significantly more 25 26 than expected by chance. Mean niche separation was also demonstrated to 26 27 increase at a greater rate than expected by chance in habitat patches with 27 28 increasing species richness. However, tests with appropriate null hypotheses 28 29 excluding interaction between species clearly demonstrated that similar 29 30 decreases in mean niche breadth with increasing species richness as had been 30 31 observed in the field. These results were important as such decreases in niche 31 32 breadth had been assumed to result from interspecific competition. Outcomes 32 33 of the same hierarchical set of null hypotheses and field observations of niche 33 34 separations demonstrate that increasing levels of interspecific interactions lead 34 35 to increased niche separations (Table 1.1). 35 36 36 37 37 38 Colwell and Winkler’s null models for null models 38 39 While most would agree that some of the methodological arguments were 39 40 unedifying (see the extended exchanges between Jared Diamond and Dan 40 28 B.J. Fox

1 Table 1.1. Niche separation measured as the slope (± one standard error) of mean 1 2 spatial niche separation as a function of the number of species present in each 2 habitat patch (n = 13, each with from 2 to 7 species) 3 3 4 4 Null model Description and interpretation of model Niche separation 5 5 6 Equal model All species equally abundant (excludes 0.018 ± 0.028 6 7 relative abundances, habitat selection (a) 7 and species interactions) 8 Total model Relative species abundances as observed; 0.062 ± 0.027 8 9 random distribution between habitats (exclude (a, b) 9 10 habitat selection and species interactions) 10 Patch model Relative abundances and distribution 0.102 ± 0.020 11 between habitats as observed; random within (b, c) 11 12 habitat patches (exclude species interactions) 12 13 Field values Observed captures of all individuals at 0.160 ± 0.011 13 14 all trap stations in all habitat patches (c) 14 (include all species interactions) 15 15 16 Data adapted from Fox (1981). 16 17 Values with the same letter (in parentheses) do not differ significantly at P = 0.05. 17 18 18 19 Simberloff in the volume by Strong et al., 1984), there were many useful 19 20 advances stemming from these and others that have led to much more focused 20 21 testing and in some cases also more focused thinking (Strong et al. 1984; 21 22 Diamond & Case, 1986). One paper had an important bearing on the design 22 23 of appropriate null models. Robert Colwell and David Winkler (1984) used 23 24 computer simulation programs to establish artificial hierarchical arrangements 24 25 of evolving species that specifically include or exclude interspecific competitive 25 26 interactions. Species subsets from these were drawn to represent communities 26 27 on imaginary islands in imaginary archipelagos, each with explicitly designed 27 28 treatments. These communities were then tested against null hypotheses which 28 29 identified three effects that can confound studies of assembly rules: (a) the 29 30 Narcissus effect (sampling from the post-competition pool underestimates the 30 31 role of competition, since its effect is already reflected in the pool); (b) the 31 32 Icarus effect (correlations between vagility and morphology can obscure the 32 33 effects of competition in morphological comparisons of mainland and island 33 34 biotas); (c) the J.P. Morgan effect (the weaker the taxonomic constraints on 34 35 sampling, the harder it becomes to detect competition) (Colwell & Winkler, 35 36 1984). The authors also emphasized the need to consider both type I and type 36 37 II errors when selecting the appropriate null hypotheses. Grant and Abbott 37 38 (1980) had already brought attention to the dangers of these problems but it 38 39 was this study by Colwell and Winkler (1984) that conclusively demonstrated 39 40 the impact these effects can have. 40 Genesis of an assembly rule 29 1 There was also a marked increase in the use of experimental ecology that 1 2 seemed to have also been influenced by this continuing controversy, which 2 3 made clear to all researchers that it was necessary to have strong experimental 3 4 and statistical support for the inferences they drew from their results. One 4 5 piece of experimental work had a very marked impact. Michael Gilpin, 5 6 Patricia Carpenter and Mark Pomerantz (1986) were able to demonstrate in 6 7 laboratory experiments that competitive interactions between species of Droso- 7 8 phila played a most important role in determining which species were able 8 9 to form viable communities. They used 28 species and found, from the 378 9 10 possible pairwise comparisons, that there were only 46 (12%) that were able 10 11 to coexist. They then excluded the strongest and weakest competitors and 11 12 used ten species from the intermediate competitors in a further set of 30 trials, 12 13 each with these ten species introduced simultaneously, but with different 13 14 initial frequencies. After 35 weeks they found that the ten-species systems 14 15 had relaxed to smaller systems: three species in 7 trials, two species in 21 15 16 trials and one species in 2 trials, so that never more than three species were 16 17 found to coexist. There are 45 possible pairwise combinations for ten species, 17 18 but only three of these were found to coexist in the 21 trials, 18 of the trials 18 19 ending with the the same pair of species. Of the 120 possible trios from ten 19 20 species only three were observed. There was a strong concordance between 20 21 the results from the ten-species trials and the outcomes of the pairwise 21 22 comparisons. The outcomes from these laboratory experiments provided 22 23 strong support for the view that assembly rules similar to those proposed for 23 24 birds by Diamond (1975) were also operating to form these Drosophila 24 25 communities. 25 26 26 27 27 An assembly rule for guilds 28 28 29 The genesis of the guild assembly rule 29 30 The author’s doctoral thesis (Fox, 1980) examined small-mammal communities 30 31 in Myall Lakes National Park, NSW, Australia. It was influenced by the work 31 32 of Diamond (1975) and M’Closkey (1978) and it considered the assembly of 32 33 the small mammals that coexisted in the patches of habitat examined in the 33 34 analysis of niche parameters and species richness (Fox, 1981). M’Closkey’s 34 35 paper was influential because it provided an elegant demonstration of the 35 36 mechanism that structured his desert rodents, determining how they were 36 37 assembled with minimum niche separation. However, the amount of infor- 37 38 mation that had to be obtained for each species made this a daunting task and 38 39 led to the proposal of a much simpler rule in which species from the same 39 40 genus were considered, from taxonomically related groups, or from guilds 40 1 1

2 shown 2 3 3 Mus 4 Sminthopsis 4

5 ry habitats). The 5 6 6 , S.m. = 7 7 8 8 9 9 10 , with the introduced 10 11 11 12 Antechinus stuartii 12 Pseudomys

13 . (Figure adapted from Fox, 1989.) 13

14 , A.s. = 14 15 15 16 16

17 , Dasyurid and 17 R. lutreolus 18 Mus domesticus 18 19 Rattus 19 20 , R.l. = 20 , M.d. = 21 21 22 22 23 23

24 Rattus fuscipes 24 25 25 26 P. novaehollandiae 26 27 27

28 , P.n. = 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 Pseudomys gracilicaudatus 36 37 37

38 , P.g. = 38 39 39 structure and order of species packing is shown for from the groups: Fig. 1.3. Mean spatial niche overlap as a function of species richness for seven habitat patches (three wet habitats and four d murina 40 in all four dry habitats. Species abbreviations are R.f. = 40 Genesis of an assembly rule 31 1 and used the term ‘functional group’ to encompass all of these classifications. 1 2 Although these data had been observed during my doctoral studies (1974– 2 3 1980), the generality of these findings was not confirmed. The author was 3 4 particularly influenced by M’Closkey’s notions that species invade ‘imagi- 4 5 nary’ assemblages because of the large niche separation, and his suggestion 5 6 that assemblages with unused resources were vulnerable. These views were 6 7 instrumental in developing the rule. 7 8 Figure 1.3 (adapted from Fox, 1980) examines how the species were assem- 8 9 bled from taxonomic groups that had entered Australia at very different times: 9 10 dasyurid marsupials (more than 38 mybp), Conylurine rodents represented by 10 11 the genus Pseudomys (more than 4.5 mybp); and native rats, Rattus (1 mybp). 11 12 When the rule was first presented (Fox, 1985a), the different evolutionary 12 13 histories of the three groups influenced the thinking on how species might 13 14 be structured in communities considering their different diets which were, 14 15 respectively; insectivore, omnivore/granivore (depending on species), and 15 16 herbivore/omnivore (depending on species). 16 17 17 18 18 19 Statement of the guild assembly rule 19 20 The rule was conceived as a resource-based rule and made the simplifying 20 21 assumption that: the most usual distribution of resources would have roughly 21 22 similar amounts of resources available to each of the three different trophic 22 23 groups considered: insectivore, omnivore, herbivore. However, the effect of 23 24 unequal resource availability was considered (see below and Fox, 1987). Based 24 25 on the assumption of equal resources, a rule was postulated that specified the 25 26 functional group from which species should come, rather than identifying the 26 27 individual species in an assemblage. For these Australian mammal assem- 27 28 blages, the rule stated: ‘There is a much higher probability that each species 28 29 entering a community will be drawn from a different genus (or other taxo- 29 30 nomically related group of species with similar diets) until each group is 30 31 represented, before the rule repeats’ (Fox, 1987: 201). The only input required 31 32 was an a priori knowledge of how the species in the pool are divided into 32 33 functional or taxonomic groups (Fox, 1989). Assemblages for which the rule 33 34 was followed were termed ‘favored’, those for which the rule was not 34 35 followed were termed ‘unfavored’. Stated mathematically: (a) ‘favored’ states 35 36 are those for which differences between the number of species from each 36 37 functional group are never more than one; or (b) ‘unfavored’ states are those 37 38 with a difference of more than one between the number of species from each 38 39 functional group. 39 40 The assembly of species reflecting this rule do so because any assemblage 40 32 B.J. Fox 1 lacking a species from any one of these groups, but with more than one from 1 2 another group (‘unfavored’), would be subject to invasion by a species able 2 3 to capitalize on the unused resource type, which would then convert the 3 4 assemblage to a ‘favored’ state. 4 5 5 6 6 7 An example of the guild assembly rule 7 8 Further analyses of heathland (Fox, 1981, 1982) and forest succession (Fox 8 9 & McKay 1981) strengthened this view, as did an analysis of patterns in 9 10 small-mammal diversity in eastern Australian heathlands (Fox, 1983). A 10 11 review of small mammal communities in Australian temperate heathlands and 11 12 forests (Fox, 1985b) finally provided me with two separate sets of data to test 12 13 the generality of this empirical rule (Fox, 1985a; Fox, 1987, 1989). 13 14 The detailed analysis of 80 forest sites (Fox, 1987) includes a much more 14 15 complete explanation of how the simulations were carried out, together with 15 16 speculation on a possible operational mechanism for the assembly rule 16 17 (discussed below). Another important part of that paper was an illustration 17 18 of how ‘favored’ and ‘unfavored’ states were determined, including a case with 18 19 a skewed resource availability curve (Fox, 1987: Fig. 1.1). Bastow Wilson 19 20 (1989) developed an idea of guild proportionality with a method for testing 20 21 whether the proportions of species from different guilds were relatively con- 21 22 stant across sites, which he applied to a New Zealand temperate rainforest. 22 23 23 24 24 25 The effect of skewed resource availability 25 26 To investigate the effect of skewed resources, and emphasize the way in which 26 27 the guild assembly rule relates to resource availability (see also the outcome 27 28 from Morris and Knight 1996, described below), an example has been built 28 29 on from the Nevada data set analyzed by Fox and Brown (1993). In this 29 30 example the resource types represented are: arthropods (insectivores), seeds 30 31 (granivores) and plant material (herbivores), but the resources are not 31 32 uniformly distributed among resource types. The distribution of available 32 33 resources is substantially skewed toward seeds. An example has been used 33 34 with the abundance of seeds three times larger than that for plant material 34 35 available to herbivores or the arthropod resources available to insectivores 35 36 (Fig. 1.4). This is a reflection of the diversity of seed eating rodents found 36 37 in this Nevada desert and other southwestern deserts. These communities have 37 38 been recognized to contain three different functional groups, each with a 38 39 different method of foraging for seeds (Brown, 1987). 39 40 Four simplistic examples of assembled communities have been shown, two 40 Genesis of an assembly rule 33 1 are ‘favored’ states, the first has five units of available resources with one 1

2 insectivore species (I1), one herbivore species (H1) and three granivore species 2

3 (G1, G2, G3), similar to the distribution of resources (Fig. 1.4(a)). For ease of 3 4 presentation these examples are shown as rectangular profiles, but the same 4 5 outcomes would be expected with more realistic bell-shaped curves (as shown 5 6 much more elegantly for Diamond’s example in Fig. 1.1, where the curves 6 7 shown are mathematical representations of the actual resource production 7 8 curves and the outcomes of subtracting the utilization curves). In these exam- 8 9 ples the amount of resource use is proportional to the area under the curve 9 10 for each species, and the resource availability is represented by the heavy line 10 11 enclosing all of the species in each example. The distribution of available 11 12 resources is represented by the length of the heavy lines below the x-axis 12 13 with appropriate labelling. The values on the y-axis representing the total 13 14 resource available. The second example shows double the available resources 14 15 (ten units), sufficient for two complete cycles assembling two insectivore 15 16 species, two herbivore species and six granivore species (Fig. 1.4(b)), each 16 17 species’ use of resources has the same area, twice as high but half as wide 17 18 as in Fig. 1.4(a), with an implication of greater specialization. 18 19 Two examples of ‘unfavored’ states are also shown. The first ‘unfavored’ 19 20 state also has ten units of resources sufficient for a total of ten species (Fig. 20 21 1.4(c)), distributed as two units of arthropods, two units of plant material and 21 22 six units of seeds (the same as in Fig. 1.4(b)), but in this case there are two 22 23 insectivore species, four herbivore species and four granivore species, shown 23 24 by the different fill and labelling. As there would be sufficient seed resource 24 25 for six granivores (as in Fig. 1.4(b)), the community will be vulnerable to 25 26 invasion by granivorous species. In addition, there are four herbivore species 26 27 present, but with sufficient resources only for two, which would lead to an 27 28 intensification of interspecific competition between herbivores. 28 29 The second ‘unfavored’ state has seven and a half units of resource avail- 29 30 able, nominally sufficient for seven species, but distributed as one and a half 30 31 units each of arthropods and plants, with four units of seeds (Fig. 1.4(d)). 31 32 However, the species occupying the site are five granivores, two insectivores, 32 33 but no herbivores (Fig. 1.4(d)). Hence, the plant resource would be under- 33 34 utilized, the community would be subject to invasion by herbivore species, 34 35 while both insectivores and granivores exceed their available resource, again 35 36 intensifying interspecific competition. 36 37 The existence of communities such as these described in Fig. 1.4, and the 37 38 behavior demonstrated with increasing species richness, (Fig. 1.3), helped 38 39 inspire a graphical model for mammalian assembly and evolution (Fox, 1987), 39 40 that can be illustrated with the following examples. The original model was 40 34 B.J. Fox 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 Genesis of an assembly rule 35 1 conceived from observations of Australian communities from heathland 1 2 including dasyurid marsupials, insectivores that can select either wet or dry 2 3 habitats for this example, with respectively brown antechinus (Antechinus 3 4 stuartii) and common dunnart (Sminthopsis murina) from Myall Lakes, NSW 4 5 (Fox, 1982, 1983). This can be thought of as a sorting process on an ecological 5 6 timescale, and for the total community the number of species that coexist is 6 7 dependent on the adequacy and abundance of resources. 7 8 However, in lowland heath at Kentbrook Heath in Victoria, two species of 8 9 antechinus are found in wet heath habitats, the swamp antechinus (A. min- 9 10 imus), that would be considered to have evolved as a wet habitat specialist, 10 11 in addition to the brown antechinus (Braithwaite et al., 1978; Fox, 1983). At 11 12 an evolutionary timescale, this should be considered a parallel process to that 12 13 at the ecological timescale (Fox, 1987) and one that can occur only when 13 14 sufficient resources are available. 14 15 An adaptation of this model is shown in Fig. 1.5, referring specifically to 15 16 the case of granivores from the desert rodent communities of the Nevada Test 16 17 Site (see Fox & Brown, 1993). Here, granivorous species (heteromyids and 17 18 cricetids) should have some advantages in terms of digestive physiology, gut 18 19 morphology and tooth morphology that allow them to efficiently use seed in 19 20 their diet, in the same way that they will be constrained from being able to 20 21 obtain sufficient benefit from eating plant material, without the ability to 21 22 obtain energy from digesting cellulose. The different foraging behaviors 22 23 exhibited by these guilds results in effective habitat partitioning, as they select 23 24 shrubby and open macrohabitats differentially, although these habitats are 24 25 intimately mixed in the southwestern deserts. 25 26 An example of within guild evolution in this case is illustrated with the 26 27 Great Basin kangaroo rat (Dipodomys microps) that has evolved adaptations 27 28 for leaf eating, with incisors adapted for peeling salt-laden epidermis cells 28 29 from saltbush before eating mesophyll cells (Kenagy, 1973). While this is an 29 30 extreme example, that shifts Dipodomys microps from the granivore guild to 30 31 31 32 Fig. 1.4. An illustration of the possible favored and unfavored states and the effects 32 33 of skewed resources. (a) Five units of available resource (one arthropod, one plant, 33

34 three seed), with ﬁve species: one insectivore (I1), one herbivore (H1) and three grani- 34 vores (G , G , G ); favored. (b) Ten units of available resource (two arthropod, two 35 1 2 3 35 plant, six seed), with ten species: two insectivores (I , I ), two herbivores (H , H ) and 36 1 2 1 2 36 six granivores (G1, G2, G3, G4, G5, G6); favored. (c) Ten units of available resource 37 (two arthropod, two plant, six seed), with ten species: two insectivores (I1, I2), four 37 herbivores (H , H , H , H ) and four granivores (G , G , G , G ); unfavored. (d) Seven 38 1 2 3 4 1 2 3 4 38 units of available resource (one and a half arthropod, one and a half plant, four seed), 39 39 with seven species: two insectivores (I1, I2), no herbivore and ﬁve granivores (G1, G2, 40 G3, G4, G5); unfavored. 40 1 1 2 2

3 ges, now 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 applied to the Nevada Test Site data. (Figure adapted from Fox, 1987.) 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39

40 Fig. 1.5. A model that speculates on a possible operational mechanism for the guild assembly rule, based Australian assembla 40 Genesis of an assembly rule 37 1 the folivore guild, it does demonstrate how on an evolutionary timescale it 1 2 is possible to produce morphological or physiological change to allow further 2 3 partitioning of the food resource, which might even in some cases produce 3 4 further speciation (Fig. 1.5). Clearly, for such specialization to occur the food 4 5 resource needs to be abundant, predictable and hence dependable. In this way 5 6 the amount of resource available will directly affect the number of groups, 6 7 subgroups and species forming the pool of species available and the number 7 8 and type of species able to coexist in any one place, in the way described for 8 9 ‘favored’ as opposed to ‘unfavored’ states. 9 10 10 11 11 12 General applicability of the guild assembly rule 12 13 In order to consider the generality of the guild assembly rule, the brief findings 13 14 from a number of tests of the rule will be summarized, noting the biogeo- 14 15 graphical area and the taxonomic groups involved. All of the studies in this 15 16 list have been published and the wide range of biogeographical regions in which 16 17 this rule is applicable greatly strengthens my confidence in the operation of 17 18 the rule. 18 19 19 20 20 21 Australia: rodents and marsupials in forest and heath 21 22 Two studies have been carried out for southeastern Australia that included 22 23 marsupials (insectivores from the family Dasyuridae); native mice (granivores 23 24 and omnivores from the old endemics, the conylurine rodents); and the more 24 25 recently arrived native rats (herbivores and omnivores from the genus Rattus). 25 26 The analyses were based on these three functional groups or guilds. One study 26 27 demonstrated highly significant departure from random assembly (P < 0.001) 27 28 for 80 small mammal assemblages sampled from eucalypt forest sites extend- 28 29 ing over a latitudinal range from 27° S to 43° S along the east coast of Aus- 29 30 tralia (Fox, 1987). The other study, over a similar latitudinal range, also 30 31 showed that the mammalian assemblages assessed from 52 coastal heathland 31 32 sites had significantly more favored states and significantly fewer unfavored 32 33 states than expected from Monte-Carlo simulations (P < 0.01), thus rejecting 33 34 the null hypothesis that they had been assembled randomly (Fox, 1989). 34 35 35 36 36 37 North America: shrews in NE forests, rodents in SW deserts and boreal 37 38 forests 38 39 On the North American continent three sets of data have been analyzed. One 39 40 study demonstrated significant departure from random assembly (P < 0.001) 40 38 B.J. Fox 1 for the soricid communities from 43 temperate forest sites in the New England 1 2 area of northeastern USA (Fox & Kirkland, 1992). In this study the shrews 2 3 were divided into three guilds based on body size. 3 4 The second study was from the southwestern deserts of USA, and com- 4 5 prised three parts (Fox & Brown, 1993). First, data were collated from over 5 6 202 sites from the Chihuahuan, Great Basin, Mojave and Sonoran deserts 6 7 with a pool of 28 species of granivorous desert rodents in three different 7 8 foraging guilds, bipedal and quadrupedal heteromyids and cricetids (Brown 8 9 & Kurzius, 1987). There were significantly more favored states observed than 9 10 the number expected from 10000 Monte-Carlo random simulations, and the 10 11 null hypothesis of random assembly was rejected (P < 0.01). Another data 11 12 set from the nearby Nevada Test site (Jorgensen & Hayward, 1965) was also 12 13 analyzed, this time at two scales: (a) three granivore guilds and (b) five rodent 13 14 guilds. Ten species from the three granivore guilds formed the pool from 14 15 which the assemblages that occupied 115 transects were drawn. Significantly 15 16 more favored states were observed than expected from random computer sim- 16 17 ulations, thus rejecting (P < 0.001) the null hypothesis (Fox & Brown, 1993). 17 18 When herbivorous and insectivorous species were added to form five guilds, 18 19 the results were similar, 74 favored states were observed while random sim- 19 20 ulations produced 41.6 ± 4.2, again the null hypothesis was rejected (P < 20 21 0.001). 21 22 The third study was conducted over an area of 100 km2 in boreal forest in 22 23 northern Ontario, Canada (Morris & Knight, 1996). Three herbivorous voles, 23 24 two diurnal chipmunks, two jumping mice and one cricetid made up the 24 25 species pool of four feeding guilds from which communities of rodents were 25 26 drawn. Favored states were observed for 67% of the assemblies when the 26 27 expected value was 28.4%, so that random assembly was rejected (P = 0.0008, 27 28 cumulative binomial test). 28 29 29 30 30 31 South America: rodents at the Valdivian rainforest–Patagonian steppe 31 32 interface 32 33 Data were collected from 31 communities in southern Chile (46° S) along 33 34 the boundary between Valdivian temperate rainforest and Patagonian steppe, 34 35 and 12 species of native sigmodontine species of rodents formed the species 35 36 pool for the analyzes (Kelt et al., 1995). Functional groups from four trophic 36 37 categories were recognized with two omnivores, six herbivores, two fungivore– 37 38 insectivores, and two granivores. Their analyzes enabled them to reject the 38 39 null hypothesis for random assembly (P = 0.001) and also demonstrate that 39 40 there was a significant level of interspecific competition with a statistical 40 Genesis of an assembly rule 39 1 power of 93–97%. Further details of these analyzes will be provided in the 1 2 section on assessment of interspecific competition, and the same techniques 2 3 are applied by Kelt and Brown (this volume). 3 4 4 5 5 6 Madagascar: lemurs in evergreen rainforests and dry forests 6 7 Jurg Ganzhorn (1997) has recently demonstrated that communities of arbo- 7 8 real lemurs from evergreen rainforest habitats in Madagascar obeyed the guild 8 9 assembly rule when compared to 10 000 Monte-Carlo simulations for neutral 9 10 models (P = 0.001). From a pool of 20 species (6 folivores, 8 frugivores 10 11 and 6 omnivores) communities were assembled with from 3 to 13 species, 11 12 indicating that the rule had operated through up to four, and in some cases 12 13 five cycles of species additions. The high degree of concordance with the rule 13 14 was quite impressive, as examples from other biogeographical areas and with 14 15 other taxonomic groups have generally only proceeded through two or in 15 16 some cases three cycles at most. The fit for dry forests was less good, although 16 17 still significant (P = 0.029), and this was attributed to recent natural and 17 18 anthropogenic changes over the 2000 years of human occupation that have 18 19 been very apparent in dry forests, but not in the undisturbed evergreen rain- 19 20 forests. Ganzhorn (1997: 534) also reported that ‘[this guild assembly rule] 20 21 is also consistent with the composition of European small mammal commu- 21 22 nities’ and cites Schröpfer (1990) as his authority. 22 23 23 24 24 25 Summary of the applicability of the guild assembly rule 25 26 This guild assembly rule has been statistically tested against Monte-Carlo 26 27 computer simulations for seven data sets of small mammal assemblages from 27 28 different biogeographical regions on four continents: Australia, North Amer- 28 29 ica, South America and Madagascar. In all cases the observed assemblages 29 30 have been shown to differ significantly from the random simulations, and to 30 31 be consistent with the composition of small mammal communities on a fifth 31 32 continent, Europe. These tests have examined a wide range of mammalian 32 33 taxa: dasyurid marsupials, lemurs, shrews, voles, heteromyid, cricetid and 33 34 sigmodontine rodents. The guilds analyzed have been based on: taxonomic 34 35 relatedeness, body size, diet, and foraging behavior. The analyzes have been 35 36 between diet guilds (insectivore, granivore and herbivore); within a diet guild 36 37 (insectivores, granivores); and finally for a combination of between and within 37 38 guild analysis (the 5-guild Nevada data set in Fox & Brown, 1993). Spatial 38 39 scales have ranged from 3500 km2 Nevada Test Site, 640000 km2 in four 39 40 southwestern deserts to the largest covering 16° of latitude along the eastern 40 40 B.J. Fox 1 Australian coast. The tests have also included a wide range of biomes from 1 2 deserts to boreal forests in North America, moist evergreen forests in north- 2 3 eastern USA, eucalypt forest and heathlands in Australia, Valdivean rain- 3 4 forest and Patagonian steppe in South America, dry forests and evergreen 4 5 rainforests in Madagascar. This great breadth of biogeographical regions, 5 6 taxonomic groups, vegetation types and spatial scales most surely attests to 6 7 the generality of this guild assembly rule. 7 8 In rejecting all these null hypotheses of random assembly one must accept 8 9 an alternative hypothesis, and the guild assembly rule, based on resource 9 10 availabilty, resource partitioning, and interspecific competition offers the best 10 11 alternative hypothesis proposed. This view is substantially strengthened by 11 12 the recent derivation of the rule from Tilman’s consumer-resource model 12 13 (Morris & Knight, 1996) described below. 13 14 14 15 15 16 Methodological considerations 16 17 The two decades since Diamond (1975) published his species assembly rule 17 18 paper have been marked by critiques and rebuttals of the methods used for 18 19 analyses of assembly rules, as I mentioned in the introduction. Many of these 19 20 have focused on the appropriateness of null hypotheses, selection of appro- 20 21 priate species pools, marginal totals constraints, initial assumptions and the 21 22 simulation techniques used in the analyses. The guild assembly rule has been 22 23 the target of similar critiques, and will no doubt continue to be subjected to 23 24 them. While these arguments are often targeted at statistical methods, they 24 25 seem to basically arise from differences of opinion on what should represent 25 26 the initial conditions or assumptions for the null hypotheses used in these 26 27 analyses, and this has been most eloquently demonstrated by Colwell and 27 28 Winkler (1984). 28 29 The major debate on the guild assembly rule comes down to whether one 29 30 should maintain the distribution of species richness across sites, which reflects 30 31 the availability of resources at those sites; or whether one should retain the 31 32 distribution of each species’ frequency of occurrence across sites, which 32 33 reflects the interactions determining which sites each species can occupy and 33 34 which species coexist at each site. The most sensible course would seem to 34 35 be to agree to differ on these points and get on with research activities that 35 36 advance our knowledge more. This is a better alternative than spending time 36 37 discussing what would seem to be relatively trivial points on which are the 37 38 most appropriate statistical techniques to use, as they are unlikely to be 38 39 resolved by these debates when the real difference is a philosophical one 39 40 about what assumptions are made in constructing null hypotheses. 40 Genesis of an assembly rule 41 1 1 2 Wilson (1995a), Fox and Brown (1995) and Wilson (1995b) 2 3 Bastow Wilson (1995a) challenged the methods used by Fox and Brown 3 4 (1993) to analyze the Nevada data set claiming they would give significant 4 5 results with randomized data. A reply to Wilson’s (1995a) critique was written 5 6 (Fox & Brown, 1995) which reaffirmed the validity of the guild assembly 6 7 rule and the methods used in the analyzes and will not be further canvassed 7 8 here. However, Wilson (1995b) wrote a second critique claiming that the 8 9 random data sets used by Fox and Brown (1995) were not random. As the 9 10 editor of Oikos would not allow a reply to this second critique, a brief 10 11 comment should be made here. 11 12 Bastow Wilson (1995b) is correct in indicating that the method used by 12 13 Fox and Brown (1995) to generate the random data set was flawed. Random 13 14 data sets had been generated by randomly allocating 115 sites to the 36 cells 14 15 in the matrix (0, 0, 0) to (2, 2, 3) (see Fox & Brown, 1995). As Wilson points 15 16 out this effectively considered all 36 cells equiprobable, which was incorrect 16 17 and hence needed to be corrected. 17 18 To rectify this, 200 new random data sets were generated, in each case 18 19 considering separately each of the 115 sites to form a matrix for the nine 19 20 species encountered, with each species identified and in the same order. For 20 21 each site, the number of species observed was retained but the presence or 21 22 absence of each species was randomly allocated across all nine species, with 22 23 no consideration of which guilds were represented, until the observed rich- 23 24 ness for the site was achieved. In this way the row constraints of the matrix 24 25 were preserved but not the the column constraints, whereas Wilson main- 25 26 tained the column constraints but not the row constraints. 26 27 For each random matrix generated, the observed guild structure (3, 2, 4 27 28 species in each of the three guilds) was reimposed across the nine species to 28 29 calculate the status of each simulated site as either favored or unfavored. Each 29 30 random data set was then analyzed against a null hypothesis with the species 30 31 pool of 3, 3, 5 species in each guild for the Nevada Test Site as used by Fox 31 32 and Brown (1993, 1995), so as to include the two additional species caught 32 33 across the whole Nevada test area, but not on any of the set of 115 transects 33 34 used in the analysis. The matrix of possible outcomes has also been con- 34 35 strained to the maximum observed number of species from each guild (2, 2, 35 36 3). In these analyses rather than use the Monte-Carlo simulation previously 36 37 used, an analytical approach developed by R.J. Luo (R.J. Luo & B.J. Fox, 37 38 unpublished data), has been used and the exact probability from the Bino- 38 39 mial test calculated. The result for three examples, all with the same matrix 39 40 constraint (2, 2, 3), but different species pools are shown in Figs. 1.6(a)–(c). 40 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 Fig. 1.6. Frequency histograms for the probability distribution that the null hypothe- 37 38 sis would be accepted when 200 randomly generated data sets for 115 sites were ana- 38 lyzed. Each random data set was generated (see text) to correct the flaw in the data 39 sets used by Fox and Brown (1995). (a) to (c) are all constrained to have maxima of 39 40 2, 2, 3 species in each guild and match the figures presented by Fox and Brown (1995) 40 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 Fig. 1.6. (continued) 37 38 with species pools of a 3, 3, 5; b 3, 2, 6; c 2, 2, 7. (d) to ( f ) all have the same species 38 pool (3, 3, 5) but show the effects of less restrictive matrix constraints, to have a max- 39 imum number of species in each guild of: (d) 2, 2, 4; and (e) 3, 2, 4; but also the 39 40 effects of a more restrictive matrix constraint ( f ) 2, 2, 2. 40 44 B.J. Fox 1 Figure 1.6(a) illustrates a relatively even probability distribution as one might 1 2 expect with the observed pool (3, 3, 5) for such random data. Probability 2 3 distributions for two more extremely skewed species pools for 11 species are 3 4 shown in Fig. 1.6(b) (3, 2, 6) and Fig. 1.6(c) (2, 2, 7), to match those illus- 4 5 trated by Fox and Brown (1995: Fig. 1). The conclusion reached is qualita- 5 6 tively little different to that reported by Fox and Brown (1995). So, while 6 7 Wilson (1995b) correctly identified a flaw in the construction of random data 7 8 sets used by Fox and Brown (1995), after correct generation, when they were 8 9 reanalyzed, they still clearly demonstrate that random data do produce 9 10 relatively even probability distributions, using the same null hypothesis as 10 11 Fox and Brown (1995), and that these may become biased with extreme 11 12 species pools. 12 13 To further illustrate the conservative nature of the null hypotheses used by 13 14 Fox and Brown (1993, 1995) another two examples have been considered 14 15 (Fig. 1.6(d), (e)), with the observed species pool (3, 3, 5), but the matrix size 15 16 constraint has been relaxed to accommodate maxima in the simulation: of 2, 16 17 2, 4, allowing up to eight species to occupy 45 cells (Fig. 1.6(d)); and of 3, 17 18 2, 4, allowing up to 9 species to occupy 60 cells (Fig. 1.6(e)), although in 18 19 fact none of the 115 sites had maxima of more than 2, 2, 3, allowing up to 19 20 seven species to be present in 36 cells. Relaxing these constraints means that 20 21 more cells (9 and 24, respectively) are added to the analysis, even though 21 22 they were never observed. However, they will have to be given a random 22 23 expectation from the null hypothesis and as more of these cells will repre- 23 24 sent unfavored states there will be an increase in the probability of rejecting 24 25 the null hypothesis. This is the reason why matrix size constraints have been 25 26 used in the past to exclude these unrealistic cells, to make the test more 26 27 conservative. As predicted, in each case the results are shifted to the left, thus 27 28 confirming that the use of these constraints on the maximum size of the matrix 28 29 does lead to more conservative tests. To emphasize this, an illustration of the 29 30 effect of constraining the matrix to maxima of 2, 2, 2, has also been included 30 31 allowing up to six species in 27 cells (Fig. 1.6(f)). There is a marked shift to 31 32 the right in the probability distribution, making it virtually impossible to reject 32 33 the null hypothesis for these random data. It should be pointed out that 113 33 34 of the sites could meet this constraint (see Fox & Brown, 1993). 34 35 One should note here that, for the observed matrix of 115 sites and nine 35 36 species, using column constraints to preserve the species frequencies means 36 37 that a great deal more of the actual structure will be smuggled into the null 37 38 hypothesis, and this is the basis of the Narcissus effect described by Colwell 38 39 and Winkler (1984). On the other hand, preserving the distribution of species 39 40 richness values for each site (row constraints) smuggles much less informa- 40 Genesis of an assembly rule 45 1 tion into the null hypothesis to be used, and is directly related to the resources 1 2 available at each site. 2 3 3 4 4 5 Stone, Dayan and Simberloff (1996) 5 6 Lewi Stone, Tamar Dayan and Dan Simberloff (1996:1013) reanalyzed the 6 7 data sets examined by Fox and Brown (1993), and they concluded: 7 8 8 Our analysis failed to find evidence that interspecific competition or deterministic 9 assembly rules shaped local community composition ...... the only possible structure 9 10 noted in our study was that attributed to three or four widespread species ...... a 10 11 study of spatial distribution alone provides little evidence for competition between 11 12 these species ...... Because ecological field and experimental studies (for review see 12 13 Brown, 1987; Kotler & Brown, 1988) strongly imply the role of interspecific com- 13 petition among granivorous desert rodents, it seems conceivable that such a process 14 14 might be better detected with more discriminating statistical techniques. 15 15 16 This is not the appropriate place for a detailed rebuttal of these comments 16 17 but it should be pointed out that the methodology used by Stone et al. (1996 17 18 and this volume) incorporates the Narcissus effect, as the authors recognize 18 19 in part. The fact that some species are widespread while others have more 19 20 restricted distributions cannot be decoupled from the relative competitive 20 21 abilities of these species. In turn, the species competitive abilities may or may 21 22 not have influenced their distributions, so that use of this distributional infor- 22 23 mation may still have the potential to introduce competitive effects into their 23 24 null model. 24 25 It should also be pointed out that, while the last part of the above quote 25 26 recognizes reality, interspecific competition plays an important role in these 26 27 communities, the authors have ignored the existence of two important stud- 27 28 ies which address the point they make (Kelt et al, 1995; Morris & Knight, 28 29 1996). Both of these are dealt with in some detail in the next section and the 29 30 reader is directed to the chapter in this volume which specifically applies 30 31 Kelt’s methodology to questions raised by Stone et al, (1996). Doug Kelt and 31 32 Jim Brown (this volume) have explicitly dealt with the geographic ranges 32 33 question by calculating separate species pools for each trapping site used in 33 34 the Fox and Brown (1993) analyses. Detailed discussion of these analyses 34 35 will be left to their chapter, but it should be pointed out that they very con- 35 36 vincingly reject the null hypothesis of no interaction (P < 0.0005), accepting 36 37 the alternative hypothesis of a maximum likelihood estimate for the mean 37 38 strength of the observed negative association of 0.33, with a power of 94% 38 39 (probability of being correct). 39 40 40 46 B.J. Fox 1 1 Further development of the guild assembly rule 2 2 3 Morris’s derivation from the MacArthur–Tilman consumer-resource 3 4 model 4 5 Doug Morris demonstrated that the guild assembly rule is a probabilistic 5 6 consequence of adding guild structure to models of consumer-resource com- 6 7 petition (Morris & Knight, 1996). The graphical models of consumer-resource 7 8 dynamics were developed initially by Robert MacArthur (1972) and then 8 9 much more fully by David Tilman (1982). The author’s guild assembly rule 9 10 had been developed in an empirical manner, as shown above, from observations 10 11 of the structure of small mammal communities occupying different patches 11 12 of habitat (Fox, 1981; 1987). A potential theoretical mechanism had been set 12 13 out, with speculation on how the rule operated (Fox, 1987). However, it was 13 14 not able to provide any theoretical derivation (although the mechanism was 14 15 linked to resource availability, with a strong implication that competition for 15 16 these resources played a major role). It was also suggested that the mecha- 16 17 nism was an extension of the niche compression hypothesis that MacArthur 17 18 and Wilson (1967) applied to the optimal use of patchy habitats. 18 19 To emphasize the main points made by Morris and Knight (1996), an 19 20 attempt has been made to devise a simplified, if not simplistic, illustration of 20 21 Tilman’s model that retains the most essential components. On a graph of the 21 22 amounts of two resources (L and N), three species each from two guilds that 22 23 specialize on each of these resources have been represented. (Fig. 1.7a). The 23 24 ratio of resource N to resource L can be shown as a vector that represents 24 25 the combined resource consumption rate for each species. The direction of 25 26 these consumption vectors will more closely parallel the direction of the 26 27 resource axis as the species become more specialized. Species from guild L 27 28 will therefore use more of resource L. For multi-guild systems one would 28 29 expect that evolutionary pressure will result in consumption vectors for 29 30 species in a guild tending to be further from the 45° line of equal resource 30 31 use and closer to the axis for the resource used by that guild. This situation 31 32 has been shown for the two-guild system (Fig. 1.7(a) with species con- 32 33 sumption vectors labeled with letters for guild L and labeled with numbers 33 34 for guild N. 34 35 For simplicity the zero net growth isocline (ZNGI) for each species has 35 36 been omitted, that would be displayed as a pair of rectangular axes repre- 36 37 senting the minimum amounts of each essential resource required for each 37 38 species’ survival (see Tilman, 1982; Morris & Knight, 1996). To concentrate 38 39 on the slope of each consumption vector the supply points (the levels of 39 40 resources available in the absence of consumption) from which each con- 40 Genesis of an assembly rule 47 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 Fig. 1.7. A simplified version of how Morris and Knight (1996) applied Tilman’s 18 19 (1982) models of consumer resource competition. (a) Species 1, 2, and 3 from guild 19 N and species P, Q, and R from guild L are represented by their consumption vec- 20 tors (lines of equal resource ratios for the two resources N and L). The area within 20 21 which supply points are available is bounded by a dashed arc. For simplicity, the zero 21 22 net growth isocline (ZNGI) has been omitted for each species, that would be displayed 22 as a pair of rectangular axes representing the minimum amounts of each essential 23 resource required for each species survival (for complete details, see Tilman, 1982, 23 24 Morris & Knight 1996). (b) Regions of stable coexistence for within-guild pairs of 24 25 species are defined by the angle between the consumption vectors, out to the dotted 25 arc. The area of each region shown can then be equated to the probability of finding 26 appropriate supply points allowing the coexistence of the two species indicated. 26 27 (c) Regions of stable coexistence for between-guild pairs of species are defined by 27 28 the angle between the consumption vectors, out to the dashed arc. The area of each 28 29 region shown can then be equated to the probability of finding appropriate supply 29 points allowing the coexistence of the two species indicated. This leads to the con- 30 clusion that the probability of finding pairs of species from the same guild coexisting 30 31 is much smaller than for pairs of species from different guilds. 31 32 32 33 33 34 sumption vector would emanate. However, a dashed arc has been used to 34 35 enclose a possible range of available supply points. Regions of stable co- 35 36 existence for pairs of species would be defined by the angle between their 36 37 consumption vectors, taking the area as extending from the dotted arc to the 37 38 point of intersection of the vectors. The point of intersection of the vectors 38 39 would be determined by the relative positions and intersection of the appro- 39 40 priate ZNGIs. For simplicity, those points have been omitted in Fig. 1.7.(a), 40 48 B.J. Fox 1 but their actual positions do not alter the conclusions that follow here (see 1 2 Morris & Knight, 1996). The area of each region of stable coexistence can 2 3 then be equated to the probability of finding appropriate supply points allow- 3 4 ing the coexistence of the two species indicated. 4 5 Representations of the relative areas of the regions that would allow coex- 5 6 istence for each of the 15 possible species pairs are shown for the six within- 6 7 guild pairs (Fig. 1.7(b)) and for the nine between-guild pairs (Fig. 1.7(c)). In 7 8 all cases the areas of the regions of coexistence are markedly smaller for 8 9 intra-guild pairs than for inter-guild pairs. This leads to the conclusion that 9 10 the probability of finding pairs of species from the same guild coexisting is 10 11 much smaller than for pairs of species from different guilds. This is the main 11 12 message that Morris and Knight (1996) convey, although they make a number 12 13 of other important points that will not be followed up here. 13 14 14 15 15 16 Interpretation of the Nevada data in terms of a consumer-resource model 16 17 A reanalysis of the data set from the Nevada test site that was used by Fox 17 18 and Brown (1993) provides an interesting illustration of, and further support 18 19 for, Morris and Knight (1996) linking the guild assembly rule to the resource 19 20 competition model. A regression technique, was developed separately by 20 21 Schoener (1974) and Pimm (Crowell & Pimm, 1976). The technique used the 21 22 density of one species as the dependent variable and focused on the use of 22 23 multiple regression to remove all of the variance associated with independent 23 24 habitat variables before allowing the densities of other potentially competing 24 25 species to enter the multiple regression equations as independent variables. 25 26 The regression coefficients for species variables were then considered as the 26 27 coefficients of competition between species. Rosenzweig et al. (1985) ques- 27 28 tioned the technique and demonstrated some inconsistencies in analyses, and 28 29 the technique fell into disuse after that. This regression technique has recently 29 30 been revived by Fox and Luo (1996). They confirmed the artefact identified 30 31 by Rosenzweig et al. (1985), but demonstrated how to overcome the problem 31 32 by making use of standardized data, thus providing a useful method to extract 32 33 competition coefficients from census data and habitat measurements. 33 34 Rodent density data were available from 95 trapping plots in four different 34 35 habitats: Larrea-Franseria (39 plots), Coleogyne (27), Grayia-Lycium (15) 35 36 and Salsola (14). Ten species were found on enough plots to be included in 36 37 these regression analyses. Two quadrupedal heteromyids (QH) Perognathus 37 38 formosus and Pg. longimembris; two bipedal heteromyids (BH) Dipodomys 38 39 merriami and D. ordii; three quadrupedal non-heteromyids (QN) Peromyscus 39 40 maniculatus, Pe. crinitis and Ammospermophilus leucurus; two folivorous 40 Genesis of an assembly rule 49 1 herbivores (F) Neotoma lepida and Dipodomys microps; and one insectivore 1 2 Onychomys torridus. The interaction coefficients resulting from this multiple 2 3 regression technique are shown in Table 1.2. There were six significant inter- 3 4 action coefficients shown for within-guild species pairs (from a total of 12 4 5 possible), all significant values were negative with a mean value of −1.820 5 6 ± 0.0295 (1 s.e.). For between-guild pairs, in the 10 equations there were 18 6 7 significant interaction coefficients (from the 78 possible), all significant val- 7 8 ues were positive (7 of these with P < 0.005) with a mean value of +2.494 8 9 ± 0.0215 (1s.e.). This is many more significant results than might have been 9 10 expected by chance. Even more important is the distribution of the results. 10 11 The strong negative outcome within-guild and the linked positive outcome 11 12 for between-guild pairs clearly demonstrates the importance of interspecific 12 13 competition. This pattern confirms the expectation from the operation of a 13 14 guild assembly rule derived from Tilman’s consumer-resource model (Morris 14 15 & Knight, 1996), as shown in Fig. 1.7. 15 16 16 17 17 18 Kelt’s inclusion of habitat component assessment of interspecific 18 19 competition 19 20 Doug Kelt, Mark Taper and Peter Meserve (1995), produced an important 20 21 development for guild assembly rules when they introduced specific alterna- 21 22 tive hypotheses to the null hypothesis that enabled them to make an estimate 22 23 of the power of their test. In addition to the total species pool that is usually 23 24 used, they introduced a geographic species pool that took account of differ- 24 25 ent geographic distributions and a habitat species pool that took account of 25 26 different habitats that might be encountered by species in any one area. While 26 27 some account had already been taken of how each species geographic distri- 27 28 bution influenced the overall or total species pool (Fox, 1987, 1989; Fox & 28 29 Brown, 1993), there had been no successful attempt to incorporate geographic 29 30 effects or to introduce habitat effects at individual sites. The inclusion of a 30 31 habitat component improved resolution for determining species pools because 31 32 habitat selection by species will influence which species may be present at 32 33 any site. In other words, trapping sites in the same area, but in different habi- 33 34 tats may have different species pools because of habitat selection. In addi- 34 35 tion to including geographic and habitat components these simulations were 35 36 run a number of times, each with a different level of interspecific competi- 36 37 tion (Θ) incorporated, ranging from Θ = −1.0 to Θ = +1.0. The values of Θ 37 38 tested then represent a range of alternative hypotheses. They were able to 38 39 reject the null hypothesis for the total species pool analysis (P = 0.001) and 39 40 demonstrate that there was a significant level of interspecific competition 40 1 1 2 2 P value 3 3 Overall 2 4 equation 4 5 5 6 6

7 Habitat 7 variables R 8 8 9 9 ns Signiﬁcant 0.169 0.0007 10 hown as dependent variables down the 10 < 0.0001). Independent variables for

11 P 11 + 12 ** 0.174* Signiﬁcant 0.348 <.0001 12

13 0.271 13 0.260** ns 0.415 0.0001 − 14 < 0.01, *** 14 P + 15 ** 0.255** Signiﬁcant 0.477 <.0001 15

16 0.271 16 − < 0.05, **

17 P 17 + ns ns ns ns Signiﬁcant 0.137 0.0012 18 ns 0.273** 0.178* ns ns 0.343 <.0001 18 0.156

19 < 0.1, * 19 P + +

20 + 20 21 21 0.168 22 − 22 23 23 > 0.1 and the interaction coefficients are shown in bold typeface with same significance 0.137 − 24 P 24 25 25 26 nsns ns ns ns ns 0.199* ns ns ns ns 0.292** ns Significant 0.257 <0.0001 ns Significant26 0.060 0.058

27 + 27

28 ns ns ns 0.464*** ns ns ns ns Signiﬁcant 0.495 <0.0001 28 0.182 29 29

30 + 30

31 0.100 31 − 32 32 ns ns ns ns nsns 0.189 ns ns ns ns ns ns ns ns Signiﬁcant 0.093 0.011 ns ns ns ns ns 0.352*** ns ns 0.346*** ns ns ns ns ns 0.171 ns ns ns ns ns 0.188 ns 0.228* ns ns ns ns ns 0.172

33 0.145 33 − Pg. for Pg. lon D. mer D. ord Pe. man Pe. cri A. leu N. lep D. mic O. tor 34 0.411*** ns ns ns 34 35 35 36 36

37 Interaction coefficients among species in local community using multiple regression technique originally developed by Crowell 37 variable Independent variables Dependent Pe. maniculatus D. merriami D. ordii O. torridus N. lepida Pe. crinitis D. microps Pg. formosus Pg. longimembris 38 A. leucurus 38 39 39 QN Flvr BH left-hand column. Interaction coefficients are marked according to their significance ( Table 1.2. and Pimm (1976) modified by Fox Luo (1996) Guild Species Inst Habitat heterogeneity was accounted for using habitat variables. Species across the row are independent variables species s QH 40 species from the same guild were forced into equations when 0.2 > symbols. 40 Genesis of an assembly rule 51 1 (with the best estimate Θ = 0.60). The statistical power of this rejection was 1 2 93–97%. For the geographic species pool analysis, they were able to reject 2 3 the null hypothesis (P = 0.023) and demonstrate that there was a significant 3 4 level of interspecific competition (Θ = 0.45) and that the statistical power of 4 5 the rejection was 56–71%. 5 6 Doug Kelt’s further development of the assembly rule confirms that inter- 6 7 specific competition is a major factor in the operation of the guild assembly 7 8 rule as described (Fox, 1987). Field removal experiments have also corrob- 8 9 orated the empirical pattern analyses (Fox & Pople, 1984; Fox & Gullick, 9 10 1989; Higgs & Fox, 1993; Thompson & Fox, 1993). Doug Kelt’s analysis 10 11 also provides support for Doug Morris’s demonstration that the guild assem- 11 12 bly rule is a probabilistic consequence of adding guild structure to models of 12 13 consumer-resource competition (Morris & Knight, 1996). Doug Kelt and Jim 13 14 Brown have explicitly applied Kelt’s methods (see chapter in this volume) 14 15 by calculating separate species pools for each trapping site used in the Fox 15 16 and Brown (1993) analyses. With this technique they use spatial distributions 16 17 to provide convincing evidence for competition between the species in these 17 18 communities, for which there is abundant experimental evidence (e.g., see 18 19 Munger & Brown, 1981; Brown & Munger, 1985; Heske et al., 1994; Valone 19 20 & Brown, 1995; and references in the Kelt and Brown chapter, this volume). 20 21 21 22 22 23 What future for guild assembly rules? 23 24 One point that should be emphasized here, is that the guild assembly rule has 24 25 both deterministic and probabilistic components, as was emphasized by Fox 25 26 and Brown (1993). Replacing the terms ‘favored’ states and ‘unfavored’ states 26 27 with terms like ‘high probability’ states and ‘low probability’ states would 27 28 reinforce this point. The message in this is that all states are possible, but 28 29 some have a higher probability of occurrence than others (see Fig. 1.7). This 29 30 seems to better represent the reality of the processes at work, on both tem- 30 31 poral and spatial scales. In the extreme case a site might be switched from 31 32 ‘high probability’ to a ‘low probability’ state by the chance addition of a 32 33 single individual, which may only persist for a short period before it dies or 33 34 moves on. On the other hand, it might be joined by other individuals of the 34 35 same species and also from another additional species, either way it could 35 36 again switch the site to a ‘high probability’ site. Hence, the site passes through 36 37 states of differing probability for different lengths of time that reflect the prob- 37 38 ability of finding those states. This is the stochastic component overlain on 38 39 an otherwise deterministic process. 39 40 Another thing to be emphasized in the guild assembly rule is the importance 40 52 B.J. Fox 1 of process. This empirically based rule grew out of an attempt to understand 1 2 the mechanisms that operate in assembling communities. The guild assembly 2 3 rule has provided a powerful tool to aid this understanding, particularly with 3 4 relation to consumer–resource competition. One of the rule’s strengths has 4 5 been its simplicity, all that is required is knowledge of which species belong 5 6 to which functional groups or guilds, the distribution of species richness across 6 7 sites and the appropriate species pool for each guild. One of the drawbacks 7 8 is that individual species in the communities are not identified, but this is the 8 9 cost of not being required to have a mass of detailed information on each 9 10 species. One point that should be emphasized: it is the principle of resource 10 11 allocation that is involved. This is the generalization of the rule, not a require- 11 12 ment that there be a similar range of guilds in all situations. This point 12 13 becomes clear when considering the way in which the question of allocation 13 14 of resources among species and guilds has been addressed by Morris and 14 15 Knight (1996), as illustrated in Fig. 1.7. 15 16 The further development of the rule by Doug Kelt provides a most inter- 16 17 esting way to proceed. By incorporating specific alternative hypotheses we 17 18 are provided with a powerful means of testing mechanisms that might be 18 19 involved. This should be limited only by the ingenuity of researchers to devise 19 20 appropriate null models and appropriate means to test them in the field. One 20 21 interesting possibility here is to incorporate the detailed approach taken by 21 22 Bob M’Closkey (1978) with that used by Doug Kelt (Kelt et al., 1995; Kelt 22 23 and Brown, this volume). If such an approach were fruitful it would certainly 23 24 provide us with an excellent opportunity to further advance our understanding 24 25 of the ways in which communities are structured. 25 26 The other area that might prove fruitful is the incorporation of the regres- 26 27 sion technique with standardized data. As has been briefly illustrated here 27 28 (Table 1.2), this technique has the ability to easily provide information on 28 29 the competition between species if additional habitat information is available. 29 30 This technique provides another interesting means of testing the degree of 30 31 interaction in assemblages to offer an independent method to that used in the 31 32 Kelt technique. 32 33 33 34 34 35 Conclusions 35 36 Twenty-five years after Robert MacArthur’s death, we are still working on 36 37 the wealth of ideas that he left as his legacy to ecology, and geographical 37 38 ecology in particular. The theory of island biogeography provided a frame- 38 39 work for the study of species richness on islands which led to Jared Diamond’s 39 40 exposition of his species assembly rules first published in the 1975 memorial 40 Genesis of an assembly rule 53 1 volume to Robert MacArthur. Diamond’s rule arose from his study of birds 1 2 from the archipelagoes near New Guinea, but these were soon followed by 2 3 an application to small mammals by Bob M’Closkey in 1978 that elegantly 3 4 demonstrated a mechanism for how the rule might operate. Despite a decade 4 5 of sometimes acrimonious debate the assembly rule concept has survived and 5 6 prospered to move on to a further stage. 6 7 In 1985 an idea for an assembly rule was presented that was based on func- 7 8 tional groups rather than individual species, these functional groups were 8 9 equivalent to guilds. The main advantage to this rule was its simplicity as it 9 10 required only a knowledge of which guilds were present and the pool of species 10 11 available for each guild, from which the communities were assembled. The 11 12 rule was tested against rigorously devised conservative null hypotheses for a 12 13 wide range of taxonomic groups, in many different habitats over a range of 13 14 spatial scales on four continents. 14 15 Doug Kelt provided a very interesting further development for the guild 15 16 assembly rule, by introducing specific alternative hypotheses to the null 16 17 hypothesis to make an estimate of the power of the test. This was a very 17 18 elegant test of the degree to which interspecific competition was involved as a 18 19 mechanism in the operation of the rule. Incorporating habitat and creating a 19 20 separate species pool for every site is a powerful extension of these analyses. 20 21 Doug Morris demonstrated that the author’s guild assembly rule is a proba- 21 22 bilistic consequence of adding guild structure to models of consumer-resource 22 23 competition, providing the necessary theoretical underpinning that had been 23 24 missing from his rule. 24 25 The most recent development has involved an application of the 25 26 Schoener–Pimm regression technique that has been recently revived by Barry 26 27 Fox and Roger Luo, using standardized data. When applied to the Nevada 27 28 communities, that already demonstrated adherence to the author’s guild 28 29 assembly rule, all six significant interaction coefficients found for within-guild 29 30 species pairs (from 12 possible) were negative. For between-guild pairs there 30 31 were 18 significant interaction coefficients (from 78 possible), all were 31 32 positive. This interesting result supports interspecific interaction as one part 32 33 of the mechanism for the guild assembly rule, and provided important field 33 34 confirmation of the theoretical derivation made by Doug Morris. 34 35 The guild assembly rule should be seen as a combination of stochastic 35 36 and deterministic components. Use of the terms ‘high probability’ states and 36 37 ‘low probability’ states in place of ‘favored’ and ‘unfavored’ should reinforce 37 38 this. The way forward offers at least three paths: (a) use of specific alternative 38 39 null hypotheses as active tests of mechanisms that might structure commu- 39 40 nities; (b) the use of habitat data and species densities with the standardized 40 54 B.J. Fox 1 regression technique to demonstrate interactive effects; (c) incorporation of 1 2 the detailed species specific information as used by Bob M’Closkey. These 2 3 all offer exciting opportunities to further develop the guild assembly rule in 3 4 the future. 4 5 5 6 6 Acknowledgments 7 7 8 I have already acknowledged the intellectual debt I owe to the people whose 8 9 ideas most influenced me in developing the guild assembly rule. However, I 9 10 must mention that Bob M’Closkey’s contribution was extremely influential, 10 11 not only in converting me from a solid-state physicist to an ecologist but also 11 12 in focusing my interest on why communities are structured as they are. 12 13 I would also like to offer my thanks to Rodger Luo for many useful and 13 14 interesting discussions and ideas about assembly rules, and for completing 14 15 the additional analyses on the Nevada data for me, particularly with his devel- 15 16 opment of the analytical solutions and construction of random data sets. I 16 17 would also like to thank Marilyn Fox, Vaughan Monamy and Doug Morris 17 18 for discussions about ideas and suggestions on reading the manuscript. I thank 18 19 Paul Keddy and Evan Weiher for the opportunity to write this chapter, and 19 20 I also appreciated the constructive critical comment from Evan Weiher. 20 21 21 22 References 22 23 Braithwaite, R.W., Cockburn, A., & Lee, A.K. (1978). Resource partitioning by 23 24 small mammals in lowland heath communities of southeastern Australia. 24 25 Australian Journal of Ecology 3: 423–445. 25 26 Brown, J.H. (1987). Variation in desert rodent guilds: patterns, processes and 26 scales. In Organization of Communities: Past and Present, ed. J.H.R. Gee & 27 P.S. Giller, pp. 185–203. Oxford: Blackwell. 27 28 Brown, J.H. & Munger, J.C. (1985). Experimental manipulation of a desert rodent 28 29 community: food addition and species removal. Ecology 66: 1545–1563. 29 Brown, J.H. & Kurzius, M. (1987). Composition of desert rodent faunas: 30 composition of co-existing species. Annals Zool. Fenn 24: 227–237. 30 31 Cody, M.L. & Diamond, J.M. (eds.) (1975). Ecology and Evolution of 31 32 Communities. Cambridge, MA: Harvard University Press. 32 Colwell, R.K. & Winkler, D.W. (1984). A null model for null models in 33 biogeography. In Ecological Communities: Conceptual Issues and the 33 34 Evidence, ed. D.R Strong, D Simberloff, L.G. Abele & A.B. Thistle, pp. 34 35 344–359. Princeton, NJ: Princeton University Press. 35 36 Connor E.H. & Simberloff, D. (1979). The assembly of species communities: 36 chance or competition? Ecology 60: 1132–1140. 37 Crowell, K.L. & Pimm, S.L. (1976). Competition and niche shifts of mice 37 38 introduced onto small islands. Oikos 27: 251–258. 38 39 Diamond, J.M. (1973). Distributional ecology of New Guinea birds. Science 179: 39 759–769. 40 40 Genesis of an assembly rule 55

1 Diamond, J.M. (1974). Colonization of exploded volcanic islands by birds: the 1 2 supertramp strategy. Science 184: 803–806. 2 Diamond, J.M. (1975). Assembly of species communities. In Ecology and 3 Evolution of Communities, ed. M.L. Cody & J.M. Diamond, pp. 342–444. 3 4 MA: Harvard University Press. 4 5 Diamond, J.M. & Case, T.J. (1986). Community Ecology. New York: Harper & 5 6 Row. 6 Fox, B.J. (1980). The ecology of a small mammal community: secondary 7 succession, niche dynamics, habitat partitioning, community structure and 7 8 species diversity. PhD Thesis, Macquarie University, Sydney. 8 9 Fox, B.J. (1981). Niche parameters and species richness. Ecology 62: 1415–1425. 9 Fox, B.J. (1982). Fire and mammalian secondary succession in an Australian 10 coastal heath. Ecology 63: 1332–1341. 10 11 Fox, B.J. (1983). Mammal species diversity in Australian heathlands: the 11 12 importance of pyric succession and habitat diversity. In Mediterranean-Type 12 Ecosystems: The Role of Nutrients, ed. F.J. Kruger, D.T. Mitchell, & J.U.M. 13 Jarvis, pp. 473–489. Berlin: Springer-Verlag. 13 14 Fox, B.J. (1985a). Small-mammal community pattern in Australian heathland. In 14 15 Fourth International Theriological Congress Abstracts, ed. W.A. Fuller, N.T. 15 16 Nietfeld, & M.A. Harris, Edmonton: ITC4. 16 Fox, B.J. (1985b). Small mammal communities in Australian temperate heathlands 17 and forests. Australian Mammalogy 8: 153–158. 17 18 Fox, B.J. (1987). Species assembly and the evolution of community structure. 18 19 Evolutionary Ecology 1: 201–213. 19 Fox, B.J. (1989). Small mammal community pattern in Australian heathland: a 20 taxonomic rule for species assembly. In Patterns in the Structure of 20 21 Mammalian Communities, ed. D.W. Morris, Z. Abramsky, B.J. Fox & Willig, 21 22 M. pp. 91–103. Lubbock, TX: Texas Tech Museum Special Publication 22 Series. 23 Fox, B.J. & McKay, G.M. (1981). Small mammal response to pyric successional 23 24 changes in eucalypt forest. Australian Journal of Ecology 6: 29–42. 24 25 Fox, B.J. & Pople, A.R. (1984). Experimental confirmation of interspecific 25 26 competition between native and introduced mice. Australian Journal of 26 Ecology 9: 323–334. 27 Fox, B.J. & Gullick, G. (1989). Interspecific competition between mice: a 27 28 reciprocal field manipulation experiment. Australian Journal of Ecology 14: 28 29 357–366. 29 Fox, B.J. & Kirkland, G.L., Jr. (1992). North American soricid communities follow 30 Australian small mammal assembly rule. Journal of Mammalogy 73: 491–503. 30 31 Fox, B.J. & Brown J.H. (1993). Assembly rules for functional groups in North 31 32 American desert rodent communities. Oikos 67: 358–370. 32 Fox, B.J. & Brown J.H. (1995). Reaffirming the validity of the assembly rule for 33 functional groups or guilds. Oikos 73: 125–32. 33 34 Fox, B.J. & Luo, J. (1996). Estimating competition coefficients in the field: a 34 35 validity test of the regression method. Oikos 77: 291–300. 35 36 Ganzhorn, J.U. (1997). Test of Fox’s assembly rule for functional groups in lemur 36 communities in Madagascar. Journal Zoology London 241: 533–542. 37 Gilpin, M.E., Carpenter, P. & Pomerantz, M.J. (1986). The assembly of a 37 38 laboratory community: multispecies competition in Drosophila. In Community 38 39 Ecology, ed. J.M. Diamond & T.J. Case, pp. 23–40. New York: Harper & 39 Row. 40 40 56 B.J. Fox

1 Grant, P.R. & Abbott, I. (1980). Interspecific competition, island biogeography and 1 2 null hypotheses. Evolution 34: 332–341. 2 Heske, E.J., Brown, J.H., & Mistry, S. (1994). Long-term experimental study of a 3 Chihuahuan desert rodent community: 13 years of competition. Ecology 75: 3 4 438–445. 4 5 Higgs, P. & Fox, B.J. (1993). Interspecific competition: a mechanism for rodent 5 6 succession after fire in wet heathland. Australian Journal Ecology 18: 6 193–201. 7 Jorgensen, C.D. & Hayward, C.L. (1965). Mammals of the Nevada test site. 7 8 Brigham Young University Science Bulletin Biology Series 6(3): 1–81. 8 9 Kelt, J.A., Taper, M.L., & Meserve, P.L. (1995). Assessing the impact of 9 competition on the assembly of communities: a case study using small 10 mammals. Ecology 76: 1283–1296. 10 11 Kenagy, G.J. (1973). Adaptations for leaf eating in the Great Basin Kangaroo Rat 11 12 Dipodomys microps. Oecologia 12: 383–412. 12 Kotler, B.P. & Brown, J.S. (1988). Environmental heterogeneity and the 13 coexistence of desert rodents. Annual Reviews of Ecology Systems 19: 13 14 281–307. 14 15 MacArthur, R.H. (1972). Geographical Ecology. New York: Harper & Row. 15 16 MacArthur, R.H. & Wilson, E.O. (1967). The Theory of Island Biogeography. 16 Princeton, NJ: Princeton University Press. 17 M’Closkey, R.T. (1978). Niche separation and assembly in four species of Sonoran 17 18 Desert rodents. American Naturalist 112: 683–694. 18 19 M’Closkey, R.T. (1985). Species pools and combinations of heteromyid rodents. 19 Journal of Mammalogy. 66: 132–134. 20 Morris, D.W., Abramsky, Z., Fox, B.J. & Willig, M. (ed.) (1989). Patterns in the 20 21 Structure of Mammalian Communities. Lubbock, TX: Texas Tech Museum 21 22 Special Publ. Series. 22 Morris, D.W. & Knight, T.W. (1996). Can consumer-resource dynamics explain 23 favored versus unfavored states of species assembly. American Naturalist 147: 23 24 558–575. 24 25 Munger, J.C. & Brown, J.H. (1981). Competition in desert rodents: an experiment 25 26 with semipermeable exclosures. Science 211: 510–512. 26 Patterson, B.D. (1990). On the temporal development of nested subset patterns of 27 species composition. Oikos 59: 330–342. 27 28 Patterson, B.D. & Brown, J.H. (1991). Regionally nested patterns of species 28 29 composition in granivorous rodent assemblages. Journal of Biogeography 18: 29 395–402. 30 Rosenzweig, M.L., Abramsky, Z., Kotler, B. & Mitchell, W. (1985). Can 30 31 interaction coefficients be determined from census data? Oecologia 66: 31 32 194–198. 32 Schoener, T.W. (1974). Competition and the form of the habitat shift. Theoretical 33 Population Biology 6: 265–307. 33 34 Schröpfer, R. (1990). The structure of European small mammal communities. 34 35 Zoologische Jahrbüch Systematik 117: 355–367. 35 36 Stone, L., Dayan, T. & Simberloff, D. (1996). Community-wide patterns unmasked: 36 the importance of species’ differing geographical ranges. American Naturalist 37 148: 997–1015. 37 38 Strong, D.R., Simberloff, D., Abele, L.G. & Thistle, A.B. (1984). Ecological 38 39 Communities: Conceptual issues and the Evidence. Princeton, NJ: Princeton 39 University Press. 40 40 Genesis of an assembly rule 57

1 Thompson, P.T. & Fox, B.J. (1993). Asymmetric competition in Australian 1 2 heathland rodents: a reciprocal removal experiment demonstrating the influence 2 of size-class structure. Oikos 67: 264–278. 3 Tilman, D. (1982). Resource Competition and Community Structure. Princeton, NJ, 3 4 Princeton University Press. 4 5 Valone, T.J. & Brown J.H. (1995). Effects of competition, colonization, and 5 6 extinction on rodent species diversity. Science 267: 880–883. 6 Wilson, J.B. (1989). A null model of guild proportionality, applied to stratification 7 of a New Zealand temperate rainforest. Oecologia 80: 263–267. 7 8 Wilson, J.B. (1995a). Null models for assembly rules: the Jack Horner effect is 8 9 more insidious than the Narcissus effect. Oikos 74: 543–544. 9 Wilson, J.B. (1995b). Fox and Brown’s ‘random data sets’ are not random. Oikos 10 72: 139–143. 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 2 1 2 2 3 Ruling out a community assembly rule: 3 4 the method of favored states 4 5 5 6 6 7 Daniel Simberloff, Lewi Stone, and Tamar Dayan 7 8 8 9 9 10 10 11 11 12 12 13 13 14 Introduction 14 15 Arguments have raged for two decades over the relative importance of inter- 15 16 specific competition and individual species’ responses to the physical environ- 16 17 ment in determining community composition and about the nature of evidence 17 18 on this matter (references in Stone, 1996). This debate is one of the most 18 19 noteworthy features of modern community ecology. Several adherents of the 19 20 view that competition plays a key role have sought support in patterns 20 21 detectable in local communities, patterns they feel reflect a governing role 21 22 for competition in the assembly of communities. The focus here is on one of 22 23 the most recent such assembly rules, that of Fox and coauthors (1985, 1987, 23 24 1989; Fox & Kirkland, 1992; Fox & Brown, 1993), who have modified the 24 25 assembly rule of Diamond (1975). Applying a null hypothesis approach (cf. 25 26 Connor & Simberloff, 1979; Gilpin & Diamond, 1982), Fox proposed that a 26 27 competitive assembly rule, described below, determines how sequential addition 27 28 constructs communities. His analyses of several data sets seem to confirm 28 29 this assembly rule and to imply for these communities that competition largely 29 30 governs their composition. 30 31 Fox and Brown (1993) applied a variant of this rule to local communities 31 32 of North American granivorous desert rodents, finding strong confirmation 32 33 of Fox’s assembly rule (1987). Many people have studied these rodent 33 34 communities (references in Brown, 1987; Kotler & Brown, 1988), and it is 34 35 evident that interspecific competition is occurring. Thus, if co-occurrence 35 36 patterns in any regional communities manifest competition, this should be 36 37 one such community. However, the fact that competition occurs between these 37 38 species need not mean that competition governs local community composi- 38 39 tion. So it would be particularly interesting to find an assembly rule that 39 40 showed that it does. Stone et al. (1996) reexamined Fox’s assembly rule using 40

58 Ruling out an assembly rule 59 1 the same data studied by Fox and Brown (1993). Their results are summa- 1 2 rized and extended here. A similar study by Fox and Kirkland (1992) of a 2 3 community of six shrew species coexisting over a wide region of the north- 3 4 eastern United States and eastern Canada is also reexamined. 4 5 5 6 6 7 Functional groups and favored states 7 8 Fox (1987, 1989) suggested that, if competition determines which species can 8 9 enter a developing local community, one can foresee the outcome of com- 9 10 munity assembly in terms of ‘functional groups’: sets of ecologically similar 10 11 species, like guilds. Fox’s strategy suggests that local community composi- 11 12 tion can be predicted in terms of numbers of species in different functional 12 13 groups without extensive knowledge of every species, so long as how to assign 13 14 species to functional groups is known. Species have often been assigned to 14 15 functional groups and guilds based solely on systematics, although in princi- 15 16 ple both categories are supposed to be determined by the types of food eaten, 16 17 and guilds by the way the food is gathered as well (Simberloff & Dayan, 17 18 1991). So, dividing a biota into functional groups is no trivial matter, though 18 19 the functional group assignments for the rodents studied by Fox and Brown 19 20 (1993) appear to have an ecological basis (see below), even though they 20 21 follow taxonomic lines. 21 22 Fox’s assembly rule is simple: ‘There is a much higher probability that 22 23 each species entering a [local] community will be drawn from a different 23 24 functional group until each group is represented, before the cycle repeats’ 24 25 (Fox 1987, p. 201). Functional groups should be equally represented in local 25 26 communities derived from a larger regional pool. The rule is based on inter- 26 27 specific competition, primarily for food (Fox & Brown, 1995): if some func- 27 28 tional group becomes disproportionately represented in a local community, 28 29 competition lowers the probability that the next species to colonize will belong 29 30 to that group and raises the probability that it will belong to one of the other 30 31 groups. 31 32 A local community is in a ‘favored state’ whenever all pairs of functional 32 33 groups have the same number of species or differ by at most one (Fox & 33 34 Brown, 1993). For example, consider the numbers of species in three differ- 34 35 ent functional groups at a hypothetical site. Favored states include (2,2,2), 35 36 (2,3,2), and (1,1,0): functional groups are evently represented. By contrast, a 36 37 local community is in an ‘unfavored state’ if the number of species in any 37 38 pair of functional groups differs by more than one. The configurations (1,0,2), 38 39 (2,1,4), and (4,3,1) are unfavored states. 39 40 If local communities are assembled according to this rule, more of them 40 60 D. Simberloff et al. 1 should be in favored states than if the species already present do not affect 1 2 which other species will be added next. Fox (1987; Fox & Brown 1993) thus 2 3 proposed a test of this assembly rule that uses the null hypothesis that species 3 4 enter local communities independently of their functional groups. He simu- 4 5 lated the assembly of each local community by randomly drawing species 5 6 from the regional pool, without replacement, and irrespective of which other 6 7 species have already been drawn. The simulated local community is complete 7 8 when the same number of species are present as are in its real analog. 8 9 Fox and his coworkers have used several algorithms to assign probabilities 9 10 that the next species drawn in a random local community comes from a par- 10 11 ticular functional group (references in Stone et al., 1996). Fox and Kirkland 11 12 (1992) made the initial probability for each functional group proportional to 12 13 the number of species in that group, but sampling was with replacement. The 13 14 focus here is on the most recent algorithm, that of Fox and Brown (1993), 14 15 who made the initial probability for each functional group proportional to the 15 16 number of species in the group and sampled without replacement. Kelt et al. 16 17 (1995), examining assembly of South American small mammal communities, 17 18 used a similar method. From the null distribution of favored states among the 18 19 simulated communities, one can test whether the observed data set differs 19 20 from expected, as would be expected if Fox’s rule is at work. 20 21 21 22 22 23 The data 23 24 Fox and Brown (1993) examined data for two regional North American mam- 24 25 mal communities: 25 26 26 (a) Data of Jorgensen and Hayward (1965) on mammals of the Nevada Test 27 27 Site, from surveys in Nevada for the ‘evaluation of the effects nuclear 28 28 weapons testing, peaceful use of nuclear weapons, and nuclear warfare 29 29 may have on mammal populations’ (Foreword, p. 1). They trapped small 30 30 mammals for six years at 115 sites in an area of 3500 km2. If Fox’s assem- 31 31 bly rule works, this finding would be remarkable, as it would mean that 32 32 the assembly rule overrides the effects of a series of nuclear explosions 33 33 near these sites four years before the survey (Stone et al., 1996)! 34 34 (b) Data of Brown and Kurzius (1987) on 28 rodent species at 202 sites across 35 35 640000 km2 of the arid Southwest. 36 36 37 Granivorous rodents have been variously partitioned into guilds (Simberloff 37 38 & Dayan, 1991). Generally, heteromyids are separated from cricetids because 38 39 the former have large external cheek pouches and different foraging behav- 39 40 ior. The quadrupedal heteromyids (pocket mice) are usually separated from 40 Ruling out an assembly rule 61 1 the bipedal ones (kangaroo rats and mice) both because they move differently 1 2 and because the former tend to forage under shrubs, while the latter forage 2 3 mostly in open areas between widely spaced plants. Stone et al. (1996) 3 4 assigned the granivorous species of each data set to functional groups exactly 4 5 as Fox and Brown (1993) did. For the Nevada site, the nine granivorous 5 6 rodents fell into three functional groups: (a) bipedal heteromyids = BH (two 6 7 species); (b) quadrupedal heteromyids = QH (three species); (c) quadrupedal 7 8 non-heteromyids = QN (four species). Fox and Brown (1993, 1995) included 8 9 in the species pool two species found at none of the sites. Stone et al. (1996) 9 10 did not; the results of Stone et al. (1996) are unlikely to be changed by whether 10 11 or not these species are seen as part of the pool. For a new analysis detailed 11 12 below, they are included, so the numbers of species in the three functional 12 13 groups are 3,3, and 5, respectively. The 28 southwestern granivorous species 13 14 were likewise partitioned into three groups: (a) bipedal heteromyids (7 species); 14 15 (b) quadrupedal heteromyids (11 species); (c) cricetids (10 species). 15 16 Fox and Kirkland (1992) used data from Rickens (1974) plus Kirkland’s 16 17 unpublished surveys of six shrew species at 43 sites in Nova Scotia, New 17 18 Brunswick, New York, Pennsylvania, and West Virginia. They assigned these 18 19 species to three functional groups of two species each, based on body mass: 19 20 large, medium, and small. We have adopted these assignments. 20 21 21 22 22 23 The original results 23 24 Of the 115 local communities at the Nevada Test Site, Fox and Brown (1993) 24 25 found 92 to be in favored states, while Stone et al. (1996) found 93. Fox and 25 26 Brown (1993), from their random-draw simulations, found an expected 26 27 number of favored states of 62.5, and the probability of a number as high as 27 28 the observed value to be P < 0.001. 28 29 For the 202 local communities in the southwestern data set, Fox and Brown 29 30 (1993) found 128 favored states, while Stone et al. (1996) found 126. Fox 30 31 and Brown (1993) calculated an expectation of 111.3 favored states and, using 31 32 their random-draw simulation, the probability of a value as high as that 32 33 observed to be P < 0.01. 33 34 Of the 43 local shrew communities, Fox and Kirkland (1992) reported that 34 35 12 were in unfavored states and 31 in favored states. Their own data (Fox & 35 36 Kirkland, 1992, Table 2.1) show 11 unfavored and 32 favored ones. Their 36 37 random draw simulations yielded an expectation of 20.8 unfavored states and 37 38 22.2 favored states, and they found the observations to differ from expecta- 38 39 tions at P = 0.02 by Fisher’s exact test. This result would be even more 39 40 extreme for the correct number of favored states. 40 62 D. Simberloff et al.

1 Table 2.1. Observed numbers of favored states and those expected for species ran- 1 2 domly assigned to functional groups for desert rodents of the Nevada Test Site and 2 American Southwest (after Stone et al. 1996) and shrews of the northeastern US 3 and eastern Canada 3 4 4 5 Observed # Expected # Standard % tail of 5 6 Data set favored states favored states deviation observation 6 7 7 Nevada 93 64.9 22.6 11.8 8 Southwest 126 105.6 18.0 13.8 8 9 Shrews 32 29.0 7.3 46.7 9 10 10 11 11 12 12 13 The matrix randomization method 13 14 Beginning with Connor and Simberloff (1979), many people have used pres- 14 15 ence–absence matrices to deal with data of this sort (references in Gotelli & 15 16 Graves, 1996). Columns represent sites and rows represent species. A ‘1’ in 16 17 the (i,j)th entry means species i is present at site j, while a ‘0’ means it is 17 18 absent. The number of species on the jth site is the jth column sum, and the 18 19 number of sites occupied by the ith species is the ith row sum. 19 20 The simulated communites are a sample from the set of all matrices having 20 21 the dimensions of the observed matrix. However, to preserve some biologi- 21 22 cal realism, we follow Connor and Simberloff (1979) and require the species 22 23 and sites of each simulated matrix to satisfy constraints: (a) A simulated site 23 24 contains as many species as its analog does in nature. Thus, some simulated 24 25 sites support more species than others do, as in the real data. These differ- 25 26 ences reflect such ecological patterns as the species–area effect. (b) A simu- 26 27 lated species occupies as many sites as its analog does in nature. Some species 27 28 are thus more widespread than others. In nature, this variation is caused by 28 29 species differences such as dispersal abilities and physical factor tolerances; 29 30 it could also reflect competitive ability. An important component of these 30 31 simulations is that species’ geographic ranges are not preserved. Real ranges 31 32 are usually continuous, while in simulated matrices species ranges tend to be 32 33 very discontinuous and can include areas far outside the real ranges (Stone 33 34 et al., 1996). 34 35 Matrices drawn equiprobably from the set of all matrices with the same 35 36 row and column sums thus represent ‘random’ communities. Because we have 36 37 incorporated these constraints to preserve realism, the ‘null model’ may also 37 38 no longer be null with respect to some hypotheses. For example, because inter- 38 39 specific competition can affect the species richness of a site or the distribution 39 40 of some species, fixed row and column sums may embed the effects of com- 40 Ruling out an assembly rule 63 1 petition in each ‘random’ community. So this test, as used by Stone et al. 1 2 (1996), is a narrow one; it is not about whether interspecific competition 2 3 occurs, but about whether, for whatever reasons, the entry of species into 3 4 local communities accords with Fox’s functional group model. 4 5 The random communities of Fox and Brown (1993) are not assembled by 5 6 filling in such a binary matrix, but, if they were, they would have column 6 7 constraints only – each site has the number of species observed on its analog 7 8 in nature. However, because species were not randomly drawn (only functional 8 9 groups), species are not restricted to particular sites or to a particular number 9 10 of sites. Thus, to the extent that species’ ranges and site occupancies in nature 10 11 reflect biological traits as noted above, the assembly algorithm is unrealistic. 11 12 Implicitly, all species in a functional group are the same. 12 13 It is often unrealistic to simulate a regional community as if all species can 13 14 occupy all sites. For example, in the southwestern data set, Dipodomys 14 15 merriami and Peromyscus maniculatus are very widespread, occupying about 15 16 half the 202 possible sites. But almost half of the species occupy fewer than 16 17 ten sites. Each functional group has at least one widespread species and sev- 17 18 eral that are much more narrowly distributed. It seems unrealistic to assume 18 19 that all species in a functional group are interchangeable, because the observed 19 20 data show otherwise. Possible reasons why different species occupy different 20 21 numbers of sites include differing dispersal abilities, differing tolerances of 21 22 physical factors, differing histories, and differing competitive abilities. The 22 23 test of Fox and Brown (1993) does not address this issue. An unrealistic 23 24 model need not be useless for testing a particular hypothesis; this would 24 25 depend on the particular nature of the lack of realism and the hypothesis. 25 26 Stone et al. (1996) found that, in this instance, the lack of realism could have 26 27 been crucial, as we discussed below. 27 28 Reanalyzing the Nevada test site and southwestern data sets by random- 28 29 izing matrices, Stone et al. (1996) fixed row and column sums. Although a 29 30 constrained row sum means that each simulated species occupies the same 30 31 number of sites as in nature, it does not limit species’ ranges, as noted above, 31 32 because each species can still occur on any site. For each data set, Stone et 32 33 al. (1996) produced 10000 random matrices and computed the number of 33 34 favored states, as well as the standard deviation and percent tail of the obser- 34 35 vation. The number of favored states in the Nevada matrix is in the 28.3% 35 36 tail, while that in the Southwest is in the 17.8% tail. So, by this method, the 36 37 null hypothesis of independent colonization would not be rejected whereas 37 38 Fox and Brown (1993, 1995) find significantly more favored states than 38 39 expected in both data sets. 39 40 Clearly, if these results are to be taken as a guide, a null hypothesis that 40 64 D. Simberloff et al. 1 species colonize sites independently of one another cannot be rejected. By 1 2 contrast, Fox’s method shows that the local communities of both data sets 2 3 include significantly more favored states than expected, just as predicted by 3 4 his assembly rule (Fox & Brown, 1993). Wilson (1995) independently pub- 4 5 lished a very similar study to ours, using matrix randomization, and came to 5 6 the same conclusion: the Fox null model will nearly always find an observed 6 7 data set significantly different from expectation. He attributes this finding to 7 8 the fact that species frequencies are rarely equal in nature. He argues that 8 9 randomly constructed data sets would show an excess of favored states, so 9 10 long as species frequencies are not all the same. He called this the ‘Jack 10 11 Horner effect’ because the obvious is demonstrated: in this case, that species 11 12 frequencies generally differ. 12 13 13 14 14 15 Is the Fox and Brown (1995) test biased? 15 16 Rejecting Wilson’s criticism of Fox and Brown (1993), Fox and Brown (1995) 16 17 sought to show that, contrary to Wilson’s assertions, randomly constructed 17 18 data sets would not show a surplus of favored states relative to expectation. 18 19 However, they chose highly non-random data sets to do this. 19 20 Their method is exemplified by their treatment of the Nevada test site data. 20 21 Observing that the maximum numbers of species in the three functional 21 22 groups BH, QH, and QN in any local community are 2,2, and 3, respec- 22 23 tively, they constructed 36 different combinations to represent the various 23 24 possible numbers of species in the functional groups in the 115 local com- 24 25 munities. That is, in their null assignment of sets of species to the local 25 26 communities, each community could have 0,1, or 2 species in the first func- 26 27 tional group, 0,1, or 2 species in the second group, and 0,1,2 or 3 species in 27 28 the third group. Thus, irrespective of species’ identities, there are 3 × 3 × 4 28 29 = 36 possible communities, and each of the 115 local communities was 29 30 uniform randomly assigned to one of these 36 combinations. Each such 30 31 random assignment of the 115 local communities constituted a random data 31 32 set. They then asked how likely it is that the Monte-Carlo method of Fox and 32 33 Brown (1993) described above would show more favored states than expected 33 34 for such a random assignment. In fact, they found the opposite – usually the 34 35 null hypothesis was accepted. 35 36 By uniform randomly assigning the communities to the 36 possible com- 36 37 binations, they were using Bose–Einstein statistics, which are appropriate only 37 38 if the species are indistinguishable (Feller, 1968), and they are not. Even if 38 39 species within a functional group are ecologically equivalent, as assumed by 39 40 Fox and Brown (1993), they are still distinguishable, in that one can be dis- 40 Ruling out an assembly rule 65 1 tinguished from the other (for example, by morphology). Thus, a random 1 2 assignment of local communities would make the probability that a commu- 2 3 nity falls in each of the 36 cells proportional to the number of possible ways 3 4 that cell can arise (Maxwell–Boltzmann statistics; Feller, 1968). Thus, for 4 5 example, the cell (0,0,0) can arise only one way, the cell (0,0,1) five ways 5 6 (because there are five species in group QN), and the cell (2,2,3) 90 ways 6

7 (because there are three species each in the other two groups, 2C3 = 3, 3C5 = 7 8 10, and 3 × 3 × 10 = 90). In fact, there are 1274 distinguishable arrange- 8 9 ments for the 36 cells. Favored states are represented by 16 cells, and these 9 10 16 cells contain 681 distinct arrangements. Thus, 681/1274 = 53.5% of all 10 11 possible arrangements are favored states, but, on average, Fox and Brown 11 12 (1995) would find only 16/36 = 44.4% of the random communities to be 12 13 favored states in any of their randomized data sets. 13 14 It is not known exactly how this fact would affect the expected number 14 15 of favored states when the Monte-Carlo algorithm of Fox and Brown (1993) 15 16 is applied using these randomized data sets as the observed data, but one 16 17 might expect it to bias the test towards finding that the expected number of 17 18 favored states is at least as large as the observed. Consider: if one happened 18 19 to assign the 115 random local communities to the 36 cells such that none 19 20 fell in favored states, one would subsequently calculate the expected number 20 21 of favored states, based on these particular observed ones, to be much greater. 21 22 This is because, in the Monte-Carlo procedure, one still randomly draws 22 23 species from groups that, in the species pool, have sizes of 3,3, and 5 species, 23 24 respectively. A truly random, Maxwell–Boltzmann assignment of the 115 24 25 local communities to the 36 cells might not show fewer random states in a 25 26 randomized set of local communities than would be expected by the Monte- 26 27 Carlo procedure, but the Bose–Einstein assignment method of Fox and Brown 27 28 (1995) is certainly biased rather than conservative. 28 29 In fact, the maximum numbers of species in the three functional groups 29 30 BH, QH, and QN are not (2,2,3), as in Fox and Brown (1995). Rather, they 30 31 are either (3,2,3) or (3,2,2), depending on whether one counts one QN species 31 32 collected too infrequently for Jorgensen and Hayward (1965) to calculate den- 32 33 sities. Apparently Fox and Brown (1995) included this species, as this inclu- 33 34 sion is necessary to find 5 QN species in the regional pool. With maximum 34 35 numbers of species in the local communities as (3,2,3), there are 4 × 3 × 4 35 36 = 48 cells, of which 18 representing favored states contain 741 of the 1456 36 37 arrangements and 30 representing unfavored states contain 715 arrangements. 37 38 So 741/1456 = 50.9% of the possible arrangements are favored states, but 38 39 only 18/48 = 37.5% of the random communities will be favored states in the 39 40 randomized data sets drawn by the Bose–Einstein method. Again, when Fox 40 66 D. Simberloff et al. 1 and Brown (1995) subsequently perform their Monte-Carlo procedure on each 1 2 randomized data set, there will be a bias towards accepting the null hypoth- 2 3 esis that the observed (random) data set has no more favored states than 3 4 expected. 4 5 5 6 6 7 Species ranges and co-occurrence patterns 7 8 The discrepancy between the results of Stone et al. (1996) and those of Fox 8 9 and Brown (1993) led Stone et al. (1996) to ask whether the apparent struc- 9 10 ture Fox and Brown detected may arise partly because their model does not 10 11 incorporate species’ differing numbers of occurrences (as hypothesized by 11 12 Wilson, 1995) and/or different ranges. If species have very different geographic 12 13 ranges, the random matrix approach might lead to an erroneous conclusion 13 14 because it would permit co-occurrences that are, in fact, very unlikely. This 14 15 would not appear to be a problem for the Nevada test site data, as the study 15 16 area is small and the ranges of all species encompass it. On the other hand, 16 17 for the southwestern data set, the fact that the study area is huge and includes 17 18 four different deserts suggests that ranges of various pairs of species overlap 18 19 only slightly or not at all. Of 378 species pairs in the set, Stone et al. (1996) 19 20 found only nine sharing more than 20 sites. Either D. merriami or P. manicu- 20 21 latus, the two most widespread species, appears in each of the nine pairs. If 21 22 these two species are considered jointly with Perognathus parvus, it is clear 22 23 how the random matrix method can produce a very unrealistic result (Stone 23 24 et al., 1996). 24 25 Pg. parvus occupies 35 sites, of which Pm. maniculatus occupies 34. One 25 26 might have expected, just from the numbers of occurrences, that Pg. parvus 26 27 and D. merriami would also co-occur on many sites, but they do not. They 27 28 co-occur on only five sites. The reason for the different co-occurrences 28 29 between Pg. parvus on the one hand, and either Pm. maniculatus or D. 29 30 merriami on the other, has to do with the species’ geographic ranges. Pm. 30 31 maniculatus has a very large range and is distributed throughout the entire 31 32 southwestern study area. However, although the range of D. merriami includes 32 33 many sites, it is much smaller than that of Pm. maniculatus, and it barely 33 34 intersects that of Pg. parvus. There were only about five sites censused by 34 35 Brown and Kurzius (1987) in the region of overlap between D. merriami and 35 36 Pg. parvus, and these are the five sites at which they co-occur. But, in the 36 37 random matrices, where both species can be placed on any site, they share 37 38 an average of 18.1 sites (Stone et al., 1996). 38 39 Now, one could argue that the reason these two species share so few sites 39 40 is competition. Given the arrangement of the sampling sites, it would be just 40 Ruling out an assembly rule 67 1 as reasonable to say that their geographic ranges are quite different, so that 1 2 the reason they share so few sites is whatever causes their ranges to differ. 2 3 This could be past or present competition, but it could also be differences of 3 4 dispersal ability, physical factor tolerances, and various historical factors. One 4 5 would need evidence independent of the ranges to judge the reasons the ranges 5 6 barely overlap. 6 7 Because of greatly differing species’ ranges, Fox and Brown’s functional 7 8 group assembly rule would also be unrealistic in exactly the same cases in 8 9 which the matrix randomization approach is unrealistic. Their null probability 9 10 that a particular functional group is drawn to populate a simulated local com- 10 11 munity is proportional to the number of undrawn species in that functional 11 12 group. This is equivalent to making all species equally likely to be drawn, 12 13 no matter how widely or narrowly distributed they are. Again, in the Nevada 13 14 test site data this problem should not greatly affect results by either method, 14 15 as the area is small and all species are part of the species pool for all sites. 15 16 For the southwestern data set, however, both methods would produce un- 16 17 realistic null distributions for at least this reason. 17 18 Brown has wagered that, if the random draws to populate local communi- 18 19 ties in the Fox–Brown method are restricted for each local community to just 19 20 those species whose ranges include the site for the southwestern data set, their 20 21 result (Fox & Brown, 1993) of an excess of observed favored states will stand 21 22 (J. Brown, pers. comm.). This will be an interesting study. 22 23 Although the Nevada test site locations were all within the geographic 23 24 ranges of all the species, it is quite possible that habitat differences between 24 25 the species would make certain random draws, by either Fox’s method or 25 26 matrix randomization, extremely unrealistic, and such lack of realism could 26 27 warp the expected number of favored states. For example, Pg. parvus 27 28 appeared in the test site study to be found in low densities in many plant 28 29 communities, but in high density only in two types, of which one is pinyon– 29 30 juniper. Pg. longimembris, on the other hand, was collected from all plant 30 31 communities except pinyon–juniper (Jorgensen & Hayward, 1965). Of the 31 32 115 local rodent communities tabulated by Jorgensen and Hayward (1965), 32 33 Pg. parvus was found in only two, while Pg. longimembris occurred in 89; 33 34 they shared only one. The description of the plant communities of these 115 34 35 sites shows that, unsurprisingly, they had few or no pinyon–juniper commun- 35 36 ities among them. Other sites not included in this tabulation included much 36 37 pinyon–juniper (Allred et al., 1963). These are two of only three species in 37 38 the QH functional group in the species pool. The particular selection of sites 38 39 for the co-occurrence study apparently predisposed strongly against any local 39 40 community containing three QH species. This predisposition probably 40 68 D. Simberloff et al. 1 increased the number of favored states relative to that in a random draw from 1 2 the species pool according to Fox’s method, irrespective of habitat. As with 2 3 the case of exclusive ranges, it may be that habitat exclusivity reflects the 3 4 ‘ghost of competition past’ (Connell, 1980), but, if this were the case, it would 4 5 have nothing to do with the operation of Fox’s assembly rule in the present. 5 6 6 7 7 8 Randomizing functional group assignments 8 9 The assembly rule of Fox and Brown (1993) rests crucially on how species 9 10 are assigned to functional groups. In fact, the only input required to test for 10 11 the operation of the rule, other than which species are where, is how the 11 12 species in the pool are divided into functional groups (Fox & Brown, 1993). 12 13 Whether an observed local community is in a favored state depends on which 13 14 functional groups its species belong to. Thus, if finding a high number of 14 15 favored states among observed local communities is evidence of Fox’s assem- 15 16 bly rule, the number of favored states should be much lower if species are 16 17 randomly assigned to functional groups. Stone et al. (1996) randomly assigned 17 18 species to functional groups for the two rodent data sets. They did this by 18 19 simply randomizing the functional group designations, maintaining the 19 20 observed numbers of species in each functional group. Then, for each set of 20 21 random functional groups, they looked at the observed local communities to 21 22 see if they were in favored states. Since every species stays on its observed 22 23 sites, the problem of unrealistic ranges and co-occurrence frequencies in the 23 24 null matrices discussed above is eliminated. 24 25 For both rodent data sets, Stone et al. (1996) randomized the functional 25 26 groups 10 000 times and recorded the distribution of favored states (Table 2.1). 26 27 Although the observed number of favored states exceeds expectation for both 27 28 data sets, the differences between observed and expected are not nearly sig- 28 29 nificant at the 0.05 level. In other words, if one randomly concocts functional 29 30 groups from the same species pool, completely independently of their biology, 30 31 the expected number of favored states among the observed local communi- 31 32 ties would not differ from the number found when the real functional groups 32 33 are used. 33 34 A similar result arises for the shrew data set of Fox and Kirkland (1992). 34 35 Here, because there are only six species in the pool, two in each functional 35 36 group, there are only 6!/2!2!2! = 90 possible assignments of species into 36 37 functional groups, and we calculated the number of favored states explicitly 37 38 for each of these, rather than simulating. Taking these assignments as 38 39 equiprobable, we found the observed 32 favored states to be approximately 39 40 that expected (29.0) if the species had been randomly assigned to functional 40 Ruling out an assembly rule 69 1 groups (Table 2.1). Of the randomized functional group assignments, 46.67% 1 2 produced at least as many favored states as were observed. 2 3 In sum, for all three data sets, there is no indication that the particular 3 4 division of the species pool into functional groups produces more favored 4 5 states among the observed local communities than a randomly chosen division 5 6 of the species pool would have. Because Fox’s assembly rule rests on parti- 6 7 tioning the species pool into functional groups of species that are especially 7 8 likely to compete with one another (Fox & Brown, 1993), the conclusion of 8 9 Fox and his coauthors for these three communities – that competition causes 9 10 an unusually high number of favored states – is unconvincing. 10 11 11 12 12 13 Effects of widespread species 13 14 As noted above, Stone et al. (1996) found that variation among species’ bio- 14 15 geographic ranges can greatly affect co-occurrence patterns. They tested the 15 16 effect of the fact that a few species have very large ranges and most are 16 17 greatly restricted on the distribution of favored states. With the matrix 17 18 randomization method, they fixed the widespread species on their actual sites 18 19 and randomized the other species. With the randomized functional group 19 20 method, they assigned the widespread species to their correct functional 20 21 groups and randomly assigned the others. Then, with either method, they 21 22 determined the expected number of favored states. For both rodent data sets, 22 23 Stone et al. (1996) found that fixing either the locations or the functional 23 24 groups of the few widespread species caused the randomizations to produce 24 25 approximately as many favored states as the observed number. For both meth- 25 26 ods, it is as if ‘unfixing’ the locations or functional groups of the widespread 26 27 species is largely responsible for the fact that the expected number of favored 27 28 states is frequently lower than the observed number in our previous simulations. 28 29 The other species simply add background ‘noise’ to this pattern. 29 30 30 31 31 32 The shared-site null hypothesis 32 33 Like Fox and Brown (1993), Wright and Biehl (1982) argued that one should 33 34 focus on particular pairs or trios of species suspected of competing. In their 34 35 view, the most efficient way to do this is to fix the numbers of occurrences 35 36 of each putative competitor, then uniform-randomly distribute these occur- 36 37 rences among all the sites. In this way, they argue, they have mimicked 37 38 different dispersal abilities among species. On the other hand, they allow the 38 39 column sums (site richnesses) to vary while maintaining the row sums (species 39 40 occurrences). That is, the numbers of species in each site are not fixed as 40 70 D. Simberloff et al. 1 species are randomly arranged on the sites. They do not view the resulting 1 2 violation of species–area relationships as a major problem. As noted by 2 3 Connor and Simberloff (1979, 1983, 1984), if one identifies in advance which 3 4 groups of species are hypothesized to compete, rather than simply scanning 4 5 a matrix for exclusively distributed pairs, it would be a fair test of the hypoth- 5 6 esis to look only at those groups. It is not so clear whether a procedure that 6 7 does not constrain numbers of species at each site is appropriate. However, 7 8 in the Nevada test site data, because all sites have so few species of any func- 8 9 tional group anyway, perhaps it is instructive to proceed as suggested by 9 10 Wright and Biehl (1982). 10 11 Each of the functional groups is identified by Fox and Brown (1993) as a 11 12 likely locus of competition, and in the 115 sites there are three species in 12 13 each of the groups BH and QH, and 4 in the group QN. (Jorgensen and 13 14 Hayward, 1965 did not tabulate densities of one species in BH, which was 14 15 thus not used by Stone et al., 1996; it is included in the present analysis.) 15 16 The explicit calculation of the probability that X sites will be occupied by 16 17 two or more species is sketched out by Wright and Biehl (1982). However, 17 18 these probabilities and the resultant tail probabilities are more easily calculated 18 19 by computer simulation. For each of the three groups, the species were 19 20 randomly distributed 10 000 times among the 115 sites, filling as many sites 20 21 for each species as it occupies in nature. 21 22 For group BH, 46 sites are occupied by two or more species (four sites 22 23 have three species each). For randomly distributed species, one would have 23 24 expected 46 or fewer pair plus trio co-occurrences 26.9% of the time (and 24 25 one would have expected four or fewer trio co-occurrences 69.7% of the 25 26 time). For group QH, 40 sites contain two species, and none contain three. 26 27 One would have expected 40 or fewer sites with two or more species 36.8% 27 28 of the time. For group QN, 21 sites contain two species (of which two con- 28 29 tain trios). One would have expected 21 or fewer sites with two or more 29 30 species 53.5% of the time (and two or fewer sites with three or more species 30 31 76.1% of the time). In short, for all three functional groups, species co-occur 31 32 about as frequently as expected if they colonize sites independently of one 32 33 another. 33 34 An analogous analysis was performed with the shrew data. For the two 34 35 species in the functional group comprising small species, which occupy 22 35 36 and 7 of the 43 sites, respectively, six sites are occupied by both species. 36 37 That is, almost all the occurrences of the least common of the two species 37 38 are on sites occupied by the other species. It is thus not surprising that, rather 38 39 than the species appearing to exclude one another, only 0.5% of the time 39 40 would they co-occur this often if they were independently distributed at 40 Ruling out an assembly rule 71 1 random. Similarly, the medium-sized species (which occur on 33 and 10 sites, 1 2 respectively) co-occur on seven sites. For independent random colonization, 2 3 one would have expected seven or fewer sites to have been occupied by both 3 4 species 41.9% of the time. The large species occupy 40 and 12 sites, respec- 4 5 tively, and they co-occur on 11 sites. Had they colonized independently and 5 6 randomly, they would have co-occurred on 11 or fewer sites 97.8% of the 6 7 time. In short, there is again no indication that species within a functional 7 8 group are excluding one another. 8 9 9 10 10 11 Discussion 11 12 No statistical test of a single presence–absence data set can conclusively 12 13 demonstrate the presence or absence of interspecific competition. There are 13 14 two main reasons (Stone et al., 1996). First, even if the data have a pattern 14 15 consistent with the workings of competition, other plausible hypotheses can 15 16 usually be found that would predict the same pattern. Second, the statistical test 16 17 of the biological hypothesis might be affected by competition (the ‘Narcissus 17 18 effect’ of Colwell and Winkler, 1984). An example is the possibility, dis- 18 19 cussed above, that row and/or column sums fixed in the matrix randomiza- 19 20 tion method used here were themselves partially determined by interspecific 20 21 competition. Thus, it is not being claimed that the rodents or shrews in these 21 22 communities do not or did not compete. For the rodents, at least, it is known 22 23 that they do from prior research cited above. This fact makes it all the more 23 24 important that these data do not implicate Fox’s assembly rule. It is also 24 25 emphasized that, in principle, one could find patterns that would be consistent 25 26 with the operation of Fox’s assembly rule. That is, one could conceivably 26 27 find a set of local communities that, when properly tested, were found to have 27 28 more favored states than expected if species colonized independently of which 28 29 other species are present or absent. All that is being claimed is that the three 29 30 communities discussed here do not show such patterns. 30 31 The abiding search for simple rules that determine community structure, 31 32 particularly rules that rest on single, relatively easily measured traits, like mor- 32 33 phological size or functional group, has so far been fruitless. Fox’s favored 33 34 state approach is one such type of rule. In addition to ignoring habitat, range, 34 35 and historical differences that might be as crucial as competition for food in 35 36 determining which species now coexist in which local communities, Fox’s 36 37 assembly rule does not see order of species entry as important. Some models 37 38 of community assembly by sequential addition of species do (e.g., Drake, 38 39 1991), building on priority effects in some empirical studies (e.g., Alford & 39 40 Wilbur, 1985; Robinson & Dickerson, 1987). Even the framework of Fox’s 40 72 D. Simberloff et al. 1 model – sequential colonization of empty sites – may not be appropriate for 1 2 some data sets. For example, in regions as large as that encompassed by the 2 3 southwestern data set, it is possible that geological events could cause whole 3 4 groups of species to be joined almost simultaneously. If competition then 4 5 acted to determine local community composition, how would this competi- 5 6 tion affect the distribution of favored states? No doubt an analog of the Fox 6 7 model could be produced for sequential deletion as for those generated by 7 8 sequential addition, but the statistical tests used by Fox and Brown (1993) 8 9 would be as problematic for communities produced by deletion as by addition. 9 10 If a simple rule turned out to implicate an assembly mechanism strongly, 10 11 our skepticism about the search would be alleviated. But Fox’s rule has not 11 12 passed this test. 12 13 The failure of simple, general assembly rules to date does not mean that 13 14 no rules govern community assembly. With sufficient knowledge of the basic 14 15 biology of each species in a local species pool, including far more informa- 15 16 tion than is usually brought to bear on the precise habitat preferences and 16 17 requirements, it would probably be possible, in a narrowly defined region, to 17 18 predict which species will be found at which site and which species will not 18 19 coexist. The necessary research, including not only matters commonly viewed 19 20 as ecological but also much that is ethological, generally falls under the 20 21 rubric of ‘natural history’. Of course, natural history is not very fashionable 21 22 nowadays, to a large extent because it is thought to be superseded by such 22 23 generalizations as assembly rules. Perhaps the failure of general assembly 23 24 rules to pass close scrutiny indicates that natural history should not be so 24 25 quickly discarded. In any event, the subtle responses of species to one another 25 26 and to minute changes in the physical environment, as well as substantial 26 27 phenotypic and genetic differences among populations of some species, 27 28 suggest that assembly rules, if they exist, will be quite local. Not only will 28 29 assembly rules not be simple, they will not be very general. 29 30 30 31 31 32 References 32 33 Alford, R.A & Wilbur, H.M. (1985). Priority effects in experimental pond 33 34 communities: competition between Bufo and Rana. Ecology 66: 1097–1105. 34 35 Allred, D.M., Beck, D.E. & Jorgensen C.D., (1963). Nevada Test Site study areas 35 36 and specimen depositories. Brigham Young University Science Bulletin, 36 Biology Series 2(4):15 pp. 37 Brown, J.H. (1987). Variation in desert rodent guilds: patterns, processes, and 37 38 scales. In Organization of Communities: Past and Present. ed. J.H.R. Gee & 38 39 P.S. Giller, pp. 185–203. Oxford: Blackwell Scientific. 39 Brown, J.H. & Kurzius, M. (1987). Composition of desert rodent faunas: 40 combinations of co-existing species. Annales Zool. Fenn. 24: 227–237. 40 Ruling out an assembly rule 73

1 Colwell, R.K. & Winkler, D.W. (1984). A null model for null models in 1 2 biogeography. In Ecological Communities. Conceptual Issues and the 2 Evidence. ed. D.R. Strong, Jr., D. Simberloff, L.G. Abele, & A.B. Thistle, pp. 3 344–359. Princeton, NJ: Princeton University Press. 3 4 Connell, J.H. (1980). Diversity and the coevolution of competition, or the ghost of 4 5 competition past. Oikos 35: 131–138. 5 6 Connor, E.F. & Simberloff, D. (1979). The assembly of species communities: 6 chance or competition? Ecology 60: 1132–1140. 7 Connor, E.F. & Simberloff, D. (1983). Interspecific competition and species co- 7 8 occurrence patterns on islands: null models and the evaluation of evidence. 8 9 Oikos 41: 455–465. 9 Connor, E.F. & Simberloff, D. (1984). Neutral models of species’ co-occurrence 10 patterns. In Ecological Communities. Conceptual Issues and the Evidence, ed 10 11 D.R. Strong, Jr., D. Simberloff, L.G. Abele, & A.B. Thistles, pp. 316–331. 11 12 Princeton, NJ: Princeton University Press. 12 Diamond, J.M. (1975). Assembly of species communities. In Ecology and 13 Evolution of Communities, ed. M.L. Cody & J.M. Diamond, pp. 342–444. 13 14 Cambridge, MA: Harvard University Press. 14 15 Drake, J.A. (1991). Community-assembly mechanics and the structure of an 15 16 experimental species ensemble. American Naturalist 137: 1–26. 16 Feller, W. (1968). An Introduction to Probability Theory and Its Applications, 3rd 17 edn., Vol. 1. New York: Wiley. 17 18 Fox, B.J. (1985). Small mammal communities in Australian temperate heathlands 18 19 and forests. Australian Mammalogy 8: 153–158. 19 Fox, B.J. (1987). Species assembly and evolution of community structure. 20 Evolution and Ecology 1: 201–213 20 21 Fox, B.J. (1989). Small-mammal community pattern in Australian heathland: a 21 22 taxonomically based rule for species assembly. In Patterns in the Structure of 22 Mammalian Communities, ed. D.W. Morris, Z. Abramsky, B.J. Fox, & M.R. 23 Willig, Special Publication 28, pp. 91–103. Lubbock, TX: Texas Tech 23 24 University. 24 25 Fox, B.J. & Brown, J.H. (1993). Assembly rules for functional groups in North 25 26 American desert rodent communities. Oikos 67: 358–370. 26 Fox, B.J. & Brown, J.H. (1995). Reaffirming the validity of the assembly rule for 27 functional groups or guilds: a reply to Wilson. Oikos 73: 125–132. 27 28 Fox, B.J. & Kirkland, Jr., G.L. (1992). An assembly rule for functional groups 28 29 applied to North American soricid communities. Journal of Mammalogy 73: 29 125–132. 30 Gilpin, M.E. & Diamond, J.M. (1982). Factors contributing to non-randomness in 30 31 species co-occurrences on islands. Oecologia 52: 75–82. 31 32 Gotelli, N. & Graves, G. (1996). Null Models in Ecology. Washington, DC: 32 Smithsonian Institution Press. 33 Jorgensen, C.D. & Hayward, C.L. (1965). Mammals of the Nevada test site. 33 34 Brigham Young University Science Bulletin, Biology Series 6(3): 1–81. 34 35 Kelt, D.A., Taper, M.A. & Meserve, P.L. (1995). Assessing the impact of 35 36 competition on the assembly of communities: a case study using small 36 mammals. Ecology 76: 1283–1296. 37 Kotler, B.P. & Brown, J.S. (1988). Environmental heterogeneity and the 37 38 coexistence of desert rodents. Annual Review of Ecology and Systematics 19: 38 39 281–307. 39 Rickens, V.B. (1974). Numbers and habitat affinities of small mammals in 40 northwestern Maine. Canadian Field Naturalist 88: 191–196. 40 74 D. Simberloff et al.

1 Robinson, J.V. & Dickerson, Jr., (1987). Does invasion sequence affect community 1 2 structure? Ecology 68: 587–595. 2 Simberloff, D. & Dayan, T. (1991). The guild concept and the structure of 3 ecological communities. Annual Review of Ecology and Systematics 22: 3 4 115–143. 4 5 Stone, L., Dayan, T. & Simberloff, D. (1996). Community-wide assembly patterns 5 6 unmasked: the importance of species’ differing geographic ranges. American 6 Naturalist 148: 997–1015. 7 Wilson, J.B. (1995). Null models for assembly rules: the Jack Horner effect is more 7 8 insidious than the Narcissus effect. Oikos 72: 139–144. 8 9 Wright, S.J. & Biehl, C.C. (1982). Island biogeographic distributions: testing for 9 random, regular, and aggregated patterns of species occurrence. American 10 Naturalist 119: 345–357. 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 3 1 2 2 3 Community structure and assembly rules: 3 4 confronting conceptual and statistical issues 4 5 5 6 with data on desert rodents 6 7 7 8 Douglas A. Kelt and James H. Brown 8 9 9 10 10 11 11 12 12 13 13 14 Introduction 14 15 Two main themes have dominated empirical research in community ecology 15 16 for the last several decades. On the one hand, the enormously successful 16 17 experiments of British plant ecologists and of intertidal ecologists such as 17 18 Connell (1961a, b) and Paine (1966) stimulated many ecologists to perform con- 18 19 trolled, replicated, manipulative field studies to investigate direct and indirect 19 20 interactions. On the other hand, the insightful observations of Hutchinson 20 21 (1957) and MacArthur (1958) stimulated many other ecologists to explore 21 22 the relationships between patterns of community organization and the mech- 22 23 anistic processes that produce them. These two approaches have tended to 23 24 diverge in the kinds of organisms and the spatial and temporal scales studied. 24 25 The experimentalists tended to work with terrestrial plants, freshwater fishes 25 26 and amphibians, and intertidal marine organisms, to manipulate on small spatial 26 27 scales, and to study ongoing, proximate ecological processes. In contrast, 27 28 those studying patterns of community organization tended to study terrestrial 28 29 reptiles, birds, and mammals, to do comparative studies over geographic spa- 29 30 tial scales, and to be concerned with evolutionary, as well as ecological, 30 31 processes and timescales. Most studies of community-level patterns and 31 32 processes have focused on two phenomena: the organization of food webs 32 33 (e.g., Cohen et al., 1993; Polis & Strong, 1996; papers in Special Feature 33 34 section of Ecology 69, pp. 1647–1676) and the structure of ‘guilds’ (sets of 34 35 functionally similar and closely related species; e.g., Cody & Diamond, 1975; 35 36 Diamond & Case, 1986; Simberloff & Dayan, 1991). The latter studies have 36 37 often invoked ‘assembly rules’ to describe the apparently non-random pat- 37 38 terns of local coexistence of species, and they have often hypothesized that 38 39 the rules reflect the ecological and evolutionary consequences of interspecific 39 40 competition. 40

75 76 D.A. Kelt and J.H. Brown 1 Much of the early, influential work on assembly rules focused on three 1 2 ‘model systems’: Caribbean Anolis lizards (e.g., Schoener, 1969, 1970; 2 3 Roughgarden, 1995), island birds (e.g., Lack, 1947; Diamond, 1975; Grant, 3 4 1986), and desert rodents (e.g., Rosenzweig & Winakur, 1969; Brown, 1975; 4 5 Reichman & Brown, 1983). There was much in common among these studies. 5 6 They relied heavily on non-experimental, comparative geographic studies in 6 7 which some variables, such as the influences of phylogenetic and paleoenviron- 7 8 mental history on the composition of the regional species pool, were kept 8 9 as constant as possible, while other variables, such as the number, identity, 9 10 morphology, diet, and habitat affinity of coexisting species, were measured, 10 11 analyzed, and interpreted. They focused on patterns of non-random assembly, 11 12 i.e., the ways in which local assemblages of species differed from a random 12 13 sample of the regional species pool. And, to explain such community struc- 13 14 ture, they often invoked interspecific competition as the most likely process 14 15 to have influenced both ecological coexistence and evolutionary divergence 15 16 of species within a guild. 16 17 In the present chapter the extensive information on the seed-eating desert 17 18 rodents of southwestern North America is used to evaluate the assembly rules 18 19 that have been proposed for these organisms. We first review the empirical 19 20 patterns of community structure found in desert rodents. In addition to sum- 20 21 marizing the existing literature, several new analyzes are presented which 21 22 provide more statistically robust conclusions. From an ecological perspective, 22 23 North American desert rodents are one of the most thoroughly studied groups 23 24 of organisms, and the extensive data on composition of local communities 24 25 are complemented by information on many other characteristics, including 25 26 systematics, phylogenetic and biogeographic history, morphology, physiology, 26 27 behavior, genetics, life history, and population dynamics (e.g., Reichman 27 28 & Brown, 1983; Genoways & Brown, 1993). Work in North America 28 29 stimulated studies of ecology of desert rodents elsewhere in the world. These 29 30 investigations revealed many similarities, but also many differences. Environ- 30 31 mental differences among the deserts and influences of phylogenetic and bio- 31 32 geographic history on the rodents have affected the ways that communities 32 33 are organized (e.g., Kelt et al., 1996). For the sake of simplicity therefore, 33 34 our attention will be confined to North American assemblages. 34 35 Second, the data and analyses from studies of these desert rodents are used 35 36 as a basis to discuss some important conceptual and statistical issues that 36 37 are fundamental to many studies of community assembly and community 37 38 structure. The conceptual issues concern the nature of assembly rules, and 38 39 the relationship between empirical patterns and underlying mechanistic 39 40 processes. Statistical and methodological issues include: the formulation of 40 Conceptual and statistical issues 77 1 null hypotheses that are, on the one hand, interesting and informative, and, 1 2 on the other hand, statistically rigorous and testable; the nature of species pools 2 3 and the problems of defining them; the spatial and temporal scales at which 3 4 the rules can be observed and the mechanisms operate; and the problems that 4 5 arise because the variables are often not independent and the hypothesized 5 6 mechanisms are not mutually exclusive. 6 7 Finally, the information on assembly rules for the desert rodents is used as 7 8 a basis to comment on two issues: on the one hand, on the contribution of 8 9 comparative geographic studies of community structure to our current under- 9 10 standing of the interactions within these guilds, and on the other hand, on the 10 11 more general contributions of the assembly rule approach to our current under- 11 12 standing of the phenomena of coexistence, convergence and divergence, and 12 13 diversity of species. 13 14 Before proceeding, it is necessary to be clear what we are talking about. 14 15 For the purposes of this chapter, assembly rules will be defined simply as 15 16 empirical patterns of community organization, synonymous with what will 16 17 also be called community structure or community organization. Thus, the 17 18 existence of an assembly rule implies a non-random pattern of community 18 19 structure. To claim the existence of a rule implies that the rule has been tested 19 20 against the null hypothesis of random assembly and rejected, usually by com- 20 21 paring the observed community organization with that expected on the basis 21 22 of random assembly from some appropriate species pool. It must be empha- 22 23 sized that assembly rules are descriptive, empirical characterizations of 23 24 community composition. While the existence of a non-random pattern implies 24 25 the operation of some deterministic mechanism, an assembly rule is not taken 25 26 as necessarily implying the operation of any particular process, such as inter- 26 27 specific competition, nor as implying the operation of any particular temporal 27 28 sequence of species assembly or successional process. In desert rodents, as 28 29 in other organisms, investigators have claimed to have demonstrated several 29 30 different kinds of assembly rules characterized by different variables, measured 30 31 in a variety of different ways. Examples of assembly rules are presented that 31 32 describe three kinds of empirical patterns: in species richness, in morphological 32 33 and functional composition, and in species identity. 33 34 34 35 35 The rules: patterns of community structure 36 36 37 Patterns of species richness 37 38 Perhaps the simplest measure of community structure is species richness, a tally 38 39 of the number of species present. Such a tally avoids the complications asso- 39 40 ciated with indices of diversity which include measures of relative abundance 40 78 D.A. Kelt and J.H. Brown 1 of the species and often creates problems with statistical analysis and inter- 1 2 pretation (see Pielou, 1969; Magurran, 1988). Both diversity indices and 2 3 simple species richness measured for a local community will depend on the 3 4 sampling effort and the number of individuals tallied. Such sampling problems 4 5 can largely be avoided, however, by restricting comparisons to communities 5 6 that have been sampled with equal effort and using some standardized con- 6 7 vention to tally only the common species. When this has been done for desert 7 8 rodent assemblages, three kinds of patterns have been documented. 8 9 9 10 Local species richness increases with increasing complexity of habitat 10 11 structure, and other characteristics of the soil and vegetation 11 12 Some variables, such as composition of the species pool and climate, can be 12 13 held approximately constant and features of the habitat allowed to vary, by 13 14 restricting comparisons to a local region. When this is done, rodent species 14 15 diversity generally varies along gradients of soil structure, from relatively few 15 16 species on shallow, rocky soils, to larger numbers on deeper, sandy soils (e.g., 16 17 Rosenzweig & Winakur, 1969; Brown, 1975; Brown & Harney, 1993). Species 17 18 richness also varies with vegetation structure. In general, there is a positive 18 19 relationship between rodent species diversity and foliage height diversity or 19 20 some other measure of structural complexity of vegetation (e.g., Rosenzweig 20 21 & Winakur, 1969; Rosenzweig, 1973; Rosenzweig et al., 1975; Price, 1978; 21 22 Thompson, 1982a). 22 23 The relationship between rodent species diversity and characteristics of 23 24 habitat heterogeneity apparently reflect the differential abilities of species to 24 25 exist and coexist in distinctive habitats and microhabitats (Price, 1978; 25 26 Thompson, 1982b). This is apparent at different spatial scales. Among habi- 26 27 tats within regions, certain species and larger functional and taxonomic groups 27 28 vary in habitat specificity. For example, the rock pocket mouse (Chaetodipus 28 29 intermedius) is typically confined to boulder-strewn hillsides, while bipedal 29 30 kangaroo rats (Dipodomys spp.) and kangaroo mice (Microdipodops spp.) are 30 31 absent from rocky soils, and the deer mouse (Peromyscus maniculatus) occurs 31 32 in an enormous range of habitats. Within these kinds of macrohabitats, rodents 32 33 forage differentially in distinctive microhabitats. For example, the bipedal 33 34 species tend to use open patches of bare ground, while pocket mice forage 34 35 under vegetative cover. As a consequence of both macro- and microhabitat 35 36 selection, highest rodent species diversity typically occurs on stabilized 36 37 sand dunes and other deep, sandy soils, where the vegetation exhibits both 37 38 vertical and horizontal heterogeneity: a spatial mosaic composed of patches 38 39 of bare ground and plants of sizes ranging from small herbs to large shrubs 39 40 or small trees. 40 Conceptual and statistical issues 79 1 1 2 Local species richness increases with increasing productivity 2 3 In arid regions productivity is closely correlated with actual evapotranspiration 3 4 and with precipitation (Rosenzweig, 1968; Brown et al., 1979). Productivity 4 5 can be varied, and other variables held relatively constant by making compar- 5 6 isons along gradients of increasing precipitation within a major desert region 6 7 with a similar species pool and among structurally similar habitat types. Brown 7 8 (1973, 1975; see also Hafner, 1977) made such comparisons along productivity 8 9 gradients, comparing sand dune habitats in the Mojave and Great Basin Deserts 9 10 and rocky hillside and sandy flatland habitats in the Sonoran Desert. He found 10 11 that about 50% of the variance in the number of common species was posi- 11 12 tively correlated with mean annual precipitation. This positive relationship 12 13 holds so long as the comparisons are restricted to desert habitats characterized 13 14 by a spatial mosaic of bare ground and isolated perennial plants (predominantly 14 15 woody shrubs and succulents). But, as precipitation and productivity continue 15 16 to increase, desert shrubland habitat is replaced by semi-arid grassland and 16 17 rodent species richness declines abruptly (Brown, 1975; Owen, 1988). 17 18 18 19 Local species richness is lower in isolated areas where the size of the 19 20 species pool is reduced 20 21 In geographically isolated desert basins, habitats with otherwise similar soils 21 22 and vegetation possess fewer species than expected on the basis of produc- 22 23 tivity (Brown, 1973, 1975). These areas have been isolated since the Pleis- 23 24 tocene by barriers of inhospitable habitats that have prevented the coloniza- 24 25 tion of several desert rodent species. The influence of this reduced species 25 26 pool is reflected in reduced species richness in local communities. 26 27 27 28 28 29 Patterns of morphological, phylogenetic, and ecological similarity 29 30 Since Darwin’s commentary (1859) and Lack’s (1948) classic study on char- 30 31 acter displacement in Darwin’s finches (see also Hutchinson, 1959), many 31 32 observers have noted that those species that coexist in local communities 32 33 appear to be more different in structure and function than closely related pop- 33 34 ulations that occur allopatrically. This pattern is readily apparent in desert 34 35 rodents, and can be characterized and quantified in several ways. 35 36 36 37 Local communities contain species that are more different in body size and 37 38 other morphological characteristics than expected by chance 38 39 Differences in size among coexisting desert rodent species have been noted 39 40 by many investigators (e.g., Grinnell & Orr, 1934; Rosenzweig & Sterner, 40 80 D.A. Kelt and J.H. Brown 1 1970; Brown & Lieberman, 1973; Brown, 1975; M’Closkey, 1976, 1978; Price, 1 2 1978, 1983). Bowers and Brown (1982) compiled data on 95 local commu- 2 3 nities of granivorous desert rodents, and showed that species of similar body 3 4 size co-occurred significantly less frequently than expected on the basis of 4 5 chance. Bowers and Brown (1982) also showed, however, that species of 5 6 similar body size had less overlap in their geographic ranges than would be 6 7 expected on the basis of chance. This is especially obvious in the case of 7 8 species in both the largest and smallest body size category, which have almost 8 9 perfectly non-overlapping geographic distributions. This is an important point, 9 10 because it means that the regional pool of species from which local commu- 10 11 nities are assembled is already structured with respect to body size (and other 11 12 attributes as shown below). 12 13 Since Bowers and Brown’s study, accumulation of larger data sets and ease 13 14 of performing computerized randomization tests of null hypotheses have per- 14 15 mitted more rigorous analyses. In Table 3.1 a similar analysis is applied to 15 16 two data sets: rodents coexisting in 115 samples of local habitat on the Nevada 16 17 Test Site (Jorgensen & Hayward, 1965), and in 201 local sites from through- 17 18 out the arid southwestern United States (Brown & Kurzius, 1987, as emended 18 19 in Morton et al., 1994). In the analysis of the southwestern US data set the 19 20 species pool for each sample site has also been restricted to just those species 20 21 whose geographic ranges overlap that locality, thus removing the influence 21 22 of the non-overlapping geographic distributions on the composition of the 22 23 regional pool. (The Nevada Test Site is such a restricted geographic area – 23 24 approximately 3500 km2 – that all of the sample sites can be assumed to be 24 25 potentially accessible to all species in the pool.) The results of these new ana- 25 26 lyses mirror those of Bowers and Brown (1982): species of similar body size 26 27 (ratio of masses < 1.5) coexist much less frequently than expected on the 27 28 basis of random associations of the species for both the regional (Nevada Test 28 29 Site) and larger geographical (southwestern North America) assemblages. 29 30 Dayan and Simberloff (1994) have shown that two regional heteromyid 30 31 faunas, from southeastern Arizona and northwestern Nevada, exhibit highly 31 32 non-random ratios of morphological traits. The most regular pattern is in the 32 33 width of the incisors, but this trait is correlated with several others, includ- 33 34 ing body size and cheek pouch volume. This result is largely complementary 34 35 to the patterns in body size (above) and other characteristics (below), but it 35 36 raises two interesting issues. First, unlike ratios of body size, which show the 36 37 clearest patterns within local communities, the ratios of incisor widths are 37 38 more even among the species in the entire regional pools than among the 38 39 actual assemblages. Second, since Dayan and Simberloff also show that the 39 40 sizes of all morphological characteristics of these rodents tend to be posi- 40 Conceptual and statistical issues 81

1 Table 3.1. Test of the null hypothesis that local coexistence and geographic overlap 1 2 or granivorous rodents are independent of body size 2 3 3 4 Frequency of pairwise co-occurrence of species of 4 5 similar size different size 5 6 (body mass ratio < 1.5) (body mass ratio > 1.5) 6 7 7 Nevada test site Observed 9 228 8 115 sites, 10 species 8 9 Expected 57.93 179.07 9 10 (Potential) (1265) (3710) 10 χ2 11 = 54.70 P < 0.001 11 12 Southwestern US Observed 217 679 12 13 201 sites, 28 species 13 Expected 270 626 14 (Potential) (4703) (10 904) 14 15 χ2 = 14.89 P < 0.001 15 16 16 17 Reanalyzed from Jorgensen and Hayward, (1965) and Brown and Kurzius, (1987). 17 Σ ×Σ 18 Expected values were calculated as expectedi = ((potentiali / potential) 18 observed), where potential is the number of potential pairwise co-occurrences of 19 i 19 similar or of different-sized species, based on the pool of species with access to 20 each site. For the Nevada test range, all sites shared a common species pool. For 20 21 the Southwestern US data set, species pools included those granivorous rodent 21 22 species whose geographic ranges (from Hall, 1981) included the site. Because the 22 number of potential co-occurrences (in parentheses) is vastly greater than the 23 observed, expected values were proportionalized to sum to the same number as to 23 24 the observed. 24 25 25 26 tively correlated, the observation that different traits appear to show clearer 26 27 patterns at different spatial scales raises intriguing questions about the under- 27 28 lying mechanisms producing the observed community structure. 28 29 29 30 Closely related species coexist less frequently than expected by chance 30 31 Because phylogenetically related species often share many ecological and mor- 31 32 phological characteristics, they might be expected to co-occur less frequently 32 33 than they would with more distantly related species. Using the Nevada test 33 34 site and the southwestern US data sets, the frequency with which closely con- 34 35 generic and more distantly related species co-occurred in local communities 35 36 was compared (Table 3.2). For both data sets, it is clear that congeners co- 36 37 occur signiﬁcantly less frequently than expected by chance. As in the previous 37 38 analysis, the species pool for each sample community in the southwestern US 38 39 analysis included only those species whose geographic ranges overlapped the 39 40 site. 40 82 D.A. Kelt and J.H. Brown

1 Table 3.2. Test of the null hypothesis that local coexistence and geographic overlap 1 2 or granivorous rodents are independent of taxonomic relatedness 2 3 3 4 Frequency of pairwise co-occurrence of species in 4 5 Same genus Different genus 5 6 6 7 Nevada test site Observed 144 372 7 115 sites, 10 species 8 Expected 100.96 415.04 8 9 (Potential) (2070) (8510) 9 10 χ2 = 22.82 P < 0.001 10 11 Southwestern US Observed 222 1686 11 12 201 sites, 28 species 12 13 Expected 312.34 1595.67 13 (Potential) (5019) (25 641) 14 14 χ2 = 31.24 P < 0.001 15 15 16 16 Reanalyzed from Jorgensen and Hayward (1965) and Brown and kurzius (1987). 17 Σ ×Σ 17 Expected values were calculated as expectedi = (potentiali / potential) 18 observed), where potentiali is the number of potential pairwise co-occurrences, 18 19 based on the pool of species with access to each site. For the Nevada test range, all 19 sites shared a common species pool. For the Southwestern US data set, species 20 pools included those granivorous rodent species whose geographic ranges (from 20 21 Hall, 1981) included the site. Because the number of potential co-occurrences (in 21 22 parentheses) is vastly greater than the observe, expected values were proportional- 22 ized to sum to the same number as to the observed. 23 23 24 24 25 25 26 Species in the same functional group coexist less frequently than expected 26 27 by chance 27 28 Fox (1987; see also Fox & Brown, 1993; Kelt et al., 1995; Fox, this volume) 28 29 developed an assembly rule for Australian small mammals that specified that 29 30 species tended to be drawn sequentially from different taxonomic or func- 30 31 tional groups. Functional groups are composed of species which use their 31 32 environment in a similar manner (see below). Fox’s rule divides all possible 32 33 combinations of functional groups into two categories: favored states, in 33 34 which the number of species in the different functional groups differs by no 34 35 more than one; and unfavored states, in which the number of species in the 35 36 different groups differs by two or more. Monte-Carlo techniques are used to 36 37 draw species at random from the pool to evaluate the null hypothesis that 37 38 neither favored nor unfavored states are observed more frequently than 38 39 expected by chance. 39 40 Fox and Brown (1993; see also Wilson, 1995; Fox & Brown, 1995) applied 40 Conceptual and statistical issues 83 1 this rule to North American desert rodents (both the Nevada Test Site and 1 2 southwestern US data sets), classifying the granivorous species a priori into 2 3 one of three functional/morphological groups: (a) bipedal with cheek pouches 3 4 (kangaroo rats and kangaroo mice); (b) quadrupedal with cheek pouches 4 5 (pocket mice); or (c) quadrupedal without cheek pouches (cricetine rodents). 5 6 For both data sets, favored states were observed significantly more frequently 6 7 and unfavored states significantly less frequently than expected by chance. 7 8 Local communities tended to be comprised of functionally dissimilar species. 8 9 But, in their analysis of the southwestern US data set, the species pool for 9 10 all sites included all of the 28 species that inhabited any of the 200 sample 10 11 sites throughout the large geographic area. 11 12 Stone et al. (1996, see also Simberloff et al. this volume) challenged the 12 13 assembly rule presented by Fox and Brown (1993), claiming that their results 13 14 were artifactual and that their analysis ‘... failed to find evidence that inter- 14 15 specific competition or a deterministic assembly rule shaped local community 15 16 composition.’ They also claimed that geographical structuring (the Narcissus 16 17 effect; Colwell & Winkler, 1984) had a much greater influence on local com- 17 18 munity structure than did local interspecific interactions. While implying that 18 19 a Narcissus effect could account for the degree to which local assemblages 19 20 follow the assembly rule (presumably obviating the analysis of Fox and 20 21 Brown, 1993), Stone et al. made no attempt to provide a more refined analysis, 21 22 claiming that ‘... detailed information on species’ ranges cannot always be 22 23 determined from the literature.’ This simply is not true. Although compiling 23 24 such information is somewhat tedious and time consuming, it is readily 24 25 accomplished. To address concerns raised by these authors, the southwestern 25 26 US data has been reanalyzed by defining species pools for each site as con- 26 27 sisting only of those species whose geographic ranges encompass the site 27 28 (this is the analysis that Simberloff et al. (this volume) claimed ‘will be an 28 29 interesting study’ and upon which he and Brown bet a beer (see Simberloff, 29 30 this volume)). This analysis, based on 2000 simulations, clearly demonstrated 30 31 that these communities were significantly non-random assemblages (Fig. 3.1a). 31 32 The observed number of favored states was significantly greater than would 32 33 be expected if species were drawn at random according to the number of 33 34 species in the regional pool in each functional group. The methodology of 34 35 Kelt et al. (1995) was also applied and simulations conducted that assumed 35 36 various levels of negative interspecific association among the species within 36 37 functional groups (Fig. 3.1a–d; see Appendix for details), to obtain a maximum 37 38 likelihood estimate for the mean strength of the observed associations. Since 38 39 this estimate was an average value over all 201 sites and 28 species, it is 39 40 most useful in allowing the statistical power of our analysis to be determined, 40 84 D.A. Kelt and J.H. Brown 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 Fig. 3.1. Frequency distributions of the number of communities expected to be in 29 30 favored states, using four levels of interspecific interaction. Each histogram represents 30 2000 iterations of the null model, in which 201 artificial communities were simulated 31 using the species pools derived from species’ geographic ranges. The vertical dotted 31 32 line indicates the observed number of favored states. Panel a demonstrates that the 32 33 null hypothesis of no interspecific interactions (θ = 0) is rejected. Increasing levels of 33 θ 34 interspecific interaction are represented by increasing values of . 34 35 35 36 i.e, the ability to detect existence of favored states. The maximum likelihood 36 37 strength of negative association was estimated as θˆ = 0.33 (Fig. 3.2). From 37 38 this, it was determined that the statistical power of our estimate was extremely 38 39 high, above 94% (Fig. 3.3). This gave us confidence that these communities 39 40 are highly structured non-random assemblages. Indeed, a relatively simple 40 Conceptual and statistical issues 85 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 Fig. 3.2. Relation between the number of iterations of 201 sites that yielded 123 19 favored states and the value of theta. The peak of this distribution represents the value 19 20 of theta that best describes the observed set of sites. This peak occurs at approxi- 20 21 mately θ = 0.33, and provides an maximum likelihood estimate of the real strength 21 of interaction in these communities. 22 22 23 23 24 model of community assembly, based on a moderate level of negative asso- 24 25 ciations among functionally similar species, described the structure of these 25 26 communities surprisingly well. 26 27 This methodology has also been applied to the Nevada test site data (Brown 27 28 et al. ms). For this data set we also developed two sets of species pools. One 28 29 species pool was based on the geographic distribution of species and was 29 30 conceptually identical to the species pools used for the southwestern US 30 31 analysis. The second set of species pools was based on the habitat require- 31 32 ments of these species, and consisted of all species found within the Nevada 32 33 Test Site who were known to occur in the type of habitat present at a given 33 34 site. Because the spatial distribution of two species (Dipodomys deserti and 34 35 Reithrodontomys megalotis) were not delineated by Jorgensen and Hayward 35 36 (1965) we have not included them in the present analyses; however, similar 36 37 analyses have been conducted with these species conservatively allocated to 37 38 sites with appropriate habitat, and the results were qualitatively identical to 38 39 those presented here. The observed number of favored states in every analy- 39 40 sis was significantly greater than expected at random (Table 3.3), suggesting 40 86 D.A. Kelt and J.H. Brown 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 Fig. 3.3. The statistical power can be estimated by determining the number of θ = 21 22 0.33 simulations that lie within the critical (5%) region of θ = 0 simulations. Using 22 23 the southwestern US data set, it can be seen that the null hypothesis of no interaction 23 is rejected with a power of about 94%. In other words, there is a 94% probability 24 of being correct when the null hypothesis of no interaction among species during 24 25 community assembly is rejected. 25 26 26 27 27 28 that even at this reduced geographical scale these communities are highly 28 29 structured and non-random assemblages. 29 30 Finally, we note that the idea that a Narcissus effect occurs for these rodents 30 31 is hardly a novel observation. Bowers and Brown (1982) pointed out that 31 32 ‘species of similar (body) size ... coexist less frequently in local communities 32 33 and overlap less in their geographic distributions than expected on the basis 33 34 of chance, suggesting that their co-occurrence is precluded by interspecific 34 35 competition.’ Additionally, as has been summarized above, there is a ten- 35 36 dency for similar species to occur less frequently than expected by chance, 36 37 whether similarity is measured in terms of body size, taxonomic affinity, or 37 38 functional group membership, and these patterns apply at the spatial scales 38 39 of both local communities and regional faunas (Brown & Bowers 1984; 39 40 Brown, 1987; Brown & Harney, 1991, this chapter). What few authors appear 40 Conceptual and statistical issues 87

1 Table 3.3. Nested subset structure of small mammal communities of 13 sand dunes 1 2 in the Mojave and Great Basin Deserts 2 3 3 4 Dune (sample site) number 4 5 Total 5 6 Species 15672348910111213 dunes 6 7 7 Dipodomys deserti XXXXXXXXXXXXX 13 8 Dipodomys merriami XXXXXXXXX 9 8 9 Perognathus longimembris XXXXXXX X 8 9 10 Peromyscus maniculatus X XXXXX 6 10 Dipodomys ordii XXXXX 5 11 Reithrodontomys megalotis XXXXX 5 11 12 Microdipodops pallidus XXX X 4 12 13 Dipodomys microps XX 2 13 Perognathus parvus X1 14 Chaetodipus penicillatus X114 15 Peromyscus crinitus X115 16 Microdipodops megacephalus X116 17 Total species 8777655332111 17 18 18 19 Reanalyzed from Brown, 1973. 19 20 As with the sites discussed in the text, these communities show highly nested 20 structure (system temperature = 7.2°, P(T < 7.2°) = 1.69 × 10−6, based on 1000 iter- 21 ations of the program (Patterson & Atmar, 1995). 21 22 22 23 23 24 to consider is that patterns occurring at local and geographical scales are 24 25 linked by both ‘top-down’ and ‘bottom-up’ processes. The former include 25 26 such factors as a Narcissus effect. However, the latter include such mecha- 26 27 nisms as interspecific interactions (including competition) which may feed 27 28 upwards to influence the structure and composition of regional biotas (e.g., 28 29 Bowers & Brown 1982), thereby producing a Narcissus effect. 29 30 30 31 31 32 Patterns in the identities of species 32 33 So far, our concern has been with characteristics of communities, total species 33 34 richness, or morphological, phylogenetic, or functional similarity among 34 35 species. But, there are also patterns in the frequency with which particular 35 36 species occur in local communities. As implied in our discussion of species 36 37 richness and habitat use, species differ in the sizes of their geographic ranges 37 38 and in the proportion of sample sites within their ranges where they occur. 38 39 Many of the assembly rules described in early studies of community struc- 39 40 ture were concerned with the associations of Latin binomials, i.e., with the 40 88 D.A. Kelt and J.H. Brown 1 patterns of occurrence and co-occurrence of particular species and combina- 1 2 tions of species (e.g., Diamond, 1975, 1984; Connor & Simberloff, 1979, 2 3 1983, 1984; Gilpin & Diamond, 1981, 1982, 1984; Diamond & Gilpin, 1982). 3 4 Two patterns are particularly apparent in desert small mammals. 4 5 5 6 The communities exhibit a nested subset structure 6 7 One pattern that appears pervasive across many communities is a nested sub- 7 8 set or core-satellite structure (e.g., Table 3.3; see Hanski, 1982; Patterson & 8 9 Atmar, 1986; Patterson, 1987, 1990; Patterson & Brown, 1991). Some species 9 10 are widely distributed and occur in many local communities, whereas other 10 11 species have more restricted distributions and occur only in a subset of the 11 12 local samples. Patterson and Brown (1991) demonstrated significant nested- 12 13 ness for the North American desert rodent fauna at the scale of major desert 13 14 regions and the entire southwestern US. Here, the 201-sample southwestern 14 15 US data set has been reanalyzed using an improved method for calculating 15 16 deviation from nestedness (Atmar & Patterson, 1995; see Atmar & Patterson, 16 17 1993) and results are similar to those presented by Patterson and Brown 17 18 (1991), except that even more of the assemblages are significantly more nested 18 19 than expected from random simulations (compare Tables 3.4 and 3.5 to Tables 19 20 1 and 2 in Patterson & Brown, 1991). 20 21 21 22 Certain pairs and other combinations of species coexist either less or more 22 23 frequently than expected by chance 23 24 Another species specific pattern of assembly is that certain combinations of 24 25 species are strongly positively or negatively associated with respect to fre- 25 26 quency of co-occurrence in local communities. In general, data on desert 26 27 rodents have not been thoroughly analyzed in this way (but see Bowers & 27 28 Brown, 1982; Brown & Kurzius, 1987; Patterson & Brown, 1991; Morton et 28 29 al., 1994), but this pattern is implicit in many of the other patterns described 29 30 above. Thus, if, in general, congeners and species of similar body size and 30 31 functional group tend to coexist less frequently than expected by chance, it 31 32 follows that particular pairs of species that belong to the same genus, are of 32 33 similar size, and/or are members of the same functional group, will often show 33 34 negative associations. Conversely, local communities will tend to be com- 34 35 prised of species that often have positive associations and are in different 35 36 genera, functional groups, and body size categories. 36 37 37 38 38 39 39 40 40 Conceptual and statistical issues 89

1 Table 3.4. Representation of 201 sites and 29 species of granivorous rodents 1 2 among deserts and habitats, in the revised data set 2 3 3 4 All habitats Desert shrub Desert grassland Sand dunes Shrubsteppe 4 5 All deserts 201,29 136,27 29,15 20,14 16,6 5 6 Great Basin 56,14 21,13 5,3 14,13 16,6 6 7 Mojave 52,14 49,14 3,5 7 Sonoran 48,14 45,14 3,1 8 Chihuahuan 45,17 21,14 24,14 8 9 9 10 Data from Brown and Kurzius (1987), as modified in Morton et al. (1994). 10 11 11 12 12 The processes: evaluating alternative mechanisms 13 13 14 Statistical, phylogenetic, and functional relationships 14 15 The previous section makes clear something that should be obvious by now: 15 16 the various patterns documented above are not independent of each other. 16 17 Species in the same genus should be closely related phylogenetically, and 17 18 they would be expected to be more similar in body size and other morpho- 18 19 logical characteristics and in ecological attributes than more distantly related 19 20 species. A general feature of desert rodent community assembly therefore, is 20 21 that species that are similar by some measures tend also to be similar in other 21 22 ways, and similar species tend to overlap in geographic ranges and to coexist 22 23 in local communities less frequently than more different species – and less 23 24 frequently than expected by chance. The fact that assemblages tend to be 24 25 formed of well-differentiated species, presumably with correspondingly dis- 25 26 tinctive requirements for resources (including macro- and microhabitats), 26 27 contributes to the patterns of nested subset structure and species diversity. 27 28 It is trendy to try to pull out the component of similarity in attributes of 28 29 species that can be attributed to phylogenetic relationships and to suggest that 29 30 the remaining similarities and differences reflect ‘ecology’. We have reserva- 30 31 tions about these approaches in general, and in particular about their utility 31 32 for disentangling the correlations among those traits that are important in 32 33 community assembly. Removing effects of phylogenetic relationships may 33 34 simultaneously remove effects of ecological similarity. While it is possible 34 35 to identify correlations between phylogenetic affinity and similarity in many 35 36 traits, it is much more difficult to attribute causality. (Do the powerful kidneys 36 37 shared by most heteromyid rodents reflect the constraint of common ancestry 37 38 or the selection of a common arid environment?) There are, however, some 38 39 cases where phylogenetic analysis could contribute importantly to under- 39 40 standing community assembly. This topic will be returned to below. 40 1 1

2 of 2 dune 3.51 3 3.57 3 − − 4 4

5 ree simulations 5 6 6

7 200 cells, the analyses 7 8 × 8 9 9 4.02 12.27°; 10 3.45 12.27°; 10 − − 11 11 12 12 13 13 14 14 15 15 16 16 3.33 21.25°; 0.83 25.08°; 3.91 0 − − 17 − 17 18 18 19 19 20 20 21 21 22 22 23 23 3.61 13.42°; 5.14 6.88°; 6.02 42.45°; + 1.1 6.02 0°; 0 11.66 16.06°; − − − −

24 − 24 25 25 26 26 27 27 28 28 Patterns of community structure in 201 sites arid North America 29 29 30 30 5.97 22.92°; 9.27 14.21°; 6.16 20.47°; 5.68 13.96°; 13.97* 14.96°; − − − − 31 − 31 Table 3.5. 32 32

33 All habitats Desert shrub Desert grassland Sand dunes Shrub-steppe 33 34 34 35 35 36 36

37 = 0.0864) and Sonoran sand dune rodents, for which the sample size precluded any statistical power. 37 38 P 38 39 39 All deserts 15.37°*; in which a different random site was removed from the data set. Presented are system temperatures, followed by the number standard deviations between observed temperatures and mean 1000 simulated systems. All analyses were significantly more structured than systems with the exception of Mojave sand for all habitats required the removal of one site analysis. The results presented this analysis are mean values from th rodents ( Chihuahuan*Because the nestedness temperature calculator (Patterson & Atmar, 1995) is limited to arrays no larger than 200 16.87°; Great Basin 12.52°; Mojave 21.26°; 40 Sonoran 15.14°; 40 Conceptual and statistical issues 91 1 It is also important to recognize that similar patterns of community struc- 1 2 ture characterize desert rodent assemblages across a range of spatial scales. 2 3 The same differences among species that are apparent at the level of commu- 3 4 nities of species coexisting within a few hectares of relatively homogeneous 4 5 habitat also tend to characterize the desert rodent faunas that inhabit much 5 6 larger geographic regions, areas of millions of square kilometers. From an 6 7 analytical perspective, this means that, while local communities represent non- 7 8 random assembly of species from the regional species pool, the regional pool 8 9 is itself a non-random assemblage of species. This is presumably because 9 10 local ecological processes influence the distributions of species on larger 10 11 spatial scales and thereby affect the composition of regional and geographic 11 12 species pools, and vice versa. More about this later. 12 13 13 14 14 15 Causal mechanisms: interspecific competition and/or allopatric speciation 15 16 So far, assembly rules have been discussed simply as empirical patterns. The 16 17 patterns have been tested against appropriate null hypotheses and shown to 17 18 require some kind of non-random assembly. As stated above, however, none 18 19 of the rules assumes the operation of interspecific competition or any other 19 20 mechanistic process. This seems the most logical and rigorous way to proceed. 20 21 Having demonstrated the patterns, the possible mechanisms can now be 21 22 evaluated as alternative, but not necessarily mutually exclusive, hypotheses. 22 23 The most obvious mechanism is interspecific competition. If competition 23 24 can be assumed to be most severe between species that are closely related 24 25 and similar in morphology, physiology, and behavior, and therefore in ecol- 25 26 ogy, then it provides a single, parsimonious explanation for nearly all of 26 27 the assembly rules described above. This includes patterns of nestedness. If 27 28 competition is greatest among similar species then species-poor communities 28 29 will be comprised largely of dissimilar species. As local diversity increases, 29 30 however, more similar species are able to coexist, and a nested pattern 30 31 emerges. On the other hand, if competition causes a checkerboard or similar 31 32 pattern of species co-occurrences, then a nested structure is not likely. In the 32 33 data set on North American desert rodents, the former scenario is true, and 33 34 competition apparently contributes to a nested community structure. 34 35 Further, the competition hypothesis does not require attempts to disentangle 35 36 the correlations among different kinds of ‘similarity’, because similarity in 36 37 any or all of the above characteristics of species (e.g., body size, phylogenetic 37 38 relationship, functional group) can be expected to increase competition and 38 39 decrease the probability of facilitated coexistence. 39 40 There is abundant experimental evidence that desert rodent species do 40 92 D.A. Kelt and J.H. Brown 1 compete (e.g., Rosenzweig, 1973; Schroder & Rosenzweig, 1975; Munger & 1 2 Brown, 1981; Freeman & Lemen, 1983; Frye, 1983; Brown & Munger, 1985; 2 3 J.S. Brown, 1989; Heske et al., 1994; Valone & Brown, 1996), and even that 3 4 competition influences community assembly by affecting probabilities of local 4 5 colonization and extinction (Valone & Brown, 1995). Thus, more recent 5 6 evidence obtained by manipulating local communities provides independent 6 7 verification of the mechanism hypothesized by earlier investigators to explain 7 8 comparative geographic patterns of community structure. 8 9 Competition is not the only mechanism that might be invoked to explain 9 10 the above patterns of faunal assembly. Another process that could explain 10 11 most of the above assembly rules is allopatric speciation. If the assumption 11 12 is made that species form in geographic isolation, that they shift their geo- 12 13 graphic distributions slowly, and diverge over time in their morphological, 13 14 physiological, and behavioral characteristics, then it follows that the most 14 15 closely related and similar species will tend not to overlap in their geographic 15 16 ranges and therefore not to coexist in local communities. Such a legacy of 16 17 allopatric speciation could account for the considerable structure observed at 17 18 a geographic scale, in the composition of regional species pools. It is hard to 18 19 see, however, how it could account for the non-random assembly of local 19 20 communities from these regional pools. 20 21 Just because it is not sufficient to account for all observed community struc- 21 22 ture, allopatric speciation should not be dismissed as an important contributing 22 23 mechanism. If it does play a significant role, we would predict that the most 23 24 closely related species are the least likely to occur together. While the negative 24 25 associations among congeneric species pairs supports this prediction, a proper 25 26 phylogenetic analysis would provide much more resolution. In particular, 26 27 while it would be generally expected that closest relatives would be most 27 28 similar in the morphological and functional characteristics that are the basis 28 29 of the above assembly rules, this need not always be true. Indeed, current 29 30 reconstructions of phylogenetic relationships among the heteromyid rodents 30 31 suggest that some species of similar body size and in the same functional 31 32 group are not particularly closely related. For example, the large kangaroo 32 33 rats (Dipodomys spp. with body mass > 100 g) which have nearly perfectly 33 34 contiguous and allopatric geographic ranges (Bowers & Brown, 1982), appear 34 35 to be more closely related to smaller kangaroo rats, with which they occur 35 36 sympatrically and coexist in local communities, than they are to each other 36 37 (Patton & Rogers, 1993). Bowers and Brown (1982) found that sister species 37 38 accounted for only a modest proportion of the pairs of species of similar body 38 39 size that showed negative associations in both overlap of geographic ranges 39 40 and frequency of local coexistence. 40 Conceptual and statistical issues 93 1 It is important to note that the allopatric speciation hypothesis is not 1 2 mutually exclusive of, and may even be complementary to, the competition 2 3 hypothesis. Everything else being equal, the most closely related species are 3 4 likely to be most similar in form and function, and hence likely to be strong 4 5 competitors. Thus, it is likely that competitive exclusion among close relatives 5 6 plays a major role, not only in restricting membership in local communities, 6 7 but also in limiting overlap in geographic ranges. This is one way that local 7 8 ecological interactions can affect the composition of regional species pools. 8 9 9 10 10 11 Causal mechanisms: ecological sorting and/or evolutionary character 11 12 displacement 12 13 A related mechanistic issue concerns the degree to which the above assembly 13 14 rules reflect some evolutionary character displacement as opposed to ecolog- 14 15 ical sorting. To what extent does the tendency of coexisting species to be 15 16 more different than expected by chance reflect adaptive divergence resulting 16 17 from, and perhaps serving to promote, coexistence? And, to what extent does 17 18 it simply reflect the ecological compatibility – the differential ability to 18 19 colonize and persist in local communities – of those species whose charac- 19 20 teristics were shaped primarily by other selective pressures. This question 20 21 warrants further attention. Brown (1975) suggested that intraspecific geographic 21 22 variation in body sizes of widely distributed species may reflect character 22 23 displacement due to local interactions with other species. However, analysis 23 24 of geographic variation in the widely distributed kangaroo rat, Dipodomys 24 25 merriami, suggested that body size was correlated with environmental prod- 25 26 uctivity but not related to geographic overlap or local coexistence with other 26 27 kangaroo rat species (J.H. Brown, unpublished data). 27 28 The issue of character displacement has been revived by Dayan and 28 29 Simberloff’s (1994a) demonstration of extremely uniform ratios of certain 29 30 morphological traits, most notably width of incisors in rodents, within regional 30 31 species pools. Since only a subset of the species in these pools coexist in any 31 32 local community, and the exact combination of species varies depending on 32 33 habitat (see above), what mechanism can account for the uniform ratios? Since 33 34 Dayan and Simberloff analyzed several morphological traits in only two 34 35 species assemblages, it is possible that the near perfectly uniform ratios in 35 36 one trait, incisor width, are due to chance. If not, then the most likely expla- 36 37 nation seems to be some kind of ‘diffuse character displacement’ as they 37 38 suggest. If incisor width plays a major role in resource use and strongly affects 38 39 interspecific competition, then having a substantial and uniform ratio of 39 40 incisor widths among all the species in the regional pool would insure that 40 94 D.A. Kelt and J.H. Brown 1 all combinations of locally coexisting species assembled from that pool would 1 2 differ by at least that critical amount. This is certainly a plausible mechanism, 2 3 and it suggests another way in which local interactions can ‘feed up’ to affect 3 4 the composition of the regional species pool. While both evolutionary char- 4 5 acter displacement and ecological sorting could play a role in structuring the 5 6 differences among the species in the regional pool, it is hard to imagine 6 7 ecological sorting alone producing such precisely uniform ratios. 7 8 Dayan and Simberloff’s findings also raise interesting questions about the 8 9 statistical correlations and functional relationships among different charac- 9 10 teristics of co-occurring species. Since the dimensions of most morphologi- 10 11 cal traits are correlated with each other and with overall body size, we need 11 12 to determine which ones most strongly influence different kinds of ecologi- 12 13 cal interactions and thereby affect coexistence, and which ones are most 13 14 subject to selection for character displacement or some other evolutionary 14 15 trend? There is abundant observational and experimental evidence that in 15 16 desert rodents both resource use, which is important in exploitative compe- 16 17 tition, and aggressive dominance, which is critical in interference competi- 17 18 tion, are correlated with body size (e.g., Rosenzweig & Sterner, 1970; Brown 18 19 & Lieberman, 1973; Brown, 1975; M’Closkey, 1976; Frye, 1983; Price, 1983). 19 20 Thus, it is perhaps not surprising that body size appears to play such a large 20 21 role in the assembly of local communities. This does not mean, however, that 21 22 other traits, such as incisor width and cheek pouch volume, that are corre- 22 23 lated with body size, do not also affect the outcome of ecological interactions 23 24 and may be subject to selection. And certainly, traits such as mode of loco- 24 25 motion (e.g., bipedal vs. quadrupedal, see functional groups above), that are 25 26 not closely correlated with body size, also affect microhabitat and resource use 26 27 and thereby strongly influence interspecific interactions and coexistence. 27 28 28 29 29 Discussion and synthesis 30 30 31 Statistical and methodological issues: distinguishing pattern from 31 32 randomness and characterizing assembly rules 32 33 The long debate over the existence of community structure and the validity 33 34 of various assembly rules (e.g., Strong et al., 1984; Diamond & Case, 1986; 34 35 Gotelli & Graves, 1995) testifies to the critical importance of rigorous 35 36 statistical methods. If an assembly rule describes a nonrandom pattern of 36 37 community organization, then it is important to test and reject the null hypoth- 37 38 esis that the apparent organization is not simply a random assemblage of 38 39 species. The human senses and brain are so attuned to perceive pattern, that 39 40 they may perceive it when it is not there. While it might seem a simple matter 40 Conceptual and statistical issues 95 1 to erect and test null hypotheses, there are many statistical and conceptual 1 2 issues that must be addressed. These are considered in detail in the excellent 2 3 recent book by Gotelli and Graves (1995). Some of these issues will be 3 4 touched upon here only insofar as they are particularly well illustrated by 4 5 studies of assembly rules in desert rodents. 5 6 6 7 There are many possible ‘null’ or alternative hypotheses 7 8 As any statistician will be quick to point out, the question of whether some 8 9 empirical data differ from a random distribution must be qualified by how 9 10 one defines random. Most tests of null hypotheses in community ecology 10 11 involve comparison of an observed distribution of values either with a 11 12 particular analytical random distribution or with a randomized distribution of 12 13 data compiled by computer simulation. An example of the former is Dayan 13 14 and Simberloff’s (1994) test for non-randon morphological ratios by com- 14 15 parison with the Barton and David statistic. An example of the latter is Fox 15 16 and Brown’s (1993) test for Fox’s assembly rule by randomly drawing species 16 17 from the pool to estimate the expected probability of favored and unfavored 17 18 states. Both approaches are equally valid, but the applications of the former are 18 19 limited to cases when a particular analytical form of the random hypothesis 19 20 can be assumed. 20 21 The issue of ‘random with respect to what?’ also means that an observed 21 22 empirical distribution can potentially be compared to multiple random dis- 22 23 tributions. This is a problem in analytical statistical tests, because different 23 24 ‘random’ distributions can be expected (e.g., normal, lognormal, negative 24 25 binomial, and broken stick, to name only a few) and different test statistics 25 26 can be applied (e.g., different parametric and nonparametric tests, each of 26 27 which of have unique assumptions). To reiterate using the above example, 27 28 Dayan and Simberloff (1994) assume that the null distribution of morpho- 28 29 logical traits is equiprobable on a logarithmic scale, and they use the Barton– 29 30 David test to compare observed and expected null distributions. This issue is 30 31 also a problem with randomization tests, because different sets of data can 31 32 potentially be randomized (e.g., different species pools can be assumed) and 32 33 different kinds of random draws can be taken from the pool (e.g., species or 33 34 their characteristics can be drawn with or without replacement, and with equal 34 35 frequencies or with frequencies intentionally biased to account for the internal 35 36 structure of the data, e.g., proportional to the number of sites inhabited by 36 37 each species or to the total number of individuals in the sample). 37 38 Related issues concern the particular question to be addressed and the 38 39 design of the study, and include: the spatial (and temporal) scale of sampling; 39 40 the level of structure that may be present at larger (or smaller) scales; and 40 96 D.A. Kelt and J.H. Brown 1 the kinds of random processes that are assumed. In randomization tests a crit- 1 2 ical step is determining the appropriate species pool. There is no single ‘right’ 2 3 answer to this question, but the null hypothesis can often be accepted or 3 4 rejected depending on what choice is made. For example, local communities 4 5 of desert rodents are more likely to appear to be nonrandom assemblages if 5 6 they are compared to some global rather than a more regional species pool, 6 7 because, as we have shown, the regional pools are themselves already ‘struc- 7 8 tured’ with respect to body sizes, congeners, and functional groups. Stone et 8 9 al. (this volume) would apply very restrictive criteria in testing for structure 9 10 of local communities, taking into account both the geographic ranges and the 10 11 relative frequency of local occurrence of species. There is nothing wrong with 11 12 this, but it deliberately ignores a high degree of structure that is already present 12 13 in the assemblages. This is not necessarily the only or most appropriate way 13 14 to evaluate the hypothesis that the communities are structured by interspecific 14 15 competition. It has been shown above how local competitive interactions can 15 16 ‘feed up’ to affect the composition of species pools at larger scales, and this 16 17 may strongly bias such restrictive tests. 17 18 18 19 Tests differ in statistical power 19 20 The power of tests of the null hypothesis depend on the nature of the test 20 21 itself (see above) and the sample sizes. Applying different tests to the same 21 22 data sets can give different apparent ‘answers’. For example, Simberloff and 22 23 Boecklen (1981; see also Dayan & Simberloff, 1994) applied the 23 24 Barton–David test to the body size distributions in several of the rodent 24 25 communities reported in Bowers and Brown (1982) and found that only a 25 26 few had more uniform size ratios than expected by chance. The Barton–Davis 26 27 test must be applied to communities one at a time, and its power to reject the 27 28 null hypothesis is low, especially if the number of coexisting species is rel- 28 29 atively small. Rejection when using the Barton–David test, however, usually 29 30 implies a highly non-random distribution. A different approach is to test for 30 31 a repeated non-random pattern of body size distributions across a large sample 31 32 of different communities (e.g., Bowers & Brown, 1982; Hopf et al., 1993; 32 33 this chapter). This method has a much greater power to detect distributions 33 34 that are less conspicuously non-random. Again, neither approach is ‘right’ or 34 35 ‘wrong’: they test for structure at different levels of organization (within 35 36 single communities vs. across multiple communities) with correspondingly 36 37 different power. However, in the aftermath of all the debate over assembly 37 38 rules, it is important to remember that the power of statistical tests vary, so 38 39 that failure to reject one null hypothesis does not necessarily mean that an 39 40 observed community is a random assemblage. 40 Conceptual and statistical issues 97 1 1 2 Many of the measures of community structure may not be independent 2 3 This is a problem at several levels. As pointed out above, there are statistical 3 4 correlations and functional relationships among many of the characteristics 4 5 used to assess community structure, e.g., body size, incisor width, functional 5 6 group, and phylogenetic relationship. This means that it is difficult to deter- 6 7 mine which trait or combination of traits provide the mechanistic basis for 7 8 the ecological sorting and evolutionary divergence that has produced the 8 9 observed structure. Probably the way to address this problem is to develop 9 10 alternative, mutually exclusive mechanistic hypotheses based on the operation 10 11 of a specific mechanism (not an easy task) and then to subject these hypoth- 11 12 eses to rigorous independent tests, using experimental methods if possible. 12 13 A second problem is that the composition of communities and the charac- 13 14 teristics of the environments in which the assemblages occur are typically 14 15 autocorrelated over both space and time. How far apart in space or distant in 15 16 time must samples be taken to insure that they are statistically independent? 16 17 A related and intertwined problem is that the same single species and 17 18 combinations of multiple species typically occur repeatedly, even in the most 18 19 widely separated samples. Thus, for example, the kangaroo rat, D. merriami, 19 20 and the deer mouse, P. maniculatus, are habitat generalists with very large 20 21 geographic ranges that occur together in many communities throughout the 21 22 southwestern desert region. How should we deal with such redundancy? The 22 23 literature of statistics and community ecology offers little guidance for 23 24 handling these problems. On the one hand, such repeated occurrences prob- 24 25 ably should not be regarded as statistically independent events, but on the 25 26 other hand, the multiple co-occurrences of certain pairs or other combinations 26 27 of species almost certainly conveys important information about their inter- 27 28 actions and compatibility, and hence about community assembly. These 28 29 difficult issues warrant attention from both statisticians and community ecol- 29 30 ogists. For the moment, we suggest that multivariate and computer simulation 30 31 techniques offers some promise for disentangling the complex patterns of 31 32 covariation. 32 33 33 34 34 35 Conceptual issues: pursuing the relationship between pattern and process 35 36 We reiterate that assembly rules are here defined to be empirical patterns 36 37 of community structure. Their real value is that they provide evidence of 37 38 interactions among species and between species and their environment that 38 39 determine the composition of assemblages. We have gone beyond the 39 40 stage of asking whether North American desert rodent communities exhibit 40 98 D.A. Kelt and J.H. Brown 1 non-random structure. With so many papers demonstrating one kind of assem- 1 2 bly rule or another, to the extent that there is still some controversy, it tends 2 3 to center around details of methodology (above) and problems of inferring 3 4 process from pattern. 4 5 So, what are the interactions and how do they operate to structure local 5 6 communities of desert rodents? The vast majority of the above assembly rules 6 7 indicate that the species that coexist are more different than expected by 7 8 chance; species in local communities are nonrandom samples from the 8 9 regional species pool, and the species in the regional pools are nonrandom 9 10 samples from the larger pool of species inhabiting the entire southwestern 10 11 desert region. The first pattern seems uniquely consistent with a process of 11 12 interspecific competition, whereas the second is consistent with mechanisms 12 13 of both competition and allopatric speciation. And the experimental evidence 13 14 for interspecific competition is overwhelming (e.g., Brown & Harney, 1993 14 15 and references therein; see also Heske et al., 1994: Valone & Brown, 1995, 15 16 1996). Recently, phylogenetic and biogeographic analyzes have provided evi- 16 17 dence of long-lasting legacies of allopatric speciation, which often occurred 17 18 when desert basins were isolated by geological events and/or climatic changes 18 19 (e.g., Riddle, 1995, 1996). Requiring further investigation, however, is the 19 20 relationship between phylogenetic, biogeographic, and ecological relation- 20 21 ships. It is possible, even likely, that interspecific competition plays a major 21 22 role in preventing the products of recent speciation events from expanding 22 23 their distributions to overlap in their geographic ranges and coexist in local 23 24 communities. 24 25 The assembly rules raise several additional questions about the process of 25 26 interspecific competition in desert rodents. For one thing, many of the assem- 26 27 bly rules describe probabilistic, rather than absolute, patterns of co-occurrence, 27 28 i.e., species in the same body size category, genus, and functional group occur 28 29 together less frequently than expected by chance, but they sometimes do co- 29 30 exist not only in the same geographic region but also in the same local 30 31 community. For example, in a clear exception to several of the above assem- 31 32 bly rules, the closely related and morphologically and functionally similar 32 33 kangaroo rats, D. merriami and D. ordii, have occurred together nearly 33 34 continuously at Brown’s experimental study site for the last 19 years (see 34 35 Brown & Heske, 1990). This does not mean that the rules are invalid or that 35 36 the two kangaroo rats do not compete. Indeed, it is easy to imagine that, even 36 37 though such species might usually be the strongest competitors, in certain 37 38 environments their relative competitive abilities would be almost evenly 38 39 matched. In such cases they could coexist almost indefinitely, especially if 39 40 the presence of each species were reinforced, either by immigration from 40 Conceptual and statistical issues 99 1 adjacent habitats where each is the superior competitor (e.g., Schroder, 1987) 1 2 or by temporal environmental shifts which alternately favor each species (e.g., 2 3 Chesson, 1986). 3 4 A more telling question concerns the generality of assembly rules. Do the 4 5 rules derived for North American desert rodents also apply to rodents in other 5 6 deserts or to other kinds of organisms? While more evidence would be helpful, 6 7 that which is available suggests that the generality is very limited. In partic- 7 8 ular, what is now known about the rodent communities inhabiting other deserts 8 9 throughout the world suggests that they are structured quite differently (Kelt 9 10 et al., 1996). There seem to be several reasons for this. First, while differences 10 11 among species may reduce interspecific competition and facilitate coexis- 11 12 tence, similarities among species may facilitate their occurrence in the same 12 13 local environments. Thus, for example, the extremely arid open deserts of 13 14 central Asia support diverse rodent communities that often contain species of 14 15 similar body size and bipedal locomotion (e.g., Shenbrot et al., 1994). Second, 15 16 different kinds of characteristics may play a role in the existence and coex- 16 17 istence of rodents in different deserts. Thus, the majority of North American 17 18 desert rodents are granivores, and coexisting species tend to differ in body 18 19 size and locomotor mode. In other deserts (Asia, Australia, South Africa) 19 20 there are few granivorous rodents, and coexistence of species appears to 20 21 depend much more on differences in diet and less on differences in body size 21 22 and locomotion (Kelt et al., 1996). Finally, the other deserts have quite dif- 22 23 ferent physical settings and geological and climatic histories. These abiotic and 23 24 historical factors have influenced the patterns of evolutionary differentiation 24 25 and ecological segregation of rodent assemblages. They are reflected not only 25 26 in the trophic differences just mentioned, but also in different patterns of mor- 26 27 phological and physiological specialization. For example, while only two gen- 27 28 era of North American desert rodents (Dipodomys and Microdipodops) are 28 29 bipedal, the more ancient and more extensive central Asian deserts contain 29 30 11 bipedal genera (Alactodipus, Allactaga, Cardiocranius, Dipus, Eremodipus, 30 31 Euchoreutes, Jaculus, Paradipus, Pygeretmus, Salpingotus, Stylodipus), and 31 32 most of them show a much higher degree of morphological specialization for 32 33 bipedal, saltatory locomotion on both hard and sandy substrates. Moreover, 33 34 most of the Old World bipedal taxa are not granivorous (Kelt et al., 1996). 34 35 Clearly, different functional groups of rodents occur in the different deserts 35 36 (e.g., there are no strict quadrupedal or bipedal granivores in Australia), and 36 37 when similar functional groups do occur, they tend to be represented by quite 37 38 different numbers of species and genera (e.g., bipedal granivores: 19–20 38 39 species of Dipodomys and two of Microdipodops in North America (Patton, 39 40 1993; Williams et al., 1993), compared to just one species of Cardiocranius, 40 100 D.A. Kelt and J.H. Brown 1 four species of Jaculus (whose diet is both granivorous and folivorous 1 2 (G. Shenbrot and K. Rogovin, personal communication)), and six species of 2 3 Salpingotus in the Middle Eastern and North African deserts (Holden, 1993; 3 4 Mares, 1993)). Thus, while Fox’s (1987, 1989, Fox & Brown 1993; this chap- 4 5 ter) approach of analyzing representation of different functional groups might 5 6 be applied in other deserts, different functional groups defined on the basis 6 7 of different criteria would have to be specified for Fox’s assembly rule to 7 8 hold. 8 9 It has been more than two decades since Diamond’s (1975) seminal paper 9 10 on community assembly rules and Rosenzweig and Winakur’s (1969) pio- 10 11 neering study of community structure in desert rodents. Despite a good deal 11 12 of work on desert rodent community ecology within the last decade, how- 12 13 ever, most of the studies have been experimental and few have been the kinds 13 14 of non-manipulative geographic comparisons that led to the formulation of 14 15 assembly rules. This emphasis seems to reflect the severe limitations of the 15 16 assembly rule approach to community ecology: difficulties in inferring process 16 17 from pattern, beyond the straightforward evidence for some form of inter- 17 18 specific competition, and lack of any broad generality in the nature and applic- 18 19 ability of the rules, despite the widespread occurrence of the predominant 19 20 underlying mechanism, interspecific competition. If there are truly general 20 21 patterns and processes that characterize the organization of communities, these 21 22 do not appear to be reflected in any very direct way in the well-documented 22 23 assembly rules that describe the structure of North American desert rodent 23 24 communities. 24 25 25 26 26 27 Acknowledgments 27 28 We thank Evan Weiher and Paul Keddy for inviting us to participate in the 28 29 symposium. Comments by Mark Lomolino and an anonymous reviewer 29 30 helped to improve the presentation. This work was partially supported by 30 31 National Science Foundation grant DEB-9553623 to J. H. Brown. 31 32 32 33 33 34 References 34 35 Atmar, W. & Patterson, B.D. (1993). The measure of order and disorder in the 35 36 distribution of species in fragmented habitat. Oecologia 96: 373–382. 36 Atmar, W. & Patterson, B.D. (1995). The nestedness temperature calculator: a 37 visual basic program, including 294 presence–absence matrices. AICS 37 38 Research, Inc., University Park, NM, and The Field Museum, Chicago, IL. 38 39 Bowers, M.A. & Brown, J.H. (1982). Body size and coexistence in desert rodents: 39 chance or community structure. Ecology 63: 391–400. 40 40 Conceptual and statistical issues 101

1 Brown, J.H. (1973). Species diversity of seed-eating desert rodents in sand dune 1 2 habitats. Ecology 54: 775–787. 2 Brown, J.H. (1975). Geographical ecology of desert rodents. In Ecology and 3 Evolution of Communities, ed. M.L. Cody & J.M. Diamond, pp. 315–341. 3 4 Cambridge, MA: Belknap Press. 4 5 Brown, J.H. & Lieberman, G.A. (1973). Resource utilization and coexistence of 5 6 seed-eating desert rodents in sand dune habitats. Ecology 54: 788–797. 6 Brown, J.H. & Munger, J.C. (1985). Experimental manipulation of a desert rodent 7 community: food addition and species removal. Ecology 66: 1545–1563. 7 8 Brown, J.H. & Kurzius, M.A. (1987). Composition of desert rodent faunas: 8 9 combinations of coexisting species. Annales Zoologici Fennici 24: 227–237. 9 Brown, J.H. & Heske, E.J. (1990). Temporal changes in a Chihuahuan Desert 10 rodent community. Oikos 59: 290–302. 10 11 Brown, J.H. & Harney, B.A. (1994). Population and community ecology of 11 12 heteromyid rodents in temperate habitats. In Biology of the Heteromyidae, ed. 12 H.H. Genoways, & J.H. Brown, Special Publication 10, pp. 618–651. 13 American Society of Mammalogists. 13 14 Brown, J.H. & Reichman, O.J. & Davidson, D.W. (1979). Granivory in desert 14 15 ecosystems. Annual Reviews of Ecology and Systematics 10: 201–227. 15 16 Brown, J.S. (1989). Coexistence on a seasonal resource. American Naturalist 133: 16 168–182. 17 Chesson, P.L. (1986). Environmental variation and the coexistence of species. In 17 18 Community Ecology, ed. J. Diamond, & T.J. Case, pp. 240–256. NY: Harper 18 19 and Row. 19 Cody, M.L. & Diamond, J. (eds). (1975). Ecology and Evolution of Communities. 20 Cambridge, MA: Belknap Press. 20 21 Cohen, J.E., R.A. Beaver, S.H. Cousins, D.L. DeAngelis, L. Goldwasser, K.L. 21 22 Heong, R.D. Holt, A.J. Kohn, L.H. Lawton, N. Martinez, R. O’Malley, 22 L.M. Page, B.C. Patten, S.L. Pimm, G.A. Polis, M. Rejmánek, 23 T.W. Schoener, K. Schoenly, W.G. Sprules, J.M. Teal, R.E. Ulanowicz, 23 24 P.H. Warren, H.M. Wilbur, and P. Yodzis. (1993). Improving food webs. 24 25 Ecology 74: 252–258. 25 26 Connell, J. (1961a). Effects of competition, predation by Thais lapillus and other 26 factors on natural populations of the barnacle Balanus balanoides. Ecological 27 Monographs 31: 61–104. 27 28 Connell, J. (1961b). The influence of interspecific competition and other factors on 28 29 the distribution of the barnacle Chthalamus stellatus. Ecology 42: 710–723. 29 Connor, E.F. & Simberloff, D. (1979). The assembly of species communities: 30 chance or competition? Ecology 60: 1132–1140. 30 31 Connor, E.F. & Simberloff, D. (1983). Interspecific competition and species co- 31 32 occurrence patterns on islands: null models and the evaluation of evidence. 32 Oikos 41: 455–465. 33 Connor, E.F. & Simberloff, D. (1984). Neutral models of species’ co-occurrence 33 34 patterns. In Ecological Communities: Conceptual Issues and the Evidence, ed. 34 35 D.R. Strong, Jr., D. Simberloff, L.G. Abele, & A.B. Thistle, pp. 317–331. 35 36 Princeton, NJ: Princeton University Press. 36 Darwin, C. (1859). The Origin of Species by Means of Natural Selection. London: 37 Murray. 37 38 Dayan, T. & Simberloff, D. (1994). Morphological relationships among coexisting 38 39 heteromyids: an incisive dental character. American Naturalist 143: 462–477. 39 Diamond, J.M. (1975). Assembly of species communities. In Ecology and 40 40 102 D.A. Kelt and J.H. Brown

1 Evolution of Communities, ed. M.L. Cody & J.M. Diamond, pp. 342–444. 1 2 Cambridge, MA: Harvard University Press. 2 Diamond, J.M. (1984). Distributions of New Zealand birds on real and virtual 3 islands. New Zealand Journal of Ecology 7: 37–55. 3 4 Diamond, J. & Case, T. (eds.) (1986). Community Ecology. NY: Harper and Row. 4 5 Diamond, J.M. & Gilpin, M.E. (1982). Examination of the ‘null’ model of Connor 5 6 and Simberloff for species co-occurrences on islands. Oecologia 52: 64–74. 6 Fox, B.J. (1987). Species assembly and the evolution of community structure. 7 Evolutionary Ecology 1: 201–213. 7 8 Fox, B.J. (1989). Small-mammal community pattern in Australian heathland: a 8 9 taxonomically based rule for species assembly. In Patterns in the Structure of 9 Mammalian Communities, ed. D.W. Morris, Z. Abramsky, B.J. Fox, & M.R. 10 Willig, Special Publication 28, pp. 91–103. Lubbock, TX: The Museum, Texas 10 11 Tech University. 11 12 Fox, B.J. & Brown, J.H. (1993). Assembly rules for functional groups in North 12 American desert rodent communities. Oikos 67: 358–370. 13 Fox, B.J. & Brown, J.H. (1995). Reaffirming the validity of assembly rules for 13 14 functional groups or guilds: a reply to Wilson. Oikos 73: 125–132. 14 15 Freeman, P.W. & Lemen, C. (1983). Quantification of competition among 15 16 coexisting heteromyids in the southwest. Southwestern Naturalist 28: 41–46. 16 Frye, R.J. (1983). Experimental field evidence of interspecific aggression between 17 two species of kangaroo rat (Dipodomys). Oecologia 59: 74–78. 17 18 Genoways, H.H. & Brown, J.H. (eds.) (1993). Biology of the Heteromyidae. Special 18 19 Publication 10, The American Society of Mammalogists. 19 Gilpin, M.E. & Diamond, J.M. (1981). Immigration and extinction probabilities for 20 individual species: relation to incidence functions and species colonization 20 21 curves. Proceedings of the National Academy of Sciences, USA 78: 392–396. 21 22 Gilpin, M.E. & Diamond, J.M. (1982). Factors contributing to non-randomness in 22 species co-occurrences on islands. Oecologia 52: 75–84. 23 Gilpin, M.E. & Diamond, J.M. (1984). Are species co-occurrences on islands non- 23 24 random, and are null hypotheses useful in community ecology? In Ecological 24 25 Communities: Conceptual Issues and the Evidence, ed. D.R., Jr., Strong, D. 25 26 Simberloff, L.G. Abele, & A.B. Thistle, pp. 297–315. Princeton, NJ: Princeton 26 University Press. 27 Gotelli, N.J. & Graves, G.R. (1995). Null Models in Ecology. Washington, DC: 27 28 Smithsonian Institution Press. 28 29 Grant, P.R. (1986). Ecology and Evolution of Darwin’s Finches. Princeton, NJ: 29 Princeton University Press. 30 Grinnell, J. & Orr, R.T. (1934). Systematic review of the californicus group of the 30 31 rodent genus Peromyscus. Journal of Mammalogy 15: 210–220. 31 32 Hafner, M.S. (1977). Density and diversity in Mojave Desert rodent and shrub 32 communities. Journal of Animal Ecology 46: 925–938. 33 Hanski, I. (1982). Dynamics of regional distribution: the core and satellite 33 34 hypothesis. Oikos 38: 210–221. 34 35 Heske, E.J., Brown, J.H., & Mistry, S. (1994). Long-term experimental study of a 35 36 Chihuahuan desert rodent community: 13 years of competition. Ecology 75: 36 438–445. 37 Holden, M.E. (1993). Family Dipodidae. In Mammal Species of the World: A 37 38 Taxonomic and Geographic Reference, ed. D.E. Wilson, & D.M. Reeder, pp. 38 39 487–499. 2nd edn. Washington, DC: Smithsonian Institution Press. 39 Hopf, F.A., Valone, T.J., & Brown, J.H. (1993). Competition theory and the 40 structure of ecological communities. Evolutionary Ecology 7: 142–154. 40 Conceptual and statistical issues 103

1 Hutchinson, G.E. (1957). Concluding remarks. Cold Springs Harbor Symposium in 1 2 Quantitative Biology 22: 415–427. 2 Hutchinson, G.E. (1959). Homage to Santa Rosalia, or Why are there so many 3 kinds of animals? American Naturalist 93: 145–159. 3 4 Jorgensen, C.D. & Hayward, C.L. (1965). Mammals of the Nevada test site. 4 5 Brigham Young University Science Bulletin, Biology Series, VI: 1–81. 5 6 Kelt, D.A., Taper, M.L. & Meserve, P.L. (1995). Assessing the impact of 6 competition on community assembly: a case study using small mammals. 7 Ecology 76: 1283–1296. 7 8 Kelt, D.A., Brown, J.H., Heske, E.J., Marquet, P.A., Morton, S.R., Reid, J.R.W., 8 9 Rogovin, K.A., & Shenbrot, G. (1996). Community structure of desert small 9 mammals: comparisons across four continents. Ecology 77: 746–761. 10 Lack, D. (1947). Darwin’s Finches. Cambridge: Cambridge University Press. 10 11 MacArthur, R.H. (1958). Population ecology of some warblers on northeastern 11 12 coniferous forests. Ecology 39: 599–619. 12 Magurran, A.E. (1988). Ecological Diversity and its Measurement. Princeton, NJ: 13 Princeton University Press. 13 14 Mares, M.A. (1993). Heteromyids and their ecological counterparts: a pandesertic 14 15 view of rodent ecology and evolution. In Biology of the Heteromyidae, ed. 15 16 H.H. Genoways & J.H. Brown Special Publication 10, American Society of 16 Mammalogists. 17 M’Closkey, R.T. (1976). Community structure in sympatric rodents. Ecology 57: 17 18 728–739. 18 19 M’Closkey, R.T. (1978). Niche separation and assembly in four species of Sonoran 19 Desert rodents. American Naturalist 112: 683–694. 20 Morton, S.R., Brown, J.H., Kelt, D.A., & Reid, J.R.W. (1994). Comparisons of 20 21 community structure among small mammals of North American and Australian 21 22 Deserts. Australian Journal of Zoology 42: 501–525. 22 Munger, J.C. & Brown, J.H. (1981). Competition in desert rodents: an experiment 23 with semipermeable exclosures. Science 211: 510–512. 23 24 Owen, J.G. (1998). On productivity as a predictor of rodent and carnivore diversity. 24 25 Ecology 69: 1161–1165. 25 26 Paine, R.T. (1966). Food web complexity and species diversity. American 26 Naturalist 100: 65–75. 27 Patterson, B.D. (1987). The principle of nested subsets and its implications for 27 28 biological conservation. Conservation Biology 1: 323–334. 28 29 Patterson, B.D. (1990). On the temporal development of nested subset patterns of 29 species composition. Oikos 59: 330–342. 30 Patterson, B.D. & Atmar, W. (1986). Nested subsets and the structure of insular 30 31 mammalian faunas and archipelagos. In Island Biogeography of Mammals, ed. 31 32 L.R. Heaney & B.D. Patterson, pp. 65–82. NY: Academic Press. 32 Patterson, B.D. & Brown, J.H. (1991). Regionally nested patterns of species 33 composition in granivorous rodent assemblages. Journal of Biogeography 18: 33 34 395–402. 34 35 Patton, J.L. (1993). Family Geomyidae. In Mammal Species of the World: A 35 36 Taxonomic and Geographic Reference, ed. D.E. Wilson & D.M. Reeder, pp. 36 469–486, 2nd edn. Washington, DC: Smithsonian Institution Press. 37 Patton, J.L. & Rogers, D.S. (1993). Cytogenetics. Biology of the Heteromyidae. In 37 38 ed. H.H. Genoways & J.H. Brown, pp. 236–258. Special Publication 10, 38 39 American Society of Mammalogists. 39 Pielou, E.C. (1969). An Introduction to Mathematical Ecology. NY: Wiley- 40 Interscience. 40 104 D.A. Kelt and J.H. Brown

1 Polis, G.A. & Strong, D.R. (1996). Food web complexity and community 1 2 dynamics. American Naturalist 147: 813–846. 2 Price, M.V. (1978). The role of microhabitat in structuring desert rodent 3 communities. Ecology 59: 910–921. 3 4 Price, M.V. (1983). Ecological consequences of body size: a model for patch 4 5 choice in desert rodents. Oecologia 59: 384–392. 5 6 Reichman, O.J. & Brown, J.H. (eds.) (1983). Biology of desert rodents. Great 6 Basin Naturalist Memoirs 7. 7 Riddle, B.R. (1995). Molecular biogeography in the pocket mice (Perognathus and 7 8 Chaetodipus) and grasshopper mice (Onychomys): the late Cenozoic 8 9 development of a North American aridlands rodent guild. Journal of 9 Mammalogy 76: 283–301. 10 Riddle, B.R. (1996). The molecular phylogeographic bridge between deep and 10 11 shallow history in continental biotas. Trends in Ecology and Evolution 11: 11 12 207–211. 12 Rosenzweig, M.L. (1968). Net primary productivity of terrestrial communities: 13 prediction from climatological data. American Naturalist 102: 67–74. 13 14 Rosenzweig, M.L. (1973). Habitat selection experiments with a pair of coexisting 14 15 heteromyid rodent species. Ecology 54: 111–117. 15 16 Rosenzweig, M.L. & Winakur, J. (1969). Population ecology of desert rodent 16 communities: habitats and environmental complexity. Ecology 50: 558–572. 17 Rosenzweig, M.L. & Sterner, P.W. (1970). Population ecology of desert rodent 17 18 communities: body size and seed husking as bases for heteromyid coexistence. 18 19 Ecology 51: 217–224. 19 Rosenzweig, M.L., Smigel, B. & Kraft, A. (1975). Patterns of food, space and 20 diversity. In Rodents in Desert Environments, ed. I. Prakash & P.K. Ghosh, 20 21 pp. 241–268. The Hague: Dr W. Junk. 21 22 Roughgarden, J. (1995). Anolis Lizards of the Caribbean: Ecology, Evolution, and 22 Plate Tectonics. NY: Oxford University Press. 23 Schoener, T.W. (1969). Size patterns in West Indian Anolis lizards. I. Size and 23 24 species diversity. Systematic Zoology 18: 386–401. 24 25 Schoener, T.W. (1970). Size patterns in West Indian Anolis lizards. II. Correlations 25 26 with the sizes of particular sympatric species-displacement and convergence. 26 American Naturalist 104: 155–170. 27 Schroder, G.D. (1987). Mechanisms for coexistence among three species of 27 28 Dipodomys: habitat selection and an alternative. Ecology 68: 1071–1083. 28 29 Schroder, G.D. & Rosenzweig, M.L. (1975). Perturbation analysis of competition 29 and overlap in habitat utilization between Dipodomys ordii and Dipodomys 30 merriami. Oecologia 19:9–28. 30 31 Shenbrot, G.I., Rogovin, K.A., & Heske, E.J. (1994). Comparison of niche-packing 31 32 and community organization in Asia and North America. Australian Journal of 32 Zoology 42: 479–499. 33 Simberloff, D. & Boecklen, W. (1981). Santa Rosalia reconsidered: size ratios and 33 34 competition. Evolution 35: 1206–1228. 34 35 Simberloff, D. & Dayan, T. (1991). The guild concept and the structure of 35 36 ecological communities. Annual Review of Ecology and Evolution 22: 36 115–143. 37 Strong, D.E., Jr., Simberloff, D., Abele, L.G., & Thistle, A.B. (eds.). (1984). 37 38 Ecological Communities: Conceptual Issues and the Evidence. Princeton, NY: 38 39 Princeton University Press. 39 Thompson, S.D. (1982a). Structure and species composition of desert heteromyid 40 40 Conceptual and statistical issues 105

1 rodent species assemblages: effects of a simple habitat manipulation. Ecology 1 2 63: 1313–1321. 2 Thompson, S.D. (1982b). Microhabitat utilization and foraging behavior of bipedal 3 and quadrupedal heteromyid rodents. Ecology 63: 1303–1312. 3 4 Valone, T.J. & Brown, J.H. (1995). Effects of competition, colonization, and 4 5 extinction on rodent species diversity. Science 267: 880–883. 5 6 Valone, T.J. & Brown, J.H. (1996). Desert rodents: long-term responses to natural 6 changes and experimental perturbations. In Long-term Studies of Vertebrate 7 Communities, ed. M.L. Cody & J.A. Smallwood, pp. 555–583. Orlando, FL: 7 8 Academic Press. 8 9 Williams, D.F., Genoways, H.H. & Braun, J.K. (1993). Taxonomy and systematics. 9 In Biology of the Heteromyidae, ed. H.H. Genoways & J.H. Brown, Special 10 Publication 10, pp. 38–196. American Society of Mammalogists. 10 11 Wilson, J.B. (1995) Null models for assembly rules: the Jack Horner effect is more 11 12 insidious than the Narcissus effect. Oikos 72: 139–142. 12 13 13 14 14 15 Appendix 15 16 The null hypothesis of community assembly states that there is a much higher 16 17 probability that a species establishing itself in a local community will come 17 18 from a functional group that is not already represented at that site, until all 18 19 functional groups are equally represented, when the cycle repeats itself. Sites 19 20 that agree with this rule are considered to be in a ‘favored’ state, while those 20 21 that are not in agreement with the rule are considered ‘unfavored’. For exam- 21 22 ple, a site with two herbivores, two carnivores, and two omnivores is in agree- 22 23 ment with the model, and is in a favored state. Similarly, a site with three 23 24 herbivores, two carnivores, and three omnivores is a favored state. In contrast, 24 25 a site containing one herbivore, two carnivores, and three omnivores would 25 26 be considered unfavored, because the model predicts that a second herbivore 26 27 should have entered the community before a third omnivore entered. Of 27 28 course, the pool of species with access to a site may be biased towards or 28 29 against certain functional groups, in which case these sites will be predisposed 29 30 towards being unfavored. The model corrects for this by comparing the observed 30 31 number of communities in favored states to a distribution of expected number 31 32 of favored states that is generated by an iterative computer model. If the 32 33 species pool for a given site is biased, this should be reflected in the distri- 33 34 bution of expected favored states. By incorporating detailed information on 34 35 the unique species pools for each site, this concern is minimized. In the present 35 36 study species were allocated to three functional groups, following the lead of 36 37 Fox and Brown (1993): bipedal heteromyids (Dipodomys and Microdipodops); 37 38 quadrupedal heteromyids (Chaetodipus, Perognathus), and cricetines (Pero- 38 39 myscus, Reithrodontomys, Baiomys, etc.). 39 40 Assembly of a simulated community proceeds as follows. The species that 40 106 D.A. Kelt and J.H. Brown 1 are present at all sites, as well as those in the pool for the sites, are charac- 1 2 terized according to their membership in these functional groups. Species are 2 3 drawn randomly from these functional groups, according to the relative sizes 3 4 of the groups. For example, if the pool for a site contained three herbivores 4 5 and a single carnivore, the probability of an herbivore entering the commu- 5 6 nity would be 3⁄4 = 0.75, while the probability of a carnivore entering the 6 7 community would be 1⁄4 = 0.25. These probabilities are re-evaluated after each 7 8 species addition, so that species are drawn from the pools without replace- 8 9 ment. Thus, if an herbivore was drawn in the above case, then the probabil- 9 10 ity that the second species entering the site is an herbivore would be 2/3 = 10 11 66.67%, while the probability of a carnivore entering would be 1/3 = 33.33% 11 12 The important point to note here is that species are not the unit of investi- 12 13 gation. This model is based not on the individual species, but on character- 13 14 istics of these species (e.g., diet, morphology, etc.). Hence, the unit of inves- 14 15 tigation is the ecological attribute that is embodied in the functional group 15 16 designation, and the model addresses the organization of communities based 16 17 on species characters. 17 18 In its simplest form, the model operates as follows. For each real site, the 18 19 number of species present is tallied. Simulated communities are assembled 19 20 and evaluated to determine if they are in a favored or unfavored state, and 20 21 the number of communities that are in favored states is recorded. This num- 21 22 ber represents one estimate of the number of communities expected to be in 22 23 favored states, and constitutes one iteration of the model. This is repeated 23 24 many times (2000 in the present paper) to produce a distribution of the 24 25 expected number of favored states among the sites examined. The observed 25 26 number of favored states is then compared to this distribution, and the prob- 26 27 ability that the observed value was obtained at random is determined by the 27 28 proportion of the expected distribution that is equal to, or more extreme, than 28 29 the observed. Because this is a one-tailed test (the alternative hypothesis is 29 30 that the observed communities are more structured than random; this implies 30 31 that the observed number of favored states should be greater than expected 31 32 at random) the null hypothesis or random assembly will be rejected if the 32 33 observed value lies in the upper 5% tail of the distribution. 33 34 The above calculations might be a trivial and unnecessary exercise in com- 34 35 puter modeling, because we could produce the exact probabilities of observed 35 36 assemblages with the multivariate hypergeometric distribution. Clearly, an 36 37 exact probability would be a better metric than the estimate produced by our 37 38 iterative model. However, a significant element has been added to our model, 38 39 which allows us to refine our analysis beyond that which could be attained 39 40 with the multivariate hypergeometric distribution. In particular, interspecific 40 Conceptual and statistical issues 107 1 interactions can be simulated by incorporating an interaction coefficient (θ) 1 2 into the model. Thus, our probability function is given as (Kelt et al., 1995): 2 3 3 4 4 (n{i} − X{i,j − 1})(1 −θ)X{i,j − 1}) 5 P(y{j} = i | X{j − 1}) = (1) 5 ∑ [(n{i} − X{i,j − 1})(1 −θ){i,j − 1}) 6 i 6 7 7 8 where i and j index functional groups and species, respectively, n{i} is the 8 9 number of species in functional group n,y{j} is a random variable, indicating 9 10 the guild to which the jth species will be placed, X{j−1} is a vector of local 10 11 species composition after the j−1 th species has entered the community, and 11 12 X{i, j−1} is a scalar which gives the number of species in functional group I 12 13 after the j−1th species has entered. The numerator gives the combined prob- 13 14 ability of a species immigrating and becoming established. The denominator 14 15 normalizes these probabilities so that they sum to one (see Kelt et al., 1995 15 16 for further details). 16 17 The coefficient θ measures the effect of species from a functional group 17 18 that is already present in a community upon the establishment of another 18 19 species from the same functional group. Positive values of θ decrease the 19 20 probability of establishment, and reflect competitive exclusion. Negative val- 20 21 ues of θ increase the probability of establishment, and reflect facilitation. 21 22 When θ is zero, community composition has no effect on the establishment 22 23 of a new species. 23 24 Finally, this model allows us to calculate two final parameters. By vary- 24 25 ing the value of θ, the strength of interaction that best fits the observed data 25 26 can be determined. That is, the value of θ that produces a distribution of 26 27 expected number of favored states that best agrees with the observed number 27 28 of favored states can be determined. This will be referred to as θˆ. This con- 28 29 stitutes a maximum likelihood estimate of the mean strength of interaction 29 30 across all sites and all species, and is clearly a general value. However, by 30 31 calculating this value, the statistical power of our model can then be estimated. 31 32 Power is the probability that a hypothesis is incorrect and therefore should 32 33 be refuted. If it is assumed that θˆ reflects the real strength of interaction among 33 34 these species, then the proportion of the distribution of expected values (num- 34 35 ber of favored states) produced with θ=θˆ that lies in the 5% critical region 35 36 of the distribution of expected values produced with θˆ (the null hypothesis) 36 37 constitutes an estimate of the power with which it can be stated that observed 37 38 communities are significantly different from the null. 38 39 39 40 40 1 4 1 2 2 3 Introduced avifaunas as natural experiments in 3 4 community assembly 4 5 5 6 6 7 Julie L. Lockwood, Michael P. Moulton, and Karla L. Balent 7 8 8 9 9 10 10 11 11 12 12 13 13 14 Introduction 14 15 In 1975, Jared Diamond hypothesized a series of rules for assembling com- 15 16 munities of birds on islands in the south Pacific. In large part, these rules 16 17 were based on incidence functions that dealt with the probability of species 17 18 occurrence as a function of species richness (Diamond, 1975). His work 18 19 sparked a heated debate and the concept of ‘assembly rules’ was largely put 19 20 to rest (Connor & Simberloff 1979; Gilpin & Diamond, 1982). Perhaps one 20 21 of the most serious, and yet least appreciated, problems with applications of 21 22 assembly rules is that they are, in practice, based principally, if not exclu- 22 23 sively, on extant community membership (e.g., Wilson & Roxburgh, 1994; 23 24 Fox & Brown, 1993). As Diamond (1975) noted, the principal goal in eluci- 24 25 dating assembly rules is to discover why some species become members of 25 26 a community and some do not. How can the true criteria for inclusion in a 26 27 community be decided without knowledge of criteria for rejection? 27 28 In order to derive meaningful assembly rules, one must begin with know- 28 29 ledge of all species once present, but now missing, from communities. In nat- 29 30 ural communities, a knowledge of this depth is generally lacking, reducing 30 31 the strength of tests for assembly rules. Communities of non-indigenous bird 31 32 species represent a companion system in which to study the effects of various 32 33 selection criteria for community membership. Within such communities there 33 34 is often a very thorough knowledge of which species failed to become com- 34 35 munity members enabling evaluation of the influence of specific mechanisms 35 36 of membership rejection. 36 37 The analyses described below were based on a data set that includes the 37 38 passeriform species introduced on four oceanic islands: Oahu, Tahiti, Saint 38 39 Helena, and Bermuda. Non-indigenous bird species have been accidentally 39 40 or intentionally introduced for centuries on these islands (Long, 1981). With 40

108 Introduced avifaunas as natural experiments 109 1 the help of ardent bird watchers past and present, very clear records have 1 2 been obtained of which species were introduced, in what manner, and when, 2 3 and if they became extinct (Moulton & Pimm, 1983). 3 4 The assembly history of each island avian community is reconstructed using 4 5 this information. With this reconstructed assembly trajectory sensible null 5 6 models can be built with which to compare the extant assemblage. Two of 6 7 the tests below were built specifically to search for patterns produced by inter- 7 8 specific competition: one of the most often cited, and controversial, criterion 8 9 for membership rejection (for reviews see Connell, 1983; Schoener, 1983). 9 10 An alternative, and equally plausible criteria for rejection, is that a particu- 10 11 lar species simply is not intrinsically capable of surviving within the new 11 12 environment (Simberloff, 1992). Thus, patterns of success are also sought, 12 13 associated with a species’ native range size which is believed to reflect a 13 14 species intrinsic ability to compensate for environmental variability (Williamson, 14 15 1996). 15 16 Results of previous studies of introduced birds have been conducted mostly 16 17 on an island by island basis (Lockwood & Moulton, 1994; Brooke et al., 17 18 1995; Moulton, 1993; Lockwood et al., 1993, 1997). Here, a much broader 18 19 view of the phenomena of introduced avian community assembly is attempted 19 20 by incorporating a large number of species (99) across four oceanic islands. 20 21 Each of the three patterns described above is addressed in more detail indi- 21 22 vidually and completely before moving to the next. First, however the species 22 23 introduced must be described, together with the islands, and the circumstances 23 24 surrounding the introduction events. 24 25 25 26 26 27 The islands 27 28 All four islands are oceanic, but they vary in degree of geographical isola- 28 29 tion. Bermuda is perhaps the least isolated recording more than a hundred 29 30 migrants every year (Wingate, 1973). Oahu and Tahiti are members of archi- 30 31 pelagos and thus may receive natural immigrants from other nearby islands, 31 32 but not from any mainland source. Saint Helena is the most geographically 32 33 isolated sitting alone over 2000 km off the Cape of Africa in the Atlantic 33 34 Ocean. It is thus doubtful that Saint Helena is subject to any regularly occur- 34 35 ring attempts at natural colonization. Bermuda is temperate and relatively 35 36 small (51 km2). Tahiti and Oahu are tropical islands with sizes of 1057 km2 36 37 and 1536 km2, respectively. Saint Helena is also tropical but relatively small 37 38 in size (122 km2). 38 39 Each island has had a significant proportion of its native flora and fauna 39 40 altered through human manipulation (Moulton & Pimm, 1983; Brooke et al., 40 110 J.L. Lockwood et al. 1 1995; Lockwood & Moulton, 1994; Lockwood et al., 1993). There remain 1 2 however, some physiographic differences between islands. For example, 2 3 Bermuda and Saint Helena are relatively low relief islands in which nearly 3 4 all native forest has been cleared (Wingate, 1973; Benson, 1950). Tahiti and 4 5 Oahu are high relief islands that still retain a small, but significant, portion 5 6 of their native forests (Moulton, 1993; Lockwood et al., 1993). 6 7 Through field observations on each island, it seems that most successful 7 8 species are utilizing suburban or disturbed habitat rather than the remaining 8 9 native habitat (JLL, MPM, R. Brooke, pers. comm.). This is supported by 9 10 others conducting avian surveys on each island (Scott et al., 1986; Crowell 10 11 & Crowell, 1976; Haydock, 1954). Not only are the avian assemblages of 11 12 human derivation, but so too are the habitats that sustain them. The intro- 12 13 duced passeriforms on each island make their living from suburban lawns, 13 14 parks, and city streets thus reducing the variability in habitat across islands. 14 15 15 16 16 17 The formulation of the species pool 17 18 Perhaps the most important aspect of determining a criteria of rejection is 18 19 determining which species one should include in the species pool. The pool 19 20 consists of all species likely to have attempted establishment during the 20 21 assembly history. The first step in formulating the pool is by far the easiest 21 22 and most enjoyable: determining the composition of the extant community. 22 23 In all the data sets described below, the extant community was determined 23 24 by combining field observations by JLL, MPM or R. Brooke with recently 24 25 published field guides (e.g., Pratt et al., 1987; Wingate, 1973). The second 25 26 step involves an in-depth exploration into the published literature. We com- 26 27 piled lists of species introduced from published references and references 27 28 contained within the following: Saint Helena; Brooke et al. (1995), Tahiti; 28 29 Lockwood et al. (1993), Bermuda; Lockwood and Moulton (1994), and Oahu; 29 30 Moulton (1993). 30 31 The results of this search revealed that, across all four islands, a total of 31 32 99 species have been introduced incorporating 131 total introduction events 32 33 (see Appendix). Of the 24 species introduced on more than one island, only 33 34 one, the Common Waxbill (Estrilda astrild), was introduced on all four 34 35 islands. These islands are clearly not geographically close enough to exchange 35 36 species. Thus, each island represents a relatively distinct assemblage with its 36 37 own unique avifaunal assembly history. 37 38 These references were also the basis for decisions regarding when species 38 39 were introduced as well as if and when they failed. On some occasions there 39 40 was a significant lag between published references concerning which species 40 Introduced avifaunas as natural experiments 111

1 Table 4.1. Success rates for each island: the number of successful species out of 1 2 the number of passeriforms introduced 2 3 3 4 Island Success rate 4 5 Saint Helena 5 out of 31 (16%) or 5 6 5 of 17 excluding 1929 introductions (29%) 6 7 Tahiti 7 out of 41 (17%) 7 Bermuda 7 out of 14 (50%) 8 Oahu 27 out of 45 (60%) 8 9 9 10 10 11 11 12 were present on what island. For example, two species on Bermuda were 12 13 introduced around 1800 but, were not referred to again until 1957 (Bourne, 13 14 1957). In this account, Bourne indicated that these species had not established 14 15 after initial introduction however, he gave no date of last sighting. Thus, each 15 16 species was counted as being present between 1800 and 1957. Such difficul- 16 17 ties do not influence the results as long as each species’ time of extinction 17 18 can be placed relative to all others. In this case, no species introduced between 18 19 1800 and 1957 subsequently went extinct. 19 20 Of the 131 introduction events, 46 are counted as successful and 85 as 20 21 failed (35%). However, the individual island success rates varied consider- 21 22 ably (see Table 4.1). Saint Helena and Tahiti had similar overall success rates 22 23 at 16% (5 of 31) and 17% (7 of 41), respectively. A total of 45 species were 23 24 released on Oahu, 27 of those are currently considered successful (60%). On 24 25 Bermuda, there were 14 exotic species introductions and half of those were 25 26 successful (50%). There is no significant tendency for the number of success- 26 27 ful species to increase with island size (P = 0.2668, r2 = 0.538, see Fig. 4.1), 27 28 although the largest island had the greatest number of successful introductions. 28 29 The first two of our three tests are sensitive to various influences on the 29 30 formation of the species pool. There was one criterion, however, that was 30 31 universally followed. Species were excluded if they invaded one of the islands 31 32 naturally or if only five or fewer individuals were known to have been released 32 33 (as these species’ fates would more likely be determined by stochastic pro- 33 34 cesses). The other influences include the phylogenetic variability seen within 34 35 the species pool, the timing and method of introduction, and the presence or 35 36 absence of a native avifauna. Each is dealt with in turn. 36 37 Colwell and Winkler (1984) point to a potential bias in the formation of 37 38 the species pool due to the amount of phylogenetic variability incorporated: 38 39 They call this the J.P. Morgan Effect. If species with extreme morphologies 39 40 are included in the pool, the remaining species will appear clumped in any 40 112 J.L. Lockwood et al. 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 Fig. 4.1. The number of species successfully colonizing islands included in our data 16 17 set does not correspond to island size. The largest island does hold the greatest number 17 18 of successful species, however. 18 19 19 20 morphologically defined space. Since morphological characters are conserved 20 21 within phylogenetic lineages, the likelihood of including an extreme morph- 21 22 ology correlates with the degree of phylogenetic variability seen within the 22 23 species pool (Colwell & Winkler, 1984). An extreme species (or group of 23 24 species) included within a group of very similar species serves to obscure 24 25 any pattern present. 25 26 In Lockwood et al. (1997) it was argued that one way to minimize the 26 27 chance for a J.P. Morgan effect would be to look only within the group of 27 28 finches introduced to Saint Helena. Finches make up the overwhelming major- 28 29 ity of all species introduced (74%) and species that were successful (84%: 29 30 Lockwood et al., 1997). Clearly, within this group competition will be the 30 31 strongest as they all feed and nest in very similar fashion (Clement et al., 31 32 1993). The presence of distantly related species in the pool may obscure any 32 33 competitively generated pattern. 33 34 The introduction events ranged in time from 1776 to the mid-1980s, with 34 35 most occurring during the late 1800s to early 1940s. Recently introduced 35 36 species (i.e., within the past 20 years) are often difficult to place as either 36 37 successful or failed. This is especially true if some species are currently 37 38 present in small numbers or are restricted to a particular habitat. We should 38 39 perhaps consider these species ‘the walking dead’ (Simberloff, 1992) rather 39 40 than successful colonizers. This is the case with several species introduced 40 Introduced avifaunas as natural experiments 113 1 to Oahu since 1960. However, most other species considered here were 1 2 released several years before the present (e.g., the last species released on 2 3 Saint Helena occurred 67 years ago). The remaining few are very clearly suc- 3 4 cessful (e.g., the Great Kiskadee was released on Bermuda in 1957 and is 4 5 currently one of the most common birds observed there; Wingate, 1973, JLL). 5 6 The modes of introduction varied across islands; however, they can be 6 7 classified into the following categories: ship or airplane stowaways, cage 7 8 escapees, and purposeful introductions (Long, 1981; Lever, 1987). No indi- 8 9 vidual or group introduced avian species on more than one of these islands. 9 10 However, on two islands (Tahiti and Saint Helena) one individual was respon- 10 11 sible for the release of the majority of species included in the species pool. 11 12 The circumstances surrounding each introduction event are unique and 12 13 influenced how we constructed each pool differently. 13 14 On Tahiti, Eastham Guild released 85% (35 of 41) of all species intro- 14 15 duced (Guild, 1938, 1940; Lockwood et al., 1993). Only two of these species 15 16 survived, although we know Guild exerted considerable effort to ensure proper 16 17 release (Guild, 1938, 1940). Guild kept individuals in an aviary for some time 17 18 after their arrival, allowing him to remove diseased or sick birds. Further, 18 19 Guild and other island residents established feeding stations to enhance the 19 20 bird’s chances for success. For these reasons, it was appropriate to include 20 21 in the pool all the species Guild released. 21 22 H. Bruins-Lich released 45% (14 of 31) of all species introduced on Saint 22 23 Helena (Haydock, 1954; Brooke et al., 1995). The argument has previously 23 24 been made that these species possibly should be excluded from the pool 24 25 (Brooke et al., 1995) as there is very little information on the release meth- 25 26 ods. Further, none of the 14 species he released was successful. Thus, unlike 26 27 Guild, the possibility cannot be dismissed that he may have, in some way, 27 28 inadvertently doomed all of these species to failure. For this reason, our 28 29 analyses were conducted with and without these 14 species. 29 30 On all islands except Bermuda, the native passeriform fauna was either 30 31 decimated before introductions occurred (Brooke et al., 1995) or severely 31 32 restricted in range or numbers (Lockwood et al., 1993; Moulton, 1993). Thus, 32 33 on these three islands interactions between native and exotic passeriforms is 33 34 minimal at best (Brooke et al., 1995; Lockwood et al., 1993; Moulton, 1993). 34 35 However, on the island of Bermuda three native passeriforms remain that are 35 36 widespread and abundant (Wingate, 1973; Lockwood & Moulton, 1994). 36 37 These are the Gray Catbird (Dumetella carolinensis), Eastern Bluebird (Sialia 37 38 sialis), and the White-eyed Vireo (Vireo griseus). Native–exotic interactions 38 39 cannot be discounted when looking for pattern within this community. The 39 40 three native species on Bermuda are included where appropriate. 40 114 J.L. Lockwood et al. 1 1 2 Priority effect 2 3 If the group of species that successfully colonized a particular island did so 3 4 when significantly fewer other introduced species were present, this commu- 4 5 nity is said to exhibit a priority effect (Moulton, 1993). Simply, species that 5 6 arrive first are less likely to encounter competitors which may preclude their 6 7 establishment. Once establishment is ensured, these same species are more 7 8 likely to become competitively dominant over later arrivals (Alford & Wilbur, 8 9 1985). This pattern would be expected if interspecific competition was a pre- 9 10 dominate criterion for membership rejection. 10 11 11 12 12 13 Methods 13 14 Moulton (1993) details the procedure for testing for a priority effect. Briefly, 14 15 we rank each species introduced to a particular island according to its date 15 16 of introduction. For each species, we calculated the number of other intro- 16 17 duced species present on the island at the time of introduction. Using a 17 18 Mann–Whitney U-test, the mean number of other species present for failed 18 19 introductions is compared to the mean for the group that was successful. Each 19 20 island is considered separately. 20 21 21 22 22 23 Results 23 24 Three of our four island avifaunas exhibits a priority effect (Table 4.2). The 24 25 most pronounced patterns are within the species introduced on Saint Helena 25 26 and Tahiti. On Tahiti, successfully introduced species initially faced an aver- 26 27 age of 13.14 other species, whereas those that failed faced an average of 38.0 27 28 (P > χ2 = 0.0032). The average number of other species faced by success- 28 29 fully introduced species on Saint Helena was 5.4, and for unsuccessful species 29 30 14.15 (P > χ2 = 0.0009: Brooke et al., 1995). As noted above, there is reason 30 31 to exclude the 1929 introductions on Saint Helena. Their inclusion certainly 31 32 inflates the priority effect. However, even if these species are excluded, the 32 33 priority effect on Saint Helena still holds (P > χ2 = 0.008). 33 34 The Bermuda avifauna tends toward a priority effect when the native 34 35 species are included as the first to arrive, which undoubtedly they were (P > 35 36 χ2 = 0.057, mean successful = 5.6, mean failed = 9.14). However, it is not 36 37 known how many other species the natives encountered when they first 37 38 arrived. One could argue that the natives should be excluded from the analy- 38 39 sis and only counted as ‘other species present’ in our calculations. If the 39 40 natives are ‘fixed’ into the analyses in this way, there is clearly no evidence 40 Introduced avifaunas as natural experiments 115

1 Table 4.2. Results of our search for a priority effect using the passeriform 1 2 assemblages on Tahiti, Saint Helena, Bermuda, and Oahu 2 3 3 4 Island Species pool used Results 4 5 Tahiti All passeriforms P = 0.0032, Yes 5 6 Saint Helena All passeriforms P = 0.0009, Yes 6 7 Passeriforms introduced before 1929 P = 0.008, Yes 7 Bermuda All passeriforms including native species P = 0.057, tendency 8 All passeriforms with native species ‘fixed’ P = 0.252, No 8 9 Oahu All passeriforms introduced before 1960 P = 0.009, Yes 9 10 All passeriforms introduced before 1981 P = 0.235, No 10 11 11 12 Three out of four island assemblages show statistical proof of a priority effect. 12 13 13 14 for a priority effect (P > χ2 = 0.252, mean successful = 7.14, mean failed = 14 15 9.14). 15 16 Moulton (1993) conducted tests for the priority effect among the intro- 16 17 duced passeriforms of Oahu. Originally, Moulton and Pimm (1983) excluded 17 18 species introduced to Oahu after 1960 as they believed the current status of 18 19 these species was uncertain. If only species introduced prior to 1960 are 19 20 examined, as is in Moulton and Pimm (1983), a priority effect (P > χ2 = 20 21 0.009, mean successful = 12, mean for failed = 19) is observed. 21 22 After more than ten years of continued monitoring of these questionable 22 23 species beyond 1983, a much better picture is available of which species have 23 24 become established and which have not (Moulton, 1993). When the list was 24 25 updated to include introductions up to 1981 (thus still excluding very recent 25 26 introductions due to their questionable status), there is no evidence for a 26 27 priority effect (P < χ2 = 0.235, mean for successful = 17.5, mean for failed 27 28 = 20). 28 29 29 30 30 31 Discussion 31 32 In three out of four independent assemblies, the presence of a priority effect 32 33 can be statistically shown. This result is consistent with the hypothesis that 33 34 interspecific competition was a predominate criterion of membership rejec- 34 35 tion in each assembly. Species which establish early likely encounter fewer 35 36 competitors and thus have a greater chance of being competitively dominant 36 37 over later arrivals (Alford & Wilbur, 1985). 37 38 With this in mind, it is not surprising that evidence for a priority effect 38 39 within the avifaunal assemblages on Oahu and Bermuda is not found con- 39 40 sistently. In each case, there is a lack of sufficient knowledge concerning 40 116 J.L. Lockwood et al. 1 which species were truly ‘early’ and which of the ‘late’ arrivals failed to 1 2 establish. It is not known which set of species may have been present when 2 3 the three Bermuda natives first colonized the island. Similarly, it is not known 3 4 how to classify the outcomes of species introduced recently on Oahu even 4 5 though the set of species they likely encountered is known. Our results depend 5 6 on how this lack of knowledge is compensated for. It is thus clear that insuf- 6 7 ficient information on the assembly history, and thus about the circumstances 7 8 of membership rejection (i.e., competition or not), influences our perception 8 9 of the assembly process. 9 10 10 11 11 12 Morphological overdispersion 12 13 Morphological overdispersion will result if the species that successfully col- 13 14 onized a community are more spread-out in morphological space and more 14 15 evenly positioned than what is expected by chance (Lockwood et al., 1993; 15 16 Moulton & Pimm, 1987; Ricklefs & Travis, 1980). Overdispersion is also 16 17 consistent with the hypothesis that interspecific competition determined 17 18 community membership. 18 19 19 20 20 21 Methods 21 22 Several morphological variables (beak width, depth, and length and wing 22 23 length) were measured on all species introduced onto each island from 23 24 museum specimens housed at the Los Angeles County Museum of Natural 24 25 History or the United States National Museum of Natural History. Each mor- 25 26 phological variable was chosen because of a believed association with some 26 27 aspect of avian ecology (Ricklefs & Travis, 1980; Grant, 1986). There are 27 28 two possible approaches when deciding how many morphological variables 28 29 to incorporate into such an analysis. The first is to include only one variable, 29 30 and preferably one that reflects a multitude of ecological characteristics (e.g., 30 31 body size). The second is to include several variables in the hopes of homog- 31 32 enizing any variability seen within any one correlation. Thus, if a species 32 33 whose body size is not highly linked to its food preference, maybe its beak 33 34 shape is, and in a multivariate analysis this would not go unnoticed (Ricklefs 34 35 & Travis, 1980; Karr & James, 1975; Findley, 1973, 1976). 35 36 These raw variables are then log-transformed to normalize the inherent 36 37 variability of such measurements and a principal components analyses is 37 38 conducted using the covariance matrix (Ricklefs & Travis, 1980). The first 38 39 two principal component scores invariably accounted for the majority of the 39 40 variance seen within the raw data in our analyses (i.e., > 90%). The first 40 Introduced avifaunas as natural experiments 117 1 principal component typically reflected body size and accounted for the major- 1 2 ity of the variability within the data. The second principal component typically 2 3 reflected the size and shape of the beak relative to the body size and accounted 3 4 for significantly less variance. 4 5 Principal components analysis was used to produce a set of uncorrelated 5 6 trait axes, upon which species can be plotted to show morphological simi- 6 7 larity. The minimal spanning tree (MST) is used as a measure of dispersion 7 8 within this morphological space and thus becomes our test statistic. The MST 8 9 is the minimum sum of the lengths of n-1 line segments that connect n points 9 10 such that no loops are created (Ricklefs & Travis, 1980). The standard devi- 10 11 ation (SDEV) of the MST line segments is used as a measure of how evenly 11 12 positioned the species are in that space. 12 13 An MST is drawn for the extant set of species on each island and then 13 14 compared to a null distribution of MST and SDEV values derived from 14 15 Monte-Carlo simulations (Ricklefs & Travis, 1980). Thus, if an introduced 15 16 community comprises ten species and it is known that 20 species were intro- 16 17 duced, MSTs and SDEVs are calculated for 1000 random selections of 17 18 ten species from the pool of 20. An observed community is considered to be 18 19 morphologically overdispersed if a large proportion (i.e., 90–95%) of null 19 20 community MST values are smaller and, simultaneously, the SDEV values 20 21 are larger (i.e., more evenly spaced) than the observed. Again, each island 21 22 avifaunal assemblage was considered separately. 22 23 The observed passeriform communities on Bermuda, Tahiti and Saint 23 24 Helena were all compared to 1000 randomly constructed communities (Lock- 24 25 wood et al., 1993, 1996; Lockwood & Moulton, 1994) when possible. The set 25 26 of successful finches introduced to Saint Helena before 1929 were compared 26 27 to all possible alternative outcomes. The extant assemblage on Oahu was com- 27 28 pared only to 200 random communities (Moulton, 1993). For a more detailed 28 29 account of methods, see Lockwood et al. (1993). 29 30 30 31 31 32 Results 32 33 Tests for morphological overdispersion among the successful passeriforms on 33 34 all four islands have previously been conducted (Lockwood et al., 1993, 1997; 34 35 Moulton, 1993; Lockwood & Moulton, 1994). Here only a brief review of 35 36 our results are provided (see Fig. 4.2 for all morphological spaces and Table 36 37 4.3 for a summary of results). Oahu, Tahiti and Bermuda all exhibited pro- 37 38 nounced overdispersion, while simultaneously being more evenly spaced 38 39 among all successful passeriforms introduced (Oahu: P = 0.035, Tahiti: P = 39 40 0.002). On Bermuda, this is true regardless of whether or not the native species 40 118 J.L. Lockwood et al. 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 Fig. 4.2. The morphological spaces defined by the first two principal component scores 33 for all island avifaunal assemblages. Principal component I indicates overall body size, 33 34 whereas principal component II indicates the size and shape of the beak relative to 34 35 body size. Because analyses used different species assemblages, the axes are only 35 broadly comparable. Graphs are redrawn from: Lockwood and Moulton (1994), 36 Lockwood et al. (1993, 1997), and Moulton and Pimm (1987). Closed diamonds 36 37 represent successful species; open squares represent failed species; stars represent 37 38 native species. 38 39 39 40 40 Introduced avifaunas as natural experiments 119

1 Table 4.3. The results of our search for the pattern of morphological 1 2 overdispersion within the four island passeriform assemblages 2 3 3 4 Island Species Pool Results 4 5 Bermuda All passeriforms including natives P = 0.006, Yes 5 6 All passeriforms ‘fixing’ natives P = 0.014, Yes 6 7 Tahiti All passeriforms P = 0.002, Yes 7 Oahu All passeriforms P = 0.035, Yes 8 Saint Helena All passeriforms P = 0.139, No 8 9 All passeriforms introduced before 1929 P = 0.301, No 9 10 Only finches P = 0.092, No 10 Only finches introduced before 1929 P = 00.076, Tendency 11 11 12 12 Again, three of the four assemblages show statistical evidence for the pattern of 13 morphological overdispersion. 13 14 14 15 15 16 were fixed in the morphological space (P = 0.006) or considered part of the 16 17 species pool (P = 0.014). 17 18 On Saint Helena overdispersion was searched for among all introduced 18 19 passeriforms and among only the group of finches to compensate for a poten- 19 20 tial J.P. Morgan Effect (see above). It is shown that the group of introduced 20 21 finches are not morphological overdispersed (P = 0.092). As noted above, 21 22 there are compelling reasons to exclude the 14 species released in 1929. The 22 23 data suggest that the finches introduced prior to 1929 are overdispersed (P = 23 24 0.076; an exact statistic from a complete permutation test). However, the set 24 25 of all species introduced are also not overdispersed (P = 0.139), nor are the 25 26 surviving species introduced before 1929 (P = 0.301). 26 27 27 28 28 29 Discussion 29 30 In three of four independent assemblies, the pattern of morphological over- 30 31 dispersion is clearly seen. Overdispersion, like the priority effect, is consistent 31 32 with the hypothesis that interspecific competition was a predominate criterion 32 33 for species rejection. Thus, using two methodological approaches, the same 33 34 general conclusion is reached: competition between existing community 34 35 members and those that are newly released often determines the fate of the 35 36 new colonizers. 36 37 Morphological overdispersion will only result if two criteria are met. First, 37 38 interspecific competition must be strong enough to drive one of the competi- 38 39 tors to extinction. Second, competition must be widespread enough to effect 39 40 all the species included in the species pool (Moulton & Pimm, 1987). Thus, 40 120 J.L. Lockwood et al. 1 determining the composition of the species pool takes on added importance. 1 2 A balance must be struck between including all species in the community 2 3 and only those that are likely to compete intensely and extensively enough 3 4 to perhaps produce pattern. This balance is sometimes tenuous as is the case 4 5 with the introduced avifauna of Saint Helena. It is not clear if the eight 5 6 phylogenetically (and thus morphologically) distinct species released on Saint 6 7 Helena significantly altered the chances that the set of finches released would 7 8 successfully colonize. Clearly, how that question was answered influenced 8 9 the results obtained. 9 10 Many problems are solved simultaneously by defining the species pool as 10 11 all the passeriforms introduced onto an oceanic island (i.e., it is known who, 11 12 how, and why species attempted colonization on the island). This is a luxury 12 13 that those working within natural systems do not enjoy. Indeed, the lack of 13 14 predators, native species and natural colonizers on oceanic islands makes the 14 15 definition of an island community all the easier. However, it does not solve 15 16 all problems of definition. Our results indicate that the definition of the species 16 17 pool, and the extent of the community itself, can play a considerable role in 17 18 how a rejection criteria is determined and therefore the assembly process itself. 18 19 19 20 20 21 Size of native range 21 22 Finally, we test a hypothesis set forth by Moulton and Pimm (1986) and 22 23 Daehler and Strong (1993) that species which have expansive native ranges 23 24 will more likely successfully colonize a new locale after introduction. Species 24 25 with larger range sizes may exhibit greater inherent ecological or physiolog- 25 26 ical flexibility or may be better dispersers (Williamson, 1996; Brown, 1995). 26 27 These species thus may be predisposed to be successful invaders no matter 27 28 where released or what other species may be present. This is consistent with 28 29 the ‘All or None’ hypothesis of Simberloff and Boecklen (1991), which has 29 30 been put forth as an alternative explanation for patterns observed above. Thus, 30 31 we explore the same data set to test this hypothesis as well. 31 32 32 33 33 34 Methods 34 35 Following the methods of Moulton and Pimm (1986), native range sizes were 35 36 estimated using maps published in Long (1981). A grid was placed over each 36 37 map and the number of intersecting points within the native range specified 37 38 by Long (1981) were counted. This number was then divided by the number 38 39 of intersections within the continent of Australia to standardize range sizes 39 40 between maps (there were two map sizes). Thus, relative range sizes were 40 Introduced avifaunas as natural experiments 121 1 calculated and not exact range sizes. All relative range sizes are presented as 1 2 the percentage of the area of Australia each species’ range incorporates. Thus, 2 3 a relative range size of 50% indicates that this species’ actual range is roughly 3 4 half the size of Australia. All species were ranked according to their relative 4 5 range size and grouped as either successful or failed. Mean range sizes are com- 5 6 pared for each group using a Mann–Whitney U test. All introduction events 6 7 across all islands are considered. Thus, if the same species was introduced 7 8 on two or more islands, each introduction was counted as an independent 8 9 event. Effects of range size were also tested for on an island-by-island basis. 9 10 10 11 11 12 Results 12 13 There were 131 independent introduction events (99 species) recorded as 13 14 occurring on the four islands. However, there were 12 species for which Long 14 15 (1981) provided no native range maps. Any introduction events associated 15 16 with these species were thus excluded. Of the remaining 118 introduction 16 17 events (87 species), 41 were successful and 77 failed. The average relative 17 18 range size of the 41 successful species was 96%, whereas the average rela- 18 19 tive range size for failed species was 109%. There was no statistical differ- 19 20 ence between these two means (P > χ2 = 0.617). 20 21 It may be that the species comprising one or more island communities do 21 22 show increased chances for success based on their native range sizes if the 22 23 island communities are considered independently. Our result above may mask 23 24 effects present at the island level. Thus, the test was conducted on an island- 24 25 by-island basis. None of the island communities showed significant correla- 25 26 tions between successful establishment and large native range (see Table 4.4). 26 27 27 28 28 29 Discussion 29 30 To some degree, all species that strive for membership in a community face 30 31 a novel situation. The species they encounter will likely not be the same as 31 32 those present in the community from which they originated. The variability 32 33 in abiotic conditions will also not be the same. This is certainly more pro- 33 34 nounced when human-mediated introduction are considered. These species 34 35 are often transported long distances and then released (Long, 1981; Lever, 35 36 1987). Thus, it may be expected that patterns regarding native range size will 36 37 be more pronounced within species introduced by human actions. 37 38 Our results, however, indicate that range size in itself is not a good pre- 38 39 dictor of successful establishment within an assembling community. There 39 40 may be species-specific attributes that will influence establishment within a 40 122 J.L. Lockwood et al.

1 Table 4.4. Island by island analyzes of the effect of native range size on the 1 2 probability of successful establishment after introduction 2 3 3 4 Average range size of Average range size Associated 4 Island successful species of failed species probability 5 5 6 Saint Helena 59% 144% P = 0.1054 6 7 Tahiti 69% 49% P = 0.2912 7 Bermuda 235% 123% P = 0.1792 8 Oahu 104% 74% P = 0.5541 8 9 9 10 None of the islands showed an effect. 10 11 11 12 12 13 particular context. Indeed, there is a growing subdiscipline surrounding the 13 14 topic (Drake et al., 1989; Williamson, 1996; Kareiva, 1996). It is not clear 14 15 that any one, or even a small suite, of these characteristics will consistently 15 16 determine a species fate when attempting to enter a community (Williamson & 16 17 Fritter, 1996; Gilpin, 1990). Our results indicate that species-specific attributes, 17 18 when represented by a ‘macro-variable’ such as range size, are not neces- 18 19 sarily good predictors of success within a unique assembly (within an island 19 20 community) or across several independent assemblies (across several island 20 21 communities). 21 22 22 23 23 24 Summary and conclusion 24 25 The determination of ‘rules’ that govern the assembly of natural communi- 25 26 ties is riddled with several methodological pitfalls. The most obvious of which 26 27 is the inability to determine a true criteria of rejection when looking only 27 28 within an extant community (Drake, 1991). However, our results indicate that 28 29 even when companion systems of introduced avifaunas are used, where species 29 30 rejections can be documented, a clear picture of those rejecting criteria cannot 30 31 always be gained. There remain questions concerning introduction timing and 31 32 the definition of the species pool and extent of community. Most analyses of 32 33 natural communities uniformly ignore such questions. 33 34 Our results, some new and some summarized here, indicate that inter- 34 35 specific competition can play a deciding role in which species will become 35 36 community members. They also point to a deficiency in the predictive ability 36 37 of species-specific attributes in determining success. Any criteria, if it is to 37 38 prove useful as an assembly rule, must provide an acceptable predictive ability 38 39 across many assembly scenarios. As yet, it is believed neither of these two 39 40 hypothesized rules completely fits the bill. 40 Introduced avifaunas as natural experiments 123 1 1 2 Acknowledgments 2 3 We wish to thank Dr Richard Banks and Dr Gary Graves at the National 3 4 Museum of Natural History, Washington, DC and Kimball Garrett of the Los 4 5 Angeles County Museum of Natural History, Los Angeles, CA for allowing 5 6 JLL and MPM access to their avian collections. This work incorporates the 6 7 knowledge of several individuals including R.K. Brooke, D. Wingate, S.K. 7 8 Anderson, and S.L. Pimm. 8 9 9 10 10 11 References 11 12 Alford, R.A. & Wilbur, H.M. (1985). Priority effects in experimental pond 12 13 communities: competition between Bufo and Rana. Ecology 66: 1097–1105. 13 14 Benson, C.W. (1950). A contribution to the ornithology of St Helena, and other 14 notes from a sea-voyage. Ibis 92: 75–83. 15 Bourne, W.P. (1957). The breeding birds of Bermuda. Ibis 99: 94–105. 15 16 Brooke, R.K., Lockwood, J.L. & Moulton, M.P. (1995). Patterns of success in 16 17 passeriform bird introductions on Saint Helena. Oecologia 103: 337–342. 17 18 Brown, J.H. (1995). Macroecology. Chicago, IL: Chicago University Press. 18 Clement, P., Harris, A & Davis, J. (1993). Finches and Sparrows: An Identification 19 Guide. London: Christopher Helm. 19 20 Colwell, R.K. & Winkler D.W. (1984). A null model for null models in 20 21 biogeography. In Ecological Communities: Conceptual Issues and the 21 Evidence, ed. D.R. Strong, Jr., D. Simberloff, L.G. Abele, & A.B. Thistle, pp. 22 344–359. Princeton, NJ: Princeton University Press. 22 23 Connell, J.H. (1983). On the prevalence and relative importance of interspecific 23 24 competition: evidence from field experiments. American Naturalist 122: 24 661–696. 25 Conner, E.F. & Simberloff, D. (1979). The assembly of species communities: 25 26 chance or competition? Ecology 60: 1132–1140. 26 27 Crowell, K.L. & Crowell, M.R. (1976). Bermuda’s abundant, beleaguered birds. 27 28 Natural History 85: 48–56. 28 Daehler, C.C. & Strong, Jr. D.R., (1993). Prediction and biological invasions. 29 Trends in Ecology and Evolution 8: 380. 29 30 Diamond, J.M. (1975). Assembly of species communities. In Ecology and 30 31 Evolution of Communities, ed. M.L. Cody & J.M. Diamond, pp. 342–444. 31 Cambridge, MA: University Press. 32 Drake, J.A. (1991). Community assembly mechanics and the structure of an 32 33 experimental species ensemble. American Naturalist 137: 1–26. 33 34 Drake, J.A., Mooney, H.A., di Castri, F., Groves, R.H., Kruger, F.J., Rejmanek, M. 34 & Williamson, M. (1989). Biological Invasions. A Global Perspective. 35 Chichester, UK: John Wiley. 35 36 Findley, J.S. (1973). Phenetic packing as a measure of faunal diversity. American 36 37 Naturalist 107: 580–584. 37 38 Findley, J.S. (1976). The structure of bat communities. American Naturalist 110: 38 129–139. 39 Fox, B.J. & Brown, J.H. (1993). Assembly rules for functional groups in North 39 40 American desert rodent communities. Oikos 67: 358–370. 40 124 J.L. Lockwood et al.

1 Gilpin, M.E. (1990). Ecological prediction. Science 248: 88–89. 1 2 Gilpin, M.E. & Diamond, J.M. (1982). Examination of the ‘null’ models of Connor 2 and Simberloff for species co-occurrences on islands. Oecologia 52: 64–74. 3 Grant, P.R. (1986). Ecology and Evolution of Darwin’s Finches, pp. 77–96. 3 4 Princeton, NJ: University Press. 4 5 Guild, E. (1938). Tahitian aviculture: acclimation of foreign birds. Avicultural 5 6 Magazine 3: 8–11. 6 Guild, F. (1940). Western bluebirds in Tahiti. Avicultural Magazine 5: 284–285. 7 Haydock, E.L. (1954). A survey of the birds of St Helena. Ibis 6: 818–821. 7 8 Kareiva, P. (1996). Special feature: developing a predictive ecology for non- 8 9 indigenous species and ecological invasions. Ecology 77: 1651–1697. 9 Karr, J.R. & James, F.C. (1975). Eco-morphological configurations and convergent 10 evolution in species communities. In Ecology and Evolution of Communities, 10 11 ed. M.L. Cody & J.M. Diamond, pp. 258–291. Cambridge, MA: Belknap 11 12 Press of Harvard University Press. 12 Lever, C. (1987). Naturalized Birds of the World. NY: Longman Scientific and 13 Technical. 13 14 Lockwood, J.L. & Moulton, M.P. (1994). Ecomorphological pattern in Bermuda 14 15 birds: the influence of competition and implications for nature reserves. 15 16 Evolutionary Ecology 8: 53–60. 16 Lockwood, J.L., Moulton, M.P. & Anderson, S.K. (1993). Morphological 17 assortment and the assembly of communities of introduced passeriforms on 17 18 oceanic islands: Tahiti versus Oahu. American Naturalist 141: 398–408. 18 19 Lockwood, J.L., Moulton, M.P. & Brooke, R.K. (1997). Morphological dispersion 19 of the introduced passeriforms of Saint Helena. Ostrich 67: 111–117. 20 Long, J. (1981). Introduced Birds of the World. London: David & Charles. 20 21 Moulton, M.P. (1993). The all-or-none pattern in introduced Hawaiian passeriforms: 21 22 the role of competition sustained. American Naturalist 141: 105–119. 22 Moulton, M.P. & Pimm, S.L. (1983).The introduced Hawaiian avifauna: 23 biogeographical evidence for competition. American Naturalist 121: 669–690. 23 24 Moulton, M.P. & Pimm, S.L. (1986). Species introduction to Hawaii. In Ecology of 24 25 Biological Invasions of North American and Hawaii, ed. H.A. Mooney & J.A. 25 26 Drake, pp. 231–249. NY: Springer-Verlag. 26 Moulton, M.P. & Pimm, S.L. (1987). Morphological assortment in introduced 27 Hawaiian passerines. Evolutionary Ecology 1: 113–124. 27 28 Pratt, H.D., Bruner, P.L. & Berrett, D.G. (1987). A Field Guide to the Birds of 28 29 Hawaii and the Tropical Pacific. Princeton, NJ: Princeton University Press. 29 Ricklefs, R.E. & Travis, J. (1980). Morphological approach to the study of avian 30 community organization. Auk 97: 321–338. 30 31 Schoener, T.W. (1983). Field experiments on interspecific competition. American 31 32 Naturalist 122: 24–285. 32 Scott, J.M., Mountainspring, S., Ramsey, F.L. & Kepler, C.B. (1986). Forest Bird 33 communities of the Hawaiian Islands: their dynamics, ecology and 33 34 conservation. Las Cruces, NM: Cooper Ornithological Society. 34 35 Sibley, C.G. & Monroe, Jr. B.L. (1990). Distribution and Taxonomy of Birds of the 35 36 World. New Haven, CT: Yale University Press. 36 Simberloff, D. (1992). Extinction, survival, and effects of birds introduced to the 37 Mascarenes. Acta Oecologica 13(6): 663–678. 37 38 Simberloff, D. & Boecklen, W. (1991). Patterns of extinction in the introduced 38 39 Hawaiian avifauna: a re-examination of the role of competition. American 39 Naturalist 138: 300–327. 40 Williamson, M. (1996). Biological Invasion. London: Chapman and Hall. 40 Introduced avifaunas as natural experiments 125

1 Williamson, M. & Fritter, A. (1996). The varying success of invaders. Ecology 77: 1 2 1661–1666. 2 Wilson, J.B. & Roxburgh, S.H. (1994). A demonstration of guild-based assembly 3 rules for a plant community, and determination of intrinsic guilds. Oikos 69: 3 4 267–276. 4 5 Wingate, D. (1973). A Checklist and Guide to the Birds, Mammals, Reptiles, and 5 6 Amphibians of Bermuda. Hamilton, Bermuda: Bermuda Audubon Society. 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 1 2 2 3 3 Relative .00000000 4 4 5 5 6 6 7 7 8 8 9 9 SS 1950 1876 1.76923076 2.30769230 10 NN native native 10 11 11 12 12 13 13 14 14 15 15 F 1853 0.46153846 16 FF 1820F 1820 1820F 1852 2.84615384 S 2.64615384 2.07692307 195016 3 17 17 18 18 19 19 20 20 21 21 22 22 S 1938 0.30769230 23 F 193823 0.46153846 24 24 25 25 26 26 All species introduced on Oahu, Tahiti, St Helena or Bermuda 27 27 28 28 29 29 SSS 1872F 1960SS 1928S S 1928 1966 1965 1908 F 1929 S 1938 S 1979 1820 F 1829 F 0.40909090 1983 1.07692307 0.61538461 0.53846153 0.23076923 0.46153846 0.30769230 30 FF 1926 1922FFF 1929 SF 1929 1931 1939 1932 30 1.30769230 1.23076923 1.23076923

31 Appendix Table 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 Acridotheres tristis Gracula religiosa Mimus gilvus Mimus polyglottos Parus varius Pycnonotus cafer Pycnonotus jocosus Zosterops japonicus Zosterops lateralis Corvus brachyrhynchos Rhipidura leucophrys Grallina cyanoleuca Dumetella carolinensis Sialia sialis Sialia mexicana Turdus merula Turdus philomelos Cyanoptila cyanomelana Erithacus rubecula Luscinia akahige Luscinia komadori Copsychus malabaricus Copsychus saularis Sturnus vulgaris 40 SpeciesPitangus sulphuratus Oahu DI Tahiti DI St Helena DI Bermuda DI40 range size 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 FF 1800 1970 3.45454545 0.15384615 10 11 11 12 12 13 13 14 14 15 15

16 FF 1929SF 1929F 1850 1929F 1929 1929FF 0.23076923 1929 1.15384615 1929 0.07692307 0.38461538 F 0.30769230 1.15384615 192916 0.46153846 0.6153846 0.53846153 17 17 18 18 19 19 20 20 21 21 22 22 23 F 1938 F 1820 FFF 1970F 1938 1.15384615 1938 23 1938 F 1929 1.30769230 0.07692307 24 24 25 25 26 26 27 27 28 28 29 29

30 SSSS 1929S 1947 S 1900 1928 1847 1871 F 1938FSS 1965S 1981 F 1965 FFF F 1965 1820F S 1870F 1965 1938 1965 1908S F 1965 S 1965 F 1938 F 1900 S 1870 F 193830 0.53846153 5.90909090 1938 0.30769230 1820 1938 F 4.53846153 0.38461538 F F 1938 1970 1929 1.38461538 S 1970 1.69230769 0.53846153 0.46153846 0.38461538 0.92307692 0.46153846 0.07692307 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 Garrulax caerulatus Garrulax canorus Leiothrix lutea Alauda arvensis Passer domesticus Passer montanus Sporopipes squamifrons Ploceus capensis Ploceus velatus Foudia madagascarensis Euplectes albonatus Euplectes progne Euplectes orix Pytilia melba Lagonosticta senegala Estrilda astrild Estrilda caerulescens Estrilda erythronotus Estrilda melpoda Estrilda melanotis Estrilda troglodytes Uraeginthus angolensis Uraeginthus bengalus Uraeginthus cyanocephala Uraeginthus granatinus Amandava amandava Amandava subﬂava Stagonopleura guttata Erythrura psittacea 40 Cettia diphone 40 1 0 0 1 2 2 3 3 Relative .00000000 4 4 5 5 6 6 7 7 8 8 9 9 10 F 196810 0.07692307 11 11 12 12 13 13 14 14 15 15 FFF 1929F 1929 1820F 1929S 1850 1776 0.76923076 0.38461538 2.38461538 0.76923076 0.15384615 0.30769230

16 ) 16 17 17 18 18

19 continued 19 20 20 21 21 22 22 23 FFF 1938F 1938F 1938S 1938 F 1938F 1899 F 1938 1938 S 1938FF 1899 1938 1938 0.07692307 0.30769230 0.0769230 23 0.23076923 0.0769230 0.07692307 0.23076923 1.38461538 1

24 Appendix Table ( 24 25 25 26 26 27 27 28 28 29 29

30 S 1964SSSF 1984 1936 1883 1962 F F 1938F S 193830 1965 1790 0.53846153 0.76923076 1.15384615 1.76923076 0.30769230 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 SpeciesErythrura cyaneovirens Oahu DI Tahiti DI St Helena DI Bermuda DI range size 40 Erythrura prasina Erythrura trichroa Neochmia modesta Neochmia ruﬁcauda Neochmia temporalis Poephila acuticauda Taeniopygia bichenovii Chloebia gouldiae Padda oryzivora Lonchura castaneothorax Lonchura cucullata Lonchura fringilloides Lonchura malabarica Lonchura malacca Lonchura punctulata Vidua macroura Vidua paradisea Vidua regia Fringilla coelebs Serinus atrogularis Serinus canaria Serinus canicollis Serinus ﬂaviventris Serinus leucopygius 40 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 . (1995), and Bermuda from 9 S 1800 2.40909090 N native 10 et al 10 11 11 12 12 13 13 14 14 15 15

16 F 1870 1.84615384 16 17 17 18 18 19 19 20 20 . (1994), Saint Helena from Brooke 21 21 22 et al 22 23 F 1938FFF 1938S 1938F 1938F 1938F 1938F 1938 1938 1938 F 1800 1.30769230 1.23076923 0.15384615 0.92307692 23 0.07692307 0.84615384 0.23076923 1.23076923 24 24 25 25 26 26 27 27 28 28 29 29 SSF 1870 1928 1931SSFF 1965F 1974S 1934 1941 1931 1929 0.76923076 0.30769230 0.15384615 S 0.76923076 1800 0.23076923 0.61538461 0.53846153 0.07692307 1.07692307 30 S 1964 F 1938 F 1929 1.07692307 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 Carduelis chloris Carduelis tristis Carpodacus mexicanus Paroaria coronata Paroaria dominicana Tachyphonus rufus Ramphocelus bresilius Ramphocelus carbo Ramphocelus dimidiatus Thraupis episcopus Tangara arthus Tangara larvata Cyanerpes cyaneus Sicalis flaveola Tiaris olivacea Passerina cyanea Passerina leclancherii Sturnella neglecta Cardinalis cardinalis Vireo griseus Total introductionsLockwood and Moulton (1994). 45* includes native species. Nomenclature follows Sibley and Monroe (1990). 41 31 17* Carduelis carduelis 40 Serinus mozambicus Total successesDI = date of introduction. F failed. S successful. Oahu dates taken from Moulton (1993), Tahiti Lockwood 2740 7 5 7 1 5 1 2 2 3 Assembly rules in plant communities 3 4 4 5 J. Bastow Wilson 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 Introduction 14 15 ‘To do science is to search for repeated patterns’ (MacArthur, 1972). There- 15 16 fore, to do community ecology must be to search for repeated community 16 17 patterns. Two basic kinds of community pattern can be envisaged, with 17 18 different causes: 18 19 19 (a) Environmentally mediated patterns, i.e., correlations between species 20 20 due to their shared or opposite responses to the physical environment. 21 21 Ecologists have long tried ‘to find out which species are commonly associ- 22 22 ated together upon similar habitats’ (Warming, 1909). Modern methods 23 23 allow more subtle questions to be examined, such as the shape of envi- 24 24 ronmental responses (e.g., Bio et al., 1998), the niche widths (e.g., Diaz 25 25 et al., 1994), and how repeatable the associations of species are (e.g., 26 26 Wilson et al., 1996c). However, the simple existence of environmentally 27 27 mediated patterns is now too obvious to need demonstrating; Warming 28 28 (1909) described it as ‘this easy task’. 29 29 (b) Assembly rules, i.e., patterns due to interactions between species, such as 30 30 competition. These patterns, when we can find them, are fascinating evi- 31 31 dence that competition, allelopathy, facilitation, mutualism, and all the 32 32 other biotic interactions that we know about in theory, actually affect 33 33 communities in the real world. 34 34 35 Of course, to make this distinction, it has to be known what processes have 35 36 caused each pattern – physical environment or biotic interactions – but that 36 37 is our task as community ecologists. Both types of process may occur. This 37 38 combination of processes can be seen as initial exclusion of species that are 38 39 unable to tolerate the physical environment (i.e., environmental filtering), fol- 39 40 lowed by the operation of assembly rules (i.e., biotic filtering; Keddy, 1992). 40

130 Assembly rules in plant communities 131

1 Table 5.1. Types of assembly rule 1 2 2 3 1 Rules based on particular named species 3 4 2 Rules based on presence/absence: 4 (a) variance in richness 5 (b) local vs. regional richness 5 6 (c) large-scale distributions. 6 7 3 Rules based on plant characters: 7 (a) texture convergence 8 (b) limiting similarity 8 9 (c) guild proportionality. 9 10 4 Rules based on species abundance: 10 11 (a) biomass constancy 11 (b) abundance-based guild proportionality 12 (c) RAD (relative abundance distribution): 12 13 (i) evenness 13 14 (ii) shape of the RAD. 14 15 15 16 16 17 The main aim here is to review the kinds of assembly rules that might exist 17 18 for plants, and to describe attempts to find examples (Table 5.1), but first the 18 19 definition of ‘assembly rule’, and some basic problems in assembly rule work 19 20 will be considered. 20 21 21 22 22 23 Definition of ‘assembly rule’ 23 24 An assembly rule definition will be followed that is based on that of Wilson 24 25 & Gitay (1995a): ‘ecological restrictions on the observed patterns of species 25 26 presence or abundance that are based on the presence or abundance of one 26 27 or more other species or groups of species (not simply the response of indi- 27 28 vidual species to the environment)’. The inclusion of ‘ecological’ indicates 28 29 that such a rule is intended to express the results of processes in ecological 29 30 time, not evolutionary processes such as character displacement. 30 31 Diamond (1975), in coining the term ‘assembly rule’, had defined rules in 31 32 terms of individual species, e.g., ‘Macropygia amboinensis and Macropygia 32 33 mackinlayi cannot both be present on any one island’. However, almost every 33 34 author since has envisaged rules of a less specific nature, making it easier to 34 35 generalize rules between communities. Some authors have bent the original 35 36 concept into rules that specify the processes by which communities are assem- 36 37 bled (Cole, 1983; Hunt, 1991); I retain the majority view that assembly rules 37 38 describe patterns that are evidence of such processes (e.g., Lawler, 1993; 38 39 Graves & Gotelli, 1993). 39 40 This definition, Diamond’s usage and any other definitions, do not include, 40 132 J.B. Wilson 1 as Keddy (1992) seems to, environmentally mediated patterns. In ecology 1 2 terms are often used, for effect, so far outside their original intent that they 2 3 lose any meaning. If this happens to ‘assembly rule’ with the inclusion of the 3 4 environmental sieve, as Keddy seems to advocate, another term will have to 4 5 be invented to cover real assembly rules, based only on species interactions. 5 6 That would be a nuisance. 6 7 7 8 8 9 Problems associated with assembly-rule work 9 10 Claims for demonstration of assembly rules have often caused considerable 10 11 debate (Weiher & Keddy, 1995). Some ecologists appear reluctant to accept 11 12 any evidence for their existence. One reason for this is belief that only exper- 12 13 imental evidence is useful in determining the importance of species inter- 13 14 actions in communities (Hastings, 1987; Goldberg, 1995; for a contrary view 14 15 see Wilson, 1995b and Wilson & Gitay, 1995a). Others are unsure that species 15 16 interactions are a significant force in structuring natural communities, and are 16 17 therefore wary of any evidence that they are (Simberloff, 1982; Wilson, 17 18 1991b). 18 19 Such caution and care is appropriate. If only it were shown more often in 19 20 ecology! Yet, it can occasionally be taken to extremes. In most guild work, 20 21 guilds are hypothesized, and hopefully then tested. This is a standard approach 21 22 in science. However, de Kroon and Olff (1995) argue that, in assembly-rule 22 23 work, the guild hypothesis has to be proved before it can be tested. 23 24 Evidence has to be examined very carefully, but the baby not thrown out with 24 25 the bathwater. Stone et al. (1996) produced valid criticisms of the assembly- 25 26 rule work of Fox & Brown (1993), but they entitled their paper: ‘Community- 26 27 wide assembly patterns unmasked ...’, as if the mistakes in Fox & Brown’s 27 28 analyzes invalidated all assembly-rule work. 28 29 There are a number of particular problems in assembly-rule work. 29 30 30 31 31 32 Problem 1: The difficulty in framing valid null models 32 33 In assembly-rule work, the null models have to be carefully chosen. When 33 34 we look for a correlation between two variates, or for a difference between 34 35 two groups (as for a t-test), it is fairly obvious what the data would look like 35 36 if there were no pattern, i.e., under a null model. With ecological communi- 36 37 ties, random assemblages may show surprising patterns (Lockwood et al., 37 38 1997), so it is far from clear what an assemblage of species would look like 38 39 were there no assembly rules. As Andrew D.Q. Agnew (pers. comm.) once 39 40 asked: ‘What does a plant community look like when it isn’t there?’. 40 Assembly rules in plant communities 133 1 Occasionally, assembly rules can be tested using standard statistical tests 1 2 (e.g., r or t). Usually, however, a special null model has to be devised, and 2 3 a corresponding randomization test used to determine significance. There are 3 4 two aspects of any such explicit null-model test: the test statistic (i.e., the 4 5 value which is compared between the observed data and the randomizations) 5 6 and the null model (i.e., in practical terms how the randomization is done). 6 7 Choosing a test statistic represents no problems of validity: any test statistic, 7 8 calculated on the observed and randomized data alike, is valid. The danger 8 9 is that, by choosing an inappropriate test statistic, evidence for an assembly 9 10 rule might be missed. From this point of view the choice of test statistic is 10 11 crucial. 11 12 Moreover, it is all too easy to use a null model which looks plausible, yet 12 13 yields answers that give no evidence on the existence of assembly rules. There 13 14 are three basic safeguards against this: 14 15 15 16 (a) The method should be tested on random datasets. A valid method should 16 17 produce significance at the 5% level in 5% of the random datasets, perhaps 17 18 fewer but certainly not more. This check can expose a lot of mistakes 18 19 (e.g., Wilson, 1987; Wilson, 1995c; Drobner et al., 1998). However, the 19 20 method of producing random datasets is important. If it is too similar to 20 21 the method used in the null-model randomization, the test will give only 21 22 5% significances by definition, and the test is of little value. 22 23 (b) The null model should include every feature of the observed data except 23 24 the one it is intended to test (Tokeshi, 1986). 24 25 There is a danger of making a null model that includes some of the 25 26 effects one is trying to test, the ‘Narcissus effect’ (Colwell and Winkler 26 27 1984). The Narcissus effect is bound to occur to some extent. For exam- 27 28 ple, a pool of extant species is considered, even though some species may 28 29 have become extinct because of competition. This gives a somewhat 29 30 conservative test. 30 31 The opposite problem (Wilson, 1995c) is a model that fails to include 31 32 obvious features, i.e., that ignores Tokeshi’s principle. For example, in 32 33 comparing islands it is pretty certain that some islands (large islands, 33 34 probably) will contain more species than others (small islands). If it is 34 35 not intended to test this feature, the island species-richnesses must be built 35 36 into the null model. Otherwise, a departure from the null model will 36 37 probably just be disclosing the obvious fact that large islands have more 37 38 species. The danger is that, if one does not notice what has happened, 38 39 this will spuriously appear to be an assembly rule, the trap into which 39 40 Fox and Brown (1993) fell. This is termed the ‘Jack Horner effect’ 40 134 J.B. Wilson 1 (Wilson, 1995c), a test in which the only conclusion from the analysis is 1 2 an already obvious fact, in the analogy that plum puddings contain plums1, 2 3 in ecology that, for example, large islands have more species. 3 4 A danger of the Narcissus effect (a conservative test, in effect an 4 5 unavoidable Type II error) should be accepted, rather than the Jack Horner 5 6 effect (a spurious ‘assembly rule’, a Type I error), especially when so 6 7 many skeptics are watching. 7 8 (c) As far as possible, the null model should represent an explicit ecological 8 9 process by which the community could conceivably have been assembled 9 10 (Manly, 1991). For example, Partel et al. (1996) used a null model phrased 10 11 in terms of the position of points on a graph, not in terms of the occur- 11 12 rence of species. They made the mistake of thinking that under an eco- 12 13 logical null model all possible values are equally likely, which is rarely 13 14 the case. This makes it very unclear what their model is testing. 14 15 Many null models are phrased in terms of the scattering of species 15 16 across quadrats or islands (e.g., Connor & Simberloff, 1979; Wilson et 16 17 al., 1987; almost all of them really). This is unrealistic, because it is indi- 17 18 viduals that are scattered onto an island (even then as part of population 18 19 processes). Yet, a model of species scattering generally has to be accepted, 19 20 because to introduce population processes would cause far more prob- 20 21 lems than it would solve. At least a model of species scattering is one 21 22 ecological step below the test statistic. 22 23 23 24 24 25 Problem 2: What assembly rule to test for 25 26 It is not known what assembly rules might be operating in any community. 26 27 Because ecologists are not plants, the contents of their rule book is unknown 27 28 to us. Therefore, it is not known what to test for (Wilson 1991b, 1994). Andrew 28 29 Agnew (pers. comm.) followed up his question: ‘What does a community look 29 30 like when it isn’t there?’ by asking: ‘What does a community look like when 30 31 it is there?’ 31 32 32 33 33 34 Problem 3: Environmental patchiness 34 35 The responses of species to the physical environment are of major interest in 35 36 ecology (e.g., Austin et al. 1990; Bio et al., 1998), but when we are seeking 36 37 assembly rules they are a nuisance, obscuring or mimicking the assembly 37 38 rules that we are looking for. Patchiness in successional stage has similar 38 39 39 40 1‘plum’ is used here in the sense of dried grape. 40 Assembly rules in plant communities 135 1 effects (Zobel et al., 1993). There is environmental variation everywhere. The 1 2 danger is that an analysis intended to test for assembly rules may merely 2 3 demonstrate that different species grow in different environments, which we 3 4 already know well (Warming, 1909). 4 5 A mitigating factor is that environmental differences, because they repre- 5 6 sent an increase in heterogeneity, usually lead to aggregated distributions, to 6 7 high variation (e.g., clustered gradient optima, differences in plant characters), 7 8 whereas most assembly rules, by their definition, lead to regular distributions, 8 9 to low variation (e.g., dispersed gradient optima, similarity in plant charac- 9 10 ters). Therefore, environmental variation often represents noise in our analyzes. 10 11 However, environmental variation can sometimes mimic assembly rules, i.e., 11 12 cause departures from the null model in the same direction as would our 12 13 assembly rule (e.g., different but equal species pools: Watkins & Wilson, 13 14 1992). 14 15 The problem arises at all scales. There can be environmental variation down 15 16 to the smallest scale we can conceive (Watkins & Wilson, 1992; Wilson et 16 17 al., 1992), and usually is. Heterogeneity is obvious if the sampling covers 17 18 several habitats (e.g. Weiher et al., 1998). On a yet larger scale, there is bio- 18 19 geographic heterogeneity (e.g., Fox & Brown, 1993; Armbruster et al., 1994). 19 20 The biggest danger is that the heterogeneity includes two or more species 20 21 pools. A null model that randomizes over all the quadrats (e.g., Fox & Brown, 21 22 1993; Weiher et al., 1998) is then obviously false before we start: all species 22 23 can not occur equally in all sites. The test therefore becomes a test between 23 24 species pools. Such a test is largely of the fundamental response of species 24 25 to the environment, i.e., a test of evolutionary patterns, and therefore arguably 25 26 not a test of assembly rules anyway. But, the greater problem is that there 26 27 are usually several quadrats in each micro-habitat/habitat/region, so that we 27 28 do not have as many independent species pools as the analysis implies. This 28 29 is related to the problem of spatial autocorrelation (see below). 29 30 The problem can be reduced by examining areas that are more uniform in 30 31 environment, but no area is ever completely uniform. Other solutions to the 31 32 problem are patch models (Fig. 5.1), in which the null-model predictions are 32 33 based on a small patch around the target quadrat (Watkins & Wilson, 1992; 33 34 Bycroft et al., 1993; Armbruster et al., 1994; Wilson & Watkins, 1994; Wil- 34 35 son & Gitay, 1995a), and accumulating variation within areas/environ- 35 36 ments/communities, but not between (e.g., Wilson & Roxburgh, 1994; Wil- 36 37 son et al. 1995a). 37 38 38 39 39 40 40 136 J.B. Wilson 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 Fig. 5.1. A patch model. The null-model prediction for each target quadrat is based 18 19 on the frequency or abundance of the species in a patch of quadrats (here nine quadrats) 19 20 centered on the target quadrat. With data comprising a line of adjacent quadrats, the 20 21 patch will be a line of quadrats centered on the target quadrat, typically seven. 21 22 22 23 Problem 4: Spatial autocorrelation 23 24 24 As in almost all ecological data, there are problems with spatial autocorrela- 25 25 tion: samples taken nearby are likely to be more similar. This does not bias 26 26 towards finding a certain result, but it affects significance testing, because it 27 27 means that the samples are not independent. The problem arises on all scales. 28 28 Samples only a few centimetres apart can be similar when the sampled area 29 29 spans a metre, but so can samples only a few miles apart when the sampled 30 30 area spans a hundred miles. 31 31 There must be thousands of demonstrations in the literature that a partic- 32 32 ular vegetational gradient correlates significantly with soil water status, pH, 33 33 elevation, or whatever. It can be guessed that more than 99% of these tests 34 34 are invalid because of spatial autocorrelation. No one minds, because no one 35 35 doubts that vegetational gradients are related to water, pH and elevation. 36 36 When assembly rules are proposed, many people do mind. They should. 37 37 Four solutions are known: 38 38 39 (a) Patch models (Watkins & Wilson, 1992). These examine only a small 39 40 group of adjacent quadrats at a time (Fig. 5.1). There is therefore little 40 Assembly rules in plant communities 137 1 opportunity for nearby quadrats to be more similar, because all the 1 2 quadrats in the patch are close to the target quadrat. In fact, such meth- 2 3 ods tend to be a bit conservative in the face of spatial autocorrelation 3 4 (Watkins & Wilson, 1992); at least this conservatism removes the prob- 4 5 lem of obtaining spurious significance because of spatial autocorrelation 5 6 (Wilson, 1995d). Armbruster et al. (1994) used what is essentially a patch 6 7 model, but on a biogeographic scale. 7 8 (b) Accumulate variation within areas (e.g., Wilson & Roxburgh, 1994; see 8 9 above). 9 10 (c) Rotation/reflection and random shifts methods (Palmer & van der Maarel, 10 11 1995). These retain the exact spatial pattern for each species, but ran- 11 12 domize the orientation or position of the pattern between species. They 12 13 are therefore immune to spatial autocorrelation. They are more restricted 13 14 in the type of sampling scheme for which they can be used, ideally a 14 15 contiguous grid or a circular transect. However, their major drawback is 15 16 that they randomize across any environmental patchiness, giving the 16 17 possibility of spurious ‘assembly rules’ (Problem (c) above). 17 18 (d) The ‘random patterns’ test (Roxburgh & Chesson, 1998). 18 19 19 20 20 21 Problem (e): Subtle effects 21 22 Problems (c) and (d) above are especially severe because the effects that we 22 23 seek in assembly-rule work are generally quite subtle, tendencies rather than 23 24 strict rules (Fox & Brown, 1993; Wilson & Roxburgh, 1994), easily over- 24 25 whelmed by sampling error and by environmental noise. 25 26 26 27 27 28 Presence/absence rules 28 29 Some assembly rules are based on the presence and absence of species in 29 30 samples. 30 31 31 32 32 33 Variance in richness 33 34 The simplest type of presence/absence assembly rule is one in terms of species 34 35 richness. Much theory (Fig. 5.2) is based on the idea that species too simi- 35 36 lar in niche cannot coexist (e.g., Pacala & Tilman, 1994). If this is true, the 36 37 number of species that can coexist locally should be limited, because there 37 38 is a limited number of niches (Ricklefs, 1987). This would give low variance 38 39 in species richness, compared to a null model. This effect has proved 39 40 surprisingly difficult to find, but it can be found, at least at a small scale. 40 138 J.B. Wilson 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 Fig. 5.2. The theory of limiting similarity, leading to the concept of niche limitation 13 14 (cf. May & MacArthur, 1972). 14 15 15 16 Watkins & Wilson (1992) sampled a number of lawns, examining richness 16 17 in 1⁄2″×1⁄2″ quadrats. For six of these sites, the observed frequency histogram 17 18 of species richness was narrower than expected on a random basis, i.e., species 18 19 richness was more constant, an assembly rule (Fig. 5.3). It is possible that 19 20 some of this effect is due to physical constraints on individual module pack- 20 21 ing (Watkins & Wilson, 1992; Palmer & van der Maarel, 1995), e.g., packing 21 22 of leaves (some people refer to plant packing; however, the non-packable unit 22 23 can vary from a leaf to a clone). Yet, up to five species (average 1.6) can be 23 24 found at a point in this lawn, so quite a few can be packed into a 1⁄2″×1⁄2″ 24 25 quadrat. And plant canopies are normally only 3–5% occupied by plant 25 26 modules. At least it’s not a simple module packing effect. 26 27 Even stronger effects of this type can be found at the smallest scale, that 27 28 of a point (Wilson et al., 1992, 1996b). It seems likely that a considerable 28 29 part of this effect is due to constraints on module packing. However, that is 29 30 not the whole story. For example, one treatment of Wilson et al. (1992) had 30 31 received selective herbicide. By the first census, 6 weeks after herbiciding, 31 32 leaf module density had already been restored to that of the control, but 32

33 species packing was signiﬁcantly less tight (i.e., RVr was higher: Fig. 5.4). 33 34 As usual in assembly-rule work, spatial heterogeneity in environment may 34 35 be a confounding factor. One way to overcome this is to record the same 35 36 quadrat thorough time. Wilson et al. (1995b) did this on limestone grassland 36 37 in Sweden. Variance in richness, when adjusted for overall year-to-year vari- 37 38 ance in richness, was signiﬁcantly less than null-model expectation at two 38 39 sites (Table 5.2). 39 40 40 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22

23 quadrats on six lawns, compared to that expected under a null model. (After 23 ″ 2 ⁄

24 1 24 25 25 ″× 2 ⁄ 26 1 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 Watkins & Wilson, 1992.) 40 Fig. 5.3. The distribution of species richness in 40 140 J.B. Wilson

1 Table 5.2. Temporal variance in richness, 1985–1989, in three sites of limestone 1 2 grassland on Öland, Sweden, sampled with 3 × 3 cm quadrats, and adjusted for 2 overall year-to-year changes in richness 3 3 4 4 Site RV P 5 r 5 6 Gettlinge 1.036 ns 6 7 Kleva 0.786 0.018 7 Skarpa Alby 0.785 0.017 8 8 9 9 RVr is an index of variance in richness: a value of 1.0 indicates variance equal to 10 that expected at random, a value < 1.0 indicates variance lower than this (a possible 10 11 assembly rule). P indicates the significance of the departure from the null-model 11 12 value of 1.0. 12 13 13 14 14 15 Local vs. regional richness 15 16 The work above sought a species-richness assembly rule at a fine scale. The 16 17 opposite approach is to work at the continental level, examining different 17 18 species pools. Connell & Lawton (1992) pointed out that niche limitation 18 19 could be seen by plotting local species richness against the size of the regional 19 20 pool. If there were niche limitation, the curve would flatten out (Fig. 5.5). 20 21 There are two problems with this approach. First, there has been uncertainty 21 22 as to how to test for this pattern (Cresswell et al. 1995; Partel et al., 1996; 22 23 Caley & Schluter, 1997), the best idea to date being to test for significant 23 24 concave-down non-linearity in regression. The second problem is that the 24 25 points must be reasonably independent, i.e., based on independent species 25 26 pools. Whereas at the local level one can sample 100, 1000 or 10000 26 27 individual quadrats, it is difficult to find many more than ten continents on 27 28 Earth with independent species pools (Lawton et al., 1993 obtained data 28 29 for six pools). Inflating the number of continents by including several taxa 29 30 does not help (Caley & Schluter, 1997); even if the taxa can be considered 30 31 as independent, there is no reason to suppose that saturation would occur at 31 32 similar pool and richness values in different groups of organisms. Unless a 32 33 method of analysis can be found that copes with overlapping pools, this 33 34 approach will not reach its potential until data can be included from other 34 35 planets. 35 36 36 37 37 38 Large-scale distributions 38 39 Most large-scale patterns can be related to the environment, the ‘easy task’ 39 40 of Warming (1909). However, there are features of zonations that could 40 Assembly rules in plant communities 141 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 Fig. 5.4. Variance in richness at a point in the Botany Lawn, University of Otago, 15 New Zealand. An RV value < 1.0 indicates a trend towards constant species richness. 16 r 16 (After Wilson et al., 1992.) 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 Fig. 5.5. The relation between local species richness and the size of the regional species 34 pool. ‘Theoretical boundary’ indicates that there logically cannot be more species 35 locally than in the regional pool. ‘Open’ shews a community-type with no limitation 35 36 to local species richness. ‘Saturating’ shews a community-type where the local 36 37 community becomes saturated with species (cf. Connell & Lawton, 1992). 37 38 38 39 39 40 40 142 J.B. Wilson 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 Fig. 5.6. The distribution of the five species of Hebe (sensu lato) occurring along an 33 altitudinal gradient on a mountain range in south-west New Zealand. (After Wilson 34 & Lee, 1993.) 34 35 35 36 36 37 indicate assembly rules. For example, some have envisaged communities as 37 38 sets of co-evolved, co-adapted species. Dice (1952) wrote: ‘All the species 38 39 which are members of a given association ... are adjusted more or less per- 39 40 fectly to one another ... [due to] interco-ordinated evolution’. More recently, 40 Assembly rules in plant communities 143 1 Gilpin (1994) envisaged ‘cohesive’, ‘self-organized’ communities; if they 1 2 met, they would not mix, but would ‘battle as coordinated armies’, each com- 2 3 munity tending to repel an invader from the enemy. Surely that would be an 3 4 assembly rule! One would then get sudden boundaries between communities 4 5 along a gradual environmental gradient, i.e., the species boundaries would be 5 6 clustered. 6 7 Of the attempts that have been made to test this, probably the best is that 7 8 of Auerbach and Shmida (1993). However, this whole approach founders on 8 9 the impossibility of defining the environmental scale, and therefore of deter- 9 10 mining whether or not the environment changes gradually (Wilson, 1994). 10 11 The problem can be overcome by asking particular questions that do not 11 12 depend on the environmental scale, e.g., whether congeners tend to exclude 12 13 each other, because they occupy similar alpha niches (Boutin & Harper, 1991). 13 14 Wilson and Lee (1993) found slight evidence for this (Fig. 5.6). Another such 14 15 question is whether species replace each other along the gradient. Dale (1984) 15 16 asked this question in a way that did not depend on the environmental scale 16 17 used (i.e., it was a non-parametric method), and found significant evidence 17 18 for a pattern, to my surprise (Wilson, 1994). 18 19 19 20 20 21 Assembly rules based on plant characters 21 22 Results based on species richness are hard to interpret, because of possible 22 23 effects of module packing. The most interesting rules are therefore those based 23 24 on the characters of the plants: limiting similarity, texture convergence and 24 25 guild proportionality (Leps, 1995; Wilson 1995a; Weiher & Keddy, 1995). 25 26 These approaches have the advantage that in the null-model analyses the 26 27 number of species per quadrat is generally held equal to that observed. There- 27 28 fore, unlike rules based on species richness, there can be no simple effect of 28 29 any constraint on module packing. For example, in guild proportionality 29 30 analyses, for a two-species quadrat there are two species in that quadrat 30 31 in both the observed and randomized data, the question is only the guild 31 32 membership of the two species present. Any difference between the observed 32 33 and randomized data cannot be due to there being only two species in that 33 34 quadrat, because this feature does not differ between the observed and the 34 35 randomized. 35 36 36 37 37 38 Limiting similarity 38 39 The limiting similarity approach looks to see whether pairs of species, that 39 40 are judged to be similar in niche, co-occur less often that one would expect 40 144 J.B. Wilson 1 on a random basis. Looking at it the other way around: do species that co- 1 2 occur have less niche overlap than expected at random? 2 3 Cody (1986) examined the leaf sizes of Leucadendron species in Cape 3 4 Province, South Africa. He determined morphological overlap between two 4 5 species as an intersection of their respective male–female morphological 5 6 ranges, an intuitively obvious if logically puzzling procedure. Cody states that 6 7 morphologically overlapping species co-occurred at a site significantly less 7 8 often than expected at random. The basis of the null model is not entirely 8 9 clear, but this represents a pioneering attempt to examine limiting similarity. 9 10 There have been many attempts to find evenly spaced (i.e., staggered) flow- 10 11 ering times within a community (e.g., Pleasants, 1980), although with many 11 12 methodological arguments (e.g., Fleming & Partridge, 1984; Pleasants, 1990). 12 13 Ranta et al. (1981) used a limiting similarity approach to this problem. They 13 14 examined the plant species in two boreal fields, and clustered them into groups 14 15 with similar pollinator visitors. In one field, species within two out of seven 15 16 groups were significantly more evenly spaced in flowering time than expected 16 17 at random (Fig. 5.7). In another field three out of six groups showed signif- 17 18 icantly even spacing. So, species that share pollinators can coexist more read- 18 19 ily if they differ in flowering time. Moreover, pairs of species that overlapped 19 20 in both pollinator visitors and flowering time usually differed in a character 20 21 such as plant height or flower color (though there was no statistical test for 21 22 this). These characters would encourage different groups of pollinators, and 22 23 therefore reduce competition for pollinators. It is not quite clear whether the 23 24 result is due to sorting in ecological time, and thus qualifies as an assembly 24 25 rule, or whether it represents ‘the ghosts of competition past’ (Connell, 1980). 25 26 Probably there are elements of both. 26 27 Weiher et al. (1998) tested for limiting similarity in the species of several 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 Fig. 5.7. The distribution of the flowering times of species in a field at Puumala, 39 Finland, group PA (the groups are based on similarity in pollinator visitors, using a 39 40 cluster analysis). (After Ranta et al., 1981.) 40 Assembly rules in plant communities 145 1 wetland habitats. For three characters, they found a significant tendency 1 2 towards even spacing, in that different quadrats tended to contain species that 2 3 spanned a wide range of values. All three characters were related to plant 3 4 size. This pattern was confirmed by guild-proportionality analyses. So, 4 5 quadrats tended to have some large species and also some small species. This 5 6 concept is related to guild proportionality in stratification (Wilson, 1989; 6 7 Wilson et al., 1995a; Wilson & Roxburgh, 1994). The problem with the 7 8 analyzes of Weiher et al. is that species occurrences were explicitly sampled 8 9 and randomized over several habitats. The problems mentioned above there- 9 10 fore arise, that this is mainly a test of species pools rather than rules for 10 11 assembly within a community, and that although there were 115 quadrats, 11 12 there were not 115 independent species pools. 12 13 In my view, a very model of how to undertake assembly-rules work is the 13 14 study by Armbruster et al. (1994) of Stylidium species in Western Australia. 14 15 These plants have a complicated pollination system involving bees and flies, 15 16 different flower shapes allowing different pollinators. Armbruster et al. used 16 17 a patch model to allow for biogeographic variation. Their null model indi- 17 18 cates how often pairs of species, that are similar in flower type, would co- 18 19 occur on a random basis (Fig. 5.8). In fact, the observed data show very few 19 20 such co-occurrences of similar pairs. Using a one-tailed test, Armbruster found 20 21 the result significant (Fig. 5.8). A two-tailed test should probably have been 21 22 used, by which P > 0.05. There was strong significance when character dis- 22 23 placement was included in the test, but, as evolutionary change, character 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 Fig. 5.8. Frequency distribution, under an ecological null model, of the number of 38 co-occurring Stylidium species pairs in an area of Western Australia that overlap in 39 pollination niche, and the observed number of such overlaps. (After Armbuster et al., 39 40 1994.) 40 146 J.B. Wilson 1 displacement does not seem to qualify as an assembly rule. Armbruster (1986) 1 2 had done a similar test using the tropical Euphorbiad Dalechampia. Again, 2 3 species assembly was significant using a one-tailed test, but not using a two- 3 4 tailed one. However, this approach looks promising. 4 5 Armbruster (1995) suggested that character displacement might occur more 5 6 often in reproductive characters, but that since the limitations to coexistence 6 7 which are the basis of assembly rules would operate over short distances, 7 8 assembly rules are more likely to be related to the vegetative characters of 8 9 the species. 9 10 10 11 11 12 Texture convergence 12 13 Texture refers to the range of plant characters in a community, irrespective 13 14 of taxon (Barkman, 1979). The characters considered are generally those 14 15 believed to be indicative of niche, e.g., leaf thickness, leaf angle, NPK con- 15 16 tent, chlorophyll content, respiration rate, rooting pattern, etc. For example, 16 17 a grassland has a different texture from a shrubland, because of differences 17 18 in leaf shape, woodiness, etc. An assembly rule in this context is observed 18 19 when biotic interactions cause convergence: similar texture in different sites, 19 20 even those on different continents. 20 21 Many workers have assumed that community convergence would have to 21 22 be the result of evolution, and have felt obliged to apologize for their lack of 22 23 information on ancestors (e.g., Orians & Paine, 1983; Schluter, 1986; Wiens, 23 24 1991a). Weiher and Keddy (1995) took the opposite extreme, assuming that 24 25 all ‘trait overdispersion’ was caused immediately by competition, i.e., by 25 26 ecological sorting. 26 27 If ecological sorting is occurring, then when two species are present that 27 28 are too similar, one of them will suffer competitive exclusion (Fig. 5.9(a)). 28 29 This assumes that there is a pool of species, of which some are eliminated, 29 30 but indeed the regional species pool, even after physical filtering, is generally 30 31 larger than the local pool, depending of the spatial scale examined (Partel et 31 32 al., 1996). Such ecological sorting will occur during colonization of empty sites. 32 33 It will continue to operate as species from the regional pool continue to invade, 33 34 either failing to establish due to suppression by superior competitors vying 34 35 for the same niche space, or causing functionally similar species already pre- 35 36 sent to succumb to competitive exclusion. If there are niches in a community 36 37 for a range of functional types, with more or less one species per niche, the 37 38 result would be expected to be convergence between comparable communi- 38 39 ties in different areas. The same pressures will act via selection in evolu- 39 40 tionary time to cause evolutionary convergence between regions (Fig. 5.9(b)). 40 Assembly rules in plant communities 147 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 Fig. 5.9. Ecological and evolutionary processes leading to convergence. The square 18 area represents functional space. Ecological and evolutionary processes can both cause 18 19 an even distribution of species in functional space. If this occurs independently in 19 20 different continents⁄sites⁄patches, the communities will converge. 20 21 21 22 Thus, convergence, in the sense of communities with a more similar dis- 22 23 tribution of species in niche space than expected on the basis of random 23 24 assortment from species pools (e.g., Wilson et al., 1994; Smith et al., 1994), 24 25 can be produced by either ecological or evolutionary processes. Comparing 25 26 continents, it is difficult to distinguish between ecological and evolutionary 26 27 convergence. However, at the community level evolutionary convergence 27 28 might be expected to be rare, because most species occur in several differ- 28 29 ent associations, and cannot coevolve simultaneously to fit in with each set 29 30 of associates (Goodall, 1966). The caveat to this argument is ecotypic adjust- 30 31 ment to different associates, in effect character displacement. 31 32 Most studies of texture convergence have compared Mediterranean shrub- 32 33 lands. The problem has been the absence of a null model, which has led some 33 34 to the defeatist view that such work is impossible (Blondel et al., 1984; Keely, 34 35 1992). However, Wilson et al. (1994) developed a suitable null model (related 35 36 to earlier work by Schluter, 1986), and used it to look for convergence in 36 37 carrs (i.e., wooded fens) in Britain and New Zealand. They measured five 37 38 functional characters related to light capture (Fig. 5.10). Species presence/ 38 39 absence data revealed no convergence. However, when species were weighted 39 40 by their abundance, convergence was seen in some variates. 40 148 J.B. Wilson 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 Fig. 5.10. Aspects of the texture of two carrs in the United Kingdom and two in New 21 Zealand. The size of the symbol indicates the abundance (photosynthetic biomass) of 21 22 the species, and the steps the hierarchy of modules, and the shape leaf shape. (After 22 23 Wilson et al., 1994.) 23 24 24 25 Some studies with animals have also failed to find texture convergence 25 26 (e.g., Wiens, 1991b). In testing assembly rules it is usually necessary to 26 27 contend with environmental noise, thus convergence may sometimes not be 27 28 demonstrable because the environments of the continents are not similar enough. 28 29 With texture convergence, there is the additional problem of historical noise, 29 30 i.e., the different evolutionary and biogeographic history of different conti- 30 31 nents may have resulted in species pools that are too different for convergence 31 32 to have been completed. A negative texture-convergence result could also be 32 33 because our analyses are not yet sophisticated enough to see convergence 33 34 above the historical and environmental noise. A recent search for texture con- 34 35 vergence in the non-Nothofagus component of Nothofagus forest in Argentina, 35 36 Chile, New Zealand, Australia and Tasmania, has demonstrated that, if 36 37 one looks at only the shape of the distribution, to control for environmental 37 38 variation, significant convergence is demonstrable (B. Smith & J.B. Wilson, 38 39 unpublished data). 39 40 Texture convergence might occur on a much smaller scale too. For exam- 40 Assembly rules in plant communities 149 1 ple, Smith et al. (1994) compared stands of Nothofagus forest a few hundred 1 2 meters apart. They obtained some evidence for convergence, although they 2 3 were aware of problems in testing convergence using abundance information 3 4 when there is species overlap. This problem has now been solved (J.B. Wilson 4 5 & B. Smith, unpublished data), and comparisons of texture in patches of com- 5 6 munities a few meters apart are currently in process (A.J. Watkins & J.B. 6 7 Wilson, unpublished data). Here, ecological convergence, not evolutionary is 7 8 clearly being examined. The question is: if two patches are compared, and, 8 9 in the second patch, a species is present in lower abundance, is that 9 10 compensated by greater abundance in a species with similar functional 10 11 characters? 11 12 12 13 13 14 Guild proportionality 14 15 The texture approach characterizes a community by the range of functional 15 16 characters within it. An alternative is to divide the species into arbitrarily 16 17 discrete guilds (= functional types, = functional groups), according to their 17 18 functional characters. Convergence can then be sought as a proportional 18 19 representation from different guilds that is constant between patches of a 19 20 community. Wilson (1989) attempted this for a gymnosperm/angiosperm rain- 20 21 forest, but found little evidence for an assembly rule. Wilson et al. (1995a) 21 22 used the same method on Nothofagus rainforests, and found a significant 22 23 tendency towards constant proportions of species from the ground and herb 23 24 guilds. 24 25 Wilson and Roxburgh (1994) sampled a species-rich, stable lawn, and 25 26 found regularity in the proportion of species from grass and forb guilds, a 26 27 rule which held whether bryophytes were counted as forbs, or omitted (Fig. 27 28 5.11(a), (b)). A similar assembly rule was found in guilds based on height 28 29 strata (Fig. 5.11(c)). 29 30 Ecologists skeptical of assembly rules have an important point, that it is 30 31 very easy to find spurious rules. The grass : forb rule of Wilson and Roxburgh 31 32 could be reproduced by confusion in identification between different species 32 33 of grass. There are sampling errors in all work, but one has to be especially 33 34 wary of errors that would mimic the effect being sought. Our result was sub- 34 35 mitted for publication only several analyzes had convinced us that this was 35 36 not the cause of our result (Wilson, 1995a). 36 37 Fox and Brown (1993), seeking an assembly rule very similar to guild 37 38 proportionality (they examined numbers, not proportions) fell into the ‘Jack 38 39 Horner’ trap (Wilson, 1995c), because they failed to include, in their null 39 40 model, differences in frequency between species. The departure from their 40 150 J.B. Wilson 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 Fig. 5.11. Variation in guild proportions on the Botany Lawn, University of Otago, 22 New Zealand, compared to that expected under a null model. An RVgp value < 1.0 23 indicates a trend towards constant guild proportions (i.e., guild proportionality). (After 23 24 Wilson & Roxburgh, 1994.) 24 25 25 26 26 27 null model can therefore be seen as just a demonstration that some species 27 28 are more frequent than others – hardly an assembly rule (Wilson, 1995c). Fox 28 29 and Brown (1993) also failed to incorporate in their null model that some 29 30 species were absent from some sites because their geographic range did not 30 31 include the whole area of sampling (Stone et al., 1996). 31 32 Sometimes, taxonomy has been used as a rough guild classification. Mohler 32 33 (1990) made a comparison with two subgenera of Quercus. For 12 of the 14 33 34 regions that he examined there was a significant tendency for the two most 34 35 dominant oak species to be from different subgenera. This was not related to 35 36 consistent pairing of particular species. There are some problems with his 36 37 data and his null model, but the result is still impressive. Mohler examined 37 38 various explanations: disease/pest pressure, niche differences in fruiting 38 39 phenology through mast fruiting, dispersal differences, etc., but could not find 39 40 any clear single explanation. The result is most easily explained by assum- 40 Assembly rules in plant communities 151 1 ing more niche similarity within subgenera than between, but it is not known 1 2 what niche axis is involved. 2 3 Similar questions, of guild proportionality, have been asked at coarser guild 3 4 levels, such as the proportions of predator and prey (including plant) species 4 5 in communities. Several claims have been made for constancy in these 5 6 proportions, which would be an impressive assembly rule (e.g., Jeffries & 6 7 Lawton, 1985; Gaston et al., 1992). Pimm (1991) gave the supposed assembly 7 8 rule of near-constant predator : prey ratios as an example of ecological theory 8 9 that could be used to guide management decisions in nature conservation. 9 10 Unfortunately for this widely touted ‘rule’, a null-model comparison shews 10 11 that predator : prey ratios are usually about as constant as expected at random; 11 12 Wilson (1996) was able to find no case of predator : prey ratios significantly 12 13 more constant than expected at random. 13 14 14 15 15 16 Intrinsic guilds 16 17 As discussed above, one of the problems in seeking assembly rules is that it 17 18 is not known what rules to seek, what is in the plants’ rule book. This is a 18 19 particular problem with rules based on plant characters. There is ignorance 19 20 not only of the assembly rules operating but also of which characters the rules 20 21 apply to. All limiting-similarity work, all texture work, and almost all guild work, 21 22 has had to start with an assumption of which characters might be important, 22 23 and should be measured. 23 24 In guild work, this problem can be overcome. If guilds are defined as groups 24 25 of species which compete more with each other than with species in other 25 26 guilds (the guild definition of Pianka, 1980, 1988), and if such guilds exist, 26 27 and if competition limits species coexistence, then there must be guild 27 28 proportionality. That is to say, if there are guilds limiting coexistence, we can 28 29 find them (Wilson & Roxburgh, 1994). An a priori hypothesis of what the 29 30 guilds are is not needed, because ‘the plants can be asked’ what guilds they 30 31 are in. To do this, the guild classification that maximizes the degree of guild 31 32 proportionality is found (of course, to avoid circularity the classification has 32 33 to be determined on one part of the data, and tested on independent 33 34 data). This allows the intrinsic guilds to be determined, i.e., the guilds intrin- 34 35 sic to the community, the guilds that the plants see, not those imposed by the 35 36 ecologist. 36 37 Correlations then might be found between the intrinsic guilds and the 37 38 characters that are known. Or no correlations might be seen, because the 38 39 significant characters are ones that have not been measured, or because 39 40 coexistence depends on complicated combinations of characters. However, 40 152 J.B. Wilson

1 Table 5.3. Intrinsic guilds in the Botany Lawn, University of Otago, New Zealand, 1 2 sampled at point scale 2 3 3 4 Guild A Guild B 4 5 Agrostis capillaris Acrocladium cuspidatum 5 6 Anthoxanthum odoratum Cerastium fontanum 6 7 Bellis perennis Cerastium glomeratum 7 Holcus lanatus Eurhynchium praelongum 8 Hydrocotyle moschata Festuca rubra 8 9 Linum catharticum Hydrocotyle heteromeria 9 10 Poa pratensis Hypochaeris radicata 10 Trifolium dubium Prunella vulgaris 11 Trifolium repens Ranunculus repens 11 12 Sagina procumbens 12 13 13 14 14 15 15 16 Table 5.4. Intrinsic guilds in a Welsh salt marsh, sampled at the scale of a 2 cm 16 17 diameter circle 17 18 18 19 Guild A Guild B 19 20 Festuca rubra Plantago lanceolata 20 21 Puccinellia maritima Salicornia europaea 21 22 Triglochin maritima Spergularia media 22 Suaeda maritima 23 Variable in guild membership: 23 24 Armeria maritima Juncus maritimus 24 25 Glaux maritima Aster tripolium 25 26 26 27 27 28 the intrinsic guilds are, within our assumptions, still the truth1, even though 28 29 there might be no idea of what plant interaction mechanism is causing them. 29 30 Using this method on a lawn, Wilson and Roxburgh (1994) found signif- 30 31 icant guild structure, and it seemed to be related to leaf shape and orienta- 31 32 tion (Table 5.3). Wilson and Whittaker (1995) used the method on a salt 32 33 marsh, and again found significant guild structure related to leaf shape and 33 34 orientation (Table 5.4). It is too soon to tell, but the possibility of a general 34 35 assembly rule emerges, if comparable guilds can be found in other commu- 35 36 nities. 36 37 One advantage of the intrinsic guild approach is that it can fail. Other meth- 37 38 ods of classifying species into guilds are bound to give guilds. With an intrin- 38 39 39 40 1i.e., P < 0.05, which counts as the truth! 40 Assembly rules in plant communities 153 1 sic guild search, if there are no guilds or there is no competitive exclusion, 1 2 the search will fail, and it often does (Wilson et al., 1995a,b,c). 2 3 3 4 4 5 Abundance-based assembly rules 5 6 Assembly rules can also be based on quantitative information, i.e., on species 6 7 abundances. 7 8 8 9 9 10 Biomass constancy 10 11 One obvious abundance-based rule, trivial almost, would be a constancy 11 12 (between patches of a community) of total biomass because of competition: 12 13 when the abundance of one species is higher, that of another or of others is 13 14 lower (Fig. 5.12). Wilson and Gitay (1995a), in a Welsh dune slack, exam- 14 15 ined variance in total biomass between quadrats, and compared it with that 15 16 expected under a null model in which the biomasses of the species were allo- 16 17 cated at random. Habitat heterogeneity makes it difficult to demonstrate such 17 18 a rule, but they did so by using a patch model. It is possible to obtain evi- 18 19 dence for competition from snapshot field data (cf. Goldberg, 1995; Wilson, 19 20 1995a). 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 Fig. 5.12. The concept of biomass constancy: when the biomass of one species is 39 40 higher, that of another or others is lower (artificial data). 40 154 J.B. Wilson 1 1 2 Abundance-based limiting-similarity / texture / guilds 2 3 The limiting similarity, texture convergence, guild proportionality and intrin- 3 4 sic guild approaches could, in theory, all be based on abundance information, 4 5 ideally biomass. No such approaches with limiting similarity are known. Plant 5 6 texture convergence work has included abundance information from the 6 7 beginning (Wilson et al., 1994). 7 8 Guild proportionality could be approached with abundance data. The ques- 8 9 tion ‘is there a constant proportion of species from each guild’ then becomes 9 10 ‘is there a constant proportion of biomass from each guild’. Wilson et al. 10 11 (1996a) calculated abundance-based guild proportionality within plots of the 11 12 Park Grass experiment. There had been no evidence of guild proportionality 12 13 using presence/absence data, and there was none using biomass data. 13 14 Preliminary searches for intrinsic guilds in semi-arid grassland in New 14 15 Zealand, using abundance (biomass) information, have given negative results 15 16 (e.g., all species in one guild: Table 5.5). 16 17 One would expect that the addition of information on abundances would 17 18 make it easier to see structure, but that does not seem to be the case so far. 18 19 19 20 20 21 Relative abundance distribution (RAD) 21 22 A notable feature of any community is the distribution of relative abundances 22 23 of the species (RAD), irrespective of species identities. To display an RAD, 23 24 we normally use a ranked-abundance plot: the log of abundance against the 24 25 rank order of abundance (Fig. 5.13). An assembly rule based on the RAD 25 26 could be very general, because the RAD of any community can be found, 26 27 and RADs can be compared worldwide. It is not necessary to know anything 27 28 about the species involved, except that they are distinct species. The RAD 28 29 is also the sort of information that is observed instinctively: if in a commu- 29 30 nity 95% of the biomass is one species, this is commented on. If there are 30 31 many species with about equal abundance, this is commented on. 31 32 32 33 33 34 Evenness 34 35 The most obvious feature of the RAD is evenness, in effect the lack of slope 35 36 on the ranked-abundance plot (although not calculated that way: Smith & 36 37 Wilson, 1996). 37 38 Grime (1973, 1979) predicted a humped-back curve for richness – an 38 39 increase and then a decrease as total community biomass increased. Using 39 40 the same logic that has been used for the humped-back richness model 40 Assembly rules in plant communities 155

1 Table 5.5. The result of an intrinsic guild search in a New Zealand semi-arid 1 2 grassland, sampled at the scale of 10 cm × 10 cm 2 3 3 4 Guild A Guild B 4 5 Agrostis sp. 5 6 Aira caryophyllea 6 7 Anthoxanthum odoratum 7 Breutelia affinis 8 Carex breviculmis 8 9 Cladonia sp. 9 10 Echium vulgare 10 Erodium cicutarium 11 Erophila verna 11 12 Geranium sessiliflorum 12 13 Gypsophila australis 13 14 Hieracium pilosella 14 Hypericum perforatum 15 Leucopogon fraseri 15 16 Minuartia hybrida 16 17 Myosotis discolor 17 Neofuscella sp. 18 Pimelea sericeo-villosa 18 19 Poa maniototo 19 20 Polytrichum juniperinum 20 Raoulia apicenigra 21 Raoulia australis 21 22 Rumex acetosella 22 23 Rytidosperma buchanani 23 24 Stellaria gracilenta 24 Trifolium arvense 25 Triquetrella papillata 25 26 Veronica verna 26 27 Vulpia dertonensis 27 Xanthoparmelia amphixantha 28 Xanthoparemelia cf. tasmanica 28 29 29 30 30 31 31 32 (Drobner et al. 1998) predicted a monotonic decrease for evenness. They 32 33 examined this in a number of communities in a high-rainfall area of New 33 34 Zealand, and obtained the relation predicted (Fig. 5.14). However, a very 34 35 similar result can be found using random numbers (Drobner et al., 1998). The 35 36 effect seems to be due to the roughly ‘geometric’ shape of most RADs, and 36 37 the fact that evenness calculations are based on the proportional relations of 37 38 the species, but total biomass is the sum of their absolute values. Here is 38 39 another illustration of the need to examine apparent assembly rules carefully, 39 40 and to test the method with random data. 40 156 J.B. Wilson 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 Fig. 5.13. A ranked-abundance plot, used to display the relative abundance distribution 18 19 (RAD) (artificial data). 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 Fig. 5.14. The relation between evenness (measured with index E′) and photosynthetic 36 37 biomass, as seen in ten distinct communities in south-west New Zealand. (After 37 Drobner et al., 1998.) 38 38 39 39 40 40 Assembly rules in plant communities 157 1 However, some interesting trends in evenness can be seen. Wilson et al. 1 2 (1996d) found that, in higher-phosphate plots of the Park Grass experiment, 2 3 evenness was higher. They also found that, in two grasslands, evenness increased 3 4 through succession, a topic on which there had been much speculation. 4 5 5 6 6 7 Shape of the relative abundance distribution (RAD) 7 8 The shape of the RAD contains information other than the slope. There are 8 9 five simple shapes possible (Fig. 5.15). S-shaped curves (b) and approximately 9 10 geometric slopes (c) seem common in the literature (e.g., Wilson, 1991a; Lee 10 11 et al., 1991; Watkins & Wilson, 1994). Concave-up slopes (a) are sometimes 11 12 found. Reverse-S (d) and concave-down (e) shapes seem very uncommon. 12 13 This is an assembly rule of a very general kind. 13 14 It has often been suggested that particular shapes of RAD will occur in 14 15 particular conditions (see Wilson, 1991a). The usual way to examine this has 15 16 been by comparison with models of RAD, for plants: Broken stick, Geometric, 16 17 General Lognormal and Zipf–Mandelbrot (Wilson, 1991a). 17 18 Watkins and Wilson (1994) examined 12 communities, with replicate 18 19 quadrats. They found highly significant differences in the RAD of different 19 20 communities, but could find no rule to predict which sort of community would 20 21 have what sort of RAD. Wilson et al. (1996d) compared a range of about 80 21 22 fertilizer treatments in the Park Grass experiment. There was no consistent 22 23 relation between the RAD and any soil factor, except for a slight tendency 23 24 for the Broken Stick model to fit less poorly where phosphate had been 24 25 applied, and that apparently just because evenness was higher. 25 26 The discovery of consistent trends in more subtle aspects of the relative 26 27 abundance pattern remains elusive. It cannot be predicted which shape of 27 28 RAD will appear when, in spite of ancient claims in the literature (see Wilson, 28 29 1991a; Watkins & Wilson, 1994). It has to be suspected that the identity of 29 30 individual species has too great an effect for general rules to be possible. 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 Fig. 5.15. Five basic shapes possible for relative abundance distributions (RADs). 40 158 J.B. Wilson

1 Conclusions 1 2 The story so far 2 3 3 Assembly rules can be found in plant communities, but they are challenging 4 4 to find. Rules can be seen for variance in richness, but interpretation of them 5 5 is ambiguous. Some rules can be seen for texture convergence, limiting sim- 6 6 ilarity and guild proportionality, and we can sometimes find intrinsic guilds. 7 7 Currently, these give the best evidence. Evidence for biomass constancy can 8 8 be found, with difficulty, but this is a trivial rule. Rules that are based on the 9 9 abundance of the species and are more subtle than biomass constancy seem 10 10 elusive. Some predictability can be seen in evenness, and in the general shapes 11 11 of the RAD that are found, but predictivity of the shape of the RAD for a 12 12 particular community is currently non-existent. 13 13 It is known that there is intense competition in natural communities. It is 14 14 known that species differ widely in their morphology, and in their physiol- 15 15 ogy. How can this not affect their ability to survive in the big wide world? 16 16 Probably the difficulty in finding assembly rules in plant communities is 17 17 because: 18 18 19 (a) there is variation at all scales in environment and in patch history, and 19 20 this represents noise in assembly-rule work, 20 21 (b) the true assembly rules are complex, e.g., because of: 21 22 variation in niche width, and 22 23 the existence of alternative plant strategies (‘there is more than one 23 24 way to kill a cat’), and 24 25 (c) the processes that produce assembly rules are subtle. 25 26 26 27 27 28 The way forward 28 29 The continued search for assembly rules requires large datasets, with high- 29 30 quality data such as the 2810 point quadrats Wilson and Roxburgh (1994) or 30 31 the 534 biomass quadrats of Wilson et al. (1996d). Data are required from 31 32 relatively homogeneous environments, and/or sampling using a grid or a 32 33 circular transect (Bycroft et al., 1993) so that patch models and the like can 33 34 be used. 34 35 More sophisticated methods of analysis are also required, especially to 35 36 utilize abundance data. For example, methods have recently been developed 36 37 for analyzing texture convergence that use the shape of the character distri- 37 38 bution, not just the mean (B. Smith & J.B. Wilson, unpublished data), and 38 39 also methods that can examine texture convergence even with complete 39 40 overlap in the species present (J.B. Wilson & B. Smith, unpublished data). 40 Assembly rules in plant communities 159 1 Methods are needed that can better partition out environmental noise and cope 1 2 with spatial autocorrelation (Watkins & Wilson, 1992; Palmer & van der 2 3 Maarel, 1995). 3 4 Insight into the operation of assembly rules might also be obtained by 4 5 examining the ecological and physiological mechanisms that are the basic 5 6 cause of the rules, such as canopy and root interactions. For example, it would 6 7 be interesting to know the species–species interactions that cause guild 7 8 proportionality (e.g., in the lawn community of Wilson & Roxburgh, 1994). 8 9 However, because the rules are usually only tendencies, it may be difficult 9 10 to see the relevant mechanisms above the noise of other processes. 10 11 11 12 12 13 Where and when will assembly rules be found? 13 14 Generalizations should be sought on where and when community structure 14 15 will be strong, giving assembly rules. Some might expect assembly rules to 15 16 operate mainly in species-rich communities, since more species–species inter- 16 17 actions are possible, and since species-rich communities may be closer to 17 18 niche saturation. However, the best evidence so far has come from quite 18 19 species-poor communities: lawn and salt marsh (Wilson & Roxburgh, 1994; 19 20 Wilson & Whittaker, 1995). 20 21 Where assembly rules exist, when do they operate? Are they continually 21 22 present? Or do they operate only in some years, for example during periods 22 23 when resources are more limiting (Wilson et al., 1995b; cf. Wiens 1977)? 23 24 24 25 25 26 Why assembly rules anyway? 26 27 If assembly rules are common, and are a significant community force, they 27 28 may prove useful for predicting responses to environmental change and to 28 29 biotic introductions, and for effective community restoration. Drake et al. 29 30 (1996) propose, albeit on rather weak evidence, the ‘Humpty Dumpty’ effect: 30 31 that it may not be possible to establish a community simply by bringing 31 32 together the constituent species, but only by knowing the assembly sequence. 32 33 It would be premature to conclude, as some do, that assembly rules are 33 34 absent. There is some evidence for them, and the search is worthwhile. Even 34 35 if they are usually not there, it is important that is discovered. 35 36 36 37 37 38 Acknowledgments 38 39 I thank for comments on a draft W.S. Armbruster, W.G. Lee, E. Weiher and 39 40 members of my research group. 40 160 J.B. Wilson 1 References 1 2 Armbruster, W.S. (1986). Reproductive interactions between sympatric 2 3 Dalechampia species: are natural assemblages ‘random’ or organized? Ecology 3 67: 522–533. 4 Armbruster, W.S. (1995). The origins and detection of plant community structure: 4 5 reproductive versus vegetative processes. Folia Geobotanica et 5 6 Phytotaxonomica 30: 483–497. 6 7 Armbruster, W.S., Edwards, M.E., & Debevec, E.M. (1994). Floral character 7 displacement generates assemblage structure of Western Australian 8 triggerplants (Stylidium). Ecology 75: 315–329. 8 9 Auerbach, M. & Shmidaa, A. (1993). Vegetation change along an altitudinal 9 10 gradient on Mt Hermon, Israel – no evidence for discrete communities. 10 Journal of Ecology 81: 25–33. 11 Austin, M.P., Nicholls, A.O., & Margules, C.R. (1990). Measurement of the 11 12 realized qualitative niche: environmental niches of five Eucalyptus species. 12 13 Ecology 60: 161–177. 13 Barkman, J.J. (1979). The investigation of vegetation texture and structure. In The 14 Study of Vegetation, ed. M.J.A. Werger, pp. 123–160. The Hague, The 14 15 Netherlands: Junk. 15 16 Bio, A.M.F., Alkemade, R., & Barendregt, A. (1998). Determining alternative 16 17 models for vegetation response analysis: a non-parametric approach. Journal of 17 Vegetation Science 9: 5–16. 18 Blondel, J., Vuilleumier, F., Marcus, L.F., & Terouanne, E. (1984). Is there 18 19 ecomorphological convergence among Mediterranean bird communities of 19 20 Chile, California, and France? Evolutionary Biology 18: 141–213. 20 Boutin, C. & Harper, J.L. (1991). A comparative study of the population dynamics 21 of five species of Veronica in natural habitats. Journal of Ecology 79: 21 22 199–221. 22 23 Bycroft, C.M., Nicolaou, N., Smith, B., & Wilson, J.B. (1993). Community 23 structure (niche limitation and guild proportionality) in relation to the effect of 24 spatial scale, in a Nothofagus forest sampled with a circular transect. New 24 25 Zealand Journal of Ecology 17: 95–101. 25 26 Caley, M.J. & Schluter, D. (1997). The relationship between local and regional 26 27 diversity. Ecology 78: 70–80. 27 Cody, M.L. (1986). Structural niches in plant communities. In Community Ecology 28 ed. J.M. Diamond & T.J. Case, pp. 381–405. NY: Harper & Row. 28 29 Cole, B.J. (1983). Assembly of mangrove an communities: patterns of geographical 29 30 distribution. Journal of Animal Ecology 52: 339–347. 30 Colwell, R.K. & Winkler, C.W. (1984). A null model for null models in 31 biogeography. In Ecological Communities: Conceptual Issues and the 31 32 Evidence, ed. D.R. Strong, D. Simberloff, L.G. Abele, & A.B. Thistle, pp. 32 33 344–359. Princeton, NJ: Princeton University Press. 33 Connell, J.H. (1980). Diversity and the coevolution of competitors, or the ghost of 34 competition past. Oikos 35: 131–138. 34 35 Connell, H.V. & Lawton, J.H. (1992). Species interactions, local and regional 35 36 processes, and limits to the richness of ecological communities: a theoretical 36 37 perspective. Journal of Animal Ecology 61: 1–12. 37 Connor, E.F. & Simberloff, D. (1979). The assembly of species communities: 38 chance or competition? Ecology 60: 1132–1140. 38 39 Cresswell, J.E., Vidal-Martinez, V.M., & Crichton, N.J. (1995). The investigation 39 40 of saturation in the species richness of communities: some comments on 40 methodology. Oikos 72: 301–304. Assembly rules in plant communities 161

1 Dale, M.R.T. (1984). The contiguity of upslope and downslope boundaries of 1 2 species in a zoned community. Oikos 42: 92–96. 2 3 de Kroon, H. & Olff, H. (1995). On the use of the guild concept in plant ecology. 3 Folia Geobotanica et Phytotaxonomica 30: 519–552. 4 Diamond, J.M. (1975). Assembly of species communities. In Ecology and 4 5 Evolution of Communities, ed. M.L. Cody & J.M. Diamond, pp. 342–444. 5 6 Cambridge, MA: Harvard University Press. 6 7 Diaz, S.; Acosta, A.; Cabido, M. (1994). Grazing and the phenology of ﬂowering 7 and fruiting in a montane grassland in Argentina – a niche approach. Oikos, 8 70: 287–295. 8 9 Dice, L.R. (1952). Natural Communities. Ann Arbor: University of Michigan Press. 9 10 Drake, J.A., Huxel, G.R., & Hewitt, C.L. (1996). Microcosms as models for 10 generating and testing community theory. Ecology 77: 670–677. 11 Drobner, U., Bibby, J., Smith, B., & Wilson, J.B. (1998). The relation between 11 12 community biomass and evenness: what does community theory predict, and 12 13 can these predictions be tested? Oikos 82: 295–302. 13 Fleming, T.H. & Partridge, B.L. (1984). On the analysis of phenological overlap. 14 Oecologia (Berlin) 62: 344–350. 14 15 Fox, B.J. & Brown, J.H. (1993). Assembly rules for functional groups in North 15 16 American desert rodent communities. Oikos 67: 358–370. 16 17 Gaston, K.J., Warren, P.H., & Hammond, P.M. (1992). Predator: non-predator 17 ratios in beetle assemblages. Oecologia, 90: 417–421. 18 Gilpin, M. (1994). Community-level competition: asymmetrical dominance. 18 19 Proceedings of the National Academy of Sciences, USA, 91: 3252–3254. 19 20 Goldberg, D.E. (1995). Generating and testing predictions about community 20 structure: which theory is relevant and can it be tested with observational data? 21 Folia Geobotanica et Phytotaxonomica 30: 511–518. 21 22 Goodall, D.W. (1966). The nature of the mixed community. Proceedings of the 22 23 Ecological Society of Australia 1: 84–96. 23 Graves, G.R. & Gotelli, N.J. (1993). Assembly of avian mixed-species ﬂocks in 24 Amazonia. Proceedings of the National Academy of Sciences, USA, 90: 24 25 1388–1391. 25 26 Grime, J.P. (1973). Competitive exclusion in herbaceous vegetation. Nature 242: 26 27 344–347. 27 Grime, J.P. (1979). Plant Strategies and Vegetation Processes. Chichester, UK: 28 Wiley. 28 29 Hastings, A. (1987). Can competition be detected using species co-occurrence data? 29 30 Ecology 68: 117–123. 30 Hunt, G.L. Jr (1991). Occurrence of polar seabirds at sea in relation to prey 31 concentrations and oceanographic factors. Polar Research 10: 553–559. 31 32 Jeffries, M.J. & Lawton, J.H. (1985). Predator-prey ratios in communities of 32 33 freshwater invertebrates: the role of enemy free space. Freshwater Biology 15: 33 105–112. 34 Keddy, P.A. (1992). Assembly and response rules – two goals for predictive 34 35 community ecology. Journal of Vegetation Science 3: 157–164. 35 36 Keeley, J.E. (1992). A Californian’s view of fynbos. In The Ecology of Fynbos – 36 37 Nutrients, Fire and Diversity, ed. R.M. Cowling, pp. 372–388. Cape Town, 37 South Africa: Oxford University Press. 38 Lawler, S.P. (1993). Direct and indirect effects in microcosm communities of 38 39 protists. Oecologia 93: 184–190. 39 40 Lawton, J.H., Lewinsohn, T.M. & Compton, S.G. (1993). Patterns of diversity for 40 the insect herbivores on bracken. In Species Diversity in Ecological 162 J.B. Wilson

1 Communities, ed. R.E. Ricklefs & D. Schluter. pp. 178–184. Chicago, IL: 1 2 University of Chicago Press. 2 Lee, W.G., Ward, C.M., & Wilson, J.B. (1991). Comparison of altitudinal 3 sequences of vegetation on granite and on metasedimentary rocks in the 3 4 Preservation Ecological District, southern Fiordland, New Zealand. Journal of 4 5 the Royal Society of New Zealand 21: 261–276. 5 6 Leps, J. (1995). Variance deficit is not reliable evidence for niche limitation. Folia 6 Geobotanica et Phytotaxonomica 30: 455–459. 7 Lockwood, J.L., Powell, R.D., Nott, M.P., & Pimm, S.L. (1997). Assembling 7 8 ecological communities in time and space. Oikos 80: 549–553. 8 9 MacArthur, R.H. (1972). Geographical Ecology: Patterns in the Distribution of 9 Species. NY: Harper & Row. 10 McIntosh, R.P. (1975). H.A. Gleason – ‘Individualistic Ecologist’ 1882–1975: his 10 11 contribution to ecological theory. Bulletin of the Torrey Botanical Club 102: 11 12 253–273. 12 Manly B.F.J. (1991). Randomization and Monte-Carlo Methods in Biology. 13 London: Chapman & Hall. 13 14 May, A.U. & MacArthur, R.H. (1972). Niche overlap as a function of 14 15 environmental variability. Proceedings of the National Academy of Sciences, 15 16 USA 69: 1109–1113. 16 Mohler, C.L. (1990). Co-occurrence of oak subgenera: implications for niche 17 differentiation. Bulletin of the Torrey Botanical Club 117: 247–255. 17 18 Orians, G.H. & Paine, R.T. (1983). Convergent evolution at the community level. 18 19 In Coevolution, ed. D.J. Futuyma & M. Slatkin, pp. 431–458. Sunderland, 19 USA: Sinauer. 20 Pacala, S.W. & Tilman, D. (1994). Limiting similarity in mechanistic and spatial 20 21 models of plant competition in heterogeneous environments. American 21 22 Naturalist 143: 222–257. 22 Palmer, M.W. & van der Maarel, E. (1995). Variance in species richness, species 23 association, and niche limitation. Oikos 73: 203–213. 23 24 Partel, M., Zobel, M., Zobel, K., & van der Maarel, E. (1996). The species pool 24 25 and its relatio to species richness: evidence from Estonian plant communities. 25 26 Oikos 75: 111–117. 26 Pianka, E.R. (1980). Guild structure in desert lizards. Oikos 35: 194–201. 27 Pianka, E.R. (1988). Evolutionary Ecology, 4th edn. NY: Harper and Row. 27 28 Pimm, S.L. (1991). The Balance of Nature? Chicago, IL: University of Chicago 28 29 Press. 29 Pleasants, J.M. (1980). Competition for bumblebee pollinators in Rocky Mountain 30 plant communities. Ecology 61: 1446–1459. 30 31 Pleasants, J.M. (1990). Null-model tests for competitive displacement: the fallacy of 31 32 not focussing on the whole community. Ecology 71: 1078–1084. 32 Ranta, E., Teräs, I., & Lundberg, H. (1981). Phenological spread in flowering of 33 bumblebee-pollinated plants. Annales Botanici Fennici 18: 229–236. 33 34 Ricklefs, R.E. (1987). Community diversity: relative roles of local and regional 34 35 processes. Science 235: 167–171. 35 36 Roxburgh, S.H. & Chesson, P. (1998). The Random Patterns test: a new method for 36 detecting species associations in the presence of spatial autocorrelation. 37 Ecology, in press. 37 38 Schluter, D. (1986). Tests for similarity and convergence of finch communities. 38 39 Ecology 67: 1073–1085. 39 Simberloff, D. (1982). The status of competition theory in ecology. Annales 40 Zoologici Fennici 19: 241–253. 40 Assembly rules in plant communities 163

1 Smith, B. & Wilson, J.B. (1996). A consumer’s guide to evenness indices. Oikos 1 2 76: 70–82. 2 Smith, B., Moore, S.H., Grove, P.B., Harris, N.S., Mann, S., & Wilson, J.B. (1994). 3 Vegetation texture as an approach to community structure: community-level 3 4 convergence in a New Zealand temperate rainforest. New Zealand Journal of 4 5 Ecology 18: 41–50. 5 6 Stone, L., Dayan, T., & Simberloff, D. (1996). Community-wide assembly patterns 6 unmasked: the importance of species’ differing geographical ranges. American 7 Naturalist 148: 997–1015. 7 8 Tokeshi, M. (1986). Resource utilization, overlap and temporal community 8 9 dynamics: a null model analysis of an epiphytic chironomid community. 9 Journal of Animal Ecology 55: 491–506 10 Warming, E. (1909). Oecology of Plants: An Introduction to the Study of Plant- 10 11 Communities. Oxford, UK: Oxford University Press. 11 12 Watkins, A.J. & Wilson, J.B. (1992). Fine-scale community structure of lawns. 12 Journal of Ecology 80: 15–24. 13 Watkins, A.J. & Wilson, J.B. (1994). Plant community structure, and its relation to 13 14 the vertical complexity of communities: dominance/diversity and spatial rank 14 15 consistency. Oikos 70: 91–98. 15 16 Weiher, E. & Keddy, P.A. (1995). Assembly rules, null models, and trait 16 dispersion: new questions from old patterns. Oikos 74: 159–164. 17 Weiher, E., Clarke, G.D.P. & Keddy, P.A. (1998). Community assembly rules, 17 18 morphological dispersion, and the coexistence of plant species. Oikos 81: 18 19 309–322. 19 Wiens, J.A. (1977). On competition and variable environments. American Scientist, 20 65: 590–597. 20 21 Wiens, J.A. (1991a). Ecological similarity of shrub-desert avifaunas of Australia 21 22 and North America. Ecology 72: 479–495. 22 Wiens, J.A. (1991b). Ecomorphological comparisons of the shrub-desert avifaunas 23 of Australia and North America. Oikos 60: 55–63. 23 24 Wilson, J.B. (1987). Methods for detecting non-randomness in species co- 24 25 occurrences: a contribution. Oecologia 73: 579–582. 25 26 Wilson, J.B. (1989). A null model of guild proportionality, applied to stratiﬁcation 26 of a New Zealand temperate rain forest. Oecologia 80: 263–267. 27 Wilson, J.B. (1991a). Methods for ﬁtting dominance / diversity curves. Journal of 27 28 Vegetation Science 2: 35–46. 28 29 Wilson, J.B. (1991b). Does Vegetation Science exist? Journal of Vegetation Science 29 2: 289–290. 30 Wilson, J.B. (1994). Who makes the assembly rules? Journal of Vegetation Science 30 31 5: 275–278. 31 32 Wilson, J.B. (1995a). Testing for community structure: a Bayesian approach. Folia 32 Geobotanica et Phytotaxonomica 30: 461–469. 33 Wilson, J.B. (1995b). Hypothesis testing, intrinsic guilds and assembly rules. Folia 33 34 Geobotanical et Phytotaxonomica 30: 535–536. 34 35 Wilson, J.B. (1995c). Null models for assembly rules: the Jack Horner effect is 35 36 more insidious than the Narcissus effect. Oikos 72: 139–143. 36 Wilson, J.B. (1995d). Variance in species richness, niche limitation, and vindication 37 of patch models. Oikos 73: 277–279. 37 38 Wilson, J.B. (1996). The myth of constant predator: prey ratios. Oecologia 106: 38 39 272–276. 39 Wilson, J.B. & Lee, W.G. (1994). Niche overlap of congeners: a test using plant 40 altitudinal distribution. Oikos 69: 469–475. 40 164 J.B. Wilson

1 Wilson, J.B. & Roxburgh, S.H. (1994). A demonstration of guild-based assembly 1 2 rules for a plant community, and determination of intrinsic guilds. Oikos 69: 2 267–276. 3 Wilson, J.B. & Watkins, A.J. (1994). Guilds and assembly rules in lawn 3 4 communities. Journal of Vegetation Science 5: 591–600. 4 5 Wilson, J.B. & Gitay, H. (1995a). Limitations to species coexistence: evidence for 5 6 competition from field observations, using a patch model. Journal of 6 Vegetation Science 6: 369–376. 7 Wilson, J.B. & Gitay, H. (1995b). Community structure and assembly rules in a 7 8 dune slack: variance in richness, guild proportionality, biomass constancy and 8 9 dominance/diversity relations. Vegetatio 116: 93–106. 9 Wilson, J.B. & Whittaker, R.J. (1995). Assembly rules demonstrated in a saltmarsh 10 community. Journal of Ecology 83: 801–807. 10 11 Wilson, J.B., Agnew, A.D.Q., & Gitay, H. (1987). Does niche limitation exist?. 11 12 Functional Ecology 1: 391–397. 12 Wilson, J.B., Roxburgh, S.H., & Watkins, A.J. (1992). Limitation to plant species 13 coexistence at a point: a study in a New Zealand lawn. Journal of Vegetation 13 14 Science 3: 711–714. 14 15 Wilson, J.B., Agnew, A.D.Q., & Partridge, T.R. (1994). Carr texture in Britain and 15 16 New Zealand: convergence compared with a null model. Journal of Vegetation 16 Science 5: 109–116. 17 Wilson, J.B., Allen, R.B., & Lee, W.G. (1995a). An assembly rule in the ground 17 18 and herbaceous strata of a New Zealand rainforest. Functional Ecology 9: 18 19 61–64. 19 Wilson, J.B., Sykes, M.T., & Peet, R.K. (1995b). Time and space in the community 20 structure of a species-rich grassland. Journal of Vegetation Science 6: 20 21 729–740. 21 22 Wilson, J.B., Steel, J.B., Newman, J.E., & Tangney, R.S. (1995c). Are bryophyte 22 communities different? Journal of Bryology 18: 689–705. 23 Wilson, J.B., Crawley, M.J., Dodd, M.E., & Silvertown, J. (1996a). Evidence for 23 24 constraint on species coexistence in vegetation of the Park Grass experiment. 24 25 Vegetatio 124: 183–190. 25 26 Wilson, J.B., Lines, C.E.M., & Silvertown, J. (1996b). Grassland community 26 structure under different grazing regimes, with a method for examining species 27 associations when local richness is constrained. Folia Geobotanica et 27 28 Phytotaxonomica 31: 197–206. 28 29 Wilson, J.B., Ullmann, I., & Bannister, P. (1996c). Do species assemblages ever 29 recur? Journal of Ecology 84: 471–474. 30 Wilson, J.B., Wells, T.C.E. Trueman, I.C., Jones, G., Atkinson, M.D. Crawley, 30 31 M.J., Dodd, M.E., & Silvertown, J. (1996d). Are there assembly rules for plant 31 32 species abundance: An investigation in relation to soil resources and 32 successional trends. Journal of Ecology 84: 527–538. 33 Zobel, K., Zobel, M., & Peet, R.K. (1993). Changes in pattern diversity during 33 34 secondary succession in Estonian forests. Journal of Vegetation Science 4: 34 35 489–498. 35 36 36 37 37 38 38 39 39 40 40 1 6 1 2 2 3 Assembly rules at different scales in plant and 3 4 bird communities 4 5 5 6 6 7 Martin L. Cody 7 8 8 9 9 10 10 11 11 12 12 13 13 14 Introduction: assembly rules, scale, purpose and application 14 15 One of the basic premises in ecology is that communities are composed of 15 16 collections of species that are subsets of a larger pool of available species (e.g., 16 17 Cody & Diamond, 1975; Diamond & Case, 1986), and that the composition 17 18 of these subsets is governed, potentially at least, by ‘assembly rules’ (Diamond, 18 19 1985). In particular, and as originally envisaged by Diamond (1985), if com- 19 20 munity size varies with, e.g., island size, vegetation structure, or some other 20 21 extrinsic factor, then such rules will also govern community amplification. 21 22 Ideally, assembly rules should cover both species composition and relative 22 23 abundance, the two more basic variables or descriptors of the community. 23 24 Conceptually, assembly rules are of two basic categories. One type, which 24 25 will be called ‘type A’, governs community membership or species compo- 25 26 sition as community size increases or decreases. Thus, type A rules address 26 27 which species are added as community size increases, e.g., with island size 27 28 (MacArthur & Wilson, 1967), or with the stature or complexity of vegetation 28 29 (MacArthur et al., 1966; Cody, 1973), or decreases in smaller or in more 29 30 isolated habitat fragments with poor recolonization potential (Cody, 1973; 30 31 Bolger et al., 1991), or decreases with time as former landbridge islands 31 32 equilibrate, either real islands (Diamond & Mayr, 1976; Diamond et al., 1976) 32 33 or habitat islands (Brown, 1971). 33 34 The extent to which there are identifiable differences between random 34 35 samples of species and individuals drawn from a species-individuals pool and 35 36 observed community replicates is the extent to which the second ‘type B’ cat- 36 37 egory of assembly rules operate. Here, the reference is not to communities 37 38 that vary in size in response to some extrinsic factor, but rather to those which 38 39 are more or less constant in size yet differ in species composition. Assuming 39 40 a fixed resource base, there may be alternative suites of species that comprise 40

165 166 M.L. Cody 1 stable or equilibrial consumer sets (e.g., Case & Casten, 1979). Assembly rules 1 2 in this case evaluate ecological compatibility; some species are compatible 2 3 by dint of some form or extent of different resource utilization and thus can 3 4 coexist as community members, whereas others are precluded in combina- 4 5 tion(s) because of their ecological similarities and resultant competition. 5 6 Assembly rules of either type, A or B, may each address different compo- 6 7 nents of diversity (see, e.g., Cody, 1975, 1986, 1993). At one level, assembly 7 8 rules might control the number of locally coexisting species, or α-diversity, 8 9 where the equilibrial species number may reflect a balance between colo- 9 10 nization and extinction rates, as on islands or in fragmented habitat patches, 10 11 or simply a matching of some subset of a wide array of potential consumers 11 12 to the resources present. At a second level, assembly rules might relate 12 13 community amplification and/or a changing species composition, i.e., species 13 14 turnover, to a changing resource base, as on islands of increasing size or in 14 15 habitats of increasingly complex structure, and thus address β-diversity. Third, 15 16 such rules might regulate species turnover and alternative community com- 16 17 position within different parts of a habitat’s range, and identify the qualifi- 17 18 cations of ecological counterparts that contribute to γ-diversity. 18 19 The determinants of community composition can be examined at several 19 20 scales, and it is an examination of assembly rules at these various scales that 20 21 is the focus of this chapter. Briefly, these are described as follows. 21 22 (a) The ‘classical’ case is that of species sets on islands, where islands 22 23 of comparable size and isolation support comparable species numbers, and 23 24 numbers are enhanced with increased island size and reduced distance from 24 25 a source biota. Then, for a given island size and isolation, one can ask the 25 26 questions: Is species number predictable? Is species composition predictable? 26 27 Are species substitutable (ecological equivalents or counterparts)? Are species 27 28 added in predictable sequence with increased size or decreased isolation? 28 29 Here, such questions are asked of plant species on islands in Barkley Sound, 29 30 Vancouver Island, British Columbia, using long-term census data that have 30 31 been collected over more than a decade. 31 32 (b) Within a single, broadly distributed habitat type, questions of habitat 32 33 patch size or isolation are less prominent. In this case, assembly rules are 33 34 sought that define and govern the extent to which communities are replicated 34 35 across the geographical range of the habitat. The relevant questions are: Are 35 36 the species’ totals, or α-diversities, constant across the habitat range? If they 36 37 are not, do they vary predictably with, e.g., latitude, habitat structure, or some 37 38 other extrinsic variable? Is species composition predictable and constant? If 38 39 not, are ecological substitutes or counterparts identifiable, and are rules 39 40 discernible that predict when the substitutions occur? 40 Scale and pattern in plant and bird communities 167 1 Reference is made here to several studies of this type that bear on these 1 2 questions, but data collected from two habitat types are discussed and empha- 2 3 sized in the chapter: mulga bushland across Australia, and oak woodland in 3 4 western North America, from Baja California north through Alta California 4 5 to Washington. 5 6 (c) Within a site and within a given habitat type, vegetation structure 6 7 typically varies, and with it the community may change at a local or patch 7 8 scale. Between years, abiotic aspects of habitat may vary (e.g., wet years vs. 8 9 dry years), and so also may the availability of candidates for community mem- 9 10 bership, for example with variable overwintering success in the birds that 10 11 constitute the breeding community. Then questions can be asked such as: Is 11 12 the assembly of the local (patch) subcommunity governed by assembly rules 12 13 that are based on the vegetation or some other surrogate of resource avail- 13 14 ability, or is it haphazard? Are the densities of species within the habitat 14 15 related to habitat suitability and its variation throughout the site, and are 15 16 species distributed patchily within habitats in accordance with habitat suit- 16 17 ability? Is the use of habitat patches by community subsets predictable, with 17 18 knowledge of the species’ habitat requirements and preferences? Is patch use 18 19 affected by the presence or density of other resource consumers? Are eco- 19 20 logical substitutes between patches readily identifiable, and what rules lead 20 21 to predicting which of alternative species or species combinations might 21 22 occupy a patch? These questions are addressed here in two habitat types in 22 23 Grand Teton National Park for which long-term data are available on species 23 24 occupancy and territory disposition. These are Grass-sage and Wet willows 24 25 habitats, with emphasis on the latter. 25 26 The theme of assembly rules at each of these different scales has been 26 27 addressed to some extent in previously published or related work, but to 27 28 different degrees. Each of the following sections begins with a perspective 28 29 on related or published work, introduces and discusses the data sets used in 29 30 the respective sections, and follows with detailed analyzes of assembly rules 30 31 in different systems at different scales. 31 32 32 33 33 34 Island systems 34 35 35 Background 36 36 37 Island distributions and nestedness 37 38 The theme of species distributions on islands has been the subject of exten- 38 39 sive research over the last several decades since its re-invigoration by 39 40 MacArthur and Wilson (1967), and the assembly rules approach to commu- 40 168 M.L. Cody 1 nity structure was formulated in this context (Diamond, 1975). A good deal 1 2 of the recent literature has addressed the notion of nestedness, or the degree 2 3 to which the floras of smaller islands are subsets of those on larger islands. 3 4 Various tests have been proposed for statistical significance of the nestedness 4 5 criterion (Cody, 1992; Patterson & Atmar, 1986; Ryti & Gilpin, 1987; 5 6 Simberloff & Levin, 1985). The phenomenon of nestedness has relevance 6 7 with respect to whether the island biota is extinction or dispersal driven 7 8 (Blake, 1991; Bolger et al., 1991; Cutler, 1991; Kadmon, 1995), and to 8 9 possible conservation strategies (Patterson 1987; Simberloff & Levin 1985; 9 10 Simberloff & Martin, 1991). Nestedness in island biotas signals type A 10 11 assembly rules, but the rules are often not transparent, especially when island 11 12 area is closely controlled. 12 13 13 14 Plant on islands in Barkley Sound 14 15 Plant distributions on islands were discussed in Cody (1992) from the point 15 16 of view of nestedness, with reference to distributions on islands in Barkley 16 17 Sound, on the outer coast of Vancouver Island, British Columbia. The climax 17 18 vegetation is Coastal Coniferous Forest. The data set is part of a long-term 18 19 study involving some 220 islands covering an island size range of over six 19 20 orders of magnitude, with isolation distances from ‘mainland’ Vancouver Island 20 21 between 0.001 to 15 km, and recording a flora of around 330 species. The 21 22 Islands have been censused repeatedly between 1981 and 1996, at intervals 22 23 of 1–4 y, providing information on species numbers and identities, immigra- 23 24 tion and extinction rates (Cody, unpublished data), and the evolution of reduced 24 25 dispersal in island populations of anemochorous species (Cody & Overton, 25 26 1996). 26 27 27 28 The mid-sized islands 28 29 To remove the most obvious source of variation in island species lists, I 29 30 consider only a set of mid-sized islands, of areas 1100–6300 m2. There are 30 31 36 such islands in the data set, on which 132 plant species have been found; 31 32 on the most diverse islands about half the species total occurs. In this larger 32 33 picture, species do not show significant nestedness with respect to island 33 34 occupied, although islands are significantly nested in terms of the species they 34 35 support (Cody, 1992). 35 36 Here, plant distributions are discussed for three habitat-specific subsets of 36 37 the island floras: forest species, shoreline species, and ‘edge’ species (a 37 38 category of largely weedy plants occupying the island perimeters between the 38 39 interior forest and the shoreline). Nestedness and assembly rules are investi- 39 40 gated in each of the three species subsets, with differences that appear to be 40 Scale and pattern in plant and bird communities 169 1 the consequence of the stability of the habitats occupied by the plants, and 1 2 the species turnover (colonization/extinction dynamics) within them. 2 3 3 4 4 5 Forest species 5 6 After subdividing the island flora by habitats, some 19 species (Fig. 6.1) con- 6 7 stitute a forest component, and by a number of alternative criteria these species 7 8 display significant nestedness (NB four of the 36 islands support none of these 8 9 forest plants; the size range of the mid-sized islands is intermediate between 9 10 smaller islands with essentially no forest plants and larger islands with many.) 10 11 Assembly rules are not obvious from a display and positive test of nest- 11 12 edness, in that inspection of the data does not reveal what they might be; 12 13 however, species likely are not accumulated on islands randomly, and posi- 13 14 tive nestedness indicate that such rules may exist. A reorganization of the 14 15 species-by-sites matrix (SSM) by plant guilds might be productive, but here 15 16 it is not (Fig. 6.1(b)). Separating plants into guilds by growth form (trees, 16 17 shrubs, ferns, and forbs – mostly monocot geophytes) does not suggest by 17 18 what means some similarly sized islands have different plants than others. 18 19 Note that if there were alternative forest species or ecological vicariants on 19 20 the islands, they should show up (with disjunct distributions) in a guild-level 20 21 reorganization of the SSM. 21 22 Eliminating the five species that occur on only one island and eight islands 22 23 with zero or just one forest species leaves 14 species and 28 islands. A re- 23 24 organization of this trimmed SSM is more useful (Fig. 6.1(c)), as the islands 24 25 now sort readily into three quite distinct groups: 25 26 (a) those with Pacific Yew (Taxus brevifolius): Class 1; 26 27 (b) those with either or both Douglas Fir (Pseudotsuga menziesii) and 27 28 Lodgepole (Shore) Pine (Pinus contorta): Class 2; 28 29 (c) and those with none of these three trees: Class 3; scrubbed islands, with 29 30 0–1 forest species, are called Class 4. 30 31 The 11 Class 1 islands with Taxus have, in addition, several shrub species 31 32 (mostly ericads) that are lacking in the other islands. Six Class 2 islands, lack- 32 33 ing Taxus and supporting Pseudotsuga/Pinus (PP), have the ferns Polypodium 33 34 vulgare (Pv) and Polystichum munitum (Pm) with lower and higher proba- 34 35 bilities, respectively. Lastly, the 11 Class 3 islands without the three trees 35 36 have the lily Maianthemum dilatatum with incidence comparable to that on 36 37 all other mid-sized islands, the fern Pm with similar probability to PP islands, 37 38 and reduced incidences of fern Pv and Boschniakia (a parasite on ericad shrubs). 38 39 The islands in these three groups are clearly distinct in species composition, 39 40 but what accounts for the distinction? The term ‘intrinsic guilds’ is suitable 40 170 M.L. Cody 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 Fig. 6.1. Species by site matrices for forest plants on 36 mid-sized islands in Barkley 38 39 Sound, British Columbia. (a) Islands ranked by species numbers and species ranked 39 by island occurrences, with significant nestedness apparent; (b) species grouped by 40 growth form; (c) islands grouped into classes, I. to r., 1: Taxus islands, 2: dry conifer 40 (Pinus/Pseudotsuga) islands, and 3: others. Class 4: few forest plants, omitted. Scale and pattern in plant and bird communities 171 1 for species groups that co-occur but without any obvious a priori reason for 1 2 their co-occurrences (cf. Wilson & Roxburgh, 1994). Note that we have already 2 3 run a strong island area filter (such that there is no longer a significant 3 4 correlation between log(area) and log(species number); r = 0.424, P > 0.05). 4 5 In fact, the Taxus islands are wetter with taller trees (see below), and deeper 5 6 soils and shade (Cody, unpublished data); in contrast, P. menziesii and P. 6 7 contorta are both indicators of dry conditions, which are found on steeper, 7 8 rockier islands on which soils do not readily accumulate, runoff is rapid, and 8 9 tree stature is in general lower; the correlation between islands class and veg- 9 10 etation height, r = −0.78, is significant). Both of the PP species are absent 10 11 from the nearby mainland, but they recur commonly on the interior, drier side 11 12 of Vancouver Island. 12 13 Sorting islands by larger (>1.3 × 103 m2) vs. smaller, taller (island summit 13 14 >7 m above the Fucus line) vs. lower, and exposed (island groups B and C, 14 15 towards the outer reaches or mouth of the sound) versus sheltered (groups A, 15 16 D and E – islands interior or close to the mainland) illuminates some of the 16 17 conditions that favor one island class over another, with their particular plant 17 18 assemblages (Fig. 6.2). Class 4 islands tend to be well represented among the 18 19 small and low islands, class 3 islands are tall and rocky, class 2 islands are 19 20 those with greater exposure, and class 1 islands are generally larger and more 20 21 sheltered. ANOVA shows that there are significant effects of all of these 21 22 factors on island class (P < 0.01). 22 23 23 24 24 25 Shoreline species 25 26 There are 29 species that are characteristically shoreline plants; they are listed 26 27 in Fig. 6.3(a) which, as with forest plants, displays significant nestedness 27 28 (Wilcoxon rank sum test, P < .05; correlation between species numbers and 28 29 island size = 0.31, NS). The most widespread shoreline plants occur on nearly 29 30 all islands (5 species on ≥30/36 islands), but only two islands support >50% 30 31 of the plant list. In Fig. 6.3(b) the matrix is rearranged by rows to reflect 31 32 plant affinity with three different shoreline substrates, respectively, 32 33 muddy/gravelly beaches, sandy beaches, and rocky shores. All islands have 33 34 rocky habitat, and the incidence of the plants of rocky shorelines does not 34 35 differ among the three island categories (0.58, 0.58, 0.57, respectively, left 35 36 to right Fig. 6.3(b)). But some islands have, in addition, sandy beaches and 36 37 others muddy beaches as well. Those islands with muddy beaches tend to be 37 38 larger (1455 m2 vs. 1326 m2, with more precipitation runoff reaching the 38 39 shoreline) but not significantly so. Such islands, however, do support signif- 39 40 icantly more shoreline species (12.22 vs. 7.81, P < 0.01 by t-test). Islands 40 172 M.L. Cody 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 Fig. 6.2. Island classes (1–4 according to forest plant associations) are differentially 21 represented among large vs. small, tall vs. low, and sheltered vs. exposed islands. The 22 proportional data (from all islands lumped together) is repeated to aid in comparisons. 22 23 23 24 24 25 with sandy beaches (left-hand two groups) show similar incidences of plants 25 26 typical of this habitat (0.36 and 0.38, respectively), but sandy beaches do not 26 27 significantly increase the total count of shoreline species on these islands. 27 28 28 29 29 30 Edge species 30 31 The third component of the island flora are referred to as ‘edge species’, those 31 32 plants that occur between the interior forest and the shoreline; they form the 32 33 most diverse category of island plants, totaling 84 species. Many of the edge 33 34 species are weedy, opportunistic and widely distributed, many have conspic- 34 35 uous dispersal abilities (e.g., anemochorous, parachuted achenes, as is many 35 36 Asteraceae and Onagraceae, minute sticky seeds as in Brassicaceae, bird- 36 37 vectored fruits as in most of the shrubs in the families Ericaceae, Caprifoli- 37 38 aceae, Rhamnaceae, and Rosaceae). Unlike the forest and shoreline floras, 38 39 the edge flora is not significantly nested (by casual inspection or by the usual 39 40 statistical tests), nor is any subset of it. Very few of the edge species are 40 Scale and pattern in plant and bird communities 173 1 widely distributed over the mid-sized islands, and most islands have but a 1 2 small proportion of the total edge flora; the mid-island average of 6.6 (± 4.2) 2 3 species constitutes just 12% of the species total in this category. 3 4 In Fig. 6.4(a) the herbaceous component of the edge flora is listed, some 4 5 54 species that exclude the woody taxa. The two largest families are Poaceae 5 6 and Asteraceae (around a dozen species each). Rearranging the matrix by 6 7 familial or growth form affinity does not generate nestedness, and an arrange- 7 8 ment of islands following that of the forest plant classes adds nothing to 8 9 pattern or conciseness of the edge species matrix. Fig. 6.4(b) reflects this for 9 10 the two largest families; the rank order of both islands and taxa is preserved 10 11 from Fig. 6.4(a). There is no obvious nestedness to the species, by appear- 11 12 ance or test, nor any suggestion of a more subtle form of assembly to these 12 13 weedy communities. 13 14 The incidences of the plant species are given in the right-most column of 14 15 the figure, and sum to the number of taxa present on the average island (e.g., 15 16 1.81 for Asteraceae, 1.75 for Poaceae). These incidences can also be used to 16 17 generate the variance in expected species richness on the islands under the 17 18 hypothesis of random distribution and contra either nestedness or disjunction 18 ∑ 19 of species over islands. The expected variance is i(Ji)(1-Ji), fide D. Ylvisaker 19 20 (pers. comm.); for the Asteraceae, observed/expected variance is 1.42/0.82, 20 21 and by the F-test (F = 1.73, df = 35, 35, P > 0.05) the random hypothesis is 21 22 upheld. For Poaceae, expected variance in species/island is 1.23, significantly 22 23 less than the observed 3.29, indicating that there are more islands than 23 24 expected with many grasses and no grasses. Some degree of nestedness in 24 25 islands is therefore indicated for species of Poaceae, although rules for their 25 26 abundance on some islands and dearth on others are still obscure. 26 27 Whereas species numbers of forest plants respond significantly to island 27 28 characteristics such as rock height and area, the edge species do not (Fig. 6.5); 28 29 island isolation or exposure do not contribute to an explanation of variance 29 30 in species numbers. Thus edge species remain quite unpredictable in number 30 31 and kind, and if there are assembly rules for them, they remain undiscovered. 31 32 32 33 33 34 Community replication in wide-ranging habitats 34 35 35 Background 36 36 37 Birds in scrub, woodland, and forest 37 38 An alternative approach to rules for community structure is by seeking type 38 39 B assembly rules via replicated censuses or repeated samples from within a 39 40 single, broad habitat type. For example, the bird communities of the scrubby 40 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 1 2 2 edness. 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36

37 ) Species-by-site matrix for shoreline plants on mid-sized islands, with islands and species ranked showing significant nest 37 a 38 38 39 39 ) species are grouped by shoreline habitats, which added on islands from right to left. b Fig. 6.3. ( 40 ( 40 176 M.L. Cody 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 Fig. 6.4. (a) Species-by-site matrix for weedy ‘edge’ plants, with no significant 34 35 nestedness. (b) Distributions of species in the two largest families of edge plants, 35 36 Asteraceae and Poaceae. The weedy composites are randomly distributed over islands, 36 but the grasses do show a degree of nestedness, although no patterns of species co- 37 occurrence are apparent in either family. 37 38 38 39 39 40 40 Scale and pattern in plant and bird communities 177 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 Fig. 6.4. (continued) 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 Fig. 6.5. The predictability of species numbers (R2, ordinate) as a function of island 33 34 area and elevation varies among different components of the island floras. It is most 34 predictable in forest species and least in weedy, edge species. 35 35 36 36 37 sclerophyllous vegetation typical of Mediterranean-type climates world wide, 37 38 and variously called chaparral, macchia, fynbos, heath or maquis, have been 38 39 extensively studied (Cody, 1973, 1974; Cody & Mooney, 1978; Cody, 1983b; 39 40 Cody, 1986; Cody, 1994a), with conclusions on the regulation of community 40 178 M.L. Cody 1 size (species numbers, α-diversity), composition (species’ relative abundances 1 2 and ecological characteristics), and membership attributes (sizes and relative 2 3 morphology, foraging behavior, etc.). When tests for assembly rules are made 3 4 across continents, phylogenetic constraints are often weak or absent (as quite 4 5 unrelated species may occupy parallel or convergently similar ecological 5 6 niches in different biogeographic regions); any case for community replication 6 7 is therefore more compelling. Analyses of bird communities within continents, 7 8 both in chaparral or its equivalent and in other habitats in the same climatic 8 9 region, e.g., renosterveld and kloof woodland in southern Africa (Cody, 9 10 1983a), or over broader geographical areas, e.g., in Eucalyptus woodlands, 10 11 forest, and rainforest in Australia (Cody, 1993), have also been made. 11 12 ‘Habitat islands’ is a phrase that refers to isolated habitat fragments 12 13 separated from each other by different vegetation; habitat islands constitute 13 14 an intermediate situation between true islands and continuously distributed 14 15 vegetation. Assembly rules for species’ retention in attenuating bird commu- 15 16 nities in increasing smaller and more isolated patches of Afromontane wood- 16 17 land were described in detail by Cody (1983b). 17 18 18 19 Mulga bushland and healthland in Australia 19 20 Analyses of replicated bird censuses across two widely distributed habitat 20 21 types in Australia, namely mulga bushland and protead heathland, were recently 21 22 published (Cody, 1994a, b), with strongly contrasting results. The Australian 22 23 heathland is a fire-prone scrub vegetation dominated by evergreen sclero- 23 24 phylls, especially Proteaceae and in particular Banksia. This habitat occurs 24 25 on the poorest soils (laterite and sand-over-laterite), more extensively in win- 25 26 ter-rainfall areas of the south and southwest, but also beyond this area, around 26 27 the periphery of the continent and especially in isolated patches near the 27 28 eastern and northern coasts. Mulga is an open bushland vegetation dominated 28 29 completely by the phyllodinious Acacia aneura, sometimes with a couple of 29 30 smaller and scarcer acacias and Myoporaceae shrubs. It burns infrequently, 30 31 occurs on the more productive red earths across interior Australia, mostly in 31 32 areas of both summer and winter rainfall with 180–220 mm annual precipi- 32 33 tation, and is nearly contiguous from the west coast to central Queensland. 33 34 In many characteristics the bird communities of these two habitats are at 34 35 opposite extremes. 35 36 Analysis of the bird communities of protead heathlands compared census 36 37 results from 17 sites continent-wide, which produced a total species list of 37 38 96 taxa. These communities varied widely in size, from 5 to 22 species, and 38 39 also in species composition from site to site, with a component of species 39 40 turnover between sites related to both differences in vegetation structure (β- 40 Scale and pattern in plant and bird communities 179 1 diversity) and distance apart of the sites (γ-diversity). Only two bird species 1 2 occur in >50% of the heathland censuses, and thereby qualify as ‘core’ species 2 3 (i.e., occur more often than would be the case if communities were assem- 3 4 bled by a random draw of species from the total species pool). Because α- 4 5 diversity averages about 12 species (and shows no significant variation across 5 6 the continent), the contribution of core species to the local α-diversity is minor 6 7 (about 1/6). Overall, community composition is unpredictable, and conforms 7 8 closely to expectations from a random selection of species according to bino- 8 9 mial null models (Cody, 1986, 1992). Thus most birds in the protead heath- 9 10 lands are rather erratic and unpredictable, have low incidence in the censuses 10 11 (>75% of the recorded species occur in <25% of the censuses), and their iden- 11 12 tities change substantially even between adjacent sites. 12 13 However, despite wide variations in community composition, some gen- 13 14 eral assembly rules do apply: 14 15 15 (a) Species richness is modest; mean α-diversity of 11.6 ± 1.9 s.d. does not 16 16 vary regionally, but 96% of its variation depends on inter-site differences 17 17 in vegetation structure, many of which are in turn related to temporal 18 18 stages in post-fire regrowth; 19 19 (b) Heathland bird density is low, averaging around four individuals per 20 20 hectare (l/H) – the lowest of all Australian shrub habitats censused; 21 21 (c) Insectivores are poorly represented, by species that are both relatively 22 22 constant (e.g., malurid wrens) and relatively rare (ca. 2 l/H); 23 23 (d) Nectarivorous birds (Meliphagidae, and especially Phyllidonyris spp.) are 24 24 dominant in both species richness and bird density; meliphagids comprise 25 25 the majority of heathland endemics and nearly 50% of the total species 26 26 list, add two to six species to local α-diversity, and generally total at least 27 27 50% of total bird biomass; 28 28 (e) Overall, species numbers and identities are sensitive to vegetation struc- 29 29 ture, which is largely a function of time since the last fire. Heathland bird 30 30 resources are successional stage specific, and their consumers are wide- 31 31 ranging resource specialists in pursuit of suitable resources (Cody, 1994a). 32 32 33 The breeding birds of the mulga bushlands are generally quite different from 33 34 those of protead heathlands; mulga species richness averages nearly twice 34 35 that of heath, and in contrast to heathlands, the communities vary in compo- 35 36 sition scarcely at all across thousands of kilometers of the habitat’s range. 36 37 With predictable community membership, species turnover between sites is 37 38 low, even cross-continentally, and unrelated to both differences in vegetation 38 39 structure (β-diversity) and distance apart of the sites (γ-diversity); peripheral 39 40 species in the mulga communities are more a product of the proximity and 40 180 M.L. Cody 1 type of adjacent habitats (Cody, 1994b). The following assembly criteria, rules 1 2 to descriptive generalities, have been formulated: 2 3 3 (a) Mulga α-diversity is narrowly constrained (mean 21.5 ± 0.8 species) 4 4 across the mulga (n = 20 census sites), with little systematic variation 5 5 attributable to geographical position or latitude, and none attributable to 6 6 variations in the physiognomic structure of the vegetation; 7 7 (b) Bird densities in mulga are around 12 indiv. ha−1, varying little among 8 8 mulga sites and generally about three times those in heathland; 9 9 (c) Core species are prominent in the community make-up, comprising 2/3 10 10 of the local α-diversity (14/21 species) and ca. 3/4 of the total bird den- 11 11 sity, even though they comprise only 26/81 species encountered in the 12 12 censuses; 13 13 (d) Ecological replacements occur in core niches, with different (often con- 14 14 generic) species substituting in different part of the habitat range; 15 15 (e) Few mulga birds are habitat endemics (perhaps 3–4 species), but the top 16 16 nine core species all reach higher densities in mulga than in other habi- 17 17 tats; 18 18 (f) Insectivores, both ground- and foliage-foraging, are the largest compo- 19 19 nent of the community, and constitute two-thirds of the total bird den- 20 20 sity; 21 21 (g) Nectarivores are very rare (ca. 0.2 indiv. ha−1), mostly utilizing mulga 22 22 mistletoes), and the 1–2 meliphagid species present are omnivorous or 23 23 frugivorous; 24 24 (h) Some assembly rules for core niches are simple (e.g., one or another crow 25 25 Corvus depending on region, one or another butcherbird Cracticus, 26 26 species interchangeable, identity unspecified); 27 27 (i) Other assembly rules for core niches are more involved, e.g., one mid- 28 28 sized omnivore species (babblers Pomatostomus spp.), with specific iden- 29 29 tity favoring the wide-ranging P. superciliosus, but reverting to P. tem- 30 30 poralis in mulga sites adjacent to woodland, or to P. hallii or P. ruficeps 31 31 within the more limited southeastern ranges of these latter two species; 32 32 or two perch-and-pounce insectivores (robins), one of which is always P. 33 33 goodenovii, the second either a Microeca or a Melanodryas species 34 34 depending on geographic region; 35 35 (j) Yet other rules, e.g., for the composition of the guild of small low-for- 36 36 aging insectivores (mostly thornbills), are complex. Eight candidate 37 37 species occur in the mulga sites (6 Acanthiza, 1 Sericornis and 1 Aphe- 38 38 locephala), and some five of these have ranges that include the average 39 39 mulga site. Exclusively summer-rainfall mulga (Northern Territory) sup- 40 40 ports two-species guilds, summer plus winter rainfall (central-western Scale and pattern in plant and bird communities 181 1 Queensland) three-species guilds, and winter rainfall sites (southwest 1 2 Australia) four to five species guilds. Guild composition varies region- 2 3 ally, with Chestnut-rumped thornbill (A. uropygialis) and Yellow-rumped 3 4 thornbill (A. chrysorrhoa) regular in most sites, Broad-tailed thornbill 4 5 (A. apicalis) more casual countrywide, Redthroat (Sericornis brunneus), 5 6 Slate-backed thornbill (Acanthiza robustirostris) and southern whiteface 6 7 (Aphelocephala leucopsis) in SW Australia, and Buff-rumped thornbill 7 8 (A. reguloides) regular, Yellow thornbill (A. nana) casual, in Queensland. 8 9 9 10 10 11 Assembly rules in oak woodland birds 11 12 Oak woodlands in Western North America 12 13 13 In this section, data from censuses of oak woodland habitats are used to 14 14 illustrate a further example of the approach. Woodlands dominated by oaks 15 15 (Quercus spp.) are widespread in western North America, and occur from the 16 16 Cape region of Baja California north through Alta California, Oregon, and 17 17 Washington to southern British Columbia, covering over 20° of latitude and 18 18 including a dozen oak species. Elsewhere, oak-dominated habitats occur also 19 19 from south-central Colorado south through the Rocky Mountains, from 20 20 Ontario south to Florida, and thence south in México into Central America, 21 21 but here only the westernmost oak woodlands along the Pacific coast will be 22 22 discussed. 23 23 Some 70 breeding bird censuses are in hand from the Pacific series of oak 24 24 woodlands, from 40 different sites, some censused more than once, in different 25 25 years). Of these, 30 censuses are the author’s, conducted mostly in May– 26 26 June 1994, on standardized 5 ha sites using standard census techniques; the 27 27 remaining censuses are the published results of others and producing 28 28 numbers and identities of breeding species, plus estimates of breeding bird 29 29 densities. 30 30 31 Overall species composition and diversity 31 32 32 The geographical distribution of 40 bird census sites in oak woodland along 33 33 the Pacific coast spans about 25° of latitude, extending from the Cape region 34 34 of Baja California north to Washington (Fig. 6.6). At these sites, α-diversity 35 35 averages 28 species, and there is no detectable trend with latitude (P = 0.34). 36 36 However, variation in α-diversity is related to variation in vegetation structure; 37 37 profile area (area under a plot of vegetation density versus height, a measure 38 38 of total vegetation biomass at the site) accounts for a modest 15% (p = 0.04) 39 39 of this variation (Fig. 6.7(a)). Total bird density also responds to the same 40 40 vegetation variable (r2 = 0.23, P =.008; Fig. 6.7(b)). 182 M.L. Cody 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 Fig. 6.6. Distribution of 40 bird census sites in oak woodlands along the west coast 21 of North America, covering about 25° of latitude. 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 Fig. 6.7. Numbers of bird species in oak woodland sites show modest increases with 36 total vegetation density (profile area, abscissa; r2 = 0.15, P = 0.04), and total bird den 36 37 sity also increases with the same woodland variable (r2 = 0.23, P = 0.008). 37 38 38 39 39 40 40 Scale and pattern in plant and bird communities 183 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 Fig. 6.8. Bird community composition among 40 oak woodland census sites differs 23 significantly from a collection of random subsamples of the avifauna. A null binomial 23 24 model P(k; n,p) is constructed with probability P = (average α-diversity)/(total species 24 25 list in all sites) = 28/112, n = 40 census sites, and k = ‘successes’, or number of sites 25 at which species are recorded. Cumulative proportion of total species (ordinate) is 26 drawn against decreasing proportion of sites occupied (abscissa) for observed and 26 27 expected functions. Over one-third of the oak woodland birds, 40/112 species, qualify 27 28 as core species in being more predictable at census sites than chance would predict. 28 29 29 30 30 31 In 70 censuses at 40 sites, 112 bird species were recorded. A null binomial 31 32 model can be used to compare species incidence over sites with that predicted 32 33 from a random species assortment (see Cody. 1994b). Here probability of k 33 34 ‘successes’ (species’ records at a site) with success probability p (= 28/112 34 35 = 0.25) in n (= 40) replicates) is computed. Any species that occurs with inci- 35 36 dence J > 0.3 is significantly more constant in the censuses than is predicted 36 37 from a random draw; expected and observed distributions are shown in Fig. 37 38 6.8. Some 40 species fall into this category, and are termed the ‘core species’. 38 39 These 40 species are shown schematically in Fig. 6.9; although they account 39 40 for only 36% of the species list, they are the major component of the oak 40 1 1

2 ch the 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 species was censused. 40 Fig. 6.9. Core species of oak woodland birds, grouped by guilds; numbers following names are the number sites at whi 40 Scale and pattern in plant and bird communities 185 1 woodland bird community, comprising an average of 80% of the α-diversity 1 2 over sites, and 82% of the total bird density. 2 3 As in the mulga, the identification of core species in oak woodlands is 3 4 compromised by the occurrence of species that are ecological replacements 4 5 in essentially the same core role or niche. For example, an oak woodland 5 6 generally supports one or two wren species, an oriole, a chickadee or titmouse, 6 7 but overall there are more candidate species for these core niches than are 7 8 supported by local site α-diversity. Here, the supernumerary species compet- 8 9 ing for limited niche representation are five wrens (House, Bewick’s, Cactus, 9 10 Canyon, Rock), two orioles (Northern, Hooded), and four parids (Mountain, 10 11 Chestnut-backed, and Black-capped chickadee, Plain titmouse). These ‘extra’ 11 12 species inflate the total list and obscure the identification of core species unless 12 13 their roles as equivalent contenders for core niches are recognized. 13 14 14 15 Oak woodland wrens 15 16 The assembly of species in particular guilds within the oak woodlands is 16 17 based on various criteria. In wrens, for example, both House and Bewick 17 18 wren are common in the woodlands overall (incidence J = 0.27, 0.28, respec- 18 19 tively), and there is tendency to favor House wren over Bewick wren at lower 19 20 latitudes in taller woodland with reduced understory. But where they are both 20 21 common (n = 25 sites with combined densities > 0.5 pr/ha) each species is 21 22 the predominant factor in the other’s density, and densities of the two species 22 23 vary inversely (r = −0.39, p = 0.05; Fig. 6.10). 23 24 24 25 Oak woodland woodpeckers 25 26 Some guilds are readily understood by reference to body size. Oak wood- 26 27 lands have three core species of woodpeckers, the large Northern flicker (142 27 28 g), the mid-sized Acorn woodpecker (83 g), and the small Nuttall’s wood- 28 29 pecker (38 g). About 3/4 of the oak woodland sites, particularly those in cen- 29 30 tral and southern California, support this three-species combination (Fig. 30 31 6.11). However, other combinations of woodpecker species occur in other 31 32 oak woodlands; in the more northerly sites a similar three-species combina- 32 33 tion substitutes Hairy woodpecker (70 g) for Acorn woodpecker and Downy 33 34 woodpecker (27 g) for Nuttall’s woodpecker, and in oak woodland sites near 34 35 desert habitats (Baja California, SW California) Gila woodpecker (70 g) fills 35 36 the mid-size woodpecker slot and Ladder-backed woodpecker (30 g) the small 36 37 size slot. Whereas the names of the woodpecker species change, their segre- 37 38 gation by body size differences of around 2× does not, and the pattern is 38 39 maintained throughout the woodlands (Fig. 6.11). 39 40 40 186 M.L. Cody 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 Fig. 6.10. The two common wrens of oak woodlands, Bewick and House wren, tend 17 to vary in density inversely (r = −0.39, P = 0.05) at 25 sites (ranked along abscissa) 18 where both species are common (combined density > 0.50 pairs ha−1). Either or 18 19 both wrens are present in most sites, and these species appear to be more or less 19 20 substitutable in the oak woodland community. 20 21 21 22 22 23 Oak woodland vireos 23 24 Vireos are slow-searching foliage insectivores; there are three species of 24 25 vireos in the oak woodlands, and all qualify as core species. They differ some- 25 26 what in body size (by an average 20%), but not as conspicuously as the wood- 26 27 peckers. The woodland sites average 1.33 vireo species; Hutton’s vireo (11.6 27 28 g) is common (J = 0.90), Warbling vireo (14.8 g) and Solitary vireo (16.6 g) 28 29 less so (J = 0.45, 0.3, respectively). Warbling vireo incidence is rather con- 29 30 stant over latitude, but that of Hutton’s vireo declines sharply to the north 30 31 whereas Solitary vireo is more common at higher latitudes (Fig. 6.12). Based 31 32 on each species’ incidence, no particular two-species combination is favored 32 33 over another, and all combinations (of one to three species) are no different 33 34 in incidence than is predicted from the incidences of individual species. 34 35 35 36 Oak woodland flycatchers 36 37 Sallying flycatchers are a conspicuous component of the oak woodland bird 37 38 community, and there are three common core species: the large Ash-throated 38 39 flycatcher (symbol L: 27 g; J = 0.80), the intermediate sized Western wood 39 40 pewee (M: 13 g; J = 0.63) and the small Western (or Pacific slope) flycatcher 40 Scale and pattern in plant and bird communities 187 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 Fig. 6.11. Almost all oak woodland sites support three woodpecker species that vary 17 by body size by a factor of ca. 2.5, small to medium, or 1.75, medium to large. How- 17 18 ever, the usual trio (left; site incidences are shown below species names) is replaced 18 19 by a similarly sized trio in northern sites (center) and again in desert-edge sites (right). 19 ‘Col.aur.’ = Colaptes auratus, ‘Mel.for.’ = Melanerpes formicivorus, ‘Mel.uro.’ = 20 M. uropygialis, ‘Pic.nut.’ = Picoides nuttallii, ‘Pic.vil.’ = P. villosus, ‘Pic.pub.’ = 20 21 P. pubescens. 21 22 22 23 23 24 24 25 (S: 10 g; J = 0.63). Four other flycatchers are rare in the censuses (as indi- 25 26 cated in Fig. 6.13). The three core species constitute the prevalent three- 26 27 species combination (14/40 sites, indicated in Fig. 13 by ‘LMS’). Only two 27 28 other species combinations are common, a two-species combination of LS (at 28 29 6/40 sites) and a single species L (at 7/40 sites; see Fig. 6.13). 29 30 The incidence of particular species’ combinations is related chiefly to veg- 30 31 etation structure in the oak woodlands (Fig. 6.14). The lowest and least dense 31 32 woodlands support only L (Ash-throated flycatcher). Most oak woodlands of 32 33 intermediate height and density have the three species LMS, but the medium- 33 34 sized species M tends to drop out in very dense woodlands which are typi- 34 35 cally LS (Fig. 6.14). Taller woodlands, with the addition of the largest Olive- 35 36 sided flycatcher (X: 32 g; J = 0.10), may have four species (XLMS), or a 36 37 trimmed two-species combination (MS) that occurs in 25 m oak woodlands 37 38 at the highest latitudes. Thus assembly rules in the flycatchers are based on 38 39 body sizes as in woodpeckers, but with preferred combinations varying largely 39 40 with the structural aspects of the woodland vegetation. 40 188 M.L. Cody 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 Fig. 6.12. The incidence of three vireo species in oak woodland varies over latitude, 19 while the mean species number (1.33) remains about constant. In 20 southerly sites, 19 20 the three vireo species occur in various 0–3 species combinations (see Obs) that are 20 21 in the same proportions as predicted (by Chi-sq., P > 0.05) from random assembly 21 (see Exp), using the species incidences Ji shown at top to generate predicted 22 frequencies. 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 Fig. 6.13. The three common flycatchers of the oak woodlands differ in size, and are 35 36 represented as large, medium and small (L, M, S; left-side of figure). Three particular 36 37 species combinations are relatively common: LMS, LS, and L, with the combination 37 LMS is especially prevalent; other combinations are observed rarely (see right-hand 38 side) or never (unlisted). 38 39 39 40 40 Scale and pattern in plant and bird communities 189 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 Fig. 6.14. The occurrence of various flycatcher assemblages is related to the wood- 19 20 land vegetation, with assemblage size increasing with vegetation height and profile 20 21 area (vegetation density), the medium-sized species squeezed out in very dense 21 22 vegetation, and high-latitude sites supporting just medium and small species. 22 23 23 24 24 25 Assembly rules and community composition within sites 25 26 26 Background 27 27 28 Insectivorous bird communities 28 29 Within- and between-site habitat variation, and concomitant variation in the 29 30 composition of bird assemblages across the changing vegetation, has been 30 31 extensively studied in Old World warblers Sylviinae, and previously reported 31 32 (Cody & Walter, 1976; Cody, 1978, 1983a, 1983b, 1984). The approach used 32 33 is to plot the territories of the breeding species within a heterogeneous site, 33 34 and then relate use of the habitat by the various species present to (a) the 34 35 vegetation within the study site and within species’ territories, (b) the habitat 35 36 requirements of the species, in terms of vegetation heights, densities, and 36 37 foraging substrates, and (c) the presence of other species with similar but 37 38 variously divergent habitat requirements. 38 39 One example of this approach was discussed by Cody and Walter (1976), a 39 40 study of the five Sylvia warblers of Mediterranean-type scrub and woodland 40 190 M.L. Cody 1 in SW Sardinia. Ranked from lowest to tallest in terms of preferred vegeta- 1 2 tion height, these are Sylvia sarda −S, S. undata −U, S. melanocephala −M, 2 3 S. cantillans −C, and S. atricapilla −A, with letters indicating species abbre- 3 4 viations. In low, meter-high vegetation (Cistus- dominated scrub at Narcao), 4 5 three species occurred (S, U, and M) with little interspecific difference in 5 6 territories classified by vegetation height and density. Yet the species pairs 6 7 S-U and U-M were spatially segregated, with a demonstrated avoidance of 7 8 interspecific overlap. In somewhat taller macchia (Erica, Arbutus, Phyllarea) 8 9 at Bau Pressiu, a fourth species (C) is added. Again, strong behavioral inter- 9 10 actions serve to segregate territories spatially between species U and M, while 10 11 species pairs S–U and U–C show weaker interactions. In yet taller macchia 11 12 at Terrubia, S drops out and A is added; here the interactions between U and 12 13 M and U and C that occur in lower vegetation disappear, while a strong inter- 13 14 action characterizes species C–A. Overall, species interactions are predictable 14 15 knowing the preferred vegetation and foraging heights of each species, and 15 16 the extent to which the vegetation within a site allows the various species to 16 17 forage at different heights (in which case their territories overlap) or constrains 17 18 them to forage at similar heights (in which case they exhibit interspecific 18 19 territoriality). 19 20 A second example is taken from the sylviine warblers in scrub (Rosa, 20 21 Juniperus) to low woodland (Quercus, Betula) habitats in southern Sweden 21 22 (Cody, 1978, 1985). Seven sylviine warblers breed in the area (with addi- 22 23 tional species present in marshlands); these are five Sylvia species (S. com- 23 24 munis −U, S. nisoria −N, S. borin −B, S. curruca −R, S. atricapilla −A), a 24 25 Hippolais (icterina −H) and a Phylloscopus (trochilus −P), with letters again 25 26 indicating species abbreviations. The species are listed in order of ascending 26 27 foraging height and preferred vegetation heights within territories, but there 27 28 are considerable overlaps among species. Within sites, the occupancy of the 28 29 habitat by each warbler species was evaluated at the scale of 15 × 15 m 29 30 quadrats, in each of which the vegetation structure (density over height) was 30 31 also measured. 31 32 As vegetation height within quadrats increases, the number of coexisting 32 33 warbler species using the patch increases from one to two to three and up to 33 34 four species. At Bejershamn, the transitions are U, to U or N, to N, N or B, 34 35 B or C, BC, BHP or ACH, and the combination ACPH is rather uncom- 35 36 mon in the tallest woodland. The species pairs U–N and N–B are strongly 36 37 interspecifically territorial in habitat intermediate between their specific pref- 37 38 erences, and the pair B–C is weakly interactive. The assembly of increasing 38 39 large suites of warblers in taller vegetation is dependent on the match between 39 40 their combined foraging height distributions and the vegetation profile (plot- 40 Scale and pattern in plant and bird communities 191 1 ting density as a function of height). Thus the stable species combinations 1 2 are products of the conformity between species’ foraging requirements and 2 3 the foraging opportunities provided by the vegetation, and given the wide 3 4 overlaps in habitat preferences and foraging height distributions, the combi- 4 5 nations are adjusted or refined by direct interspecific interactions. 5 6 6 7 Birds in grass-sagebrush, Grand Teton National Park 7 8 Emberizid finches have been the subject of a long-term study in a grass-sage- 8 9 brush habitat by Cody (1974, 1996). Four species breed in a 4.7 ha site in 9 10 Grand Teton National Park, Wyoming: White-crowned sparrow Zonotrichia 10 11 leucophrys −W, Brewer’s sparrow Spizella breweri −B, Vesper sparrow 11 12 Pooecetes gramineus −V, and Savannah sparrow Passerculus sandwichensis 12 13 −S, with a fifth, Chipping sparrow Spizella passerina, of sporadic occurrence. 13 14 At a scale of 15 × 15 m quadrats, each of 208 quadrats was classified as being 14 15 core, secondary, or marginal with respect to the habitat preferences of the 15 16 emberizids. Within the site, W is restricted to areas of the tallest sage and S 16 17 to the grassier parts of the site, and their territories are located in similar posi- 17 18 tions among years. The habitat requirements, core plus secondary, of both B 18 19 and V are satisfied over about 50% of the study area, and these two species 19 20 are the commoner of the four. But for both species their preferred habitat is 20 21 interspersed with less suitable vegetation, and the disposition of their territories 21 22 changes somewhat from year-to-year. 22 23 The densities of the emberizid finches varies among years, most obviously 23 24 in response to wetter or drier conditions at the site. Savannah sparrow, for 24 25 example, is rarer is drier years, when its occupancy of quadrats drops to 11% 25 26 (vs. up to 36% in wet years). But its habitat occupancy depends also on the 26 27 densities of, and habitat use by, the other species present; these sparrows are 27 28 less tolerant of shared habitat (with B and V) in dry years than in wet years. 28 29 Although habitat choice is similar in Brewer’s and Vesper sparrows, the two 29 30 species tend to oscillate in density asynchronously, with V commoner in drier 30 31 years and B in the wetter. The two species demonstrate an aversion to terri- 31 32 torial overlap in years when the overall sparrow density at the site is high, 32 33 but a significant tendency to shared use of their joint core quadrats when 33 34 densities are low; in years of intermediate density, there is neither positive 34 35 nor negative association of the two species within the site. In this study, the 35 36 assembly of the finches at the quadrat scale is understood by a hierarchical 36 37 set of factors: (a) the extrinsic conditions that affect species’ densities in the 37 38 area (perhaps a function of overwintering success); (b) the extrinsic conditions 38 39 within the site, likely a function of late spring–summer weather conditions 39 40 there; (c) distinct habitat preferences in some species (W, S) and similar 40 1 1 2 2 3 3 4 ed in feet, 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 15 m quadrats as indicated.

23 × 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 with grassy willow-free areas, pools and streams, 15 m 40 Fig. 6.15. Mapped variation in vegetation height the Wet Willows site, Grand Teton National Park. Height contours are record 40 Scale and pattern in plant and bird communities 193 1 requirements in others (B, V), all potentially satisfied at the site; (d) behav- 1 2 ioral interactions that moderate interspecific territorial overlap, interactions 2 3 that vary between years with local conditions and with local sparrow densities 3 4 (Cody, 1996). 4 5 5 6 6 The willows site, Grand Teton National Park 7 7 8 The habitat, birds, and methodology 8 9 Close by the grass-sagebrush site just discussed is a very different habitat that 9 10 similarly has been studied over a long time period, a marshy site dominated 10 11 by various species of willows 1–3 m high and interspersed with small streams, 11 12 ponds, grassy and sedgy glades (Cody, 1974, 1996). The bird community of 12 13 the 3.3 ha willows site has been recorded intermittently since 1966 (and was 13 14 earlier studied by Salt, 1957a, b). The site supports two main bird guilds: 14 15 five species of paruline warblers (Yellow warbler Dendroica petechia. Com- 15 16 mon yellowthroat Geothlypis trichas. Wilson’s warbler Wilsonia pusilla, 16 17 MacGillivray’s warbler Oporornis tolmei and Northern waterthrush Seiurus 17 18 novaboracensis, with the first three common and the last two rare; and six 18 19 emberizid finches (Song sparrow Melospiza melodia, Lincoln’s sparrow 19 20 Melospiza lincolnii, Fox sparrow Passerella iliaca, White-crowned sparrow, 20 21 Clay-colored sparrow Spizella pallida, and Savannah sparrow), with the first 21 22 four common and the last two rare; the willows community is rounded out 22 23 with a flycatcher, magpie, wren, thrush, and a hummingbird. 23 24 The site is mapped by vegetation height (Fig. 6.15), and the frequency dis- 24 25 tribution of willow heights within, and its variation among, 15 × 15 m quadrats 25 26 forms the basis of this study. Thus the organization of the willows bird com- 26 27 munity and its assembly guides are studied at this local ‘patch’ resolution, at 27 28 which scale habitat use is assessed and territory locations plotted. 28 29 29 30 The emberizid sparrows 30 31 Six emberizid sparrow species have bred within the site over the last 30 years 31 32 (Song −S, 20 g, Fox −F, 32 g, Lincoln’s −L, 17.4 g, White-crowned −W, 32 33 29 g, Savannah −V, 19 g and Clay-colored Sparrow −C, 12 g), with the first 33 34 four species common and consistently present, the penultimate sporadic, and 34 35 the last common before the early 1970s but not present since. All species 35 36 feed mostly on the ground except the Clay-colored and White-crowned spar- 36 37 rows, which often forage somewhat higher in the vegetation. The common 37 38 species show extensive similarities in quadrat use (see plots of species over 38 39 principal components of vegetation, Fig. 6.16(a)), although discriminant func- 39 40 tion analysis (DFA) reveals differences in the mean habitat preferences of 40 194 M.L. Cody 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 Fig. 6.16. (a) Factor analysis (principal components) of vegetation characteristics for 34 emberizid sparrows in the willows site, showing distinct habitat preference in Savan- 34 35 nah sparrow for grassy areas (high F{1}scores), a less distinct preference for grassy 35 36 openings in White-crowned sparrow, and similar habitat choice in the remaining three 36 species. (b) Discriminant function analysis of the same data show that mean habitat 37 preferences differ among species (95% centroids) in all except Song and Fox 37 38 sparrows. 38 39 39 40 40 Scale and pattern in plant and bird communities 195 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 Fig. 6.17. Core habitat (50% confidence ellipses) is shown for the four common ember- 22 23 izid sparrows at the willows site. Numbers of occupied quadrats Q, by species, are 23 shown at left, while the numbers of quadrats potentially supporting various 1–4 species 24 combinations are shown at the right. See text for further discussion of actual species 24 25 combinations found. 25 26 26 27 all species except Song and Fox sparrows (see habitat centroids, Fig. 6.16(b)). 27 28 Given that habitat preferences overlap extensively among species, predicting 28 29 the identity of a quadrat’s occupant based on vegetational characteristics is 29 30 successful in just 38% (90/237) of the cases. 50% confidence ellipses were 30 31 used on 1994 data to define core habitat (Fig. 6.17) for the four common 31 32 emberizids (F,L,S,W), which amounts to 97 quadrats or two-thirds of the site. 32 33 Of this combined core, half is core habitat for all four species (47/97), while 33 34 most of the remainder is core for either the trio FLS (14 quadrats) or the duo 34 35 FS (12 quadrats); see Fig. 6.17 for the complete breakdown. 35 36 Given that core habitat is identified and the numbers of quadrats that fall 36 37 within different categories of core habitat known, the next question was how 37 38 each species occupies its core habitat, and specifically whether its occupancy 38 39 is affected by (a) whether its core quadrats are co-occupiable by other spar- 39 40 rows and (b) whether they are actually occupied by other species. The results 40 196 M.L. Cody

1 Table 6.1. Analysis of habitat occupancy by emberizid sparrows in the Tetons wet 1 2 willows habitat 2 3 3 4 (a) Lincoln’s sparrow core habitat: Q = 71; 26/71 occupied (prop. 0.366) 4 Quadrats occupied 5 Core type # Q # Occ Prop. #Exp by + L L′ #Exp 5 6 FSLW 47 14 .298 17.20 FSW 2 6 2.93 6 7 FLS 14 6 .429 5.12 ∪(FS,FW,SW) 5 17 8.05 7 LW 3 1 .333 1.10 ∪(F,S,W) 8 6 9.88 8 L 7 5 .714 2.56 None 8 6 5.12 8 9 Chi-sq. = 1.24, df = 3, NS Chi-sq. = 3.2, df = 3, NS 9 10 10 S 7 29 13.12 (i) 11 (i) Chi-sq. = 4.49, 4.71, df = 1, p < .05 S′ 19 16 12.81 (i) 11 12 12 13 F 11 29 14.64 (ii) 13 (ii) Chi-sq. = 1.43, 1.85, NS F′ 15 16 11.35 (ii) 14 14 15 (iii) Chi-sq. = 2.57, NS SF 5 19 8.78 (iii) 15 16 (iv) Chi-sq. = 20.79, p <<.05; S′F′ 12 0 4.29 (iv) 16 (v) Chi-sq. = 1.79, NS S or F 9 26 12.8 (v) 17 17 18 (b) Song sparrow core habitat: Q = 81; 47/81 occupied (prop. 0.580) 18 19 Quadrats occupied 19 Core type # Q # Occ Prop. #Exp by + S S′ #Exp 20 FSLW 47 26 .553 27.26 FLW 1 2 1.74 20 21 FLS 14 8 .571 8.12 ∪(FL,FW,LW) 12 14 15.08 21 22 FSW 6 5 .833 3.48 ∪(F,L,W) 21 8 16.82 22 23 FS 12 7 .583 6.96 None 13 8 12.18 23 SW 2 1 .500 1.16 24 Chi-sq. = 0.71, df = 4, NS Chi-sq. = 2.04, df = 3, NS 24 25 25 26 L 8 16 13.92 26 Chi-sq. = 8.70, df = 3, p < .05 L′ 39 18 33.06 27 27 28 F 11 15 14.64 28 ′ 29 Chi-sq. = 3.10, DF = 3, NS F 15 16 11.35 29 30 FL 5 5 8.78 30 31 Chi-sq. = 0.26, 0.01, 0.01, NS F′ L′ 13 9 4.29 31 32 ForL 28 21 12.8 32 33 (c) Fox sparrow core habitat: Q = 81; 46/81 occupied (prop. 0.568) 33 34 Quadrats occupied 34 ′ 35 Core type # Q # Occ Prop. #Exp by + F F #Exp 35 FSLW 47 25 .532 26.70 LSW 1 1 1.14 36 FLS 14 11 .786 7.95 ∪(LS,LW,SW) 14 7 11.93 36 37 FSW 6 3 .500 3.41 ∪(L,S,W) 25 19 24.99 37 38 FS 12 5 .417 6.82 None 6 8 7.95 38 F 2 2 1.00 1.14 39 Chi-sq. = 2.46, df = 4, NS Chi-sq. = 0.85, df = 3, NS 39 40 40 Scale and pattern in plant and bird communities 197

1 Table 6.1. (continued) 1 2 2 3 (d) White-crowned sparrow core habitat: Q = 62; 21/62 occupied (prop. 0.339) 3 4 Quadrats occupied 4 Core type # Q # Occ Prop. #Exp by + W W′ #Exp 5 FSLW 47 13 .277 15.93 FLS 1 4 1.70 5 6 FSW 6 3 .500 2.03 ∪(FL,FS,LS) 9 11 6.78 6 7 LW 3 1 .333 1.02 ∪(F,L,S) 11 21 10.85 7 SW 2 0000 0.68 None 0 5 1.70 8 W 4 4 1.00 1.36 8 9 Chi-sq. = 2.20, df = 1, NS Chi-sq. = 2.42, df = 2, NS 9 10 10 11 Q = # of quadrats, #Occ = # of quadrats occupied, #Exp = expected number of 11 12 quadrats occupied. Species abbreviations are L: Lincoln’s sparrow, F: Fox sparrow, 12 S: Song sparrow, W: White-crowned sparrow. The ∪ is the set theory symbol for 13 union. 13 14 14 15 15 16 of this analysis are shown in Table 6.1. For each species, the occupancy rate 16 17 (quadrats occupied/quadrats available) of its core habitat is not affected by 17 18 whether or not its core is also core for the other three species, for various 18 19 two-species combinations, or for other species individually (Table 6.1, left 19 20 side). Further, in no species does occupancy of core habitat deviate from ran- 20 21 dom expectations depending on whether its core quadrats, grouped by num- 21 22 bers of co-occupant species, are actually occupied by the trio, various duos 22 23 or various other single species (Table 6.1, upper right for each species). 23 24 However, two species, Lincoln’s and Song sparrow, do show a significant 24 25 dependence on each others’ distributions and tending to avoid quadrat co- 25 26 occupancy with the other (Table 6.1, lower right for the two species). 26 27 Lincoln’s sparrow has a significantly higher occupancy rate in quadrats with- 27 28 out Song sparrows (S′) and especially those without Song and Fox sparrow 28 29 (S′F′), and a significantly lower occupancy rate where Song sparrow is present 29 30 (+S). The corresponding analysis for Song sparrow occupancy shows a sig- 30 31 nificantly reduced quadrat use where Lincoln’s sparrow is present (+L); Fig. 31 32 6.18 graphically summarizes these results. 32 33 Overall, the assembly of emberizid species’ combinations in various habi- 33 34 tat quadrats is understood by (a) species specific habitat preferences, which 34 35 are distinct among species for the most part, but broadly overlapping in the 35 36 four commoner species, especially in Song and Fox sparrows (S, F), less 36 37 similar in Lincoln’s sparrow (L); (b) Fox sparrow is a much larger species, 37 38 which may explain why it appears not to interact with S or L; (c) Song and 38 39 Lincoln’s sparrows are similarly sized, and tend to avoid shared use of their 39 40 common core habitat. (d) Further evidence supports a behavioral interaction 40 198 M.L. Cody 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 Fig. 6.18. Two emberizid species, Song and Lincoln’s sparrow, avoid interspecific co- 22 23 occupancy of quadrats via behavioral interactions. The mean quadrat occupancy rate 23 for each species is shown on ordinate at the right, and each species’ occupancy rate 24 is shown in the figure (open dots for Song sparrow, closed for Lincoln’s sparrow) as 24 25 a function of quadrat occupancy by the other species (at bottom, abscissa). Note that 25 26 Lincoln’s sparrow occupies significantly more quadrats in its core habitat when Song 26 (−S) or both Song and Fox sparrows (−S,F) are absent, significantly fewer when Song 27 sparrow is present (+S), and is unaffected by Fox sparrow presence. Song sparrow 27 28 shows a corresponding but weaker avoidance of Lincoln’s sparrow, occupying sig- 28 29 nificant fewer of its core quadrats when Lincoln’s sparrow is absent (−L). * indicates 29 a significant deviation from the expected occupancy rates. 30 30 31 31 32 between S and L, as the two species occasionally display interspecific aggres- 32 33 sion, with song duels and chasing between allospecific territorial neighbors. 33 34 Lastly, (e) in dry years Lincoln’s sparrow is more common and Song spar- 34 35 row densities are lower, while in wet years the opposite is true; in any one 35 36 year the two tend to occupy the site in a complementary fashion. 36 37 37 38 The paruline warblers 38 39 The suite of warblers breeding in the wet willows site is comprised of three 39 40 10 g species: Yellowthroat (H), Yellow (Y), and MacGillivray’s warbler (G), 40 Scale and pattern in plant and bird communities 199 1 the slightly smaller Wilson’s warbler (W, 8 g), and the larger (and rare) North- 1 2 ern waterthrush (N, 18 g). In rank density, Y and H are abundant, W is usually 2 3 common but varies in density among years, and G and N are both scarce at 3 4 the site, the former present most years and the latter only in wet years (Cody, 4 5 1996). While the waterthrush forages on or near the ground in the more water- 5 6 logged quadrats, the other four species are foliage insectivores within the 6 7 willows vegetation. Wilson’s warbler does a little aerial flycatching, but apart 7 8 from that all four of the species engage in similar foraging behavior (but 8 9 employed at distinctly different heights above; see below). 9 10 Like the emberizids, the parulines also show interspecific differences in 10 11 habitat preference, but with much overlap. A factor analysis of vegetation 11 12 characteristics shows the distributions occupied quadrats for each species (Fig. 12 13 6.19(a)). The waterthrush is most distinct, with a preference for watery sites 13 14 in which only the tallest willows survive. MacGillivray’s warbler prefers taller 14 15 willows with less open water, and Yellowthroat preferentially occupies the 15 16 shorter willows. The results of DFA on the warbler habitat preferences are 16 17 shown in Fig. 6.19(b). The centroids depicted are 95% confidence ellipses 17 18 around the mean; while the means are fairly distinct among species, there is 18 19 overall extensive interspecific overlap, and just 95/243 quadrats (39%) can 19 20 be correctly classified to specific occupant. 20 21 An analysis of the core habitats for the three more common species (Y, H, 21 22 W) was undertaken as for the emberizids (above). It shows there are no inter- 22 23 actions among species over habitat; a quadrat is occupied or not indepen- 23 24 dently of whether another one or two species also include the quadrat within 24 25 their territories. 25 26 The assembly of warblers in habitat patches as represented by different 26 27 quadrats is determined largely by habitat structure. The basis of the assem- 27 28 bly rules is provided by the foraging height distributions of the different 28 29 species, which rank G-W-Y-H from higher to lower in foraging activity within 29 30 the vegetation; Fig. 6.20 (left-hand side; from Cody, 1996) illustrates the dif- 30 31 ferences among species in this predominant aspect of foraging ecology. The 31 32 warblers select habitat patches within the site that provide vegetation of 32 33 heights corresponding to their foraging activity. Thus the representation of 33 34 tall vegetation in quadrats occupied by high-foraging MacGillivray’s warbler 34 35 significantly exceeds the site average, that of low vegetation in Common 35 36 yellowthroat quadrats, a low-foraging species, is higher than the site average, 36 37 and Yellow warbler, which is nearly ubiquitous within the plot, is least distinct 37 38 in quadrats occupied and forages at intermediate heights that are well repre- 38 39 sented throughout the site (Fig. 6.20, right-hand side). 39 40 In summary, different quadrats support combinations of warbler species in 40 200 M.L. Cody 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 Fig. 6.19. (a) Factor analysis (principal components F{1}, F{2}) of vegetation char- 35 acteristics at the willows site, and the distributions of paruline warblers over these 35 36 factors. Northern waterthrush shows a distinct habitat preference for wet habitat (high 36 37 F{2} scores) and for taller vegetation (high F{1} scores), whereas MacGillivray’s 37 warbler occupies tall vegetation but in dry quadrats. Rather similar habitat preferences 38 characterize the remaining three species. (b) Discriminant function analysis of the 38 39 same data shows that mean habitat preferences tend to be distinct among all species 39 40 (95% centroids shown), except that MacGillivray’s and Wilson’s warblers exhibit a 40 considerable overlap. Scale and pattern in plant and bird communities 201 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 Fig. 6.20. Foraging height distributions differ significantly (by Chi-squared tests) in 35 36 the four common foliage insectivorous warblers (left-hand side). MacGillivray’s 36 37 warbler forages highest in the vegetation, Common yellowthroat lowest, with Wilson’s 37 and Yellow warblers at intermediate heights. Foraging heights correspond to the pre- 38 dominant vegetation heights in the quadrats occupied by each species (right-hand side); 38 39 ‘relative habitat choice’ is measured here by the deviation of vegetation in occupied 39 40 quadrats from the site as a whole, in standard deviation units. Off-scale points labeled 40 with arrows. 202 M.L. Cody 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 Fig. 6.21. Quadrats support various wabler combinations in accordance to the distri- 17 18 bution of vegetation height within them. The common combinations are circled, indi 18 cating the numbers of quadrats in which these combinations occur. 19 19 20 20 21 accordance with the availability of willows in the required height classes. 21 22 Favored combinations are those supported by willows that are low (with H 22 23 only), low to intermediate (H + Y), of intermediate height (Y only) or are 23 24 intermediate to tall (Y + W; see Fig. 6.21.) The three species combination 24 25 (H+Y+G) is also common. Other combinations are rare, and the low inci- 25 26 dence of MacGillivray’s warbler at the site is readily explained by the dearth 26 27 of willows over 2m in height (viz. Fig. 6.15). 27 28 28 29 29 30 Conclusions 30 31 In this chapter the application of assembly rules to communities has been 31 32 discussed in three general classes of comparison, at three different levels of 32 33 resolution. Data from a variety of studies were reviewed for each of the three 33 34 classes, which were illustrated further and discussed in detail using additional 34 35 studies and examples: (a) amongst islands with discrete plant species assem- 35 36 blages, (b) in oak woodland bird communities replicated across a consider- 36 37 able range of the habitat type, and (c) within a single heterogeneous willows 37 38 site, in which the composition of emberizid sparrow and paruline warbler 38 39 guilds was assessed at the level of individual quadrats. Clearly, assembly rules 39 40 can be sought at each of these different levels of resolution. 40 Scale and pattern in plant and bird communities 203 1 Patterns of species co-occurrence on islands were used to identify a pos- 1 2 teriori island characteristics that support one or another assemblage of forest 2 3 plants. While α-diversity is not closely constrained on the islands, differences 3 4 in island size, elevation, and sheltered vs. exposed positions contribute to 4 5 species turnover and hence to β-diversity. In other components of the flora, 5 6 the weedy edge species and especially the Asteraceae allow for no predictable 6 7 species assemblages, and species replacements among islands contribute not 7 8 to β- but to γ-diversity. 8 9 Oak woodland bird communities, on the other hand, appear tightly 9 10 regulated in α-diversity, although the community composition does change 10 11 throughout the range of the oaks. Some species replacements appear unpre- 11 12 dictable, with the substitution of ecologically related species unrelated to shifts 12 13 in vegetation structure, latitude, or other geographical aspects of site position 13 14 (e.g., wrens, vireo guilds in the southern sites). In some cases the composition 14 15 of favored assemblages is related to variation in the structure of the wood- 15 16 land (flycatchers) or the geographical position of the site (woodpeckers). Thus 16 17 changes among sites in community composition contribute variously to both 17 18 β- and γ-diversity. 18 19 Within sites, ecologically related species may coexist by dint of morpho- 19 20 logical or behavioral differences, and their coexistence moderated by the char- 20 21 acteristics of the habitat at the quadrat scale. Species that are particularly 21 22 close, ecologically, may coexist only in certain years (e.g., high productivity 22 23 years) or in certain sites or that permit their divergence, but otherwise they 23 24 segregate by habitat use. Thus favored combinations will include one or the 24 25 other but not both species together. Old World sylvine warblers in various 25 26 pair-wise combinations, Brewer’s and Vesper sparrows, and Song and Lincoln’s 26 27 sparrows all provided examples of this phenomenon, and further illuminated 27 28 conditions under which their coexistence might in fact be permitted. 28 29 29 30 30 31 References 31 32 Blake, J.G. (1991). Nested subsets and the distribution of birds on isolated 32 woodlots. Conservation Biology 5: 58–66. 33 Bolger, D.T., Alberts, A.C., & Soulé M.E. (1991). Occurrence patters of bird 33 34 species in habitat fragments: sampling, extinction, and nested species subsets. 34 35 American Naturalist, 137: 155–166. 35 36 Brown, J.H. (1971). Mammals on moutaintops: nonequilibrium insular 36 biogeography. American Naturalist 105: 467–478. 37 Case, T.J. & Casten, R. (1979). Global stability and multiple domains of attraction 37 38 in ecological systems. American Naturalist 113: 705–714. 38 39 Cody, M.L. (1973). Parallel evolution and bird niches. In Mediterranean-Type 39 Ecosystems: Analysis and Synthesis, ed. F. diCastri & H.A. Mooney 40 Ecological Studies 7, Chapter 3, pp. 307–338. Wien: Springer-Verlag. 40 204 M.L. Cody

1 Cody, M.L. (1974). Competition and the Structure of Bird Communities. 1 2 Monographs in Population Biology 7, Princeton NJ: Princeton University 2 Press. 3 Cody, M.L. (1975). Towards a theory of continental species diversities. In Ecology 3 4 and Evolution of Communities, ed. M.L. Cody & J. Diamond Chapter 10, pp. 4 5 214–257. 5 6 Cody, M.L. (1978). Habitat selection and interspeciﬁc interactions among the 6 sylviid warblers of England and Sweden. Ecology Monographs 351–396. 7 Cody, M.L. (1983a). Bird species diversity and density in Afromontane woodlands. 7 8 Oecologia 59: 210–215. 8 9 Cody, M.L. (1983). Continental diversity patterns and convergent evolution in bird 9 communities. In Ecological Studies 43, ed. F. Kruger D.T. Mitchell & J.U.M. 10 Jarvis Chapter 20, pp. 347–402. Wien: Springer-Verlag. 10 11 Cody, M.L. (1985). Habitat selection in the sylviine warblers of western Europe 11 12 and north Africa. In Habitat Selection in Birds, ed. M.L. Cody, Chapter 3, pp. 12 85–129. Orlando, San Diego: Academic Press. 13 Cody, M.L. (1986). Diversity and rarity in Mediterranean ecosystems. Ch. 7 In 13 14 Conservation Biology: The Science of Scarcity and Diversity, ed. M. Soulé, 14 15 pp. 122–152. Sunderland, MA: Sinauer Assoc. 15 16 Cody, M.L. (1992). Some theoretical and empirical aspects of habitat 16 fragmentation. Proceedings of the Society of California Academy of Science 17 Symposium, Occidental College, CA. 17 18 Cody, M.L. (1993). Bird diversity patterns and components across Australia. In 18 19 Species Diversity in Ecological Communities: Historical and Geographical 19 Perspectives, ed. R.E. Ricklefs & D. Schluter, Chapter 13, pp. 147–158. 20 Chicago, IL: University of Chicago Press. 20 21 Cody, M.L. (1994a). Bird diversity components in heathlands: comparisons SW and 21 22 NE Australia. Tasks in Vegetation Science, ed. R. Groves & M. Farragitaki, 22 vol. XX, pp. 47–61. Holland: Kluwer. 23 Cody, M.L. (1994b). Mulga bird communities. I. Species composition and 23 24 predictability across Australia. Australian Journal of Ecology 19(6): 206–219. 24 25 Cody, M.L. (1994c). Bird communities of Australian Eucalyptus and north- 25 26 temperate Quercus woodlands. Journal für Ornithologie 135: 468. 26 Cody, M.L. (1996). Birds of the central Rocky Mountains. In Long-Term Studies of 27 Vertebrate Communities. ed. M.L. Cody, & J. Smallwood Chapter 11, pp. 27 28 291–342. Orlando, San Diego: Academic Press. 28 29 Cody, M.L. & Diamond, J.M. (1975). Ecology and Evolution of Communities. 29 Cambridge, MA: Belknap Press of Harvard University Press. 30 Cody, M.L. & Walter, H. (1976). Habitat selection and interspeciﬁc territoriality 30 31 among Sardinian Sylvia warblers. Oikos 27: 210–223. 31 32 Cody, M.L. & Mooney, H.E. (1978). Convergent and divergent evolution in 32 chaparral, plant and bird communities on four continents. Annual Review of 33 Ecology and Systematics 9: 265–321. 33 34 Cody, M.L. & Overton, J.McL. (1996). Evolution in achene morphology of 34 35 composite plants on small islands. Journal of Ecology 84(1): 53–61. 35 36 Cutler, A. (1991). Nested faunas and extinction in fragmented habitats. 36 Conservation Biology 5: 496–505. 37 Diamond, J.M. (1975). Assembly of species communities. In Chapter 14, Ecology 37 38 and Evolution of Communities, ed. M.L. Cody, & J.M. Diamond, pp. 342–444. 38 39 Cambridge, MA: Belknap Press of Harvard University Press. 39 Diamond, J.M. & Mayr, E. (1976). Species–area relation for birds of the Solomon 40 40 Scale and pattern in plant and bird communities 205

1 Archipelago. Proceedings of the National Academy of Sciences, USA 73: 1 2 262–266. 2 Diamond, J.M., Gilpin M.E. & Mayr, E. (1976). Species–distance relation for birds 3 of the Solomon Archipelago: the paradox of the great speciators. Proceedings 3 4 of the National Academy of Sciences USA, 73: 2160–2164. 4 5 Diamond, J. & Case, T.J. (1986). Community Ecology. NY: Harper & Row. 5 6 Kadmon, R. (1995). Nested species subsets and geographic isolation: a case study. 6 Ecology 76: 458–465. 7 MacArthur, R.H., Recher, H. & Cody, M.L. (1996). On the relation between habitat 7 8 selection and bird species diversity. American Naturalist 100: 319–332. 8 9 MacArthur, R.H. & Wilson, E.O. (1967). The Theory of Island Biogeography. 9 Monographs Population Biol. 1. Princeton, NJ: Princeton University Press. 10 Patterson, B.D. (1987). The principle of nested subsets and its implications for 10 11 biological conservation. Conservation Biology 1: 323–334. 11 12 Patterson, B.D. & Atmar, W. (1986). Nested subsets and the structure of insular 12 mammalian faunas and archipelagoes. Biological Journal of the Linnaean 13 Society 28: 65–82. 13 14 Ryti, R.T., & Gilpin, M.E. (1987). The comparative analysis of species occurrence 14 15 patterns on archipelagos. Oecologia 73: 282–287. 15 16 Salt, G.W. (1957a). An analysis of avifaunas in the Teton Mountains and Jackson 16 Hole, Wyoming. Condor 59: 373–393. 17 Salt, G. W. (1957b). Song, Lincoln’s and fox sparrows in a Tetons willow thicket. 17 18 Auk 74: 258. 18 19 Simberloff, D. & Levin, B. (1985). Predictable sequences of species loss with 19 decreasing island area – and birds in two archipelagoes. New Zealand Journal 20 of Zoology 8: 11–20. 20 21 Simberloff, D. & Martin, J-L. (1991). Nestedness of insular avifaunas; simple 21 22 summary statistics masking complex species patterns. Ornis Fennica 68: 22 178–192. 23 Wilson, B. & Roxburgh, X. (1994). Oikos 69: 267–276. 23 24 Wilson, J.B. & Roxburgh, S.H. (1994). A demonstration of guild-based assembly 24 25 rules for a plant community, and determination of intrinsic guilds. Oikos 69: 25 26 267–276. 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 7 1 2 2 3 Impact of language, history, and choice of 3 4 system on the study of assembly rules 4 5 5 6 6 7 Barbara D. Booth and Douglas W. Larson 7 8 8 9 9 10 10 11 11 12 12 13 13 14 Introduction 14 15 Assembly rules provide a useful framework that allows ecologists to make 15 16 predictions about community change and development over time. However, 16 17 three features of the discussion about ecological assembly rules lessen its 17 18 potential impact on ecologists and those who apply ecology to management. 18 19 First, the current language of the idea has become unnecessarily complex. It 19 20 is not always clear what researchers mean when they refer to assembly rules. 20 21 Some argue that assembly rules only apply to biotic interactions (Wilson & 21 22 Gitay, 1995) whereas others consider all constraints on community develop- 22 23 ment (Keddy, 1992). Some workers look for rules that apply at the small 23 24 scale, or to speciﬁc species distributions, whereas others look for larger, more 24 25 general patterns (Drake et al., 1993). 25 26 Second, the current discussion appears to have little regard for the history 26 27 or origin of work on the constraints on community development. Diamond 27 28 (1975) is usually cited as having coined the phrase ‘assembly rule’, but in 28 29 reality, many early ecologists such as Clements (1916, 1936) and Gleason 29 30 (1917, 1926) addressed many of the questions currently being considered. By 30 31 ignoring their work, we run the risk of reinventing ideas already well estab- 31 32 lished in the literature. 32 33 Finally, the primary evidence used to develop and test ideas on assembly 33 34 rules is often derived from complex ecosystems that do not necessarily 34 35 follow simple trajectories (Drake, 1991). Thus, community ecologists may 35 36 be starting with the most complex systems available. An alternative strategy 36 37 may be to start with simple natural systems where assembly rules are easier 37 38 to ﬁnd. After being able to observe them, it may be possible to expand the 38 39 horizon to more complex systems. It is suggested that communities in un- 39 40 productive, stressful environments are more likely to be predictable and 40

206 Impact of language, history and choice of system 207 1 therefore are good systems to use for preliminary investigations of assembly 1 2 rules. 2 3 In this chapter, it is argued that, if workers persist in ignoring these 3 4 problems, little progress will be made in the study of assembly rules. Each 4 5 of these three points will be addressed below. In each case the objective is 5 6 to simplify the debate, not to complicate it. It is important to remember that 6 7 if the history of science and the way humans participate in it are ignored the 7 8 mistakes of the past are very likely to be repeated. 8 9 9 10 10 11 Language considerations 11 12 Since the publication of Diamond’s book chapter (Diamond, 1975), there has 12 13 been a proliferation of papers trying to promote, refute, or test the idea that 13 14 there are sets of constraints (rules) on community formation and maintenance 14 15 (assembly). Some argue that particular combinations of species are the result 15 16 of the non-random consequences of competition (Diamond & Gilpin, 1982; 16 17 Fox, 1987). Others suggest that the observed patterns could also be explained 17 18 by chance (Connor & Simberloff, 1979) or historical factors (Drake, 1990a). 18 19 Others take the view that assembly rules should be discussed within a gram- 19 20 matical context that includes many other equally vague but potentially use- 20 21 ful terms such as ‘organism’, ‘species’, or ‘ecosystem’ (Haefner, 1978, 1981). 21 22 Some have argued that the assembly rules can include any constraint on the 22 23 species pool (Keddy, 1992), while others have a more narrow view that assem- 23 24 bly rules can only include constraints placed upon species by other species 24 25 (Lawton, 1987; Wilson & Gitay, 1995; Wilson & Whittaker, 1995). Clearly, 25 26 there is no general concensus as to what assembly rules are and how they 26 27 should be defined, although usage of the term falls under two general schools 27 28 of thought: those who believe assembly rules should describe the mechanisms 28 29 of biotic interactions only, and those who recognize that a community is 29 30 formed from a species pool and that biotic and abiotic mechanisms will act 30 31 to determine which of these species will form the community (Table 7.1). 31 32 Even within these groups, there is some variation in the definition. Here we 32 33 take the latter view. 33 34 Many researchers have tried to articulate assembly rules for various eco- 34 35 logical and artificial communities. In more controlled environments the rules 35 36 generated are general in nature and can be used to establish principles of com- 36 37 munity assembly. For example, Drake (1991) showed, in laboratory experi- 37 38 ments, that larger systems with greater total productivity are more prone to 38 39 historical events than small systems with low productivity. He found that 39 40 species invasion sequence was more likely to have an effect on the species 40 208 B.D. Booth and D.W. Larson

1 Table 7.1. Some definitions and statements used to define or describe assembly 1 2 rules 2 3 3 4 Biotic interactions 4 Diamond (1975). ‘Do properties of each species considered individually tell us 5 most of what we need to know in order to predict how communities are 5 6 assembled from the total species pool.’ p. 385 6 7 ‘It seems likely that competition for resources is a major factor underly- 7 ing assembly rules.’ p. 423 8 Lawton (1987). ‘I take the view that assembly rules arise from species interactions.’ 8 9 Wilson & Watkins (1994). ‘... generalized constraints on species interactions’ 9 10 Wilson & Whittaker (1995). ‘... generalized restrictions on species presence or 10 11 abundance that are based on the presence or abundance of one or several 11 other species or types of species (not simply the response of individual 12 species to the environment).’ 12 13 13 Biotic and abiotic interactions 14 Haefner (1981). describing the ‘ecosystem assembly problem’. ‘Construct an algo- 14 15 rithm such that, given an arbitrary collection of environmental factors, the 15 16 output of the algorithm is a list of species associated with the environ- 16 ment.’ 17 ‘The model must be general by being applicable to any given environ- 17 18 ment and species pool ... [and] produce specific predictions of the spatial 18 19 distributions of particular species.’ 19 Keddy (1989). ‘Given (a) a species pool and (b) an environment, can we predict 20 the abundance of the organisms actually found in that environment?’ p. 20 21 156 21 22 Roughgarden (1989). ‘A community’s membership is structured by the transport 22 23 processes that bring species to it, and it is structured by the population 23 dynamics, including species interactions of its members; and furthermore, 24 it is rarely possible to focus on only one of these sides. That is to say, a 24 25 community reflects both its applicant pool and its admission policies.’ p. 25 26 218 26 Drake (1990b). ‘Given a species pool and environment, what rules influence com- 27 munity structure?’ 27 28 Drake et al. (1993). ‘... how mechanisms [ecological processes and environmental 28 29 variation] function to produce ecological patterns.’ 29 Grover (1994). ‘When local communities are assembled from a regional species 30 pool, assembly rules state which of the species from this pool can coex- 30 31 ist.’ 31 32 32 33 33 34 34 35 composition of more productive systems. Similarly, Law and Morton (1993) 35 36 used computer simulations to show that the predictability of community 36 37 composition decreased as the number of species increased, and that histori- 37 38 cal factors influencing the order of invasion were extremely important to 38 39 the outcome. These generalities can now be tested in suitable ecological 39 40 communities. 40 Impact of language, history and choice of system 209 1 Some workers have described constraints in field situations; however, these 1 2 tend to be limited in scope as they are species or trait specific and rarely 2 3 describe the entire community. Fox and Brown (1993), for example, found 3 4 in rodent communities in southwestern deserts of North America that each 4 5 new species entering a community will tend to come from a different func- 5 6 tional group. When all functional groups are filled, the rule repeats. They 6 7 were able to apply this rule to different groups of desert rodents and at 7 8 different spatial scales; however, it applies only to rodents and focuses just 8 9 on competitive interactions of rodents. Similarly, van der Valk (1981) showed 9 10 that the invasion of wetlands involves the primary assembly rule that plants 10 11 must be able to tolerate flooding during germination. This rule is useful as it 11 12 focuses on the mechanisms important to establishment and allows us to predict 12 13 which species may become part of the community; however, is limited 13 14 because it does not take competition into consideration nor does it predict 14 15 final species composition and abundance. One further difficulty with empirical 15 16 field studies is determining whether the identified mechanisms are important 16 17 to community development or whether they are simply important in main- 17 18 taining the current community. This has been discussed extensively by Drake 18 19 (1991). 19 20 In other settings we find that: ‘[The assembly rules] demonstrated are not 20 21 strong ones. Either there are many other processes at work, obscuring the 21 22 assembly rules, or more powerful methods are required to find the rules’ 22 23 (Wilson & Watkins 1994). Other assembly rules are defined and then con- 23 24 cluded to be ephemeral or contextual in their appearance (Wilson et al., 24 25 1995b), or are defined in the introductions of papers and then never men- 25 26 tioned in the Discussion sections (Wilson et al., 1996). 26 27 What seems to have been missed is that the original chapter that offered 27 28 the term assembly rule simply took advantage of an often used technique of 28 29 ecological pedagogy: when faced with a truth that more or less applies at 29 30 least some of the time, try to present that truth in the form of a ‘law’ or, 30 31 when even greater uncertainty is involved, as a ‘rule’. Hence Bergmann’s 31 32 rule predicts that homeotherms in polar regions should be relatively larger 32 33 than similar organisms near the equator, Allen’s rule claiming that appendages 33 34 should be shorter in polar regions, and Gloger’s rule stating that animals from 34 35 hot dry regions ought to be less pigmented than ones from wet cold habitats 35 36 (Pianka, 1983). Few practising ecologists expect such rules to do more than 36 37 describe very general trends that might reflect some underlying physiologi- 37 38 cal or evolutionary principle. That is their purpose. But from time to time, 38 39 there are those who try to test empirically some of the predictions from these 39 40 rules, only to find that the truth is obscured by the variance (Peters, 1988). 40 210 B.D. Booth and D.W. Larson 1 By trying to make assembly rules that apply all of the time, we are prevent- 1 2 ing ourselves from establishing general principles that apply most of the time. 2 3 Some of the difficulty in finding general rules that apply most of the time 3 4 is due simply to the complexity of ecological systems, and the different scales 4 5 at which they are examined. Further, ecologists should not be embarrassed 5 6 to derive rules or laws to describe patterns or processes that apply only part 6 7 of the time, or for part of the world. As Diamond and Gilpin (1982) also tried 7 8 to show, ecologists should not permit a single philosophical view to be used 8 9 to dissect the structure of nature. The evidence from the assembly rules debate 9 10 suggests that ecologists are trying to search for greater clarity in the word- 10 11 ing of various assembly rules based on the available data. However, the debate 11 12 continues over how best to do this. Diamond and Gilpin (1982), and later 12 13 Keddy (1989), correctly pointed out that, even the decision about what null 13 14 model to test patterns against requires more information than we have about 14 15 the structure and function of ecosystems. Wilson (1994) argues that actual 15 16 organizational features of communities should be tested against null (or alter- 16 17 nate) models. He does not address the issue that few statistical tests would 17 18 be able to discriminate among any of the individual species distribution pat- 18 19 terns he plotted. Thus, the choice of a particular null model could entirely 19 20 explain the finding of an assembly rule in a New Zealand rain forest (small 20 21 plants and the herb guild in the ground flora will be relatively constant whereas 21 22 large, less frequent plants will not) (Wilson et al., 1995a). 22 23 So what scientific progress has been made by the use of new terminology 23 24 in this area – progress that would not have been made if simple and clear 24 25 terminology had been used? The debate has certainly resulted in the genera- 25 26 tion of more primary data, and despite Keddy’s assertion that we have enough 26 27 primary data (Keddy, 1989), many believe that we do not. Other than this 27 28 vastly greater amount of information on the composition of specific plant and 28 29 animal communities world-wide, what new and consistently verifiable pre- 29 30 dictions have been made because of the new language? Few. What has actually 30 31 been accomplished is that workers previously exhausted with 70 years of 31 32 debate over the complex controls of community composition in succession 32 33 have been re-energized to study essentially the same problem dressed in new 33 34 clothes. 34 35 One of the questions that seems important here is this: what new questions 35 36 does the idea of assembly rules permit that were not possible with the orig- 36 37 inal language of succession or even the language that was reviewed by Drury 37 38 and Nisbet (1973)? It can be argued that if the word constraint is used to 38 39 substitute for the word rule and the word development is used to substitute 39 40 for the word assembly, then the current discussion of assembly rules is 40 Impact of language, history and choice of system 211 1 reduced to a discussion of developmental constraints on community struc- 1 2 ture: an idea fully explored by Clements, Gleason and many other ecologists. 2 3 Wilson and Whittaker’s (1995) discussion of guild proportionality would 3 4 likely not surprise any of the past workers in the area of plant or animal com- 4 5 munities undergoing succession because the idea of there being developmental 5 6 constraints on the form and number of plants and animals was widely accepted 6 7 at the turn of the century. 7 8 The language of the assembly rules debate has hidden the initial intention 8 9 of the term, and has added to the confusion that ecologists, conservationists 9 10 and land managers have over whether the idea encapsulates new and impor- 10 11 tant concepts and applications. In the next section, the development of the 11 12 ideas of constraints on community structure is discussed in an attempt to 12 13 remind ourselves of the already established principles. 13 14 14 15 15 16 Historical considerations 16 17 The idea of ecological or community assembly rules is not new, nor is it fair 17 18 to attribute the idea to Diamond (1975) as many have done (Connor & 18 19 Simberloff, 1979; Fox, 1987; Keddy, 1989; Wilson, 1991, 1994; Drake et al., 19 20 1993; Wilson et al., 1994, 1995a,b). What is new is the emergence of the 20 21 term ‘assembly rules’, used initially within the context of animal ecologists 21 22 trying to explain the constraints on species abundance and composition on 22 23 tropical islands. The attempt to explain these island biogeographic patterns 23 24 itself was not new and is dated at least to the summary work by MacArthur 24 25 and Wilson (1967). As is often the case in biology, these complicated species 25 26 composition patterns – both their genesis and maintenance – were being 26 27 explored by one group of ecologists, animal ecologists, without explicit 27 28 discussion of the fact that the problem had already been examined at length 28 29 by other groups, most notably the plant ecologists. For example, in 1916 29 30 Clements wrote: 30 31 Reasons why plants appear at certain stages – Migrules are carried into an area more 31 32 or less continually during the course of its development. This is doubtless true of per- 32 33 mobile seeds, such as those of aspen. As a rule, however, species reach the area con- 33 34 cerned at different times, the time of appearance depending chiefly upon mobility and 34 35 distance. As a consequence, migration determines in some degree when certain stages 35 will appear. The real control, however, is exerted by the factors of the habitat, since 36 36 these govern ecesis1 and hence the degree of occupation. The habitat determines the 37 37 38 38 1‘Ecesis is the adjustment of the plant to a new home. It consists of three essential processes, 39 germination, growth and reproduction. It is the normal consequence of migration ...’ Clements 39 40 (1916) (p. 68). 40 212 B.D. Booth and D.W. Larson

1 character of the initial stage by its selective action in the ecesis of migrules.... After 1 2 the initial stage the development of succeeding ones is predominantly, if not wholly, 2 3 a matter of reaction, more or less affected by competition. In addition, some stages 3 owe their presence to the fact that certain species develop more rapidly and become 4 4 characteristic or dominant, while others which entered at the same time are growing 5 slowly. (p. 102) 5 6 6 7 In this text, taken from the classic work ‘Succession’, there is a concern for 7 8 the very problem considered by Diamond (1975) and other island biogeog- 8 9 raphers before him: the question of what constraints determine if, and when, 9 10 a species will be successful at a certain site. 10 11 Very shortly after Clements wrote the paragraph quoted above, H.A. Glea- 11 12 son wrote in 1917 and reiterated the following in 1926: 12 13 It has sometimes been assumed that the various stages in a successional series follow 13 14 each other in a regular and fixed sequence, but that frequently is not the case. The 14 15 next vegetation will depend entirely on the nature of the immigration which takes 15 16 place in the particular period when environmental change reaches the critical stage. 16 Who can predict the future for any one of the little ponds considered above? In one, 17 17 as the bottom silts up, the chance migration of willow seeds will produce a willow 18 thicket, in a second a thicket of Cephalanthus may develop, while a third, which 18 19 happens to get no shrubby immigrants, may be converted into a miniature meadow 19 20 of Calamagrostis canadensis. A glance at the diagram of observed successions in the 20 21 Beach Area, Illinois, as published by Gates, will show at once how extraordinarily 21 22 complicated the matter may become, and how far vegetation may fail to follow simple, 22 pre-supposed successional series. (p. 21). 23 23 24 Here is the idea of historical artifact or randomness being important to species 24 25 invasion patterns. In addition, what is really interesting about these quota- 25 26 tions from Clements and Gleason, and the vast literature that grew out of the 26 27 debate between the ‘organismic’ and the ‘individualistic’ view of succession, 27 28 is that the primary evidence used to support the two warring schools of thought 28 29 was identical, or nearly so. 29 30 The fight continued through the 1960s when Odum (1969) attributed some 30 31 measure of Clements’ superorganismic thought to the process of succession, 31 32 as revealed by the new buzzwords: ‘The strategy of ecosystem development’ 32 33 (Odum, 1969). Odum believed that succession was ‘community controlled’ 33 34 – that there were some rules governing community development. This idea 34 35 was discussed by many, including Grime (1977, 1979) who used a relatively 35 36 simple triangular model to describe the rate of biomass accumulation and the 36 37 sequence of life-forms along successional pathways. Drury and Nisbet’s brave 37 38 1973 attempt to tow ecologists out of the gumbo of rhetoric about the actual 38 39 patterns and mechanisms behind ecological succession was applauded by 39 40 some (Golley, 1977), but ignored by most. Drury and Nisbet promoted the 40 Impact of language, history and choice of system 213 1 idea that patterns resulted from differential growth, colonizing ability, and 1 2 longevity of constituent species, rather than from autogenic processes where 2 3 earlier plants ameliorated conditions for later colonizing species as Odum had 3 4 suggested. Connell and Slatyer (1977) argued that competitive interactions 4 5 (between sessile organisms and with herbivores, predators and pathogens) 5 6 were an important force and outlined three models of mechanisms (facilita- 6 7 tion, tolerance, inhibition) that produce successional change. It was at this 7 8 time that the new idea of ecological communities being controlled by assem- 8 9 bly rules was first offered by animal ecologists concerned with the control of 9 10 species mixtures. 10 11 In the history of ideas about how communities are assembled and main- 11 12 tained, we see that many of the same problems identified at the beginning of 12 13 ecology are still with us, unsolved and (perhaps) forever clouded by complex 13 14 details that are the reality of nature. The core of the current discussion repeats 14 15 similar debates over the last century. 15 16 16 17 17 18 The choice of system used to search for assembly rules 18 19 The last feature of the discussion of assembly rules is the part that attracted 19 20 us to this debate and to a symposium on the subject in the first place. Regard- 20 21 less of the set of language used, why is it that specific sets of constraints on 21 22 community composition are so hard to consistently find in real plant 22 23 communities? Again, a return to the historical literature may help: Under the 23 24 chapter dealing with ecesis causes, Clements (1916) noted: 24 25 25 26 Properly speaking, competition exists only when plants are more or less equal. The 26 relation between ... dominant tree and secondary herbs on the forest floor [is not com- 27 petition]. The latter has adapted itself to the conditions made by the trees, and is in 27 28 no sense a competitor of the latter. Indeed, as in many shade plants, it may be bene- 28 29 ficiary. The case is different, however, when the seedlings of the tree find themselves 29 30 alongside the herbs and drawing upon the same supply of water and light. They meet 30 31 upon more or less equal terms, and the process is essentially similar to the competi- 31 tion between seedlings alone, on the one hand, or herbs on the other. The immediate 32 32 outcome will be determined by the nature of their roots and shoots, and not by the 33 dominance of the species. Naturally, it is not at all rare that the seedling tree suc- 33 34 cumbs. When it persists, it gains an increasing advantage each succeeding year, and 34 35 the time comes when competition between tree and herb is replaced by dominance 35 36 and subordination. This is the course in every bare area and in each stage of the sere 36 37 which develops upon it. The distinction between competition and dominance is best 37 seen in the development of a layered forest in a secondary area, such as a burn. All 38 the individuals compete with each other at first in so far as they form intimate groups. 38 39 Within the growth of shrubs, the latter become dominant over the herbs and are in 39 40 turn dominated by the trees. Herbs still compete with herbs, and shrubs with shrubs, 40 214 B.D. Booth and D.W. Larson

1 as well as with younger individuals of the next higher layer. Within the dominant 1 2 tree-layer, individuals compete with individuals and species with species. Each layer 2 3 exemplifies the rule that plants similar in demands compete when in the same area, 3 while those with dissimilar demands show the relation of dominance and subordina- 4 4 tion. (page 72) 5 5 6 This text sets out the idea that community development will have a signifi- 6 7 cant degree of organization by life form, but that organization will be strongly 7 8 scale dependent at the species level. This has been confirmed by Wilson et 8 9 al. (1995b). That life-form itself can be predicted is, to a certain degree, a 9 10 morphological tautology. Küppers (1985) showed that the invasion sequence 10 11 and final structure of European hedgerows was controlled almost completely 11 12 by the architecture of the plants shading the ground. Physiological differences 12 13 among the plants had no role in regulating the final configuration of these 13 14 systems. This text of Clements also implies that complex systems will be 14 15 strongly influenced by the sequence of species introductions, as has been 15 16 shown by several ecologists (Abrams et al., 1985; Robinson & Dickerson, 16 17 1987; Drake, 1991; Drake et al., 1993). McCune and Allen (1985) showed 17 18 that only 10% of canopy tree composition in very productive west-coast conif- 18 19 erous forest could be predicted by site factors, whereas the other 90% were 19 20 attributable to historical (and other unknown) factors. In other words, the 20 21 details of the species composition, including the dominant species, are impos- 21 22 sible to predict because the number, timing, and consequences of competitive 22 23 interactions and other historical events are too great to compute. 23 24 In contrast to the complex and unpredictable species composition that 24 25 applies to productive forest or grassland, Clements notes that for unproductive 25 26 communities: 26 27 The selective action of bare areas upon the germules brought into them is exerted by 27 28 ecesis.... The two extremes, water and rock, are the extremes for ecesis, the one 28 29 impossible for plants whose leaves live in the air and the light, the other for those 29 30 whose roots must reach water. The plants which can ecize in such extremes are nec- 30 31 essarily restricted in number and specialized in character, but they are of the widest 31 distribution, since the habitats which produced them are universal. From the stand- 32 point of ecesis, succession is a process which brings the habitat nearer the optimum 32 33 for germination and growth, and thus permits the invasion of an increasingly larger 33 34 population. The fundamental reason why primary succession is long in comparison 34 35 with secondary, lies in the fact that the physical conditions are for a long time too 35 36 severe for the vast majority of migrants, as well as too severe for the rapid increase 36 of the pioneers. (page 71) 37 37 38 This selection of text suggests that a small array of species that share an abil- 38 39 ity to tolerate environmental extremes will occur in habitats that are ‘bare’. 39 40 Without stating it explicitly, Clements suggested that bare areas are more 40 Impact of language, history and choice of system 215 1 likely to have communities of organisms that share many features in com- 1 2 mon, and that these features permit the slow and relentless increase in bio- 2 3 mass over long periods of time during which invasions of the habitat by other 3 4 organisms are not possible. In bare areas, interactions among taxa, and espe- 4 5 cially between each taxon and the physical environment, ought to be more 5 6 easily predicted. These same interactions that take place in productive habi- 6 7 tats where there are many more species and therefore more interactions, will 7 8 not be as easy to predict. 8 9 The predictability of succession in unproductive areas was also discussed 9 10 by Grime (1979). He wrote: 10 11 11 Where the productivity of the habitat is low, the role of ruderal plants and competi- 12 tors in secondary succession is much contracted, and stress-tolerant herbs, shrubs, and 12 13 trees become relatively important at an earlier stage. The growth form and identity of 13 14 the climax species varies according to the nature and intensity of the stresses occur- 14 15 ring in the habitat.... Under more severe stress, such as that occurring in arctic and 15 16 alpine habitats, ruderal and competitive species may be totally excluded and here both 16 primary and secondary successions may simply involve colonization by lichens, certain 17 17 bryophytes, small herbs, and dwarf shrubs. (page 150) 18 18 19 Grime observed that the successional trajectories for (in his terms) highly 19 20 stressed communities appear to be very flat, or perhaps even non-existent. 20 21 Plant biomass remains low throughout and moves from ruderals to stress- 21 22 tolerant species, bypassing the middle phase of intense competition. This idea 22 23 was also thoroughly explored and supported by Svoboda and Henry (1987), 23 24 who showed that many arctic tundras behave as systems that undergo non- 24 25 directional and non-replacement succession. Communities of plants and animals 25 26 that develop in such sites engage in more or less permanent low-level inter- 26 27 actions with each other. As a result, they show little historical contingency, 27 28 and little ability to exploit the full array of developmental, successional or 28 29 assembly options that more productive systems show. Schulman (1954) and 29 30 Bond (1989) also presented the idea that many conifer woodlands contain 30 31 communities of slow-growing organisms that are poor competitors with more 31 32 aggressive higher plants. These non-competitive communities also have 32 33 enormous capacity for displaying individual plant and community longevity. 33 34 Baker (1992) shows that timescales of nearly 2000 years are needed if one 34 35 wishes to detect pulses of recruitment in Bristlecone Pine forests. This is also 35 36 true for other conifers (Barden, 1988; Larson, 1990) and even some plant 36 37 communities in tropical latitudes (Aplet & Vitousek, 1994). 37 38 38 39 39 40 40 216 B.D. Booth and D.W. Larson 1 1 2 Constraints on community structure of the Niagara Escarpment 2 3 It appears to us that an unproductive, stressed community (one that is ‘bare’) 3 4 might be ideal to examine constraints on community structure. These systems 4 5 are likely to be deterministic, stable, and not prone to control by historical 5 6 events. This led us to our work on the Niagara Escarpment. Cliffs of the Nia- 6 7 gara Escarpment, southern Ontario, Canada, support one of the oldest and 7 8 least disturbed forests in the world (Larson & Kelly, 1991; Kelly et al., 1994) 8 9 (Fig. 7.1(a),(b)). Uneven-aged stands of Thuja occidentalis (eastern white 9 10 cedar) occur on the cliffs with individual tree ages extending to 1890 years. 10 11 Older generations of cedar are represented in the coarse woody debris which 11 12 accumulates and persists for over 3500 years in the talus at the base of the 12 13 cliff face. In addition to the tree component of the community that has been 13 14 precisely dated, community composition is also surprisingly predictable over 14 15 many spatial scales (Larson et al., 1989; Cox & Larson, 1993; Gerrath et al., 15 16 1995) and timescales (Ursic et al., 1997) (Fig. 7.2). This is true for animals 16 17 as well (Matheson, 1995). In fact, for almost all of the plant groups so far 17 18 examined, the species richness, species composition, and plant/animal abund- 18 19 ance on cliff faces are predictable from site to site and over time. This 19 20 constancy of the form of the community is present along the entire Niagara 20 21 Escarpment, even though the cliff face community itself is surrounded by 21 22 lush deciduous forest in the south near Niagara Falls, and near-Boreal forest 22 23 in the north near Sault Ste. Marie, Michigan, USA. At the life-form level, 23 24 and sometimes even at the level of genus, this predictability may apply to 24 25 cliffs globally. The cliffs therefore appear to support an extremely persistent 25 26 and permanent ecosystem. 26 27 The aspect of community composition on cliffs of the Niagara Escarpment 27 28 that is the most surprising is that only a single species of conifer occurs as 28 29 the dominant canopy-forming tree, namely T. occidentalis. Recent work by 29 30 Young (1996) and Walker (1987) shows that such dominance by Thuja (along 30 31 with its predictable array of herbs, lichens and mosses) on limestone cliffs 31 32 has been a feature of this community for perhaps the entire interglacial and 32 33 post-glacial period (ca. 30 000 years). This community of plants still persists 33 34 throughout some southern US states (North Carolina, Virginia, Ohio, Ten- 34 35 nessee, Kentucky, Pennsylvania) when the limestone cliffs face due north, 35 36 but even in settings where cliff aspects deviate from this, the dominant cliff 36 37 tree simply switches to Juniperus virginiana. Most of the other plant species 37 38 are retained. 38 39 The question therefore, becomes, ‘why does such a predictable community 39 40 of plants emerge from a landscape with so many sources of “migrules”?’ If 40 Impact of language, history and choice of system 217 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 (a) 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 (b) 36 36 Fig. 7.1. (a) Aerial and (b) ground view of the Niagara Escarpment, Ontario, Canada. 37 (Courtesy of Uta Matthes-Sears.) 37 38 38 39 39 40 40 218 B.D. Booth and D.W. Larson 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 Fig. 7.2. Illustration of the predictable nature of the cliff face community and the less 21 predictable nature of plateau and talus communities along the Niagara Escarpment. 21 22 The cliff face community supports the same species of higher plants (hp), ferns (f) 22 23 and mosses (m) from North to South and the same microclimate conditions. Con- 23 versely, plateau forests species such as red oak (Quercus rubra), white ash (Fraxinus 24 americana), and sugar maple (Acer saccharum) are common in the South, while aspen 24 25 poplar (Populus tremuloides) and white spruce (Picea glauca) are common in the 25 26 North. The talus slope supports a greater diversity of higher plants in the South and 26 27 more mosses in the North. 27 28 28 29 we consider the approach suggested by Keddy (1992), the question can be 29 30 answered following three steps. First, by assembling a list of potential 30 31 colonists (the species pool). Second, by collecting morphological, physio- 31 32 logical and ecological information about each to form a trait matrix. Finally, 32 33 the resultant community can be found by determining what traits, and there- 33 34 fore what species, will be deleted from the species pool. In our case we 34 35 decided to narrow the search by focusing experimental work on just the tree 35 36 species that anchors the entire system. Thus a new question emerges: namely, 36 37 how does the cliff environment filter all trees, except Thuja from the pool of 37 38 possible colonists? 38 39 39 40 40 Impact of language, history and choice of system 219 1 1 2 Methods 2 3 To answer this question, two cliffs typical of the Niagara Escarpment were 3 4 selected for study (Milton and Dufferin) and transects were set up running 4 5 from the plateau, down the clifff face and into the talus. Samples were 5 6 collected from the three communities. This design allowed comparisons of 6 7 assembly characteristics of the adjacent, yet highly distinct communities. The 7 8 seed rain (seeds arriving at a site) was sampled by placing ‘sticky traps’ (mod- 8 9 eled on Werner, 1975 and Rabinowitz and Rapp, 1980) in the plateau and 9 10 talus, and at meter intervals down the cliff face in five transects at both sites 10 11 over two years. Similarly, soil samples were collected to examine the seed 11 12 bank (seeds being deposited in soil and litter) from the three positions in five 12 13 transects at both sites in the spring and fall over a three-year period. The low 13 14 rate of natural seedling recruitment had already been determined in demo- 14 15 graphic work (Larson & Kelly, 1991). To examine seedling establishment we 15 16 planted newly germinated seeds of seven tree species at a third cliff site. Forty 16 17 randomized locations were used for each species. Individual plants were 17 18 followed over time, and the rate and timing of mortality was compared among 18 19 taxa. After three years, surviving seedlings were harvested and biomass and 19 20 root: shoot ratios calculated. These experiments allowed us to establish a 20 21 species pool for the cliff face and to determine at what stages species were 21 22 removed from the pool. 22 23 23 24 24 25 Results 25 26 A variety of tree seeds arrived on the cliff in the seed rain (Table 7.2) and 26 27 became incorporated into the seed bank (Table 7.3). Species lists of the three 27 28 communities are almost identical. Seed rain and seed bank density was higher 28 29 in the talus than on the plateau or cliff face (Figs. 7.3 and 7.4). In addition, 29 30 although many species were present, seed density of tree species other than 30 31 T. occidentalis and B. papyrifera was very low (Figs. 7.3 and 7.4). There- 31 32 fore, the seed rain and seed bank acted only as partial constraints on com- 32 33 munity composition at these earliest of stages in plant development. Early 33 34 establishment patterns of the various forest trees also differed among species. 34 35 A variety of tree species became established on the cliff face and survived 35 36 into their second year, however, the most interesting results from the seedling 36 37 establishment experiments came when the plants were entering their third 37 38 year (Fig. 7.5). By this time a slow filtering of the non-viable taxa was evident. 38 39 Some seedlings of most species survived until this time; however, while 39 40 mortality of cedars decreased, the mortality of other species continued. Some 40 220 B.D. Booth and D.W. Larson

1 Table 7.2. List of tree species found in the plateau, cliff face and talus seed rain at 1 2 the Milton (M) and Dufferin (D) sites in 1993 and 1994 2 3 3 4 Latin name Common name Plateau Cliff face Talus 4 5 Acer saccharum Sugar Maple M D M D M D 5 6 Betula papyrifera White Birch M D M D M D 6 7 Betula alleghaniensis Yellow Birch M DM DM D7 Ostrya virginiana Ironwood M D M D M D 8 Quercus rubra Red Oak M D M D 8 9 Thuja occidentalis Eastern White Cedar M D M D M D 9 10 10 11 11 12 Table 7.3. List of tree species found in the plateau, cliff face and talus seed bank at 12 13 the Milton (M) and Dufferin (D) sites from 1993 to 1995 13 14 14 15 Latin name Common name Plateau Cliff face Talus 15 16 Acer saccharum Sugar Maple M DMD 16 17 Betula papyrifera White Birch M D M D M D 17 18 Betula alleghaniensis Yellow Birch M D M D M D 18 Fagus grandifolia Beech M D M D 19 Fraxinus americana White Ash M DM DM D 19 20 Ostrya virginiana Ironwood M D M D M D 20 21 Quercus rubra Red Oak M DMD 21 Thuja occidentalis Eastern White Cedar M D M D M D 22 22 23 23 24 24 25 of these patterns were hinted at by the end of the first year, but it was not 25 26 until the third year that the similarity between the field manipulations and the 26 27 natural Niagara Escarpment vegetation became clear. Therefore, a variety of 27 28 tree species can arrive on the cliff face in the seed rain, become incorporated 28 29 into the seed bank and become established as seedlings on the cliff face. How- 29 30 ever, at each of these stages there is a filtering or removal of individuals from 30 31 the community that contributes to final species composition. 31 32 The next step in the investigation was to determine what traits of cedar 32 33 allow it to be successful on cliff faces, and what traits of other species result 33 34 in them being filtered out of the community. Results to date are preliminary. 34 35 At the end of the experiment, in the summer of 1996, all the trees remaining 35 36 alive from the seedling experiments were harvested. Seedling characteristics 36 37 and the biomass of roots and shoots was measured. Most of the surviving 37 38 seedlings were exceptionally small plants. All species except white pine and 38 39 tamarack were less than 3 cm tall and were less than approximately three 39 40 grams in weight after two years growth (Table 7.4). White birch was 1.5 cm 40 Impact of language, history and choice of system 221 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 Fig. 7.3. Seed rain density (number of seeds per square metre) of eastern white cedar 37 (Thuja occidentalis), white birch (Betula papyrifera), yellow birch (Betula alleghan- 38 iensis), sugar maple (Acer saccharum), and ironwood (Ostrya virginiana) on the 38 39 plateau, cliff face and talus at two sites (Milton and Dufferin) in 1993 and 1994. 39 40 40 222 B.D. Booth and D.W. Larson 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 Fig. 7.4. Seed bank density (number of seeds per square metre) of eastern white cedar 37 38 (Thuja occidentalis), white birch (Betula papyrifera), yellow birch (Betula alleghan- 38 iensis), sugar maple (Acer saccharum), and ironwood (Ostrya virginiana) on the 39 plateau, cliff face and talus at two sites (Milton and Dufferin). Seed bank samples 39 40 were collected in the spring and fall from 1993 to 1995. 40 Impact of language, history and choice of system 223 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 Fig. 7.5. Seedling survivorship (lx) of eastern white cedar (Thuja occidentalis), hem- 38 lock (Tsuga canadensis), tamarack (Larix laricina), white pine (Pinus strobus) and 39 white spruce (Picea glauca) on natural cliff faces over a three-year period. Seeds were 39 40 planted in May 1993 on crevices (squares) and ledges (circles). 40 224 B.D. Booth and D.W. Larson

1 Table 7.4. Table of seed weight, second and third year survivorship, and character- 1 2 istics of second year seedlings planted on the cliff face. Species planted were yellow 2 birch (Betula alleghaniensis), white birch (Betula papyrifera), tamarack (Larix 3 laricina), white spruce (Picea glauca), white pine (Pinus strobus), eastern white cedar 3 4 (Thuja occidentalis) and hemlock (Tsuga canadensis) 4 5 5 6 Seed Final Final Longest Final 6 7 mass survivorship height root biomass Root: 7 Species (mg) year 2 year 3 (cm) (cm) (g) shoot 8 8 9 Betula papyrifera 0.3 1% – 1.5 11.0 0.030 1:1.7 9 10 Betula alleghaniensis 1.0 0% – – – – – 10 Thuja occidentalis 1.3 2% 2% 2.8 6.2 0.032 1:2.7 11 Larix laricina 1.4 3% 2% 4.7 5.3 0.047 1:1.8 11 12 Picea glauca 2.0 1% 0.3% 3.0 7.3 0.029 1:1.6 12 13 Tsuga canadensis 2.4 4% 0% – – – – 13 14 Pinus strobus 17.2 0.3% 0% 5.9 9.1 0.147 1:2.0 14 15 15 16 16 17 tall (Table 7.4). In addition, most surviving cedar trees showed much smaller 17 18 root systems compared to shoots than was expected. Root–shoot ratios in 18 19 healthy surviving cedars were about 1:2.7 while values for other species were 19 20 less than 1:2 (Table 7.4). These experimental cedars has similar root: shoots 20 21 ratios to naturally occuring cedars (Matthes-Sears et al., 1995). White pine 21 22 seedlings, which have the largest seeds and seedlings, were completely 22 23 excluded by the end of the experiment. Many of these seedlings appeared 23 24 healthy, but pushed themselves out of their small crevices. In addition, the 24 25 large-seeded species also suffered heavy seed predation. Hemlock and yellow 25 26 birch were also excluded by the end of the experiment. 26 27 27 28 28 29 Significance 29 30 The experiments were set up to explore the species pool on cliffs, and then 30 31 to explore the constraints against the retention of species in the woody plant 31 32 component of the community. The results show that the small array of trees 32 33 that occur on cliffs is not totally restricted at the stage of seed rain or seed 33 34 bank. Starting at the germination stage, however, the relationship between the 34 35 size and potential productivity of the plant comes into play. Large-seeded 35 36 species, those with high growth rates, or those with limited capacity to grow 36 37 large shoots with small roots were selected against. By the end of the exper- 37 38 iment, the woody plant composition was similar to that shown during the late 38 39 stages of primary succession on the limestone cliff faces of abandoned lime- 39 40 stone quarries (Ursic et al., 1997). This suggests that the deletion constraints 40 Impact of language, history and choice of system 225 1 or assembly rules that operate on natural cliff faces are the same as those 1 2 operating on artificial quarry cliff faces. 2 3 Crawley (1987) has shown that cliffs are habitats with almost zero invasi- 3 4 bility. In fact, this appears to be not the case. Instead, it appears that most 4 5 organisms on cliffs have low survival, and that only certain physiological, 5 6 anatomical, morphological, and developmental characteristics permit the 6 7 colonization of cliffs. If conditions on the cliff change, then the characteris- 7 8 tics that permit species to colonize cliffs will change. If productivity increases, 8 9 then cliff species may grow faster; however, other species may then become 9 10 established and outcompete them. 10 11 One might ask, what rules can be derived from these results? We offer the 11 12 following: 12 13 (a) Any component of a local biota has access to cliffs. 13 14 (b) Propagules of the local biota may persist on cliffs in a dormant phase. 14 15 (c) The size of propagule, and the size, growth rate, and competitive ability 15 16 of a seedling are all negatively related to survival on cliffs. 16 17 (d) Any agent that increases the potential growth rate of cliff species increases 17 18 the likelihood of new invasion by other components of the species pool. 18 19 This may result in a faster growth rate of species, but in a less predictable 19 20 community. 20 21 (e) Any agent that decreases the potential growth rate of cliff species 21 22 decreases the likelihood of new invasions from the species pool. A slower 22 23 growing, but more predictable community is the result. 23 24 (f) Cliff communities are permanent and non-survivable for most taxa. 24 25 25 Unlike many of the other ecological settings within which assembly rules 26 26 have been sought, cliffs and other low productivity systems (such as caves 27 27 or arctic tundra) offer a set of advantages that ecologists can exploit. First, 28 28 the processes that occur on cliffs do so very slowly, lessening the likelihood 29 29 that key events in the establishment of the community will be missed. Second, 30 30 soil is nearly absent, so the historical contingency imposed by earlier site 31 31 conditions of that soil do not apply. Third, because the environments start off 32 32 ‘without things’, it is relatively easy to conduct resource addition experiments 33 33 (as already done Matthes-Sears et al., 1995). Thus, if one wished to experi- 34 34 mentally examine how assembly constraints vary with productivity, the 35 35 experiments would be easy to carry out. 36 36 37 37 38 Conclusions 38 39 It seems compelling to consider what low-productivity environments offer to 39 40 ecologists by way of overall conclusions and recommendations for future 40 226 B.D. Booth and D.W. Larson 1 work. If community ecology, at least in the area of assembly rules, has 1 2 suffered from anything, it has been (a) the failure to keep the language simple 2 3 and unpretentious, and (b) the failure to be aware of history. But, these two 3 4 problems probably cannot be resolved even by drawing attention to them. A 4 5 possible solution is to simplify the way assembly theory is tested. By using 5 6 familiar forest, grassland, or wetland systems as test sites, ecologists are 6 7 almost universally examining systems with high and similar amounts of pro- 7 8 ductivity. As Drake (1991) and others have very clearly shown, such systems 8 9 will have enormous amounts of unknown (and perhaps unknowable) histor- 9 10 ical contingency, therefore these systems cannot ever permit a detailed test- 10 11 ing of the validity of the concept. Of all the factors cited by Kodric-Brown 11 12 and Brown (1993) as being important to explain the remarkably predictable 12 13 species composition of fish communities in Australian springs, the lack of 13 14 historical differences in their development was the most important. 14 15 We conclude that by examining systems at (or very near) the limits of sur- 15 16 vival for most organisms, the processes can be viewed more clearly without 16 17 historical contingency. Thus, low productivity ecosystems that start off with 17 18 their ‘backs against the wall’ may actually be the best places to use as test 18 19 systems. Ecosystems that have so far been marginalized in the literature may 19 20 actually offer the opportunity to study the dynamic aspects of community ecol- 20 21 ogy. They are sufficiently simple and predictable that the organizational 21 22 ‘rules’ that apply to them may be easy to find, and it may be possible to test 22 23 them empirically. Clements (1916) was aware of this opportunity 80 years ago. 23 24 24 25 25 26 Acknowledgments 26 27 The authors would like to thank Drs Richard Reader, Tom Nudds and Brian 27 28 Husband for many discussions that led up to this text, and Evan Weiher and 28 29 Paul Keddy for being brave enough to let us say some of these things. The 29 30 research work was supported by the Natural Sciences and Engineering 30 31 Research Council of Canada. 31 32 32 33 33 34 References 34 35 35 Abrams, M.D., Sprugel, D.G., & Dickmann, D.I. (1985). Multiple successional 36 pathways on recently disturbed jack pine sites in Michigan. Forest Ecology 36 37 Management 10: 31–48. 37 38 Aplet, G.H. & Vitousek, P.M. (1994). An age-altitude matrix analysis of Hawaiian 38 rainforest succession. Journal of Ecology 82: 137–148. 39 Baker, W.L. (1992). Structure, disturbance, and change in the bristlecone pine 39 40 forests of Colorado, USA. Arctic and Alpine Research 24: 17–26. 40 Impact of language, history and choice of system 227

1 Barden, L.S. (1988). Drought and survival in a self-perpetuating Pinus pungens 1 2 population: equilibrium or nonequilibrium? American Midland Naturalist 119: 2 253–257. 3 Bond, J.W. (1989). The tortoise and the hare: ecology of angiosperm dominance 3 4 and gymnosperm persistence. Biological Journal of Linnaean Society London 4 5 36: 227–249. 5 6 Clements, F.E. (1916). Plant Succession. Publication Number 242. Washington: 6 The Carnegie Institution. 7 Clements, F.E. (1936). The nature and structure of the climax. Journal of Ecology 7 8 22: 9–68. 8 9 Connell, J.H. & Slatyer, R.O. (1977). Mechanisms of succession in natural 9 communities and their role in community stability and organization. American 10 Naturalist 111: 1119–1144. 10 11 Connor, E.R. & Simberloff, D. (1979). The assembly of species communities: 11 12 chance or competition? Ecology 60: 1132–1140. 12 Cox, J.E. & Larson, D.W. (1993). Spatial heterogeneity of vegetation and 13 environmental factors on talus slopes of the Niagara Escarpment. Canadian 13 14 Journal Botany 71: 323–332. 14 15 Crawley, M.J. (1987). What makes a community invasible? In Colonization, 15 16 Succession and Stability, ed. A.J. Gray, M.J. Crawley, P.J. Edwards, pp. 16 429–453. Oxford: Blackwell Scientiﬁc Publications. 17 Diamond, J.M. (1975). Assembly of species communities. In Ecology and 17 18 Evolution of Communities, ed. M.L. Cody & J.M. Diamond, pp. 342–444. 18 19 Cambridge (USA): Belknap Press, Harvard University Press. 19 Diamond, J & Gilpin, M.E. (1982). Examination of the ‘null’ model of Connor and 20 Simberloff for species co-occurrences on islands. Oecologia 52: 64–74. 20 21 Drake, J.A. (1990a). Communities as assembled structures: do rules govern pattern? 21 22 Trends in Ecology and Evolution 5: 159–163. 22 Drake, J.A. (1990b). The mechanisms of community assembly and succession. 23 Journal of Theoretical Biology 147: 213–233. 23 24 Drake, J.A. (1991). Community-assembly mechanics and the structure of an 24 25 experimental species ensemble. American Naturalist 137: 1–26. 25 26 Drake, J.A., Flum, T.E., Witteman, G.J., Voskuil, T., Hoylman, A.M., Creson, C., 26 Kenny, D.A., Huxel, G.R., Larue, C.S., & Duncan, J.R. (1993). The 27 construction and assembly of an ecological landscape. Journal of Animal 27 28 Ecology 62: 117–130. 28 29 Drury, W.H. & Nisbet, I.C.T. (1973). Succession. The Arnold Arboretum Journal 29 54: 331–368. In Ecological Succession. Benchmark papers in Ecology ⁄5, ed. 30 F.B. Golley, (1977), pp. 287–324. Stroudburg: Dowden, Hutchinson & Ross. 30 31 Fox, B.J. (1987). Species assembly and the evolution of community structure. 31 32 Evolutionary Ecology 1: 210–213. 32 Fox, B.J. & Brown, J.H. (1993). Assembly rules for functional groups of North 33 American desert rodent communities. Oikos 67: 358–370. 33 34 Gerrath, J.F., Gerrath, J.A., & Larson, D.W. (1995). A preliminary account of 34 35 endolithic algae of limestone cliffs of the Niagara Escarpment. Canadian 35 36 Journal of Botany 73: 788–793. 36 Gleason, H.A. (1917). The structure and development of the plant association. 37 Bulletin of the Torrey Botany Club 44: 463–481. 37 38 Gleason, H.A. (1926). The individualistic concept of the plant association. Bulletin 38 39 of the Torrey Botany Club 53: 7–26. 39 Golley, F.B. (1977). Ecological Succession. Benchmark papers in Ecology ⁄5. 40 Stroudburg: Dowden, Hutchinson & Ross. 40 228 B.D. Booth and D.W. Larson

1 Grime, J.P. (1977). Evidence for the existence of three primary strategies in plants 1 2 and its relevance to ecological and evolutionary theory. American Naturalist 2 111: 1169–1194. 3 Grime, J.P. (1979). Plant Strategies and Vegetation Processes. Chichester: John 3 4 Wiley. 4 5 Grover, J.P. (1994). Assembly rules for communities of nutrient-limited plants and 5 6 specialist herbivores. American Naturalist 143: 258–282. 6 Haefner, J.W. (1978). Ecosystem assembly grammares: generative capacity and 7 emprical adequacy. Journal of Theoretical Biology 73: 293–318. 7 8 Haefner, J.W. (1981). Avian community assembly rules: the foliage-gleaning guild. 8 9 Oecologia 50: 131–142. 9 Keddy, P.A. (1989). Competition. London: Chapman & Hall. 10 Keddy, P.A. (1992). Assembly and response rules: two goals for predictive 10 11 community ecology. Journal Vegetation Science 3: 157–164. 11 12 Kelly, P.E., Cook, E.R., & Larson, D.W. (1994). A 1397-year tree-ring chronology 12 of Thuja occidentalis from cliffs of the Niagara Escarpment, southern Ontario, 13 Canada. Canadian Journal of Forestry Research 24: 1049–1057. 13 14 Kodric-Brown, A., & Brown, J.H. (1993). Highly structured ﬁsh communities in 14 15 Australian desert springs. Ecology 74: 1847–1855. 15 16 Küppers, M. (1985). Carbon relations and competition between woody species in a 16 central European hedgerow. Oecologia 66: 343–352. 17 Larson, D.W. (1990). Effects of disturbance on old-growth Thuja occidentalis at 17 18 cliff edges. Canadian Journal of Botany 68: 1147–1155. 18 19 Larson, D.W. & Kelly, P.E. (1991). The extent of old-growth Thuja occidentalis on 19 cliffs of the Niagara Escarpment. Canadian Journal of Botany 69: 1628–1636. 20 Larson, D.W., Spring, S.H., Matthes-Sears, U., & Bartlett, R.M. (1989). 20 21 Organization of the Niagara Escarpment cliff community. Canadian Journal of 21 22 Botany 67: 2731–2742. 22 Law, R. & Morton, R.D. (1993). Alternative permanent states of ecological 23 communities. Ecology 74: 1347–1361. 23 24 Lawton, J.H. (1987). Are there assembly rules for successional communities? In ed. 24 25 A.J. Gray, M.J. Crawley, & P.J. Edwards, Colonization, Succession and 25 26 Stability, pp. 225–244. Oxford: Blackwell Scientiﬁc Publications. 26 MacArthur, R.H. & Wilson, E.O. (1967). The Theory of Island Biogeography. 27 Monographs in population biology. Princeton, NJ: Princeton University Press. 27 28 McCune, B. & Allen, T.F.H. (1985). Will similar forests develop on similar sites. 28 29 Canadian Journal of Botany 63: 367–376. 29 Matheson, J.D. (1995). Organization of forest bird and small mammal communities 30 of the Niagara Escarpment, Canada. MSc. thesis, University of Guelph. 30 31 Matthes-Sears, U., Nash, C.H., & Larson, D.W. (1995). Constrained growth of trees 31 32 in a hostile environment: the role of water and nutrient availability for Thuja 32 occidentalis on cliff faces. International Journal of Plant Science, 156: 33 311–319. 33 34 Odum, E.P. (1969). The strategy of ecosystem development. Science 64: 262–270. 34 35 Peters, R.H. (1988). The relevance of allometric comparisons to growth, 35 36 reproduction and nutrition in primates and man. In Symposium on Comparative 36 Nutrition, ed. K. Blaxter & I. MacDonald, pp. 1–19. London: Libbey. 37 Pianka, E.R. (1983). Evolutionary Ecology. NY: Harper and Row. 37 38 Rabinowitz, D. & Rapp, J.K. (1980). Seed rain in a North American tall grass 38 39 prairie. Journal of Applied Ecology 17: 793–802. 39 Robinson, J.V & Dickerson, Jr., J.E. (1987). Does invasion sequence affect 40 community structure. Ecology 68: 587–595. 40 Impact of language, history and choice of system 229

1 Roughgarden, J. (1989). The structure and assembly of communities. In: 1 2 Perspectives in Ecological Theory. ed. J. Roughgarden, R.M. May, & S.A. 2 Levin, pp. 203–226. NJ: Princeton University Press. 3 Schulman, E.R. (1954). Longevity under adversity in conifers. Science 119: 3 4 396–399. 4 5 Svoboda, J. & Henry, G.H.R. (1987). Succession in marginal arctic envrironments. 5 6 Arctic and Alpine Research 19: 373–384. 6 Ursic, K., Kenkel, N.C., & Larson, D.W. (1997). Revegetation dynamics of cliff 7 faces in abandoned limestone quarries. Journal of Applied Ecology 34: 7 8 289–303. 8 9 van der Valk, A.G. (1981). Succession in wetlands: a Gleasonian approach. 9 Ecology 62: 688–696. 10 Walker, G.L. (1987). Ecology and population biology of Thuja occidentalis L. in its 10 11 southern disjunct range. PhD thesis, University of Tennessee, Knoxville. 11 12 Werner, P.A. (1975). A seed trap for determining pattern of seed deposition in 12 terrestrial plants. Canadian Journal of Botany 53: 810–813. 13 Wilson, J.B. (1991). Does vegetation science exist? Journal of Vegetation Science 13 14 2: 189–190. 14 15 Wilson, J.B. (1994). Who makes the assembly rules? Journal of Vegetation Science 15 16 5: 275–278. 16 Wilson, J.B. & Watkins (1994). Guilds and the assembly rules in lawn 17 communities. Journal of Vegetation Science 5: 590–600. 17 18 Wilson, J.B. & Gitay, H. (1995). Limitations to species coexistence: evidence for 18 19 competition from field experiments using patch model. Journal of Vegetation 19 Science 6: 369–376. 20 Wilson, J.B. & Whittaker, R.J. (1995). Assembly rules demonstrated in a saltmarsh 20 21 community. Journal of Ecology 83: 801–808. 21 22 Wilson, J.B., Agnew, A.D.Q., & Partridge, T.R. (1994). Carr texture in Britain and 22 New Zealand: community convergence compared with a null model. Journal 23 of Vegetation Science 5: 109–116. 23 24 Wilson, J.B., Allen, R.B., & Lee, W.G. (1995a). An assembly rule in the ground 24 25 and herbaceous strata of a New Zealand rain forest. Functional Ecology 9: 25 26 61–64. 26 Wilson, J.B., Sykes, M.T., & Peet, R.K. (1995b). Time and space in the community 27 structure of a species-rich limestone grassland. Journal of Vegetation Science 27 28 6: 729–740. 28 29 Wilson, J.B., Wells, T.C.E, Trueman, I.C., Jones, G., Atkinson, M.D., Crawley, 29 M.J., Dodd, M.E., & Silvertown, J. (1996). Are there assembly rules for plant 30 species adundance? An investigation in relation to soil resources and 30 31 successional trends. Journal of Ecology 84: 527–538. 31 32 Young, J.M. (1996). The cliff ecology and genetic structure of northern white cedar 32 (Thuja occidentalis L.) in its southern disjunt range. MSc thesis, University of 33 Tennessee, Knoxville. 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 Part II: 1 2 2 3 Other perspectives on community assembly 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 8 1 2 2 3 On the nature of the assembly trajectory 3 4 4 5 James A. Drake, Craig R. Zimmermann, Tom Purucker, and 5 6 Carmen Rojo 6 7 7 8 8 9 9 10 10 11 11 12 12 13 Introduction 13 14 Assembly is a basic device of nature where a broad range of entities, spanning 14 15 many levels of scale, interact across time and space to produce oberserved 15 16 pattern. Entities such as individuals, phenotypes, populations, guilds, and 16 17 higher levels of organization like hierarchical structures are all subject to the 17 18 processes of assembly. Because all biological systems are assembled in a 18 19 dynamical sense, any generality in the process would prove valuable to our 19 20 understanding of nature. Arguably, such an understanding is essential to fully 20 21 appreciate the action of mechanisms as they are played out in evolutionary 21 22 and ecological time. 22 23 Community assembly is ultimately driven by the invasion (e.g., speciation, 23 24 immigration) and extinction of species played out against a complex back- 24 25 ground of environmental constraint. While the environment acts as a filter, 25 26 eliminating some species and promoting others, it also provides spatio- 26 27 temporal complexities which serve as a resource upon which ecological 27 28 strategies can be built. Assembly processes and rules which operate within 28 29 one environment may exhibit entirely different outcomes as a function of 29 30 even minor environmental variation. Despite such obvious descriptions of the 30 31 course of nature, the essence of the assembly trajectory remains little more 31 32 than an elusive metaphor. Here, the character of the assembly trajectory is 32 33 evaluated and a general framework offered which serves to interface the oper- 33 34 ation of ecological mechanisms and the mechanics of community assembly 34 35 within the more general realm of complex systems. 35 36 36 37 37 38 An assembly perspective 38 39 In order to begin our exploration into the nature of the assembly trajectory, a 39 40 perspective is first presented from which we base our arguments. It is suggested 40

233 234 J.A. Drake et al. 1 that a variety of phenomena serve to regulate the assembly processes in space 1 2 and time. The outcome of an assembly process is the nature of the action of 2 3 specific mechanisms. This perspective is developed in this chapter by outlining 3 4 several of these phenomena and their respective action during community 4 5 assembly. 5 6 Ecological communities are aggregations of entities (e.g., individuals, 6 7 populations, guilds) variously meshed and integrated across many levels of 7 8 scale. Hence, it is likely that the process of assembly is both manifested and 8 9 regulated at many levels of scale. A direct consequence of this biologically 9 10 driven complexity is that assembly processes, although regulating system 10 11 organization, are often invisible or distorted at many levels of observation. 11 12 For example, if indeed forbidden species combinations (sensu Diamond, 12 13 1975) exist, it is unlikely that an examination focused exclusively on the 13 14 species directly involved can resolve the reasons for the observed patterns. 14 15 Competition may, indeed, be operating, but to demonstrate the direct mech- 15 16 anism of exclusion is simply not enough. It must be known what factors lend 16 17 competition its dynamical character, given the situation at hand. This dynami- 17 18 cal character comes in part from assembly processes operating synergistically 18 19 and anatgonistically at a multitude of scales replete with positive and nega- 19 20 tive feedbacks (Drake et al., 1996). An understanding of the full dynamics 20 21 comes only when the dynamical step, here the inclusion/exclusion of a species, 21 22 is placed in the perspective of the entire system assembly. 22 23 This multiplicity of control is illustrated by presenting a simple scenario. 23 24 Consider generic communities which are subject to colonizing populations of 24 25 different species. The trajectory1 produced by fitting together populations in 25 26 time and space can derive its directionality from a variety of assembly 26 27 processes operating at different levels of scale. This directionality may be 27 28 simply observed as changes in species relative abundance or in the whole- 28 29 sale substitution of extant species. Temporal dynamics could range from ran- 29 30 dom to quasiequilibrium behavior or even exhibit criticality (Bak, 1996) or 30 31 chaos. As invasions occur at points along the trajectory wholsesale changes 31 32 in the direction and nature of the trajectory can occur as the community 32 33 exhibits sensitivity to a new set of initial conditions. Small differences in 33 34 some parameter, say the fecundity of colonizing individuals, have been shown 34 35 to produce substantive changes in some assembly trajectories but not others 35 36 (Drake, 1991; Drake et al., 1993). 36 37 At many points along the trajectory, however, assembly processes result 37 38 38 1At this point we define the trajectory as a sequence of community states. We acknowledge the 39 powerful role played by environmental heterogeneity and offer assembly against that background. 39 40 We modify this definition subsequently. 40 On the nature of the assembly trajectory 235 1 in robust trajectories. Here, variation in individual fecundity may not be an 1 2 issue and, despite large variance in the fecundity of an invader, the system 2 3 responds indifferently (Drake, 1991). In this case, the rules which govern 3 4 assembly, if indeed rules are operating, do not operate at the level of variance 4 5 in invader fecundity. Even a local population may be a trivial level of orga- 5 6 nization if the system is governed by metascale processes and the incorporation 6 7 of the population into a community occurs at that scale. These are clearly 7 8 essential aspects of assembly, but to date these processes have barely been 8 9 explored. It is necessary to ask at what levels of scale is assembly operating 9 10 and what is the nature of the interface between the component dynamics and 10 11 the the context of whole system response. 11 12 12 13 13 14 Assembly as a general mechanic of nature 14 15 Assembly is not a mechanism but a mechanic. It is the process of fitting 15 16 together the dynamically variable pieces which comprise a system, pieces 16 17 operating at disparate levels of spatial and temporal scale. Variation in the 17 18 pieces result from the action of ecological mechanisms. The mechanisms of 18 19 ecology are familiar to all: species interactions like competition or predation 19 20 and successional processes like inhibition or facilitation. Such mechanisms 20 21 and processes form the nuts and bolts of contemporary ecological thought 21 22 and their dynamical nature in ecological and evolutionary time form the pre- 22 23 vailing ecological paradigm. Mechanisms operate against a background of 23 24 dynamic constraint driven by assembly mechanics, however, and consequently, 24 25 often appear idiosyncratic when viewed outside this context (Drake, 1991; 25 26 Drake et al., 1996). 26 27 Assembly mechanics are the regulatory agents and processes which define 27 28 the suite of plausible system stages or transitions through which a system can 28 29 proceed. As such, assembly comprises the historical events and processes 29 30 which have acted to drive the particular system to its present state. Notably, 30 31 assembly mechanics can define how the operation of mechanism becomes 31 32 manifest. Variability in the operation of a mechanism can result from either: 32 33 33 • constraints which vary in time and space, 34 34 • indeterminism in the action of the mechanism which is a direct product 35 35 of some assembly trajectories, or 36 36 • environmental and stochastic events which modify the action of the mech- 37 37 anism. 38 38 39 The outcome of the operation can also serve as a mechanic for future states. 39 40 For example, a case where competition has resulted in some specific pattern 40 236 J.A. Drake et al. 1 of resource utilization can be easily conceived which may, in itself, drive sub- 1 2 sequent assembly steps. An understanding of the operation of any mechanism 2 3 therefore, must also include an understanding of the mechanics which vari- 3 4 ously permit and modify its operation. 4 5 5 6 6 7 The levels of assembly 7 8 At this point, we define an assembly rule as: 8 9 an operator which exists as a function or consequence of some force, dynamical neces- 9 10 sity, or context which provides directionality to a trajectory. The nature of this direc- 10 11 tion includes movement toward a specific state, some subset of all possible states, or 11 12 a dynamical realm of definable character. Operators are the mechanics of community 12 13 assembly. 13 14 The dichotomy between mechanism and operator is an essential distinction. 14 15 An assembly rule exists as an ‘if-then-else’ or ‘yes-no-maybe’ type of gram- 15 16 matical switch, while a mechanism is something like competition manifest in 16 17 terms of the classic R* (Tilman, 1988). In ecological terms, assembly rules 17 18 define reachable and unreachable community states, the community being 18 19 some complete set of species (bacteria on up) exhibiting limited membership 19 20 (see McIntosh, 1995). 20 21 So defined, assembly rules appear to operate on two fundamental levels. 21 22 The first level is categorical – defined by the presence of functioning levels 22 23 of organization (e.g., individuals, populations, guilds) operating at various 23 24 levels of scale. Assembly at this level is defined by the character of the pieces 24 25 and the functional level of organization at which such pieces operate within 25 26 the context of assembly processes. Assembly rules at this level may include 26 27 things like the result of competition thus far, although not competition itself. 27 28 The manner in which the operation of competition cascaded through the 28 29 community will define the suite of plausible subsequent states. 29 30 The second level of assembly is topological – defined as the dynamical 30 31 fabric upon which the system operates. One can visualize this fabric as a 31 32 complex changing landscape or manifold which maps all plausible assembly 32 33 space. We do not view this fabric in static terms, that is there can be no fixed 33 34 template against which ecology functions. A trajectory which becomes unsta- 34 35 ble will warp the realized manifold into a new topology of plausible states. 35 36 As such, this fabric – the set of all possible states and transitions – reflects 36 37 the complex, evolving nature of the assembly process. 37 38 The dynamics and rules of assembly at the topological level are of a decid- 38 39 edly different nature than the rules which govern the categorical level. The 39 40 fabric may be composed of a handful of trajectories that possess a common 40 On the nature of the assembly trajectory 237 1 solution or exhibit similar dynamics. Some regions may initiate self-organizing 1 2 behavior, while others do not. 2 3 The fabric might include an edge between ordered and chaotic regimes – 3 4 the point of maximum information and diversity – at which trajectories 4 5 approach or reach a critical state (Bak, 1996; Hastings et al., 1993; Jorgen- 5 6 son, 1995; Solé & Manrubia, 1995). These are all higher-ordered components 6 7 of trajectory directionality which are not necessarily, and perhaps never, a 7 8 direct product of interspecific interactions. Further, the domain may be 8 9 characterized by behavior that is either: 9 10 10 • deterministic – yielding a specific community configuration, 11 11 • probabilistic – yielding a density function of plausible states, 12 12 • chaotic, or entirely random and unpredictable (Hastings et al., 1993). 13 13 14 Exposing the relationship between rules at the topological and those at the 14 15 categorical level is critical to understanding the structure and organization in 15 16 assembling ecological systems. 16 17 Given this view of the nature of systems and assembly, where and when 17 18 are rules of assembly likely to exist? At the categorical level, assembly rules 18 19 may be manifest in patterns of species co-occurrence. For example, certain 19 20 combinations of species are simply so unstable, due perhaps to competition, 20 21 that it is unlikely such species combinations will be observed in nature 21 22 (Diamond, 1975). Trajectories containing that species set as a solution exhibit 22 23 instability when realized. Here, the system can either evolve a higher-ordered 23 24 structure or collapse to a former configuration. Interestingly, the species which 24 25 initiate system-wide collapse can also function as an assembly mechanic 25 26 because the subsequent community state may very well be a state which 26 27 cannot otherwise be reached (cf. ‘humpty-dumpty effects’, Lewin, 1992). 27 28 Whatever the scenario, assembly rules take the form of the manner in 28 29 which the manifestation of the mechanism directs subsequent assembly 29 30 steps. 30 31 Rules may also be architectural in nature because some food web topolo- 31 32 gies are either observed frequently or are impossible (Higashi & Burns, 1991; 32 33 Martinez, 1992; Sugihara, 1985). Here, rules of assembly enforce specific 33 34 interaction patterns apparently without regard to the species involved. Clearly, 34 35 assembly pervades every aspect of biological structure and organization, but 35 36 the general nature of control within and among systems remains elusive. For 36 37 example, the way in which assembly leads to top-down or bottom-up control 37 38 might offer a solution to controversies over the nature of short-term ecolog- 38 39 ical contol. Imagine the uility of being able to switch a system from top-down 39 40 to bottom-up control. 40 238 J.A. Drake et al.

1 Self-organization and the assembly trajectory 1 2 2 3 Ecological systems are non-equilibrium, nonlinear, spatially extended, dissi- 3 4 pative systems at all but the most trivial levels of scale. Yes, the local forest 4 5 looks much the same from day to day but such apparent constancy belies a 5 6 world of massive change. It must be accepted that there can never be a 6 et al 7 ‘balance of nature’ at least in terms of equilibrium (Drake ., 1994; Grover 7 8 & Lawton, 1994). Indeed, the fact that evolution and extinction are not rare 8 9 events, but commonplace, quashes arguments of ecological stability and con- 9 10 stancy. This realization demands that alternative approaches and explanations 10 11 be turned to questions of organization and structure in ecological communi- 11 12 ties (Brown, 1994, 1995). The characteristics listed above would seem to 12 13 qualify ecological systems as one of the very best candidates for a so-called 13 14 complex systems approach. The proof is in the pudding and while an arsenal 14 15 of complexity-oriented tools is available, compelling demonstrations of the 15 16 utility of such an approach are needed. 16 17 17 18 Ecological consequences of non-linearity 18 19 19 20 Dissipative systems, unlike conservative systems, give rise to time-irreversible 20 21 processes which invite a powerful role for history. In an open energy environ- 21 dissipative structures 22 ment, these processes can result in the formation of – 22 23 structures which are created and maintained at a thermodynamically far-from- 23 24 equilibrium state by the continuous uptake and transformation of energy. 24 25 Entropy, as a byproduct of internal metabolic processes, is exported out of 25 26 the system by the exchange of matter and energy across system boundaries 26 27 (Brooks & Wiley, 1988; Johnson, 1995; Kauffman, 1993; Nicolis & Prigogine, 27 28 1989; Prigogine, 1980). When the emergent structure varies, energy flow 28 29 varies, and when energy input changes so does the emergent structure. Dis- 29 30 sipative structures do not exist without reference to the entities contained in 30 31 the system, so clearly species substitutions matter. 31 32 An essential characteristic of many nonequilibrium, dissipative systems is 32 33 the operation of self-organization: the emergence of integrated structures 33 34 through nonlinear interaction and feedback mechanisms between system com- 34 35 ponents (Goodwin, 1987; Haken, 1988). Such structures are characterized by 35 36 spatial and temporal regularities which could not be predicted with even a 36 37 precise knowledge of initial conditions. Common themes generated by self- 37 38 organization form attractors, which provide the coarse weave of the assembly 38 39 fabric. These attractors define the possible states that this structure can attain 39 2 40 even before mechanisms like competition and predation are played out . 40 2A reassessment of the null model may be in order. On the nature of the assembly trajectory 239 1 Phase transitions between attractors occur when a critical value in one or 1 2 more system parameters make the current system configuration unstable. The 2 3 original attractor dissipates and a new attractor, stable with respect to the fluc- 3 4 tuation, restructures the system under a new set of dynamic constraints. 4 5 Attractors, therefore, are ephemeral entities which form, fluctuate, and dissi- 5 6 pate as the self-organizing system evolves. This does not mean that attrac- 6 7 tors form and dissipate willy-nilly although such a situation is concievable. 7 8 Rather specific attractors represent recurrent themes in ecological assembly. 8 9 Coupled with the dynamical realities of competition, predation, and other 9 10 mechanisms and processes, these attractors represent solutions to the opera- 10 11 tors or rules of assembly. 11 12 Self-organizing dynamics are only partially deterministic. Chance plays an 12 13 essential role at all scales – a phenomenon well known to ecologists. Indeed, 13 14 it has been argued that self-organizing processes have a absolute necessity 14 15 for random fluctuations at some point in their histories (Pattee, 1987; Yates, 15 16 1987). Regardless of its ultimate source, biologists have long contended that 16 17 stochasticity plays a role in biological systems (Chesson, 1986; Chesson & 17 18 Case, 1986; Gilpin & Soulé, 1986; Goodman, 1987; Sale, 1977, 1979). It now 18 19 appears likely that chance plays a fundamental role in the emergence of order 19 20 in all dissipative systems. 20 21 It is asserted that the assembly trajectory must be approached as a self- 21 22 organizing process built from a variety of components (e.g., individuals, pop- 22 23 ulations, guilds) which are themselves historically derived and self-organized. 23 24 It is recognized, however, that not all community assemblages have been 24 25 shaped by self-organization. Communities likely exist on a continuum from what 25 26 are essentially random assemblages up to highly integrated self-organized 26 27 structures. Two factors which may be fundamental to whether a given system 27 28 exhibits self-organization are environmental constancy and component diver- 28 29 sity. 29 30 First, some degree of environmental constancy may be required. Highly 30 31 volatile systems as a result of frequent natural or anthropogenic disturbance 31 32 may not self-organize. Variable disturbance frequency and magnitude can 32 33 strongly affect which community state is attained (Connell, 1978; Paine & 33 34 Levin, 1981). Where the rate of disturbance is low compared to rate of 34 35 community development, disturbance may act to reset the system back over 35 36 one or more attractors. Forest gap dynamics demonstrate such a repeating 36 37 cycle of community states. Cyclical disturbance can also act to maintain a 37 38 trajectory within the basin of a given attractor. Fire-climax communities, for 38 39 example, involve evolutionary adaptions by the constituent community 39 40 members such that the presence of periodic fires has been incorporated as a 40 necessary factor in the persistence of the community state. Disturbance, in 240 J.A. Drake et al. 1 this case, could be constructively considered as the cessation or suppression of 1 2 fires. Finally, very frequent, intense or highly variable disturbances at any point 2 3 during assembly process may prevent a self-organized state from emerging 3 4 at all but the longest timescales, spans of time meaningless to the system. 4 5 Transient species and the ensuing variations in system dynamics may also 5 6 adversely affect self-organization. Communities which experience high rates 6 7 of disrupting invasions may never settle onto a self-organized trajectory. 7 8 Huxel (1995), for example, found that communities subjected to simultane- 8 9 ous multiple invasions never attained a stable, invasion-resistant state. This 9 10 is in stark contrast to the structural patterning which occurs when species 10 11 invade over a period of time allowing community dynamics to be fully or 11 12 largely expressed before subsequent invasion (Drake, 1991, 1993). 12 13 Secondly, some minimal species diversity and abundance may be neces- 13 14 sary for self-organization to occur. If so, species-poor systems or systems in 14 15 the very early stages of development would be expected to exhibit assemblages 15 16 that are less structured in nature. In other cases, the available species pools 16 17 might simply have ‘holes’ wherein species that would otherwise have a key 17 18 role in the assembly process are missing. How efficiently self-organization 18 19 can proceed in lieu of these missing species may depend on the strength and 19 20 resilience of the attractors involved. 20 21 These exceptions notwithstanding, a self-organization approach provides a 21 22 useful framework for understanding the formation of complex ecological sys- 22 23 tems. Ecological succession, for example, may be a self-organizing assembly 23 24 process under the influence of one or more attractors. These attractors not 24 25 only influence the assembly process within each seral stage but also govern 25 26 transitions between stages. Transitions can initiated in two ways. First, tran- 26 27 sitions can be biotically initiated by the addition of new hierarchical levels 27 28 or by key additions, substitutions, or extinctions of species within existing 28 29 levels. Restructuring of the system, at this point, can occur at all scales and 29 30 levels. The effect of adding a predator, for example, on restructuring food 30 31 web topologies is well known. Predators can alter the interactions between 31 32 competing consumers, such that consumer coexistence is either favored or 32 33 prohibited (Caswell, 1978; Connell, 1971; Paine, 1966). The alteration of 33 34 interaction at the consumer level can cascade downward altering interactions 34 35 at lower levels (e.g., producers and abiotic interactions). Circular feedback, 35 36 ultimately, can also affect the predator-level itself. A predator which pro- 36 37 motes multiple species coexistence, for example, provides the opportunity for 37 38 other predators to become established. 38 39 Phase transitions can also be abiotically initiated as autogenic processes of 39 40 the assembling community alter the environment in such a way that fosters 40 On the nature of the assembly trajectory 241 1 the entry of more competitive species. Connell and Slayter (1977) termed this 1 2 process facilitation. Such transitions could be accompanied by relatively dis- 2 3 crete transition phases wherein large species turnover occurs as food-web 3 4 hierarchies crumble and are rebuilt. The system may approach a self-organized 4 5 critical state at the phase transition. At this point the system is either driven 5 6 to a more complex, higher-order structure or it is returned to a simpler state. 6 7 The region around the critical phase transition may not be a favorable place 7 8 to stay. Kaufmann (1993), for example, found that his model systems resided 8 9 in the ordered regime just away from the order-chaos boundary. 9 10 Of course, not all assembly patterns show discrete transitions and varying 10 11 degrees of discontinuity in species membership are often exhibited. Some pat- 11 12 terns show discrete turnover of a core of dominant species only, while others 12 13 display a gradual turnover of all species throughout the process. In the first 13 14 case, a phase transition between two attractors can be imagined that only 14 15 affects the community core in the short term. Replacement of peripheral 15 16 species then gradually occurs as the trajectory moves deeper into the basin 16 17 of the governing attractor. In the second case, the system may be under the 17 18 influence of an attractor with a shallow, but wide basin of attraction. Critical 18 19 parameter changes do not occur and a phase transition is not incurred. Rather, 19 20 such a trajectory might slowly spiral into the attracting basin producing the 20 21 observed gradual species replacement. Finally, the relationship between time 21 22 and space mentioned earlier begs a comparison between the temporal behavior 22 23 discussed here and similar spatial transitions (Milne et al., 1996). Exploration 23 24 of this relationship is left to the so inclined reader. 24 25 25 26 26 27 Evidence for self-organization 27 28 While the process of self-organization presents a theoretically compelling 28 29 framework, the extent and manner in which these processes are responsible 29 30 for the patterns of nature has yet to be fully appreciated. Indeed, the rela- 30 31 tionship between self-organization, natural selection, and the mechanisms and 31 32 assembly operators of ecology are simply unknown despite a growing theo- 32 33 retical effort (e.g., Depew & Weber, 1995; Kauffman, 1993). Demonstrations 33 34 of self-organization at higher-levels of organization generally rely on a 34 35 statistical similarity between the system at hand and the behavior of a model 35 36 (e.g., cellular automata) which exhibits self-organization. While such attempts 36 37 represent bold explorations of the nature of the biological world, they remain 37 38 correlative. Experimenation is needed. 38 39 These cautions offered, a variety of studies have implicated self-organization 39 40 as a fundamental process in the production of pattern in many biological and 40 242 J.A. Drake et al. 1 physical systems. In fact, a database search using the term self-organization 1 2 reveals an absolute explosion of research spanning a wide variety of physi- 2 3 cal, biological, and social systems. Solé and Manrubia (1995), for example, 3 4 observed that the fractal-like patterns of rainforest canopy gap distribution 4 5 were patterns readily recreated with a simple cellular automata model. Their 5 6 model exhibited self-organization and an approach to a critical state3, result- 6 7 ing in patterns which bore a striking similarity to those observed in nature. 7 8 Similarly, Keitt and Marquet (1996) suggest that the patterns of species extinc- 8 9 tion in the Hawaiian avifauna can be explained by nonequilibrium dynamics 9 10 leading to a such a critical state. Plotnick and McKinney (1993) have sug- 10 11 gested that patterns of species extinction observed in the fossil record are also 11 12 consistent with models of species turnover which exhibit self-organization 12 13 and approaches to a self-organized critical state. 13 14 Experimental explorations of the process of self-organization have been 14 15 conducted in a wide variety of physical and biochemical systems where the 15 16 process has taken on an almost matter-of-fact air. Experimental verification 16 17 is clearly a more daunting task at higher-levels of organization where the 17 18 elements of the system (e.g., species, populations, guilds) are variable and pro- 18 19 cess function on very long timescales. Regardless of the logistical difficulties 19 20 of verification, the assembly trajectory exhibits properties and processes which 20 21 are consistent with a self-organized structure. It is suggested that laboratory 21 22 and microcosm analyses are the best candidates for initial demonstrations of 22 23 self-organization in ecological systems. 23 24 If, indeed, self-organization withstands further scrutiny, the implications 24 25 for ecology and biology in general are staggering. How much of the pattern 25 26 and process is a result of self-organization and the operators of the assembly 26 27 trajectory? In what follows we offer some thoughts and ideas, couched in 27 28 terms of a general theory of assembly. 28 29 29 30 30 31 Probabilistic elements of assembly 31 32 With each point of an assembling trajectory, we associate an assembly oper- 32 33 ator in terms of a probability function – perhaps a logic switching function 33 34 – which defines the suite of plausible directions the assembly trajectory can 34 35 take. The assembly function at each point is a unique expression of the relative 35 36 roles of constraint and chance as defined by: 36 37 37 38 3A critical state is an attractor which exhibits correlations at all length and spatial scales (Bak 38 et al., 1988). In terms of species extinctions or turnover the frequency and magnitude of collapse 39 follows a power-law distribution of the form D(s) ≈ s−α, where s is the size of collapse, D(s) is 39 40 the frequency of occurrence of collapse size s, and α is the spectral exponent. 40 On the nature of the assembly trajectory 243 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 Fig. 8.1. Graphical representation of three assembly functions, which display behavior 27 28 ranging from fully stochastic to fully deterministic. 28 29 29 30 30 • of available species, their ecologies, and relative abundance; 31 31 • the basin portrait as defined by the system’s parameter space; 32 32 • the historical inertia of the system; and 33 33 • vagaries of the environment. 34 34 35 Three assembly functions are offered ranging from fully random to fully 35 36 deterministic. The first state (Fig. 8.1(a)) demonstrates fully random behavior. 36 37 Here the state is not under the influence of any attractor and all plausible 37 38 future states are equally probable. A successive string of such states would, 38 39 thus, constitute a random walk. The second state (Fig. 8.1(b)) may be under 39 40 the constraining influence of one or more attractors. While behavior, at this 40 244 J.A. Drake et al. 1 point, can become somewhat more predictable, varying degrees of sensitiv- 1 2 ity to initial conditions can still maintain a highly probable nature toward 2 3 future states. Chaotic attractors, for example, demonstrate high sensitivity to 3 4 initial conditions. Here, two arbitrarily close points can diverge at an 4 5 exponential rate. The last state (Fig. 8.1(c)) shows a totally constrained, fully 5 6 deterministic assembly function. This trajectory is locked deep within the 6 7 basin of an attractor. Such an attractor could be considered a climax state as 7 8 long as system parameter changes are insufficient to perturb the system out 8 9 of its basin. Clearly, system behavior runs the gamut from that which is 9 10 completely random to that which is fully deterministic. However, there is 10 11 every reason to believe that most communities exhibit a composite of these 11 12 behaviors. We would expect some degree of constraint and stochasticity to 12 13 play a role at all steps in the assembly process. 13 14 Beginning at some initial point in state space, that initial state can be 14 15 coupled with each of the plausible subsequent states as defined by the assem- 15 16 bly probability function at that point. Repeating this process at progressive 16 17 time steps will produce a Markovian-like assembly tree of branching alterna- 17 18 tive assembly paths. For a given elapsed time interval, a path probability can 18 19 be determined for each alternative trajectory by simply taking the product of 19 20 the function-defined probabilities at each time step within that interval. The 20 21 probability that an assembling community will reach some defined point at 21 22 the end of some time interval is the sum of the path probabilities for all 22 23 trajectories which reach that point. Given some degree of stochasticity at each 23 24 step, it is easy to imagine that, for any given assembling community, there 24 25 could exist a very large number of alternative assembly paths. While some 25 26 could be markedly different in nature, many others would simply be random 26 27 variations on a theme. Path probabilities, consequently, are likely to be quite 27 28 small and prediction of any given trajectory would be largely impossible. 28 29 For any given elapsed time, we can determine the instantaneous probabil- 29 30 ity density of the state space for the assembling community based on the sum 30 31 of all alternative path probabilities for each point in state space. If early stages 31 32 of assembly are largely random in nature, the probability density at any point 32 33 will be low as alternative trajectories are spread out over relatively large 33 34 regions of state space. If the assembly process is self-organizing, however, 34 35 these trajectories will begin to coalesce into smaller regions as they fall under 35 36 the influence of various basins of attraction. Therefore, increasing the amount 36 37 elapsed time would produce a probability density displaying regions of higher 37 38 and higher probability. Consequently, predictions could be made as to where 38 39 in state space an assembling community is likely to be found at any given 39 40 time. Unfortunately, the best hope is for rather fuzzy predictions. 40 On the nature of the assembly trajectory 245 1 1 2 Elements of a general theory of assembly 2 3 A conceptual view of the nature of the assembly trajectory has been offered 3 4 by evaluating constraint and context at levels of scale where controlling fac- 4 5 tors might emerge. Themes common across systems point to elements of a 5 6 general theory of assembly. Such a theory must accommodate the existence 6 7 and operation of a wide range of dynamics from pure determinism to sheer 7 8 chance as well as higher-order processes such as self-organization and 8 9 approaches to critical states. Clearly, the specifics of each ecological system, 9 10 from the biology of constituent populations to ensembles of particular species 10 11 which resist invasion, are system dependent and change with time. Compe- 11 12 tition here is not competition there. Top-down control works here but not 12 13 there. Diversity is thusly maintained here but not there. Vagaries aside, how- 13 14 ever, biological systems display commonalties which point to the essential 14 15 elements of structure. Here, a set of informal propositions, corollaries and 15 16 comments are offered which we believed will have broad application. 16 17 Proposition 1: All biological systems are assembled, historically contingent structures. 17 18 Corollary 1.1: The structure and organization of the present state cannot be 18 19 fully understood without reference to the past. 19 20 Corollary 1.2: The configuration of subsequent community states is strongly 20 21 affected by the nature and character of the presently observed state. 21 Corollary 1.3: Myriad factors such as disturbance are capable of erasing or 22 rewriting history. 22 23 23 Proposition 2: Biological systems are generally non-linear, dissipative, spatially 24 24 extended, dynamic systems which are far from equilibrium over most of the plausible 25 parameter and state space. 25 26 Corollary 2.1: Dissipative systems are time irreversible. Therefore, an under- 26 27 standing of a dissipative system requires an understanding of its evolution 27 28 and history. 28 29 Corollary 2.2: Non-linear systems can exhibit sensitivity to initial conditions. 29 Such systems can quickly amplify small differences in system parameter 30 values such that assembly trajectories diverge at an exponential rate. This 30 31 insures that anything less than a perfect measurement of some state can lead 31 32 to a rapid deterioration in predictive ability (See Corollary 2.5). 32 33 Corollary 2.3: Feedbacks are highly significant. 33 34 Corollary 2.4: Dynamics can range from pure chance to stable equilibrium to 34 deterministic chaos. 35 35 Corollary 2.5: Predictability decays exponentially where sensitive dependence 36 on initial conditions exists. Thus, at some levels of scale prediction beyond 36 37 the short term is difficult, if not impossible. 37 38 Proposition 3: Directionality in the assembly trajectory is governed by critical depen- 38 39 dence on the interplay between events, processes and structures which emerge in space 39 40 and time. 40 246 J.A. Drake et al.

1 Corollary 3.1: Ecological systems (e.g., food webs) are dynamic across scales. 1 2 Short term studies necessarily capture but a glimpse of reality. 2 3 Corollary 3.2: When the system is spatially extended (e.g., metapopulation 3 processes) the trajectory is also spatially extended. Visualize a complex land- 4 4 scape which itself operates as a point on a trajectory within another landscape. 5 Corollary 3.3: Disturbance frequency and magnitude interplay with assembly 5 6 in a complex fashion. 6 7 Corollary 4.2: Stochasticity plays a role at many points along the trajectory. 7 8 The role played by stochasticity on future states reflects the nature of the 8 9 assembly fabric at that point. At points, stochastic fluctuations are immate- 9 rial while, at other points, they dominate the system 10 Corollary 3.4: Some assembly scenarios can uniquely expose dynamical realms 10 11 which are not possible in other scenarios even given the same species and 11 12 environments. 12 13 Comment: The extent to which such a system or a variable of that system ex- 13 14 hibits sensitive dependence is a function of properties which emerge during 14 assembly. Specific properties of the system can either initiate/amplify or 15 15 eliminate/dampen sensitive dependence. Such control develops as a function 16 of system context. For example, during the assembly of an experimental 16 17 community we have observed trajectories where variation in the fecundity 17 18 of an invader led to large community impacts, while such variation in other 18 19 trajectories was largely immaterial (Drake, 1991; Drake et al., 1993). Such 19 20 dynamics are suggestive of the operation of an assembly operator. 20 21 Proposition 4: The action of mechanism is under the control of the assembly opera- 21 22 tor. This operator is context sensitive which can lead to variation in the manifesta- 22 tion of the mechanism. 23 23 24 Proposition 5: Self-organization provides selectable structures. Assuming that self- 24 organizing is occurring as is competition among a set of species along some assembly 25 25 trajectory, what relationship is there between self-organization (the nature, direction 26 and product of the SO trajectory) and the occurrence of competition? 26 27 27 Proposition 6: No single type, nor cadre of types, of dynamical behavior is (are) capa- 28 ble of characterizing ecological reality. Rather, nature is a complex blend of dynamics 28 29 spanning and combining a wide range of elements each operating with different 29 30 constraints at a variety of spatio-temporal scales. 30 31 Corollary 6.1: Structure can be a product of past deterministic events. At other 31 32 times, structure is simply a product of chance. Unfortunately, the structures 32 which result from both extremes can be indistinguishable. 33 Corollary 6.2: Just as the nature of control varies widely across space even 33 34 within the confines of a given ecological system (e.g. a lake or grassland), 34 35 control also varies with time. 35 36 36 37 37 38 Prospectus 38 39 One can either approach ecological systems for what they are or take solace 39 40 in the notion that pieces equal the whole. Contemporary views are largely 40 On the nature of the assembly trajectory 247 1 based on information obtained from highly reductionistic analyses. As with 1 2 most complex systems, the general analytical approach has been one of a 2 3 reduction of components, properties and processes to a logistically manage- 3 4 able level. Unfortunately, nature proceeds without regard to the logistic 4 5 limitations of the observer. Faster hardware and increasingly sophisticated 5 6 measurement devices generally serve to temporarily rescue the reductionistic 6 7 approach until the limits of that level of resolution have again been reached. 7 8 A far more serious problem faced by the reductionistic approach, however, 8 9 is that the focus of reduction – the system’s pieces and mechanisms – fre- 9 10 quently do not themselves possess many of the critical ‘system-level’ prop- 10 11 erties which influence structure and process. Understanding mechanism is 11 12 undeniably an essential ingredient in understanding system properties, but an 12 13 understanding of mechanism alone is not sufficient to explain the dynamics 13 14 of ecological systems. Biological systems are more than simply an additive 14 15 function of the system’s components and even a perfect knowledge of mech- 15 16 anism cannot illuminate the nature of higher ordered phenomena. Such prop- 16 17 erties are emergent and their detection hinges on analytical scales capable of 17 18 exposing those properties. Emergent properties form the essence of the system 18 19 and exert a strong controlling effect on both the nature of the system and its 19 20 component parts. 20 21 We offer that the manner in which assembly mechanics interface with, and 21 22 modify, the expression of a mechanism is perhaps even more essential to 22 23 advancing our understanding of nature than is the further refinement of pure 23 24 mechanism. This is not a call to abandon the mechanistic approach, but rather 24 25 to enhance that effort by conducting experiments within a developing theo- 25 26 retical context of both assembly and the general nature of complex systems. 26 27 Assembly rules are the interface between the topological level of assembly 27 28 space and the categorical level of functioning components. Assembly rules 28 29 are the quantifiable expression of the assembly mechanic and their detection 29 30 can only come by first understanding the interface between these levels – the 30 31 interface is where ecology occurs. 31 32 Several aspects of assembly behavior which we have focused on suggest 32 33 that there are higher-ordered processes, in excess of basic ecological mech- 33 34 anism, which structure community states. At this point, however, it is difficult 34 35 to specify precisely what the states are which comprise the assembly trajectory. 35 36 There is undoubtedly a hierarchy of significant and insignificant levels of 36 37 organization which are expressed along the trajectory. Yet, what levels of 37 38 organization play the most dominant role in defining structure changes along 38 39 the course of the trajectory? Are there recognizable phase transitions during 39 40 the course of assembly that provide information about the course of system 40 248 J.A. Drake et al. 1 development or, perhaps, even allow prediction of the final state? While it 1 2 has been argued that ecological systems are indeed complex non-linear sys- 2 3 tems, systems inherently difficult to predict, tractable questions do exist if the 3 4 system is approached with an adequate epistemological framework. While 4 5 some tentative elements of such a framework have been offered here, the 5 6 answers are not all here. Ecology is far too complex and hard-fought ground 6 7 remains. Further progress will hinge on the development of a general and 7 8 robust theory of complex systems, self-organization, and the creation of 8 9 emergent order. 9 10 10 11 11 12 References 12 13 Bak, P. (1996). How Nature Works: The Science of Self-Organized Criticality. 13 14 Copernicus. 14 15 Brooks, D.R. & Wiley, E.O. (1988). Evolution as Entropy: Toward a Unified 15 16 Theory of Biology. Chicago: University of Chicago Press. 16 Brown, J.H. (1994). Complex ecological systems. In Complexity: Metaphors, 17 Models, and Reality, ed. G. Cowan, D. Pines, & D. Meltzer, SFI Studies in the 17 18 Science of Complexity, Proc. Vol. XIX, Addison-Wesley. 18 19 Brown, J.H. (1995). Macroecology. Chicago, IL: University of Chicago Press. 19 Cambel, A.B. (1993). Applied Chaos Theory: A Paradigm for Complexity. Boston, 20 MA: Academic Press. 20 21 Caswell, H. (1978). Predator-mediated coexistence: a non-equilibrium mode. 21 22 American Naturalist, 112: 127–154. 22 Chesson, P.L. (1986). Environmental variation and the coexistence of species. In: 23 Community Ecology, ed. J. Diamond & T.J. Case. NY: Harper & Row. 23 24 Chesson P.L. & Case, T.J. (1986). Overview: non-equilibrium community theories, 24 25 chance, variability, history, and coexistence. In Community Ecology, ed. J. 25 26 Diamond & T.J. Case, NY: Harper & Row. 26 Connell, J.H. (1971). On the role of natural enemies in preventing competitive 27 exclusion in some marine animals and in rain forest trees. In Dynamics of 27 28 Populations, ed. P.J. den Boer & G. Gradwell. Wageningen, The Netherlands: 28 29 Center for Agricultural Publishing and Documentation. 29 Connell, J.H. (1978). Diversity in tropical rainforests and coral reefs. Science, 199: 30 1302–1310. 30 31 Connell, J.H. & Slayter, R.O. (1977). Mechanisms of succession in natural 31 32 communities and their role in community stability and organization. American 32 Naturalist, 111: 1119–1144. 33 DeAngelis, D.L., Post, M.W., & Travis, C.C. (1986). Positive Feedback in Natural 33 34 Systems. New York: Springer-Verlag. 34 35 Depew, D.J. & Weber, B.H. (1995). Darwinism Evolving: Systems Dynamics and 35 36 the Genealogy of Natural Selection. Cambridge, MA: MIT Press. 36 Diamond, J.M. (1975). Assembly of species communities. In Ecology and 37 Evolution of Communities, ed. M.L. Cody & J.M. Diamond, Cambridge, MA: 37 38 Harvard University Press. 38 39 Drake, J.A. (1991). Community assembly mechanics and the structure of an 39 experimental species ensemble. American Naturalist 137: 1–26. 40 Drake, J.A., Witteman, G.J., & Huxel, G.R. (1992). Development of biological 40 On the nature of the assembly trajectory 249

1 structure: critical states, and approaches to alternative levels of organization. In 1 2 Biomedical Modeling and Simulation, ed. J. Eisenfeld, D.S. Levine, & M. 2 Witten, pp. 457–463. North Holland: Elsevier. 3 Drake, J.A., Flum, T.E., Witteman, G.J., Voskull, T., Hoylman, A.M., Creson, C., 3 4 Kenney, D.A., Huxel, G.R., LaRue, C.S., & Duncan, J.R. (1993). The 4 5 construction and assembly of an ecological landscape. Journal of Animal 5 6 Ecology 62: 117–130. 6 Drake, J.A., Flum, T.E., & Huxel, G.R. (1994). On defining assembly space: a 7 reply to Grover and Lawton. Journal of Animal Ecology 63: 488–489. 7 8 Drake, J.A., Hewitt, C.L., Huxel, G.R., & Kolasa, J. (1996). Diversity and higher 8 9 levels of organization. In Biodiversity: A Biology of Numbers and Differences, 9 ed. K. Gaston, Oxford: Blackwell Scientific. In Press. 10 Gilpin, M.E. & Soulé, M.E. (1986). Minimum viable populations: the processes of 10 11 species extinctions. In Conservation Biology: The Science of Scarcity and 11 12 Diversity, ed. M.E. Soulé, pp. 13–34. Sunderland, MA: Sinauer Associates. 12 Goodman, D. (1987). The demography of chance extinction. In Viable Populations 13 for Conservation, ed. M.E. Soulé, pp. 11–34. Cambridge: Cambridge 13 14 University Press. 14 15 Goodwin, B.C. (1987). Developing organisms as self-organizing fields. In Self- 15 16 organizing Systems: The Emergence of Order, ed. F.E. Yates, NY: Plenum 16 Press. 17 Grover, J. & Lawton, J.H. (1994). Experimental studies on community convergence 17 18 and alternative states: comments on a paper by Drake et al. Journal of Animal 18 19 Ecology 63: 484–487. 19 Haken, H. (1988). Information and Self-organization. Berlin: Springer-Verlag. 20 Hastings, A., Hom, C.L., Ellner, S., Turchin P., & Godfray, H.C.J. (1993). Chaos 20 21 in ecology: is mother nature a strange attractor? Annual Review of Ecology 21 22 Systematics 24: 1–33. 22 Higashi & Burns (1991). Theoretical Studies of Ecosystems: The Network 23 Perspective. Cambridge: Cambridge University Press. 23 24 Huxel, G.R. (1995). Influences of Community Assembly, Unpublished PhD 24 25 Dissertation, University of Tennessee, Knoxville, 128 pp. 25 26 Johnson, L. (1995). The far-from-equilibrium ecological hinterlands. In Complex 26 Ecology, ed. B.C. Patten & S.E. Jorgensen, NJ: Prentice Hall, Englewood 27 Cliffs. 27 28 Jorgensen, S.E. (1995). The growth rate of zooplankton at the edge of chaos: 28 29 Ecological models. Journal of Theoretical Biology, 175: 13–21. 29 Kauffman, S.A. (1993). The Origins of Order: Self-Organization and Selection in 30 Evolution. NJ: Oxford University Press. 30 31 Keitt, T.H. & Marquet, P.A. (1996). The introduced Hawaiian avifauna 31 32 reconsidered: evidence for self-organized criticality. Journal of Theoretical 32 Biology 182: 161–167. 33 Lewin, R. (1992). Complexity: Life at the Edge of Chaos. Maxwell MacMillian. 33 34 Martinez, N.D. (1992). Constant connectance in community food webs. American 34 35 Naturalist 139: 1208–1218. 35 36 McIntosh, R.P. (1995). H.A. Gleason’s ‘Individualistic Concept’ and theory of 36 animal communities: a continuing controversy. Biology Review 70: 317–357. 37 Milne, B.T., Johnson, A.R., Keitt, T.H., Hatfield, C.A., David, J., & Hraber, P.T. 37 38 (1996). Detection of critical densities associated with piñon-juniper woodland 38 39 ecotones. Ecology 77: 805–821. 39 Nicolis, G. & Prigogine, I. (1989). Exploring Complexity: An Introduction. NJ: 40 W.H. Freeman and Company. 40 250 J.A. Drake et al.

1 Paine, R.T. (1966). Food web complexity and species diversity. American 1 2 Naturalist 100: 65–75. 2 Paine, R.T. & Levin, S.A. (1981). Intertidal landscapes: disturbance and the 3 dynamics of pattern. Ecological Monographs, 51: 145–178. 3 4 Pattee, H.H. (1987). Instabilities and information in biological self-organization. In 4 5 Self-organizing Systems: The Emergence of Order, ed. F.E. Yates, NY: 5 6 Plenum Press. 6 Plotnick, R.E. & McKinney, M.L. (1993). Ecosystem organization and extinction 7 dynamics. Palaios 8: 202–212. 7 8 Prigogine, I. (1980). From Being to Becoming: Time and Complexity in the 8 9 Physical Sciences. San Francisco: W.H. Freeman and Co. 9 Sale, P.F. (1979). Recruitment, loss and coexistence in a guild of territorial coral 10 reef fishes. Oecologia 42: 159–177. 10 11 Sale, P.S. (1977). Maintenance of high diversity in coral reef fish communities. 11 12 American Naturalist 111: 337–359. 12 Solé, R.V. & Manrubia, S.C. (1995). Are rainforests self-organized in a critical 13 state? Journal of Theoretical Biology 173: 31–40. 13 14 Sugihara, G. (1985). Graph theory, homology, and food webs. Proceedings of the 14 15 Symposium Applied Mathematics 30: 83–101. 15 16 Tilman, D. (1988). Plant Strategies and the Dynamics and Structure of Plant 16 Communities. NJ: Princeton University Press. 17 Yates, F.E. (1987). General Introduction. In Self-organizing Systems: The 17 18 Emergence of Ordered, ed. F.E. Yates, NT: Plenum Press. 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 9 1 2 2 3 Assembly rules as general constraints on 3 4 community composition 4 5 5 6 6 7 Evan Weiher and Paul Keddy 7 8 8 9 9 10 10 11 11 12 12 13 13 14 Introduction 14 15 It has been more than 20 years since Jared Diamond focused attention on 15 16 assembly rules for communities (Diamond, 1975), and his three principal 16 17 questions about community assembly are still topical and largely unsolved. 17 18 The questions are: (a) To what extent are the component species of a 18 19 community mutually selected from a larger species pool so as to ‘fit’ with 19 20 each other? (b) Does the resulting community resist invasion, and if so, how? 20 21 (c) To what extent is the final species composition of a community uniquely 21 22 specified by the properties of the physical environment, and to what extent 22 23 does it depend on chance events? These questions are really about how com- 23 24 munities, or assemblages, are selected as subsets of a species pool (Fig. 9.1). 24 25 For any region and taxa, one could define a species pool of potential mem- 25 26 bers of an assemblage of species. At the largest scale, this would simply be 26 27 an exhaustive list of all species (i.e., the flora and fauna), while at smaller 27 28 scales the pool might be a list of birds, beetles, fish, or plant species that 28 29 could inhabit a given site or habitat. Communities would be composed of 29 30 some subset of these smaller lists. Assembly rules are explicitly defined con- 30 31 straints on community structure – in other words, assembly rules set limits 31 32 on which species can be a part of locally coexisting subsets of the species 32 33 pool (Fig. 9.1). Of all the possible assemblages, some will conform to the 33 34 rules and will therefore have a greater likelihood of existence. Assemblages 34 35 that have a large deviation from the assembly rules may exist for short periods 35 36 of time, but will likely be replaced by assemblages that more closely conform 36 37 to the contraints. 37 38 There has been a fair amount of discussion about the existence of assem- 38 39 bly rules. In fact, some of it continues in this volume. These discussions may 39 40 arise out of basic confusion about the term: are assembly rules a wooly 40

251 252 E. Weiher and P.A Keddy 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 Fig. 9.1. A conceptual model of community assembly from a large species pool. 19 Assemblages that do not conform to the assembly rules are unlikely to persist and can 19 20 be thought of as having an unstable high energy state. Assemblages that conform to 20 21 the rules have found low energy, local solutions to the constraints set by the assembly 21 rules. 22 22 23 23 24 concept (sensu Peters, 1980) or are they a rigorous predictive tool? It has 24 25 been suggested that it is time to (a) accept that in some places and for some 25 26 taxa assembly rules do exist, (b) start to define assembly rules and (c) begin 26 27 testing them for their utility for making predictions (Weiher & Keddy, 1995). 27 28 Sometimes the very term ‘assembly rules’ causes confusion. Some argue that 28 29 assembly rules are only about patterns caused by biotic interactions (e.g., Wil- 29 30 son, this volume), but many different kinds of factors (both biotic and abi- 30 31 otic) can lead to similar patterns (Connor & Simberloff, 1979; Schluter, 1984; 31 32 Huston, 1997). Some might say that the term was coined by Diamond to mean 32 33 patterns caused by interactions; however, Diamond currently does not insist 33 34 upon such a strict definition of assembly rules (Diamond, pers. com.). Assem- 34 35 bly rules are simply about constraints on composition, regardless of mecha- 35 36 nism. The other main confusion comes from the notion that assembly rules 36 37 are akin to a mechanistic recipe that describes the necessary steps for build- 37 38 ing a community. Assembly rules are not recipes for building communities. 38 39 Rather, they are a set of limits that constrain how species can come together 39 40 to form assemblages. As limits, or constraints, assembly rules are analogous 40 Assembly rules as general constraints 253 1 to the rules of chemical stoichiometry, which limit the ways atoms can come 1 2 together to form molecules. And, like the rules of basic chemistry, assembly 2 3 rules are based on functional, characteristic traits, and not on the names given 3 4 to the entities (be they elements or species). 4 5 In this chapter (a) several strategies for finding empirical assembly rules are 5 6 outlined, (b) some initial attempts at using assembly rules for prediction are 6 7 described, and (c) an explanation of how the results suggest that ecological 7 8 communities can be thought of as complex adaptive systems, with properties 8 9 common to both physical energy dissipating systems and species evolving in 9 10 an adaptive landscape is given. 10 11 11 12 12 13 13 Strategies for finding assembly rules 14 14 15 Null models and consistent patterns of assembly 15 16 Diamond’s interest in patterns of community assembly prompted Dan Sim- 16 17 berloff and Ed Connor to investigate the idea using testable null hypotheses 17 18 about composition. Simberloff and Connor were interested in finding out 18 19 whether the apparent patterns that Diamond found were more than chance 19 20 events. Their null model approach (Simberloff & Connor, 1981) allowed them 20 21 to compare actual communities to simulated communities that were con- 21 22 structed in the absence of species associations, if species were independent 22 23 of one another. In some respects, they were constructing random communities. 23 24 Simberloff and Connor’s idea of the community null model both revolu- 24 25 tionized community ecology and caused a rather bitter debate over methods 25 26 and interpretations (e.g., Strong et al., 1984). While the debate lingers (even 26 27 within this volume), a consensus is emerging. Ecologists are finding ever 27 28 more evidence supporting assembly rules (Weiher & Keddy, 1995; Gotelli & 28 29 Graves, 1996; Wilson, this volume). It has been argued (Weiher & Keddy, 29 30 1995) that it is time to move on from the simple question of whether assem- 30 31 bly rules exist and begin asking what are the assembly rules, how do they 31 32 vary along gradients and among taxa? 32 33 The use of null models has led to several important ideas about the struc- 33 34 ture of communities. Some of the most important advances have come when 34 35 the focus has been on traits, rather than on species names. The approach was 35 36 developed by animal ecologists who were looking for patterns of morpho- 36 37 logical overdispersion that are consistent with limiting similarity (Ricklefs & 37 38 Travis, 1980; Moulton & Pimm, 1987; see also Lockwood et al., this vol- 38 39 ume). Some animal assemblages are overdispersed (e.g., Travis & Ricklefs, 39 40 1983; Brown & Bowers, 1984; Dayan & Simberloff, 1994), which means 40 254 E. Weiher and P.A Keddy 1 they are assembled according a set of trait-based assembly rules that call for 1 2 some degree of limiting similarity. In a review of animal assemblages, Miles 2 3 and Ricklefs (1994) found a positive relationship between species richness 3 4 and the morphological volume occupied by the assemblage, while the den- 4 5 sity of species packing (nearest morphological neighbor) remains constant. 5 6 This means that as animal species are added to assemblages, they tend to 6 7 enter at the morphological periphery. Again, this corresponds to a view that 7 8 animal assemblages are structured according to assembly rules that call for 8 9 limiting similarity. 9 10 The prevailing view among plant ecologists has been that environmental 10 11 factors act like sieves or filters, which remove species from assemblages if 11 12 they lack certain requisite traits (e.g., Grubb, 1977; Grime, 1979; Box, 1981; 12 13 Southwood, 1988; Keddy, 1992). This view leads to the idea that coexisting 13 14 species might be morphologically underdispersed, which is a possibility that 14 15 has received little attention in the literature on assembly rules (Weiher & 15 16 Keddy, 1995). Trait dispersion has not been widely studied in plant assem- 16 17 blages. There are, however, a few cases of apparent trait overdispersion. Floral 17 18 traits have been shown to be overdispersed among congeners (e.g., Arm- 18 19 bruster, 1986; Armbruster et al., 1994). There is some tendency for dominant 19 20 oaks to be from different subgenera (Mohler, 1990), which is also suggestive 20 21 of limiting similarity. Bastow Wilson has found evidence for guild propor- 21 22 tionality, but the patterns are not strong (e.g., Wilson & Roxburgh, 1984; 22 23 Wilson, this volume). 23 24 Investigations of morphological dispersion in temperate riverine (palustrine) 24 25 wetlands have shown that assemblages can be simultaneously overdispersed in 25 26 some traits (mostly those associated with size) and underdispersed in others 26 27 (plasticity and form) (Fig. 9.2; Weiher et al., 1998). This means that wetland 27 28 plant assemblages conform to both the template-filter model and to the 28 29 limiting similarity model. In the first case, traits are clumped because species 29 30 had to pass some sort of environmental filter which likely acts as an exter- 30 31 nal pressure, limiting the morphological volume occupied by an assemblage. 31 32 In the latter case, the traits are overdispersed due to some sort of internal 32 33 pressure (like competition), which keeps morphological nearest-neighbor 33 34 distances large. The results strongly agree with the notion of capacity rules, 34 35 which limit the presence of species according to their ability to tolerate abiotic 35 36 conditions, and allocation rules, which limit the presence of species accord- 36 37 ing to their ability to get along with their neighbors (Brown, 1987). 37 38 The degree to which wetland plant assemblages are under or overdispersed 38 39 depends on resources. In order to assess overall morphological dispersion, 39 40 the first four principal components of 11 measured traits were used. The prin- 40 Assembly rules as general constraints 255 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 Fig. 9.2. Patterns of morphological dispersion for wetland plants. A mixture of over- 22 23 and underdispersed traits mean that assembly rules constrain community composition 23 in ways that conform to both limiting similarity and environmental filtering (Weiher 24 et al., 1998). 24 25 25 26 26 27 cipal components corresponded to size, leafshape, vegetative reproduction and 27 28 spread, and plasticity). Mean nearest neighbor Euclidean distance in PC- 28 29 defined trait space was positively correlated with soil phosphorus concen- 29 30 tration (Fig. 9.3; Weiher et al., 1998). We also know that above-ground 30 31 competition intensity increases with soil resources in these wetlands (Twolan- 31 32 Strutt & Keddy, 1996) and this supports the idea that competition is driving 32 33 trait dispersion. 33 34 While null models have increased our understanding of community com- 34 35 position, there are still astoundingly few explicit assembly rules. Many of the 35 36 tests using null models provide very little information about what the assem- 36 37 bly rules are. For instance, the checkerboard test (e.g., Stone & Roberts, 1990; 37 38 Graves & Gotelli, 1993; Weiher et al., 1998) tells if some pairs of species 38 39 co-occur less often than expected by chance. A significant checkerboard test 39 40 means that assembly rules are at work, but it tells nothing about the assembly 40 256 E. Weiher and P.A Keddy 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 Fig. 9.3. Morphological dispersion, measured as mean nearest-neighbor distance, 13 increases with soil fertility in herbaceous wetland vegetation. Eleven traits were 14 collapsed into four principal components (Weiher et al., 1998). 14 15 15 16 16 17 rules. Tests of trait dispersion may increase our understanding of community 17 18 composition, but they also do not tell us what the assembly rules are. Wilson’s 18 19 guild proportionality is a test that can potentially supply a simple and pow- 19 20 erful assembly rule (Wilson, this volume). When successful, the method 20 21 provides an explicit rule: one can expect a constant proportion of species in 21 22 each assemblage to belong to a certain guild. However, there are still few 22 23 cases where such rules have been found. We failed to reject the null hypoth- 23 24 esis for eight guilds of wetland plants, but this was partly due to the prob- 24 25 lem of multiple hypothesis testing (Weiher et al., 1998). Plant assemblages 25 26 may simply not be structured according to guild proportionality, but it is too 26 27 soon to give up on such an elegant approach. 27 28 In our opinion, null models can best be used to increase our understanding 28 29 of community structure, but they rarely provide us with defined assembly 29 30 rules. Attempts can be made to find significantly low variances in functional 30 31 groups or traits, and thus explicitly define rules. Unfortunately, more often 31 32 than not, these rules are hard to find and do not provide much predictive power. 32 33 33 34 34 35 Trait–environment linkages 35 36 In plant ecology there has been a rich history of finding associations between 36 37 plant species or their traits and environmental conditions. The earliest exam- 37 38 ples come from von Humbolt and the first biogeographers (e.g., see McIntosh, 38 39 1985; Bowler, 1993), and from ecophysiologists (e.g., Mooney, 1972; Fitter 39 40 & Hay, 1987). More recently, there has been interest in how specific plant 40 Assembly rules as general constraints 257 1 traits vary along gradients (e.g., Givnish, 1987; Tilman, 1988; Keddy, 1992; 1 2 Gaudet & Keddy, 1995) and in the relationships between guilds or functional 2 3 groups and environmental factors (e.g., Cody, 1991; Steneck & Dethier, 1994; 3 4 Schulze & Mooney, 1994). 4 5 It should be made clear that, while some patterns are obvious (such as the 5 6 lack of spiny succulents in temperate wetlands), others are not. Moreover, 6 7 many ecological patterns are understood in only a qualitative manner, such 7 8 as the apparent increase in plant capacity for lateral spread as disturbance 8 9 rates increase or in the height of dominant vegetation as soil resources 9 10 increase. These patterns of species replacement are an issue that is central to 10 11 both ecology in general and for the realm of assembly rules. 11 12 In order to investigate trait–environment linkages, plant traits were plotted 12 13 against various edaphic factors. The riverine wetlands previously described 13 14 by Day et al. (1988), Moore et al. (1989), and Weiher et al. (1998) were 14 15 again used. Fertility and elevation are the two main gradients that affect these 15 16 herbaceous wetlands. The wetlands vary from fertile protected cattail marshes, 16 17 to wave swept sandy shorelines with high species diversity, to tussocky wet 17 18 meadows. Trees and shrubs form the upper margin of the wetlands, while at 18 19 the lower end the wetlands grade into submersed aquatics and Scirpus acu- 19 20 tus. Plant trait data from Weiher et al. (1998) were used 20 21 The plan for defining trait–environment linkages was to define how the 21 22 constraints on plant traits vary along the main environmental gradient (soil 22 23 resources). To do this, the first focal point focus was on the edges of the mor- 23 24 phological space occupied by each assemblage. For each trait, the maximum 24 25 and the minimum value for each assemblage was determined. In this manner, 25 26 the morphological edges of the assemblages were defined. To assess the effect 26 27 of soil resources, each set of assemblage trait maximas and minimas was plot- 27 28 ted as a function of soil resources (soil P, N, loss on ignition, pH, and N:P 28 29 ratio). Nearest-neighbor distances in morphological space were determined in 29 30 order to measure the dispersion of traits in each assemblage (using Euclid- 30 31 ean distances). Lastly, guild proportion were determined (the proportion of 31 32 species richness from each functional guild). The best relationships were 32 33 nearly always with soil [P]µg g−1 and this is what has been presented here. 33 34 It was hoped that the relationships could be modeled with General Linear 34 35 Models, but this was not the case and a series of apparently weak, noisy, 35 36 often non-linear stepped or triangular-shaped ‘patterns’ were found. Figure 36 37 9.4 shows only a selection of the results because some the patterns were 37 38 repeated for several traits (presumably because some traits are correlated). 38 39 For brevity, the main types of patterns are shown. Similar patterns were also 39 40 found for crown and leaf shape mean nearest neighbor distances, for maxima 40 258 E. Weiher and P.A Keddy 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 Fig. 9.4. Examples of trait–environment relationships for wetland plant assemblages. 27 The upper or lower limits are putative assembly rules that constrain community com- 27 28 position. The dashed trend lines were used as target values in the simulation model. 28 29 The dependent variables are: (a) maximum above ground biomass, g, (b) minimum 29 capacity for lateral spread (distance to the farthest attached ramet × number of rhi- 30 zome apicies, cm), (c) minimum height plasticity (coefficient of variation in height), 30 31 (d) mean nearest-neighbor distance in traits with strongest associations with the prin- 31 32 cipal component of 11 measured traits (above ground biomass, form index of leaf 32 33 shape, capacity for lateral spread, and height plasticity), (e) proportion of species rich- 33 ness in the matrix tolerator guild (short plants with high capacity for lateral spread), 34 and (f) proportion of species richness in the interstitial reed guild (compact plants with 34 35 very small canopy, small or absent leaves). 35 36 36 37 of height, crown, and leafshape, and for minima in height, crown, and a four 37 38 dimensional combination of traits that had heaviest loadings on the four prin- 38 39 cipal components of the trait space (biomass, leafshape, capacity for lateral 39 40 spread, and height plasticity). 40 Assembly rules as general constraints 259 1 There has been growing interest in non-linear relationships among 1 2 ecological variables (Brown & Maurer, 1987; Keddy, 1994; Brown, 1995). 2 3 Such patterns often resemble phase diagrams and when plotted, they often 3 4 appear as filled triangular areas (such as the relationship between average 4 5 population density and individual body mass for North American birds, 5 6 Brown & Maurer, 1987) or curves that set upper limits (such as the rela- 6 7 tionship between standing crop and species density, e.g., Tilman & Pacala, 7 8 1993). Conceptually, such patterns likely come from some set of constraints 8 9 that limit allowable community states. Within such constraints, anything may 9 10 be possible. 10 11 Fourteen patterns were found like those shown in Fig. 9.4 for trait min- 11 12 ima, maxima, and nearest-neighbor distances (n = 33). If patterns for traits 12 13 that are significantly correlated are excluded, then eight patterns remain. It 13 14 seems unlikely that these patterns are the result of chance, but an important 14 15 next step is to develop methods for testing the significance of these patterns 15 16 using random assemblages. 16 17 17 18 18 19 Communities as energy dissipating systems 19 20 One quite remarkable generality emerges: in nearly every case, the patterns 20 21 become more tightly constrained as soil fertility increases. It is possible that 21 22 these patterns could be partly the result of fewer sampling points and lower 22 23 diversity at high soil [P]. Alternatively, this relationship corresponds to what 23 24 would be expected if ecological communities are functioning as non-linear 24 25 energy dissipating systems (Schneider, 1987; Ulanowicz & Hannon, 1987). 25 26 A general feature of such self-organizing systems is that as the energy avail- 26 27 able to do work increases, the order or structure of the system tends to also 27 28 increase (Schneider & Kay, 1995). A simple way of looking at this idea is 28 29 that it takes work to move the system from an unstructured (high entropy) 29 30 state to a more ordered state. The idea is also analogous to information the- 30 31 ory, where increased ‘information flow’ results in greater structure (Margalef, 31 32 1968). These ideas have, for the most part, been applied to ecosystem-level 32 33 phenomena such as with energy flows and qualitative changes in energy 33 34 (Schneider & Kay, 1995; Ulanowicz, 1997). If these ideas are applied to com- 34 35 munity-level phenomena and patterns, then increased mineral resources likely 35 36 means increased capacity to do work (production), and as resources increase, 36 37 the constraints on pattern become tighter. There may be more community 37 38 structure where there are higher mineral resources. Results from our experi- 38 39 ment using wetland microcosms also support this idea (Weiher & Keddy, 39 40 1995). In the experiment, a mixture of 20 species were grown in 12 treatments 40 260 E. Weiher and P.A Keddy 1 of soil texture, litter, and hydrology at both low and high fertilization. After 1 2 five years of growth, the high fertilization treatments tended to tightly con- 2 3 verge on one community type while the less productive treatments had more 3 4 variability. The more energy there is to dissipate, the more structure is 4 5 imposed on the system. 5 6 Another observation is that some of the apparent constraints on trait- 6 7 environment patterns look like a series of steps rather than a linear divide 7 8 (Figure 9.4 (a), (b), (c), (f)). Initially, these patterns seemed problematic, given 8 9 the great interest in linear models (e.g., Peters, 1991). However, such stepped 9 10 patterns are also what one might expect if looking at the problem as a study 10 11 in non-linear thermodynamics. The effects of added energy to a system often 11 12 do not cause linear changes in its structure. Rather, energy can usually be 12 13 added with little or no effect, until some critical point is reached. At that crit- 13 14 ical point there is a sudden (non-linear) change in the system. A classic exam- 14 15 ple that is often used is the spontaneous organization of Bénard cells in fluids 15 16 that are heated from the bottom and cooled at the top (Atkins, 1994). When 16 17 the energy gradient increases to some critical point, hexagonal circulation 17 18 cells spontaneously appear. When the energy gradient is increased, the sys- 18 19 tem maintains itself until another threshold is reached. At this point the 19 20 circulation cells spontaneously divide into smaller cells that increase the rate 20 21 of energy dissipation (Schneider & Kay, 1994). The discontinuities in the 21 22 trait–environment patterns may also be interpreted similarly. At low soil 22 23 fertility there may be a single weak attractor in the n-dimensional species 23 24 space, such that many different outcomes are allowed. Resources can increase 24 25 to some critical point without changing the rules because the attractor does 25 26 not change. When that critical point is reached, the system jumps to a new 26 27 attractor, and the allowable patterns change. 27 28 Another expectation from non-linear thermodynamics is that it takes ever- 28 29 increasing levels of energy to push the system into each successive state of 29 30 increasing order or structure. Our results hint at this idea as well. At low 30 31 resource levels it may take only small increases in resources to reach the 31 32 critical point, which is associated with the first tightening of constraints in 32 33 Fig. 9.4 (a) and (b). Small additional increases in soil P cause little or no fur- 33 34 ther change until the next point of tightening is reached. Note that each suc- 34 35 cesive ‘step’ in the pattern has some tendency to be wider than the previous 35 36 one. 36 37 The ideas presented here owe a great debt to Schneider, Kay, and Ulanow- 37 38 icz who pioneered the notion of applying thermodynamic concepts to com- 38 39 plex living systems. Clearly, large concepts are being mapped onto a small 39 40 set of observations; no explicit hypotheses have as yet been tested. The degree 40 Assembly rules as general constraints 261 1 of agreement between the patterns observed and the expectations from non- 1 2 linear thermodynamics was too great and too interesting to ignore. The value 2 3 in this is that it suggests non-linear thermodynamics should be considered as 3 4 a framework for posing hypotheses. 4 5 Given the range of trait–environment patterns in Fig. 9.4, there are several 5 6 methodological choices for seeking assembly rules based on trait–environ- 6 7 ment relationships. General linear modeling and Loess curves are well estab- 7 8 lished, but their utility in addressing noisy patterns is doubtful. Null models 8 9 can be used to find statistically significant patterns of consistency, but are 9 10 probably most applicable to homogeneous patches (Wilson, this volume). 10 11 There are at least two important steps to take regarding these patterns. One 11 12 is to develop statistical tests to investigate whether they differ from chance 12 13 expectation. The other is to test whether such patterns have any use in 13 14 predicting species composition. 14 15 15 16 16 17 Using trait–environment patterns as assembly rules 17 18 So far, different ways of finding patterns in assemblages have been discussed. 18 19 Implicit in the discussion was the idea that such patterns can be used as assem- 19 20 bly rules. Patterns can be used as assembly rules only if accurate predictions 20 21 can be made with them. It is questionable whether a list of putative constraints 21 22 on allowable assemblages can actually be used to make useful predictions. 22 23 While the patterns themselves may be ecologically interesting, they are not 23 24 very useful unless we can make predictions. It would be good to be able to 24 25 take a short list of abiotic conditions and predict assemblage composition. It 25 26 might be useful to predict the most probable assemblages, or quantify the 26 27 number of equiprobable assemblages, each of which may represent a differ- 27 28 ent endpoint of assembly (cf. Law & Morton, 1993). The predictions might 28 29 also be in terms of the species that are most likely to occur. Predictions regard- 29 30 ing how a known assemblage might change, given certain alterations in either 30 31 the abiotic conditions or by adding species to the species pool, are also worth- 31 32 while goals. 32 33 In order to make predictions using trait–environment relationships, we dev- 33 34 eloped an optimization model that is based on the idea of the ‘fitness landscape’. 34 35 Fitness landscapes were developed to understand and model evolution 35 36 (Wright, 1932). In such models, fitness is defined by an algorithm based on 36 37 traits or genotypes, and these are either measured or are initially assigned at 37 38 random. When individuals reproduce, they form novel combinations of alle- 38 39 les and traits, and, over time, the individuals with the highest fitness will tend 39 40 to predominate by finding fitness peaks in the landscape (Templeton, 1981; 40 262 E. Weiher and P.A Keddy 1 Kauffman, 1995). This kind of approach is very good at finding fitness peaks 1 2 in any large, complex fitness landscape. There is no guarantee that the global 2 3 highest fitness will be found because species and natural systems are selected 3 4 for survival rather than optimized and because they have the baggage of past 4 5 solutions which limits their ability to find wildly novel phenotypes. 5 6 In order to investigate the possibility of making predictions using a large 6 7 group of weak patterns, it was necessary to use an optimization approach to 7 8 find assemblages that have a high degree of agreement with the patterns. Over 8 9 time, it became more and more clear that the analogy of an assemblage as a 9 10 ‘species’, evolving over a poorly understood adaptive landscape was very 10 11 compelling. From a numerical point of view, it probably does not matter if 11 12 the evolution of species or the assembly of communities is being discussed. 12 13 In order to adopt the metaphor however, we must dance very close to a 13 14 Clementsian view of communities. This may raise eyebrows. Therefore, it 14 15 should be pointed out that is not our contention that community assembly is 15 16 like the development of an organism, which reaches some stable, adult, climax 16 17 stage. Rather, community assembly is like the evolution of a species; it is 17 18 ever adapting to a changing world within the constraints of assembly rules 18 19 and the limits imposed by its own past. 19 20 Species (genotypes) evolve to local fitness peaks on an adaptive land- 20 21 scape. Exactly which local peak is reached is determined in part by histori- 21 22 cal constraint. It is extremely unlikely that elephants will evolve wings, 22 23 no matter how much their fitness might increase. Similary, communities 23 24 assemble to local fitness peaks on an adaptive landscape. Exactly which local 24 25 peak is reached is also determined in part by historical constraint (Drake, 25 26 1990). Because it is somewhat too Clementsian to talk of assemblage ‘fit- 26 27 ness’, instead assemblage ‘deviation’ from the trait-based patterns described 27 28 above will be referred to. Assemblages with low deviations are favored over 28 29 those with higher deviations. Therefore, assemblage may be described as 29 30 moving across an assembly landscape and finding local minima (with low 30 31 deviation). 31 32 Low points in assembly space nicely correspond to the idea of low energy 32 33 states. Therefore, it might be useful to refer to assemblages with high devi- 33 34 ations from the observed patterns as also having a high energy state (as in 34 35 Fig. 9.1). Assemblages at high energy states will move to a lower energy 35 36 state by altering their composition. Some species may enter, others may fail 36 37 to persist, and these changes are how assemblages move on the adaptive land- 37 38 scape to find basins of low deviation from the assembly rules. 38 39 Assembly rules define the topography of the adaptive landscape for assem- 39 40 blages. The landscape itself is an n-dimensional species space, which for now 40 Assembly rules as general constraints 263 1 can be defined according to species presences and absences. In the future, 1 2 species abundances should also be incorporated. If there are no assembly 2 3 rules, then there is no topography on the adaptive landscape and all combi- 3 4 nations of species are equally favored. Complexity theory suggests that, if 4 5 the assembly rules call for many trade-offs (that conforming to one rule means 5 6 deviation from another), the topography will tend to flatten into an apparently 6 7 ruleless plain (Kauffman, 1993). The greater the topography, the stronger and 7 8 more consistent are the rules. 8 9 Does the topography of the assembly landscape vary with resources? Is it 9 10 pitted with many local minima or are there smooth transitions? How many 10 11 minima are there? Are they clumped in one part of the landscape or are there 11 12 many strong attractor basins? Such questions are basic to understanding 12 13 community assembly. 13 14 As a first step toward addressing these questions, the patterns described 14 15 above have been used to build a model of community assembly. Please note 15 16 that a mechanistic simulation model has not been developed. Rather, it is an 16 17 exploratory model based on a collection of admittedly weak empirical 17 18 patterns. Our initial questions regarding the model include: 18 19 19 (a) does the model find assemblages that ‘fit’ the empirical patterns? 20 20 (b) do the assemblages make any sense at all, given the natural history of the 21 21 species? 22 22 (c) do the assemblages have a reasonable number of species (i.e., is species 23 23 diversity an emergent property of the rules)? 24 24 25 25 26 26 27 An optimization model for community assembly 27 28 The model starts with the input of abiotic resource levels and conditions (soil 28 29 P in µg g−1, and soil percentage organic content as loss on ignition). From 29 30 this input, ‘target’ values are determined from the empirical trait-environment 30 31 patterns (these are shown as dashed lines in Fig. 9.4). Fourteen such patterns 31 32 were used, as noted above. Tolerance values were set at the edges of the 32 33 observed patterns. The tolerance values are the putative assembly rules that 33 34 are used to constrain composition. The model chooses a random set of species 34 35 with richness between 3 and 25. Three were chosen because of an historical 35 36 accident (three species are required to calculate an among-species variance) 36 37 and 25 because it increased the speed of the model (initial runs of the model 37 38 revealed that assemblages with species richness greater than about 25 either 38 39 declined to lower values or the assemblages had a very high deviation). 39 40 The model then checks whether each trait falls outside the tolerance value 40 264 E. Weiher and P.A Keddy 1 (the assembly rule). Whenever it does, a standardized partial deviation was 1 2 calculated as: 2 3 3 ( target – observed )/target. 4 | | 4 5 Partial deviations were standardized in order to minimize the effects of numer- 5 6 ical size bias on the amount of deviation from each rule. The partial devia- 6 7 tions were summed as the total assemblage deviation. Then the model searches 7 8 the local species space for the lowest total deviation. This was done by trying 8 9 all possible single species additions to the assemblage, trying all possible 9 10 single species deletions from the assemblage, and trying all possible combi- 10 11 nations of single species additions and deletions (or more simply, all possible 11 12 replacements). The assemblage with the lowest total deviation was then 12 13 adopted as the best fit, and the new local assembly space was searched for a 13 14 new minimum total deviation. While this search engine may not be the best 14 15 for finding the global optimum, it has the positive quality of randomly sam- 15 16 pling the species space. After 10 to 20 iterations, the search would generally 16 17 fail to turn up an assemblage with lower total deviation, which means the 17 18 model had found a local minimum. This assemblage was saved, and the model 18 19 started with a new random assemblage. The process was repeated until 1000 19 20 local minima were found. In general, the model quickly found many minima 20 21 with low and identical deviation scores. Increasing the number of searches to 21 22 10000 or more simply led to a greater abundance of these putative global 22 23 minima, but it does not find lower deviation scores. This approach was 23 24 adopted to sample local minima and to get a general idea of the topography 24 25 of the adaptive landscape. 25 26 The degree to which real assemblages may get stuck in local minima is 26 27 highly dependent on the number of possible concurrent establishment and 27 28 extirpation events. In our model, we allowed for the simultaneous deletion of 28 29 one species with the addition of another. For real plant communities this is 29 30 not reasonable because the number of simultaneous events is not limited. In 30 31 wetlands, the majority of seeds arrive in the fall, overwinter, and therefore 31 32 functionally have a simultaneous arrival time. Similarly, disturbances such as 32 33 floods, burial, ice scour, or trampling can extirpate many species simultaneously. 33 34 Because of this, not all local minima found by our model are considered to 34 35 be ecologically significant minima. If the number of possible simultanous 35 36 events were increased, then assemblages could escape from many local minima 36 37 and move on to better ones. 37 38 38 39 39 40 40 Assembly rules as general constraints 265 1 1 2 Results of the simulation model 2 3 Random assemblages have much higher deviation scores than do the local 3 4 minima, and the local minima have deviations that approach those of natural 4 5 assemblages. This means that random assemblages do not conform to the 5 6 assembly rules and that the search engine is finding assemblages that gener- 6 7 ally conform to the rules. 7 8 The simulated assemblages roughly conform to our expectations based on 8 9 species natural histories. At low soil phosphorus (1 µg g−1), the best 100 9 10 assemblages (with lowest total deviation) were surprisingly reasonable (Table 10 11 9.1). These assemblages had the same total deviation score, and all had ten 11 12 ‘core’ species which are indeed commonly found on infertile, sandy shore- 12 13 lines. Each assemblage also included three additional species that tend to be 13 14 interchangeable ecological equivalents of sandy shorelines. A good natural 14 15 historian would note however that Phalaris arundinacea is the wrong grass; 15 16 Spartina pectinata is what would have been expected. The simulated assem- 16 17 blages are compelling, but none exactly matches any observed assemblage. 17 18 This is partly due to a lack of elevation data, which allows the model to mix 18 19 upper and lower wetland species. At high soil fertility (30 µg P g−1) the model 19 20 also predicted startlingly reasonable assemblages (Table 9.1). Cattails and 20 21 sagittate-leafed species are predicted, as is Boehmeria cylindrica, an annual 21 22 common in cattail stands. Once again, the model chose the wrong grass 22 23 species. At intermediate soil fertility, a vast number of low deviation assem- 23 24 blages are found, but there are no simple trends to report. One interpretation 24 25 is that at intermediate fertility species composition is poorly constrained by 25 26 assembly rules. Another, equally valid interpretation is that our assembly rules 26 27 are woefully inadequate for predicting these intermediate community types. 27 28 Robert Ulanowicz (1997) has argued that there is no need to find a precise 28 29 casual mechanism for everything in nature, and that probablism and gross 29 30 relationships are just as valuable. 30 31 The agreement with real assemblages at both ends of the fertility gradient 31 32 means that the assembly rules do indeed constrain community compostition 32 33 and they have some utility in making predictions. Assemblages with equal 33 34 deviations suggest that alternative equivalent endpoints exist and that at least 34 35 some species are ecological equivalents. 35 36 The simulations were in qualitative agreement with actual patterns of 36 37 species diversity, but the predictions were generally too high (Fig. 9.5). The 37 38 ubiquitous ‘humped’ species richness – productivity relationship with low 38 39 richness at both ends of the productivity gradient emerged from the model. 39 40 The pattern also took the form of an upper-limit function, with a filled in 40 266 E. Weiher and P.A Keddy

1 Table 9.1. Simulated community composition using trait-based assembly rules. 1 2 These results summarize the best 100 assemblages for each simulation 2 3 3 −1 4 P = 30 µg g 4 P = 1 µg g−1 Incidence The four most common assemblages 5 5 6 Core species: 1 Boehmeria cylindrica 6 7 Carex vesicaria 1 Carex lenticularis 7 Dulichium arundinaceum 1 Sagittaria latifolia 8 Eleocharis acicularis 1 Sagittaria rigida 8 9 Eupatorium perfoliatum 1 Typha glauca 9 10 Hypericum boreale 1 10 Lindernia dubia 12Boehmeria cylindrica 11 Phalaris arundinacea 1 Eleocharis erythropoda 11 12 Potentilla palustris 1 Pontederia cordata 12 13 Scirpus acutus 1 Sagittaria latifolia 13 14 Scirpus cyperinus 1 Typha angustifolia 14 15 + three of the following: 3 Equisetum fluviatile 15 Eleocharis smallii 0.905 Mentha arvensis 16 Epilobium ciliatum 0.571 Sagittaria latifolia 16 17 Juncus nodosus 0.333 Rumex verticillatus 17 18 Drosera intermedia 0.238 Spartina pectinata 18 Eriocaulon septangulare 0.190 19 Juncus alpinus 0.143 4 Boehmeria cylindrica 19 20 Juncus subtilis 0.143 Juncus effusus 20 21 Leersia oryzoides 0.143 Sagittaria latifolia 21 Ranunculus flammula 0.143 Pontederia cordata 22 Taraxicum officinale 0.095 Typha glauca 22 23 Juncus bufonius 0.048 23 24 Juncus pelocarpus 0.048 24 25 25 26 26 27 distribution of points. With increasing fertility, the model constrains diversity 27 28 by setting limits on morphological nearest-neighbor distances. With decreas- 28 29 ing fertility, the model constrains diversity by limiting species composition 29 30 to relatively small species with short nearest-neighbor distances. The results 30 31 are qualitatively satisfying because they suggest the trait-based assembly rules 31 32 may be useful for prediction and that species diversity may be an emergent 32 33 property of assembly rules. However, the model’s imprecision shows that we 33 34 still have much to do. 34 35 One way to visualize the topography of the adaptive landscape is to show 35 36 how deviation varies with soil fertility and species richness (Fig. 9.5). The 36 37 contours show minimum total deviation scores. The assemblages seek mini- 37 38 mal deviation by adding, deleting, or replacing species. Conceptually, assem- 38 39 blages with high deviation scores should not persist and should ‘evolve’ toward 39 40 potentially more stable, low deviation endpoints (cf. Fig. 9.1). At both low 40 Assembly rules as general constraints 267 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 Fig. 9.5. Species richness as an emergent property from trait-based assembly rules. 20 The simulation model overpredicts richness, but there is qualitative agreement with 20 21 the expected humped and filled-in relationship between richness and productivity. 21 22 22 23 and high fertility, species richness is constrained to low values (Fig. 9.6). The 23 24 flat, mildly undulating central region shows that at intermediate fertility 24 25 species richness is poorly constrained. It also shows that there are many weak 25 26 attractors, or small depressions, which suggests that there are many possible 26 27 endpoints. As soil fertility increases, the terrain becomes steeper and more 27 28 rugged, which is suggestive of a thermodynamic tightening of constraints. 28 29 29 30 30 31 Conclusions 31 32 Assembly rules are explicit constraints that limit how assemblages are selected 32 33 from a larger species pool. Evidence can be sought for assembly rules using 33 34 a variety of tools, from null models to gradient analysis, but insight is max- 34 35 imized when the focus is on guilds and species traits. When relationships 35 36 between plant traits and soil fertility were investigated, a series of weak phase- 36 37 space diagrams were found, which were highly suggestive of a non-linear 37 38 thermodynamic model of self-organized energy dissipating systems. Com- 38 39 munity ecologists should make use of these recent conceptual advances in 39 40 complexity theory and ecosystem ecology. 40 268 E. Weiher and P.A Keddy 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 Fig. 9.6. The adaptive landscape for simulated wetland plant communities. Diversity 22 23 is shown as a function of soil fertility; contours show minimum total deviation scores. 23 In the model, assemblages move vertically to minimal deviations by species addition, 24 deletion, or replacement. Note the flat central region where diversity is poorly 24 25 constrained. 25 26 26 27 27 28 If assembly rules are going to be useful, then attempts must be made to 28 29 use them to predict composition. Trait–environment patterns were used as 29 30 putative assembly rules in an optimization model that used the analogy of a 30 31 species evolving on an adaptive landscape. A pragmatic search for a model- 31 32 ing tool led communities to be viewed as self-organized entities that change 32 33 their constituent membership so as to minimize their deviation from a black 33 34 box of assembly rules. The qualitative success of the model suggests that 34 35 trait–environment relationships have a place in the search for assembly rules 35 36 and therefore have some utility in making predictions. As is usual with work 36 37 on assembly rules, it seemed much easier to find new metaphors and con- 37 38 ceptual insights into the workings of ecological communities than to find 38 39 accurate, reliable predictions. 39 40 40 Assembly rules as general constraints 269

1 Acknowledgments 1 2 2 Thanks to James Kay for giving an enlightening seminar. Many thanks to 3 3 Michael Ell for programming the simulation model. 4 4 5 5 6 References 6 7 Armbruster, W.S. (1986). Reproductive interactions between sympatric 7 8 Dalechampia species: are natural assemblages ‘random’ or organized? Ecology 8 9 67: 522–533. 9 10 Armbruster, W.S., Edwards, M.E., & Debevec, E.M. (1994). Floral character 10 displacement generates assemblage structure of Western Australian 11 triggerplants (stylidium). Ecology 75: 315–329. 11 12 Atkins, P.W. (1994). The 2nd Law: Energy, Chaos, and Form. NY: Scientiﬁc 12 13 American Books. 13 Bowler, P.J. (1993). The Norton History of the Environmental Sciences. NY: WW 14 Norton. 14 15 Box, E.O. (1981). Macroclimate and Plant Forms: An Introduction to Predictive 15 16 Modeling in Phytogeography. London: Dr W. Junk Publishers. 16 Brown, J.H. (1987). Variation in desert rodent guilds: patterns, processes, and 17 scales. In Organization of Communities: Past and Present. ed. J.H.R. Gee & 17 18 P.S. Giller, pp. 185–203. Blackwell, London. 18 19 Brown, J.H. (1995). Macroecology. Chicago, IL: The University of Chicago Press. 19 20 Brown, J.H. & Bowers, M. (1984). Patterns and processes in three guilds of 20 terrestrial vertebrates. In Ecological Communities: Conceptual Issues and the 21 Evidence. ed. D. Strong, D. Simberloff, L. Abele, & A. Thistle, pp. 282–296. 21 22 Princeton, NJ: Princeton University Press. 22 23 Brown, J. H. & Maurer, B. A. (1987). Evolution of species assemblages: effects of 23 energetic constraints and species dynamics on the diversiﬁcation of the North 24 American avifauna. American Naturalist 130: 1–17. 24 25 Cody, M.L. (1991). Niche theory and plant growth form. Vegetatio 97: 39–55. 25 26 Connor, E.F. & Simberloff, D. (1979). The assembly of species communities: 26 chance or competition. Ecology 60: 1132–1140. 27 Day, R.T., Keddy, P.A., McNeill, J., & Carleton, T. (1988). Fertility and 27 28 disturbance gradients: a summary model for riverine marsh vegetation. Ecology 28 29 69: 1044–1054. 29 30 Dayan, T. & Simberloff, D. (1994). Morphological relationships among coexisting 30 hereromyids: an incisive dental character. American Naturalist 143: 462–477. 31 Diamond, J.M. (1975). Assembly of species communities. In Ecology and 31 32 Evolution of Communities, ed. M.L. Cody & J.M. Diamond, pp. 342–373. 32 33 Cambridge, MA: Belknap Press. 33 Drake, J.A. (1990). Communities as assembled structures: do rules govern pattern? 34 Trends in Ecology and Evolution 5: 159–164. 34 35 Fitter, A.H. & Hay, R.K.M. (1987). Environmental Physiology of Plants. London: 35 36 Academic Press. 36 Gaudet, C.L. & Keddy, P.A. (1995). Competitive performance and species 37 distribution in the shoreline plant communities: a comparative approach. 37 38 Ecology 76: 280–291. 38 39 Givnish, T.J. (1987). Comparative studies of leaf form: assessing the relative roles 39 40 of selective pressures and phylogenetic constraints. New Phytologist 106: 40 131–160. 270 E. Weiher and P.A Keddy

1 Gotelli, N.J. & Graves, G.R. (1996). Null Models in Ecology. Washington, DC: 1 2 Smithsonian Institution Press. 2 Graves, G.R. & Gotelli, N.J. (1993). Assembly of avian mixed-species ﬂocks in 3 Amazonia. Proceedings of the National Academy of Sciences 90: 1388–1391. 3 4 Grime, J.P. (1979). Plant Strategies and Vegetation Processes. Chichester, UK: 4 5 John Wiley. 5 6 Grubb, P.J. (1977). The maintenance of species-richness in plant communities: the 6 importance of the regeneration niche. Biological Reviews 52: 107–145. 7 Huston, M.A. (1997). Hidden treatments in ecological experiments: re-evaluating 7 8 the ecosystem function of biodiversity. Oecologia 110: 449–460. 8 9 Kauffman, S. (1995). At Home in the Universe. NY: Oxford University Press. 9 Keddy, P.A. (1992). Assembly and response rules: two goals for predictive 10 community ecology. Journal of Vegetation Science 3: 157–164. 10 11 Keddy, P.A. (1994). Applications of the Hertzprung–Russell star chart to ecology: 11 12 reﬂections on the 21st birthday of geographical ecology. Trends in Ecology 12 and Evolution 9: 231–234. 13 Law, R. & Morton, R.D. (1993). Alternative permanent states of ecological 13 14 communities. Ecology 74: 1347–1361. 14 15 McIntosh, R.P. (1985). The Background of Ecology. Cambridge, UK: Cambridge 15 16 University Press. 16 Margalef, R. (1968). Perspectives in Ecological Theory. Chicago, IL: University of 17 Chicago Press. 17 18 Miles, D.B. & Ricklefs, R.E. (1994). Ecological and evolutionary inferences from 18 19 morphology: an ecological perspective. In Ecological Morphology: Integrative 19 Organismal Biology, ed. P.C. Wainwright & S.M. Reilly, pp. 13–41. Chicago, 20 IL: University of Chicago Press. 20 21 Mohler, C.L. (1990). Co-occurrence of oak subgenera: implications of niche 21 22 differentiation. Bulletin of the Torrey Botanical Club 117: 247–255. 22 Mooney, H.A. (1972). The carbon balance of plants. Annual Review of Ecology 3: 23 315–346. 23 24 Moore, D.R.J., Keddy, P.A., Gaudet, C.L., & Wisheu, I.C. (1989). Conservation of 24 25 wetlands: do infertile wetlands deserve higher priority? Biological 25 26 Conservation 47: 203–217. 26 Moulton, M.L. & Pimm, S.L. (1987). Morphological assortment in introduced 27 Hawaiian passerines. Evolutionary Ecology 1: 113–124. 27 28 Peters, R.H. (1980). From natural history to ecology. Perspectives in Biology and 28 29 Medicine 23: 191–202. 29 Peters, R.H. (1991). A Critique for Ecology. Cambridge, UK: Cambridge 30 University Press. 30 31 Ricklefs, R.E. & Travis, J. (1980). A morphological approach to the study of avian 31 32 community organization. Auk 97: 321–338. 32 Schluter, D. (1984). A variance test for detecting species associations with some 33 example applications. Ecology 65: 998–1005. 33 34 Schneider, E.D. (1987). Schrödinger shortchanged. Nature 328: 300. 34 35 Schneider, E.D. & Kay, J.J. (1994). Complexity and thermodynamics: towards a 35 36 new ecology. Futures 24: 626–647. 36 Scheider, E.D. & Kay, J.J. (1995). Order from disorder: the thermodynamics of 37 complexity in biology. In What is Life? The Next Fifty Years. ed. M.P. 37 38 Murphy & L.A.J. O’Neill, pp. 161–172. Cambridge, UK: Cambridge 38 39 University Press. 39 Schulze, E.D. & Mooney, H.A. (eds.) (1994). Biodiversity and Ecosystem Function. 40 NY: Springer-Verlag. 40 Assembly rules as general constraints 271

1 Simberloff, D. & Connor, E.F. (1981). Missing species combinations. American 1 2 Naturalist 118: 215–239. 2 Southwood, T.R.E. (1988). Tactics, strategies and templets. Oikos 52: 3–18. 3 Steneck, R.S. & Dethier, M.N. (1994). A functional group approach to the structure 3 4 of algal-dominated communities Oikos 69: 476–498. 4 5 Stone, L. & Roberts, A. (1990). The checkerboard score and species distributions. 5 6 Oecologia 85: 74–79. 6 Strong, D., Simberloff, D., Abele, L., & Thistle, A. (eds.) (1984). Ecological 7 Communities: Conceptual Issues and the Evidence. Princeton, NJ: Princeton 7 8 University Press. 8 9 Templeton, A.R. (1981). Mechanisms of speciation – a population genetic 9 approach. Annual Review of Ecology and Systematics 12: 23–48. 10 Tilman, D. (1988). Plant strategies and the dynamics and structure of plant 10 11 communities. Princeton, NJ: Princeton University Press. 11 12 Tilman, D. & Pacala, S. (1993). The maintenance of species richness in plant 12 communities. In Species Diversity in Ecological Communities. ed. R.E. 13 Ricklefs & D. Schluter, pp. 13–25. Chicago, IL: University of Chicago Press. 13 14 Travis, J. & Ricklefs, R.E. (1983). A morphological comparison of island and 14 15 mainland assemblages of neotropical birds. Oikos 41: 434–441. 15 16 Twolan-Strutt, L. & Keddy, P.A. (1996). Above- and below-ground competition 16 intensity in two contrasting wetland plant communities. Ecology 77: 259–270. 17 Ulanowicz, R. (1997). Ecology, The Ascendant Perspective. NY: Columbia 17 18 University Press. 18 19 Ulanowicz, R.E. & Hannon, B.M. (1987). Life and the production of entropy. 19 Proceedings of the Royal Society B 232: 181–192. 20 Weiher, E. & Keddy, P.A. (1995). Assembly rules, null models, and trait 20 21 dispersion: new questions form old patterns. Oikos 74: 159–164. 21 22 Weiher, E., Clarke, G.D.P., & Keddy, P.A. (1998). Community assembly rules, 22 morphological dispersion, and the coexistence of plant species. Oikos 81: 23 309–322. 23 24 Wilson, J.B. & Roxburgh, S.H. (1994). A demonstration of guild-based assembly 24 25 rules for a plant communtiy, and determination of intrinsic guilds. Oikos 69: 25 26 267–276. 26 Wright, S. (1932). The roles of mutation, inbreeding, crossbreeding, and selection 27 in evolution. Proceedings of the XI International Congress of Genetics 1: 27 28 356–366. 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 10 1 2 2 3 A species-based, hierarchical model of island 3 4 biogeography 4 5 5 6 6 7 Mark V. Lomolino 7 8 8 9 9 10 10 11 11 12 12 13 13 14 Introduction 14 15 MacArthur and Wilson’s (1963, 1967) Equilibrium Theory has served as a 15 16 paradigm for the field of ecological biogeography for nearly three decades 16 17 (see also Munroe, 1948, 1953; Brown & Lomolino, 1989). The theory pro- 17 18 vided an insightful alternative to the prevailing, ‘static theory’ of island 18 19 biogeography (see Dexter, 1978), which held that insular community struc- 19 20 ture was fixed in ecological time, resulting from unique immigration and 20 21 extinction events. Either a species made it to the island in question, or it did 21 22 not. Once it arrived, the species found adequate resources for survival, and 22 23 did so in perpetuity, or it failed to establish a population. In contrast, 23 24 MacArthur and Wilson held that insular community structure was dynamic 24 25 in ecological time, resulting from recurrent immigrations and extinctions. As 25 26 an island filled with species, immigration rate declined and extinction rate 26 27 increased until an equilibrial number of species was attained. At this level, 27 28 species richness would remain relatively constant, while species composition 28 29 would continue to change as new species replaced those becoming extinct. 29 30 MacArthur and Wilson (1967: 19–21) acknowledged that ‘a perfect bal- 30 31 ance between immigration and extinction might never be reached, ... but to 31 32 the extent that the assumption of a balance has enabled us to make certain 32 33 valid new predictions, the equilibrium concept is useful.’ The equilibrium 33 34 theory has indeed served as a worthy paradigm, stimulating much research 34 35 on structure of isolated communities. It has, however, also been the subject 35 36 of much criticism (e.g., see Sauer, 1969; Pielou, 1979; Gilbert, 1980), which 36 37 will not be dwelt on here. Extending MacArthur and Wilson’s insights to 37 38 develop a more general, multi-scale theory of insular community structure is 38 39 the challenge at hand. Here a model will be presented that is based on 39 40 MacArthur and Wilson’s fundamental assumption that insular community 40

272 A biogeographical model 273 1 structure results from the combined effects of recurrent immigration and 1 2 extinction. Like their model, the one developed here is limited in temporal 2 3 scale to ecological time. Also like their model, it is assumed that immigra- 3 4 tion and extinction are deterministic with respect to island characteristics. 4 5 MacArthur and Wilson’s model, however, assumed that immigrations and 5 6 extinctions were stochastic with respect to species characteristics. That is, 6 7 their model assumed that all species were equivalent. The model presented 7 8 here assumes that species differ with respect to factors affecting immigration 8 9 and extinction. Many patterns in insular community structure result from, not 9 10 despite, interspecific differences in abilities to immigrate to and survive on 10 11 islands. While the model is species-based, it is hierarchical in that it addresses 11 12 the potential influences of immigration and extinction across ecological and 12 13 spatial scales ranging from distributions of individual species within an arch- 13 14 ipelago, to differences in community assembly, nestedness and species rich- 14 15 ness patterns among archipelagoes. For the sake of simplicity, this chapter 15 16 will focus on biogeographic patterns of insular animals and the potential 16 17 importance of processes occurring within ecological time. 17 18 First, a focal species model will be explained; it was originally presented 18 19 during a symposium on mammalian biogeography in 1984 (Lomolino, 1986). 19 20 This model describes how the insular distribution of a focal species should be 20 21 influenced by factors affecting immigration and extinction; specifically, isolation 21 22 and area, respectively. After describing the expected form of this insular distri- 22 23 bution function (IDF), a summary of empirical patterns of insular distributions 23 24 of selected species will be given. The focal species model serves as the 24 25 fundamental level of a hierarchical model extending to community and inter- 25 26 archipelago scales. Linkage between these levels is achieved by exploring how 26 27 species, islands and archipelagoes covary with respect to factors affecting the 27 28 two fundamental biogeographic processes, i.e., immigration and extinction. 28 29 In some, perhaps many cases, this hierarchical model may provide a 29 30 relatively simple explanation for apparent anomalies or fundamental ambi- 30 31 guities of biogeography. As Brown (1995) observed in a recent discussion of 31 32 MacArthur and Wilson’s model, ‘many advances in insular biogeography 32 33 come from developing alternative models to explain cases in which the pre- 33 34 dictions do not hold.’ The model presented here may explain many of these 34 35 special cases. More importantly, rather than providing all the answers, this 35 36 model should generate some new predictions and identify critical gaps in our 36 37 understanding of the forces structuring isolated communities. This model 37 38 should, therefore, provide some useful insights for conservation biology, which 38 39 has largely become a challenge to conserve native biotas on ever-shrinking 39 40 ‘islands’ of their natural habitats. Potential applications of the species-based 40 274 M.V. Lomolino 1 model for conserving biodiversity will be discussed in a forthcoming paper 1 2 (see also Lomolino, 1986). 2 3 3 4 4 5 A focal species model of island biogeography 5 6 This focal species model is based on three assumptions: 6 7 7 (a) Species are common on those islands where their persistence time exceeds 8 8 the time between immigrations. 9 9 (b) Persistence time of the focal species should increase with island area. 10 10 (c) Immigration time should increase with island isolation. 11 11 12 Given this, the minimum area requirements of each focal species should 12 13 increase with isolation. This pattern can be illustrated on a coordinate system 13 14 where one axis represents island isolation and the other axis represents island 14 15 area (Fig. 10.1; see also Alatalo, 1982). The focal species may be present on 15 16 small islands if they are close enough to a source population such that high 16 17 immigration rates can compensate for high extinction rates (cf. ‘mass effect’ 17 18 – Schmida & Wilson, 1985; Auerbach & Schmida, 1987; ‘rescue effect’ – 18 19 Brown & Kodric-Brown, 1977). Conversely, the focal species may be present 19 20 on distant islands if they are large enough such that low extinction rates 20 21 compensate for low immigration rates. As a result of these compensatory 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 Fig. 10.1. A simple, linear Insular Distribution Function (IDF) for a hypothetical 36 37 species. The IDF delineates those combinations of area and isolation where extinc- 37 38 tion and immigration rates are equal. The focal species is expected to occur on islands 38 that exceed this ratio (i.e., islands that fall above the IDF – darkened symbols) and 39 absent from those that fall below it (open symbols). The actual form of the IDF is 39 40 expected to be non-linear (see Fig. 10.3). 40 A biogeographical model 275 1 effects, insular distribution functions (IDFs; Fig. 10.1), should have a positive 1 2 slope. 2 3 The form of the IDF depends, in turn, on the form of two fundamental 3 4 relationships: the extinction–area and immigration–isolation functions. Most 4 5 theoretical work indicates that population persistence increases as an expo- 5 6 nential function of island area (MacArthur & Wilson, 1967; Goel & Richter- 6 7 Dynn, 1974). In contrast, the immigration–isolation relationship has not 7 8 received nearly as much attention. Nonetheless, it is known that the factors 8 9 affecting vagilities and, for passive immigrators, persistence and survival 9 10 during rafting, tend to be correlated with physical traits of individuals such 10 11 as fat stores, buoyancy and lift, or metabolic rates. These traits tend to be 11 12 normally or log-normally distributed within species (Calder, 1984; Peters, 12 13 1983). The precise form of frequency distributions of these traits is not an 13 14 issue here. It is only important that immigration abilities (active or passive) 14 15 of most individuals tend to cluster about some modal value. Simply put, nearly 15 16 all individuals can immigrate beyond some critical minimum distance, but 16 17 few can immigrate beyond some maximal distance (Fig. 10.2(a). Thus, immi- 17 18 gration rate of a focal species should be a negative sigmoidal function of iso- 18 19 lation (Fig. 10.2(b); see also MacArthur & Wilson, 1967: 56, 127–128). More 19 20 to the point, the time between immigrations by this species should be a pos- 20 21 itive sigmoidal function of isolation. Thus, to maintain insular populations, 21 22 area requirements (and persistence times) of the focal species will increase, 22 23 again as a positive sigmoidal function of isolation (Fig. 10.3). Because per- 23 24 sistence time should be an exponential function of area, increasing more 24 25 rapidly for the large islands, the slope of the IDF should be more shallow 25 26 than that of the immigration curve (i.e., increasing less rapidly for the more- 26 27 isolated islands; Fig. 10.3, solid vs. dashed lines). Yet, the general form of 27 28 the IDF should still be a positive sigmoidal – increasing relatively slowly for 28 29 the near islands, more rapidly for those of intermediate isolation, and less 29 30 rapidly again for the more isolated islands. Because the IDF delineates the 30 31 combinations of area and isolation where the focal species immigrates as 31 32 frequently as it suffers extinction, its insular populations should be present 32 33 on islands that fall above the curve and absent on those that fall below it (Fig. 33 34 10.3). 34 35 The IDF has two readily interpretable components: an intercept and a 35

36 generalized, albeit variable, slope. The intercept (Amin in Fig. 10.3) is a mea- 36 37 sure of area required to maintain a population of the focal species on the 37 38 ‘mainland’, while the slope is a measure, actually an inverse measure, of its 38 39 immigration abilities. Given the expected variation in resource requirements 39 40 and immigration abilities among species, insular distribution patterns may 40 276 M.V. Lomolino 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 Fig. 10.2. (a) Frequency distribution of immigration abilities (distances) among indi- 27 28 viduals of a hypothetical focal species. Because immigration abilities are inﬂuenced 28 by a combination of many physiological, morphological and behavioral traits, they 29 are expected to be distributed as normal or log-normal functions, clustering about 29

30 some intermediate value (Dmode). Here, nearly all individuals can immigrate beyond 30 some minimal distance (D ), while very few can immigrate to the very distant islands 31 near 31 (i.e., those beyond Dfar). (b) Given immigration abilities of individuals are distributed 32 as depicted in Fig. 10.2(a), then immigration rate (the number of propagules reach- 32 33 ing an island per time) of the focal species should decrease as a negative sigmoidal 33 34 function of isolation (dashed line), while the time between immigrations should be a 34 positive sigmoidal function of isolation (solid line): increasing slowly for the near 35 islands (region a), more rapidly for islands of intermediate isolation (region b), and 35 36 again more slowly for the more isolated islands (region c). 36 37 37 38 38 39 39 40 40 A biogeographical model 277 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 Fig. 10.3. The general form of the insular distribution function (IDF). The IDF delin- 13 14 eates those combinations of area and isolation for which immigration equals extinc- 14 tion. The focal species is expected to occur on islands that fall above the IDF. Amin 15 represents the minimum area required to maintain populations of this species on the 15 16 mainland. The dashed line depicts the IDF expected if persistence time was a linear 16 17 function of area, while the solid line depicts the expected IDF given that persistence 17 time should increase as an exponential function of island area. The influence of 18 isolation on insular distribution of this species should be difficult to detect in regions 18 19 a and c (i.e., if sampling is limited to the very near or very isolated islands; see 19 20 Fig. 10.2). 20 21 21 22 22 23 vary widely, even within a group of taxonomically or ecologically similar 23 24 species (Fig. 4.10 and 10.5). In fact, depending on the range in area and iso- 24 25 lation surveyed, the same species may exhibit what appear to be qualitatively 25 26 different patterns within the same archipelago. Fig. 10.6 illustrates how, given 26 27 surveys across different ranges in biogeographic variables, the same species 27 28 may exhibit distributions that appear to be limited by just area, but markedly 28 29 different areas (boxes A and B in Fig. 10.6), or by just isolation (box C). 29 30 Alternatively, this species may appear ubiquitous on small islands, but rare 30 31 on larger islands (boxes D and E). Given a broader range in biogeographical 31 32 variables sampled, these fundamental ambiguities should clear to reveal the 32 33 insular distribution function, i.e., the combined influence of immigration and 33 34 extinction. 34 35 The sigmoidal form of the IDF also can account for some other biogeo- 35 36 graphic ambiguities or anomalies. For example, distributions of some species 36 37 or groups of species appear to be uninfluenced by isolation. The relaxation 37 38 faunas of the Great Basin and Bass Straits may be exemplary cases (Fig. 38 39 10.7). While the apparent absence of isolation effects for most of these species 39 40 seems counter to the fundamental assumption of the equilibrium theory, it is 40 278 M.V. Lomolino 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 Fig. 10.4. The predicted effects of interspecific differences in resource requirements 14 and immigration abilities on insular distribution functions. In comparison to the IDF 14 15 of a hypothetical focal species (solid line), IDFs should shift upward for species with 15 16 greater resource requirements (dashed line) and rightward for species with greater 16 immigration abilities (dotted line). 17 17 18 18 19 19 20 consistent with predictions of the species-based model. That is, for most non- 20 21 volant mammals studied, the islands of the Great Basin and Bass Straits may 21 22 encompass a region where isolation effects are difficult to detect (grey box 22 23 of Fig. 10.7 (c); see also Fig. 10.3, region c). Here, species distributions should 23 24 be largely a function of area, but not isolation. In addition, most island bio- 24 25 geography studies are not well designed to detect the effects of isolation. 25 26 While these natural experiments include islands that vary by three or four 26 27 orders of magnitude in area, isolation typically varies by just one or two 27 28 orders of magnitude. Thus, isolation effects may be detectable only when 28 29 the structure of these isolated communities is compared to that of the less- 29 30 isolated ones (see Lomolino & Davis, 1997). 30 31 In summary, much of what appear to be biogeographic ambiguities or 31 32 anomalies may stem from the sigmoidal nature of the IDF and the tendency 32 33 for biogeographic studies to sample just part of its bivariate space. If the IDF 33 34 was a linear function, the effects of area and isolation would be equally 34 35 detectable throughout the archipelago. Again, the form of the IDF derives 35 36 from two fundamental biogeographic functions, the immigration–isolation 36 37 relationship and the extinction–area relationship. Because these functions are 37 38 not linear, much of what is studied in biogeography is strongly scale depen- 38 39 dent. Thus, to better understand the geographical basis of biodiversity, island 39 40 biogeography can gain many important insights from comparisons of patterns 40 A biogeographical model 279 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 Fig. 10.5. Empirical distribution patterns of three species inhabiting islands, or island- 35 like habitats: the bog copper butterfly (Michaud, 1995), the Ouachita Mountain shiner 35 36 (Taylor, 1997), and the masked shrew (Lomolino, 1993; darkened symbols depict 36 37 presence, open symbols depict absence). Distribuitons of each species were signifi- 37 cantly associated with both area and isolation (P < 0.05; linear regression; sigmoidal 38 IDFs were drawn by inspection). 38 39 39 40 40 280 M.V. Lomolino 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 Fig 10.6. Effects of different sampling regimes on apparent patterns of distributions 29 of a hypothetical species (darkened symbols depict presence, open symbols depict 30 absence). Depending on the region sampled, the same species may appear to exhibit 30 31 a variety of distribution patterns with respect to island area and isolation (see text). 31 32 32 33 33 34 among species and across spatial and ecological scales. These between-scale 34 35 comparisons are the subject of the following sections. 35 36 The influence of immigration and isolation on insular distributions may be 36 37 studied using multivariate analyses – linear regression, logistic regression or 37 38 discriminant analysis (see also similar methods of analyzing insular distrib- 38 39 utions by Schoener & Schoener, 1983; Simberloff & Martin, 1991). In such 39 40 cases, the dependent variable expresses presence or absence of the focal 40 A biogeographical model 281 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 Fig. 10.7. Insular distribution patterns of non-volant mammals on two isolated arch- 35 ipelagoes (a) relaxation fauna of montane forest islands of the Great Basin, North 35 36 America, after Brown, 1978; (b) relaxation fauna of the islands of Bass Straits, Aus- 36 37 tralia, after Hope, 1973. Species occur on islands that fall above the lines (see 37 38 Lomolino, 1986). Although distributions of these species appear to be uninfluenced 38 by isolation, this may be an artifact of the sampling regime. That is, isolation effects 39 may have been detected if less-isolated islands were included in these studies ((c) 39 40 darkened symbols depict presence, open symbols depict absence). 40 282 M.V. Lomolino

1 species (e.g., arbitrarily set at 1 or 0, respectively), and independent variables 1 2 are isolation and area. Because the IDF is not a linear function, it probably 2 3 will be necessary to transform one or both independent variables before con- 3 4 ducting the analyzes. Given the diversity of possible patterns (Fig. 10.6), the 4 5 appropriate transformations may best be assessed by inspection. The output 5 6 of these analyzes can be used to estimate the statistical significance of isola- 6 7 tion or area effects and to calculate the slope and intercept of the IDF (for 7 8 a more detailed description of these methods, see Lomolino, 1986; Lomolino 8 9 et al., 1989). More general and possibly more robust tests of the influence 9 10 of isolation and area may be achieved using a randomization approach. 10 11 Basically, it could be tested whether islands occupied by the target species 11 12 are significantly clustered with a portion of biogeographic space (i.e., 12 13 area–isolation plots). To test whether clustering is significant, nearest neigh- 13 14 bor distances of observed distributions can be compared to those when the 14 15 occupancy of the focal species is randomly distributed across the archipelago 15 16 (see Manly, 1991). 16 17 17 18 18 Community structure 19 19 20 Linkage to the focal species model 20 21 One way to extend the focal species model to a community level is simply 21 22 to overlay IDFs of different species on one graph to form what I referred to 22 23 as a community spectrum (e.g., Fig. 10.8(a); Lomolino, 1986). While this 23 24 descriptive approach has some heuristic value, a more rigorous and insight- 24 25 ful model requires explicit linkage among hierarchical levels (see O’Neill et 25 26 al., 1986; Wiens et al., 1986; Allen, 1987; Hengeveld, 1987). Specifically, to 26 27 predict patterns in species composition, we need to know how the two com- 27 28 ponents of the IDF, its intercept and generalized slope, covary among species. 28 29 Alternatively, to predict patterns in species richness, we need to assess the 29 30 form of frequency distributions of slopes and intercepts. 30 31 Again, the intercept of the IDF is a measure of resource requirements and 31 32 the slope is a measure, actually, an inverse measure of immigration ability. 32 33 Immigration abilities may either increase with resource requirements, 33 34 decrease with resource requirements, or the two traits may be uncorrelated 34 35 among species. For the sake of simplicity, it is assumed here that the latter 35 36 case will be intermediate with respect to resultant patterns in community struc- 36 37 ture. Therefore, the following discussion focuses on implications of the two 37 38 alternative patterns of covariation (positive or negative) of slopes and inter- 38 39 cepts on insular community structure. 39 40 How should immigration abilities and resource requirements covary? 40 A biogeographical model 283 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 Fig 10.8. Effects of two alternative patterns of covariation of resource requirements 11 and immigration abilities on assembly and nestedness of insular communities. Solid lines 12 depict IDFs for three hypothetical species (A, B, C; or E, F, G) in two archipelagoes. 12 13 (a) Where better immigrators require more resources (larger islands) to maintain their 13 14 populations, relatively complex assembly of insular communities is expected depend- 14 15 ing on the size and isolation of islands sampled. (b) In contrast, where better immi- 15 grators require fewer resources to maintain their insular populations, then assembly 16 sequences should be constant (i.e., first species E, then F, and then G). 16 17 17 18 18 19 Physiological considerations may provide some useful insights here. Within 19 20 a group of similar species (e.g., non-volant mammals or passerine birds), 20 21 larger species tend to require more energy, but they also tend to exhibit greater 21 22 capacities for active migration and greater endurance during stressful condi- 22 23 tions (e.g., dehydration or starvation during active or passive immigration; 23 24 Calder, 1984; Lomolino, 1988, 1989). Therefore, for many faunal groups, 24 25 resource requirements and immigration abilities may be positively correlated 25 26 – better immigrators may require more resources to maintain their insular 26 27 populations. That is, the intercept and slope of the IDF should be inversely 27 28 correlated. This pattern of covariation, however, is not likely to be a rule for 28 29 all species groups. Indeed, some faunal groups, such as micro-snails that 29 30 depend on aerial dispersal for long-distance colonization (see Vagvolgyi, 30 31 1975), may exhibit the opposite pattern; i.e., good immigrators may be small 31 32 and thus require fewer resources to maintain insular populations. This is 32 33 certainly an area where more empirical studies are sorely needed. 33 34 34 35 35 36 Assembly and nestedness of insular communities 36 37 As illustrated in Fig. 10.8, two alternative patterns of covariation have very 37 38 different implications with respect to assembly of insular communities 38 39 (Diamond, 1975). Here assembly ‘rules’ are defined as patterns in how species 39 40 composition varies along biogeographical gradients (e.g., increasing area or 40 284 M.V. Lomolino 1 isolation) and gradients in species richness. Positive covariation of slopes and 1 2 intercepts (i.e., where better immigrators require fewer resources to maintain 2 3 their populations) implies a remarkably simple and orderly pattern of assem- 3 4 bly (species E first, then F, then G; Fig. 10.8(b). In contrast, where better 4 5 immigrators tend to require more resources, this relatively simple model pre- 5 6 dicts a diverse collection of communities. Indeed, depending on which ranges 6 7 in area and isolation are sampled, different groups of these hypothetical 7 8 islands are expected to be inhabited by almost all possible combinations of 8 9 species (species A, or B, or C; A and B, B and C, or all three species; Fig. 9 10 10.8(a). However, within the limited biogeographical space included in most 10 11 surveys, even these species may exhibit simple patterns of assembly. For 11 12 example, if sampling was limited to just the larger islands of Fig. 10.8(a), 12 13 species accumulation sequences (with decreasing isolation) should always be 13 14 C first, then B, and then C. 14 15 In summary, the species-based, hierarchical model predicts the following 15 16 assembly rules for insular faunas. 16 17 17 (a) Where better immigrators tend to require fewer resources, regardless of 18 18 the biogeographic space sampled, communities should accumulate species 19 19 in order of increasing resource demands and decreasing immigration 20 20 abilities (i.e., first species E, then F, then G; Fig. 10.8(b). 21 21 (b) On the other hand, where better immigrators tend to require more 22 22 resources (Fig. 10.8(a), accumulation sequences will vary depending on 23 23 the range in area and isolation sampled (e.g., first A, then B, then C on 24 24 the near islands; first C, then B, then A on the large islands). 25 25 26 Therefore, the species-based model provides a relatively simple explanation 26 27 for what may otherwise seem perplexing and contradictory results. Where 27 28 better immigrators tend to require more resources to maintain their popula- 28 29 tions, the same species pool may exhibit contradictory patterns of assembly 29 30 for surveys conducted in different archipelagoes, or different regions of the 30 31 same archipelago. Only when the results of these studies are combined, does 31 32 the ambiguity clear to reveal the combined importance of immigration and 32 33 extinction at the community level. Assembly ‘rules’ should differ, but in a 33 34 manner consistent with the resource requirements and immigration abilities 34 35 of the pool of focal species. 35 36 The alternative forms of covariation in resource requirements and immi- 36 37 gration abilities also have important implications with respect to nestedness 37 38 of insular communities. Nestedness, as first described by Darlington (1957) and 38 39 later developed by Patterson and Atmar (1986), refers to the tendency for 39 40 more depauperate communities to form proper subsets of richer communities 40 A biogeographical model 285 1 (Fig. 10.9; see also Patterson & Brown, 1991). Alternatively, nestedness may 1 2 be viewed as the tendency for communities on smaller or more isolated islands 2 3 to form proper subsets of those on larger or less isolated islands. 3 4 Nestedness of insular communities can result from differential immigration 4 5 or differential extinction (Fig. 10.9). That is, nestedness implies that immigra- 5 6 tion and extinction vary in an orderly manner among islands and among species. 6 7 Of course, both of these fundamental biogeographic processes, immigration 7 8 and extinction, may contribute to nestedness. If, however, slopes and inter- 8 9 cepts of IDFs exhibit negative covariation among species (good immigrators 9 10 tend to be poor survivors), then immigration and extinction can have con- 10 11 founding effects on nestedness. This should be obvious from inspection of 11 12 Fig. 10.8(a). Sampling islands along gradients of isolation or area can be sim- 12 13 ulated by moving along horizontal or vertical lines within the bivariate space. 13 14 Perfect nestedness occurs where, as we move along a gradient of decreasing 14 15 area or increasing isolation, once a species is omitted, it remains absent from 15 16 all smaller or more isolated communities. More simply put, archipelagoes 16 17 comprised of perfectly nested communities are those where their species’ 17 18 IDFs do not intersect (intercepts and slopes exhibit positive covariation). 18 19 Therefore, perfect or near perfect nestedness is only expected where better 19 20 immigrators require fewer resources (smaller islands) to maintain their insular 20 21 populations (Fig. 10.8(b), or where biogeographical surveys include only a 21 22 limited range in area or isolation (e.g., just the near islands of Fig. 10.8(a). 22 23 23 24 24 25 Patterns in species richness 25 26 Perhaps more than any other ecological pattern, the species–area relationship 26 27 is of fundamental importance to development of island biogeography theory 27 28 (Fig. 10.10). Indeed, MacArthur and Wilson (1967:8–9) wrote that ‘theories, 28 29 like islands, are often reached by stepping stones. The species–area curves 29 30 are such stepping stones’. The primary objective of MacArthur and Wilson’s 30 31 theory, of course, was to explain the general tendencies for species richness 31 32 to be higher on larger and less-isolated islands. The model presented here, 32 33 while species based, provides an alternative explanation for patterns in species 33 34 richness. Given its hierarchical nature, and provided appropriate linkage 34 35 between species – and community-levels, this model should accurately pre- 35 36 dict the general form of the species–area and species–isolation relationships. 36 37 37 38 The species–area relationship 38 39 First note that the models presented in Fig. 10.8 predict patterns in richness 39 40 as well as composition and nestedness. These figures, however, fall short of 40 286 M.V. Lomolino 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 Fig. 10.9. (a) Darlington’s (1957: 485) ‘diagram to compare a simple immigrant 27 28 pattern, formed by dispersal of several groups of animals from the mainland for 28 different distances along a series of islands, and a relict pattern, formed by partial 29 extinction of an old fauna formerly common to all the islands’. Darlington’s immi- 29 30 grant pattern assumes that the species differ in their immigration abilities (A being 30 31 the best, D being the least vagile immigrator), while his relict pattern assumes that 31 32 species are equivalent (i.e., extinction is not selective). (b) Pattern of nestedness 32 expected for faunas where insular community structure is determined solely by selec- 33 tive extinctions and interspecific differences in resource requirements (resource 33 34 requirements are highest for species Q and lowest for species M; see Patterson & 34 35 Atmar, 1986; Patterson & Brown, 1991). 35 36 36 37 37 38 38 39 39 40 40 A biogeographical model 287 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 Fig. 10.10. Alternative forms of the species–area relationship. The sigmoidal form is 16 based on observed distributions of body size and on bioenergetic considerations (see 17 text), and is consistent with the small island effect (i.e., a tendency for species rich- 17

18 ness to be independent of area for the very small islands; i.e., those smaller than Amin 18 19 – see Niering, 1963). Inset depicts the frequency distribution of resource (area require- 19 ments) among species (based on the assumption that resource requirements increase 20 in proportion to body size). 20 21 21 22 22 23 describing the form of the species-area and species-isolation relationships. To 23 24 accomplish this, it is necessary to assess how the intercept and slope of the 24 25 IDFs are distributed across the species pool. That is, all else being equal, 25 26 species–area curves are determined by the frequency distribution of the IDF 26 27 intercepts, which, again, are measures of resource levels required to establish 27 28 and maintain insular populations. If the intercepts are uniformly distributed 28 29 among species, then the species–area relationship should be linear. At every 29 30 increment in area, another species should be added. Of course, it is known 30 31 that this relationship is less than linear (Fig. 10.10). This suggests that the 31 32 frequency distribution of IDF intercepts is not uniform, but skewed such that 32 33 most species have relatively low resource requirements. 33 34 Alternatively, inductive means can be used to approach this question, again 34 35 drawing on the rich literature in physiological ecology. Resource requirements 35 36 of animals (e.g., energy, water, nutrients and home range size) are positively 36 37 correlated with body size among species. For most animal taxa, however, the 37 38 frequency distribution of body size is strongly skewed, with many more 38 39 smaller than larger species (Brown, 1995; Calder, 1984). Thus, as area 39 40 increases, the rate at which an island exceeds the resource requirements of 40 288 M.V. Lomolino

1 additional species should decrease, yielding the general form of the species- 1 2 area relationship (Fig. 10.10). If pressed, allometric equations could be manip- 2 3 ulated to derive the parameters, or at least an exponent for one expression of 3 4 the species–area relationship – the power model (which as Gould, 1979 4 5 pointed out is an allometric formula). This exponent, often referred as z, is a 5 6 measure of the how the slope of the species-area relationship changes (typi- 6 7 cally declines) with increasing area. As z-values decrease from 1.0 to 0.0, the 7 8 slope of the species–area relationship levels off more rapidly (see Lomolino, 8 9 1989; Rosenzweig, 1995). 9 10 Insights from physiological ecology are again drawn on to predict z-val- 10 11 ues, here using mass as the common currency. To answer the question of how 11 12 species should accumulate with increasing area, the question is ﬁrst asked 12

13 how species accumulate with increasing mass. That is, how does Ss, the 13 14 number of smaller species, scale with body mass. Brown and Nicolleto (1991) 14 15 provide some useful data here. Re-analyzing their data on body mass of North 15

16 American land mammals, we find Ss scales with mass raised to the 0.234 16 17 power (r2 = 0.76). To convert this species–mass to a species–area accumula- 17 18 tion function, the question is now asked how resource requirements (RR) 18 19 scales with mass. Jurgens and Prothero (1991) report that lifetime RR of mam- 19 20 mals increases as a function of mass raised to the 0.87 to 0.93 power (for 20 21 birds the exponent ranges from 0.88 to 0.94, which is not significantly dif- 21 22 ferent from 1.0; Jurgens & Prothero, 1991). Home range size of different 22 23 mammals scales as an approximately linear function of mass (see Lindstedt 23 24 et al., 1986; see also Harestad & Bunnell, 1979; Schoener, 1968; Calder, 24 25 1984; Peters, 1983). Assuming that resource requirements of insular popula- 25 26 tions increase in a linear fashion with home range size (or roughly propor- 26 27 tional to life-time energy budgets), and that the level of insular resources (bio- 27 28 mass and productivity) increases as a linear function of island area, area 28 29 requirements (A) can be substituted for mass (M) to obtain the following. 29 30 30 S ∝ M0.23 (again, after Brown & Nicolleto, 1991) 31 s 31 therefore, S ∝ Area0.23 32 s 32 33 These results are eerily close to Preston’s (1962) canonical value (z = 0.26) 33 34 for the species–area relationship. This similarity may be more than just a 34 35 coincidence, given that frequency distributions of body size (the basis of the 35 36 above derivation) and abundances (the basis of Preston’s method) may be 36 37 causally related: simply put, larger species tend to be more rare. However, it 37 38 would be foolhardy to make much of the precise values obtained above. The 38 39 scaling relationships used to generate these results will vary somewhat among 39 40 taxonomic groups. Similarly, the assumption that insular resources vary as a 40 A biogeographical model 289 1 linear function of area, while intuitively appealing, remains largely untested. 1 2 Given this, z-values may vary considerably among biotas. The quagmire of 2 3 predicting how z-values should vary among specific groups or regions will 3 4 be avoided. However, because the scaling relationships (i.e., exponents in 4 5 allometric equations) tend to be conservative, the model provides an alterna- 5 6 tive, species-based explanation for the general tendency for z-values to be 6 7 conservative, ranging from 0.1 to .5 for most biotas. 7 8 8 9 The species–isolation relationship 9 10 In comparison to the species–area relationship, which is an accumulative func- 10 11 tion, the species–isolation relationship can be viewed as one of attenuation. 11 12 How does species richness of insular communities attenuate as isolation 12 13 increases? Of the two patterns, the species area-relationship has received the 13 14 lion’s share of attention. The species–isolation relationship, in contrast, seems 14 15 like a neglected sibling. The author is unaware of any debates over its general 15 16 form, let alone controversies over the meaning of precise, ‘canonical’ values 16 17 of its exponents or coefficients. 17 18 MacArthur and Wilson (1967: 125–128) predicted that the species–isola- 18 19 tion relationship would take one of two forms, depending on the nature of 19 20 immigration. If immigration is passive with constant directionality, such as 20 21 that for wind-dispersed, terrestrial propagules, then immigration rate (num- 21 22 ber of propagules reaching an island per time) should be a negative expo- 22 23 nential function of isolation (i.e., immigration should be proportional to e−I). 23 24 On the other hand, MacArthur and Wilson predicted that, if immigration was 24 25 active, or passive but on logs or other ‘rafts’ with normally distributed per- 25 26 sistence times, then immigration should be a normal function of isolation (i.e., 26 27 immigration should be proportional to e−(I × I )). 27 28 Observed species–isolation relationships, when significant, tend to be 28 29 consistent with one of these predictions. MacArthur and Wilson’s (1967) rea- 29 30 soning, however, may be problematic because the equilibrium model treated 30 31 the biota as a collection of homogeneous species. On the contrary, even for 31 32 taxonomically or ecologically similar faunas, immigration abilities are likely 32 33 to vary markedly among species. Assuming that this is the more logical 33 34 alternative, the form of the species–isolation relationship can be derived by 34 35 asking how immigration abilities should be distributed among species. 35 36 The following argument is analogous to that used to derive the form of the 36 37 IDF. Immigration abilities (passive or active) of most species will probably 37

38 exceed some minimal distance (Dnear of Fig. 10.11). This distance depends on 38 39 the physiological and behavioral characteristics of the species group in ques- 39 40 tion and the nature of the immigration ﬁlter (sensu Simpson, 1940; see also 40 290 M.V. Lomolino 1 next section on inter-archipelago comparisons). Beyond this distance, rich- 1 2 ness should decline at a rate determined by the distribution of immigration 2 3 abilities among species. Thus, we need to know how the slopes of the IDFs 3 4 are distributed among species. If they were uniformly distributed, then, as 4 5 isolation increased, species richness would decline as a linear function of iso- 5 6 lation. Again, it is known that this relationship tends to be non-linear. Based 6 7 on physiological considerations, the frequency distribution of immigration 7 8 abilities will probably be log-normal or log-skewed, with most species exhibit- 8 9 ing relatively limited vagilities. 9 10 The above prediction is based on the assumption that immigration abilities 10 11 of animals, like many species traits, are correlated with body size, which tends 11 12 to be distributed as a log-normal or log-skewed function (Brown, 1995; Peters, 12 13 1983). That is, for a great diversity of taxonomic groups, most individuals 13 14 tend to be of relatively small size. The species–isolation relationships should 14 15 therefore, approximate a negative sigmoidal function (Fig. 10.11). Species 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 Fig. 10.11. Species–isolation curve derived from physiological considerations (see 34 35 text) and frequency distributions of body size for species of a given faunal group. 35 Species richness should remain relatively high until isolation exceeds the immigra- 36 36 tion ability of the least vagile species (Dnear, which marks the end of region a). Beyond 37 this distance (region b), species richness should decline rapidly (see Fig. 10.12). 37 38 Finally, species richness should remain relatively low and appear independent of iso- 38 lation for the most distant islands (i.e., those in region c). Inset depicts the frequency 39 distribution of immigration abilities (distances) among species (based on assumption 39 40 that immigration abilities are correlated with body size). 40 A biogeographical model 291 1 richness should remain high until isolation exceeds the immigration capacity 1

2 (Dnear) of the least vagile species. Beyond this minimal critical isolation, 2 3 species richness should decline rapidly until isolation approximates the immi- 3 4 gration range of the modal species. As isolation increases beyond this point, 4 5 richness should asymptotically approach zero. It would be unwise to make 5 6 any specific predictions regarding ‘canonical’ exponents or coefficients of this 6 7 relationship. Indeed, in comparison to patterns for active immigrators, passive 7 8 immigrators as a group are likely to have very different physiological rela- 8 9 tionships, immigration functions and species–isolationship curves. However, 9 10 because propagule dispersal tends to be log-normally distributed across dis- 10 11 tance (with most landing near the source), species–isolation relationships for 11 12 these groups should still approximate a negative sigmoidal in arithmetic space 12 13 (Fig. 10.11). 13 14 In short, there is much to be learned about immigration. At least one 14 15 important inference, however, can be drawn from the expected pattern. The 15 16 sigmoidal nature of the immigration function may, at least in part, accounts 16 17 for the relative paucity of well-documented isolation effects. Biogeographers 17 18 are unlikely to detect the effects of isolation unless their studies encompass 18 19 a broad range in isolation. While most studies span three or four orders of 19 20 magnitude variation in area, they seldom include islands that vary more than 20 21 an order of magnitude in isolation. If these islands are all relatively close or 21 22 all distant (i.e., falling within the ranges of isolation where the s-i slope is 22 23 near zero, regions ‘a’ and ‘c’ in Fig. 10.11, inset), then isolation effects may 23 24 be difficult to detect. Aside from this sampling artifact, most species–isola- 24 25 tion relationships should belong to the same family of sigmoidal curves whose 25 26 slopes vary as a function of vagilities of the species and nature of the immi- 26 27 gration filter. On pp. 293–295, it will be considered how species–isolation 27 28 and species–area curves should vary among taxa and among archipelagoes. 28 29 29 30 Patterns in species richness: summary 30 31 The species-based model presented here provides a relatively simple expla- 31 32 nation for the species–area and species–isolation relationships. Unlike the 32 33 equilibrium theory, this model does not assume a balance of immigrations 33 34 and extinction, but it does assume that species differ and in an orderly pattern. 34 35 That is, the forms of species–area and species–isolation curves may derive 35 36 from the general tendency for body size and related physiological and eco- 36 37 logical characteristics to be distributed as log-normal or log-skewed functions. 37 38 Given this and the assumptions, based on bioenergetic considerations, that 38 39 resource requirements and immigration abilities are correlated with body size, 39 40 then the species–area relationship should take the form of a positive sigmoidal, 40 292 M.V. Lomolino 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 Fig. 10.12. Body sizes of species within a faunal group tend to be log-normally or 17 log-skewed, with most species exhibiting relatively small body size (solid line; see 17 18 Brown, 1995). Consequently, accumulation functions of the number of species with 18 19 larger or smaller size tend to be sigmoidal functions (dashed and dotted lines). Because 19 resource requirements and immigration abilities may be correlated with body size, 20 these accumulation functions may have important implications for understanding pat- 20 21 21 terns in species richness of insular faunas (Spool = total number of species in the source, 22 or mainland, pool). 22 23 23 24 tracking the cumulative number of smaller (less resource intensive) species, 24 25 while the species–isolation relationship should be a negative sigmoidal, track- 25 26 ing the number of larger (more vagile) species (Fig. 10.12). 26 27 The possible sigmoidal form of the species–area relationship is seldom dis- 27 28 cussed. The conventional models (power and semi-log models) typically pro- 28 29 vide a good approximation of the species–area relationship of most insular 29 30 biotas. Indeed, the power and semi-log models are qualitatively similar to the 30 31 sigmoidal model (Fig. 10.10), with one exception. For all three models, rich- 31 32 ness increases rapidly for islands of intermediate area, then more slowly and, 32 33 perhaps, imperceptibly for the larger islands. However, only the sigmoidal 33 34 model predicts that richness increases slowly for the very small islands as 34 35 well. Thus, if biogeographic surveys are limited to smaller islands (i.e., those 35 36 too small to satisfy resource requirements of most species), species richness 36 37 may appear to be independent of area. This phenomenon, termed the ‘small 37 38 island effect’, has only occasionally been observed for some insular biotas 38 39 (Niering, 1963; see also Dunn & Loehle, 1988; MacArthur & Wilson, 39 40 1967:30–33; Wiens, 1962; Whitehead & Jones, 1969; Woodroffe, 1986). It 40 A biogeographical model 293 1 is likely to remain a rarely reported phenomenon as most biogeographic 1 2 surveys are designed to include islands large enough to satisfy resource 2 3 requirements of at least a few focal species. Given this, and the ease at which 3 4 species–area curves can be estimated using power or semi-log models, it is 4 5 unlikely that biogeographers will switch to a sigmoidal model. Again, these 5 6 more conventional models typically provide a good approximation of the 6 7 species–area relationship. However, where surveys encompass a relatively 7 8 broad range in area and include many small islands, a sigmoidal model may 8 9 prove superior to conventional models. 9 10 10 11 11 12 The inter-taxa and inter-archipelago scales 12 13 An additional tier can be added to the hierarchical model by noting how 13 14 species-groups vary in ways affecting the fundamental biogeographic 14 15 processes of immigration and extinction. Biogeographical studies often focus 15 16 on groups composed of taxonomically and functionally similar species (e.g., 16 17 amphibians, freshwater fish or passerine birds). Thus, focal groups differ in 17 18 immigration abilities and resource demands and, more importantly, these 18 19 differences are predictable. For example, birds and bats tend to be more vagile 19 20 than non-volant mammals and amphibians. Focal groups comprised of 20 21 endothermic, large, eusocial or carnivorous species should require more 21 22 resources (larger islands) than groups composed of ectothermic, small, asocial 22 23 or herbivorous species. 23 24 These differences among groups of species can be modelled as shifts in 24 25 IDFs. That is, the prediction is that within an archipelago, IDFs should be 25 26 shifted to the right (reduced slopes) for focal groups composed of more vagile 26 27 species (Fig. 10.13(a)). Similarly, the prediction is a downward shift in IDFs 27 28 (reduced intercepts) for faunal groups composed of species with relatively 28 29 low resource demands (small, ectothermic, and⁄or herbivorous species; Fig. 29 30 10.13(b)). 30 31 Alternatively, species or community level patterns may be compared across 31 32 archipelagoes instead of across taxonomic groups. To model the effects of 32 33 differences among archipelagoes, we must first assess the relative productivity 33 34 and filter severity of the different archipelagoes. Here, relative productivity 34 35 refers to the ability of insular ecosystems to support populations of the focal 35 36 species group (see applications of the species-energy theory to island bio- 36 37 geography, Wright, 1983; Wylie & Currie, 1993). By filter severity the 37 38 reference is to impedance to immigration, which should be associated with 38 39 the nature of the isolating medium or matrix. For example, Aberg and his 39 40 colleagues have shown that the distribution of the hazel grouse (Bonasa 40 294 M.V. Lomolino 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 Fig. 10.13. Effects of differences in characteristics of faunal groups (e.g., birds vs. 23 24 non-volant mammals) and differences among archipelagoes on distribution patterns 24 of insular faunas. The curves depict insular distribution functions (IDFs) of repre- 25 sentative species for a given faunal group. (a) IDFs should shift leftward (increased 25 26 slope) in archipelagoes with relatively strong (more isolating) filters or, equivalently, 26 27 for faunal groups composed of relatively poor immigrators. On the other hand, IDFs 27 should shift to the right in archipelagoes with relatively weak filters or for faunal 28 groups composed of relatively good immigrators. (b) IDFs should shift upward (higher 28 29 intercept) in archipelagoes with higher relative productivity (RP) or, equivalently, for 29 30 faunal groups composed of species with relatively high resource requirements (RR; 30 31 e.g., endotherms vs. ectotherms, carnivores versus herbivores). Relative productivity 31 refers to the ability of insular ecosystems to support populations of the focal species, 32 and should be a function of primary productivity and trophic characteristics of the 32 33 faunal group. 33 34 34 35 bonasia) among fragmented forest of Sweden is strongly influenced by char- 35 36 acteristics of isolating habitats (Aberg et al., 1995). In landscapes dominated 36 37 by agricultural fields, populations of these birds evidence a strong and highly 37 38 significant effect of isolation. In contrast, the effect of geographic isolation 38 39 on grouse populations surrounded by more suitable second growth forests 39 40 was only marginally significant (Fig. 10.14). 40 A biogeographical model 295 1 In summary, the prediction is shifts in IDFs among archipelagoes equiv- 1 2 alent to those associated with differences among focal groups; downward 2 3 shifts for archipelagoes with higher relative productivities (equivalent to 3 4 species groups with lower resource demands), and leftward shifts in IDFs for 4 5 archipelagoes with more severe immigration filters (or species groups with 5 6 lower vagilities; Fig. 10.13). Where the isolating matrix is so severe that it 6 7 forms a barrier to immigration for most species, we then expect a relaxation 7 8 fauna, with most species limited to the relatively large islands and little evi- 8 9 dence of isolation effects (see Fig. 10.7). Again, these predictions point to 9 10 some potentially fruitful directions for future research, which would focus on 10 11 the key linkages between community and archipelago (or inter-taxon) scales. 11 12 For example, do interarchipelago differences in productivity and filter sever- 12 13 ity covary in nature? That is, do archipelagoes composed of islands with rel- 13 14 atively high productivity also tend to be situated in regions with less isolat- 14 15 ing filters? Do faunal groups composed of relatively good immigrators (e.g., 15 16 birds versus amphibians; bats vs. non-volant mammals) also tend to be com- 16 17 posed of species with high resource demands? Research addressing these 17 18 questions should provide some important insights into the forces influencing 18 19 biological variation over a broad range of geographical scales. Such research, 19 20 however, will be especially challenging as it requires fundamental changes 20 21 in the nature and scale of ecological research – changes that call for increased 21 22 emphasis on processes operating at large spatial scales (see Edwards et al., 22 23 1994; Brown, 1995). 23 24 24 25 25 26 Species interactions and realized insular distributions 26 27 The hierarchical model presented above accounts for a diversity of biogeo- 27 28 graphic patterns across a range of spatial and ecological levels. There are 28 29 other interesting ecological patterns, however, that cannot be accounted for 29 30 by the present version of the model. Checkerboards, or more generally, exclu- 30 31 sive distributions of species has interested biogeographers and community 31 32 ecologists for at least the past three decades (e.g., see Diamond, 1975; chapters 32 33 in this volume). There remains much controversy over the effects of inter- 33 34 specific interactions at relatively coarse, geographic ranges. At a local or 34 35 within-island scale, however, many studies indicate that distributions among 35 36 habitats are strongly influenced by interspecific interactions (e.g., see Connell, 36 37 1961; Paine, 1966; Grant, 1972; Menge, 1972; Lubchenco, 1978; Werner, 37 38 1979; Munger & Brown, 1981; Schoener, 1983; Brown & Gibson, 1983; 38 39 Lomolino, 1984). The question remains whether interspecific interactions also 39 40 influence species distributions at coarser scales (e.g., distributions of insular 40 296 M.V. Lomolino 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 Fig. 10.14. (a) Effects of differences in immigration filters on insular distribution func- 35 36 tions (IDFs) of Hazel grouse (Bonasa bonasia) in fragmented forest of Sweden (Aberg 36 37 et al., 1995). As predicted by the model of Fig. 10.13, the IDF for this species shifted 37 38 leftward (increased slope) when forests were surrounded by less hospitable habitats 38 (agricultural fields [triangles] vs. second growth, managed forest (circles); presence 39 and absence is indicated by filled and unfilled symbols, respectively). (b) Differences 39 40 in filter strength should be evidenced by shifts in species richness curves. As pre- 40 A biogeographical model 297 1 populations among islands; see Simberloff & Connor, 1981; Connor & 1 2 Simberloff, 1979). 2 3 While the species-based model in its present form does not include the 3 4 potential influence of interspecific interactions, it does describe the biogeo- 4 5 graphical space (ranges of isolation and area) over which those interactions 5 6 may occur. Borrowing from niche theory, this biogeographic space can be 6 7 referred to as the fundamental (vs. realized) distribution. If interspecific inter- 7 8 actions are intense enough to reduce persistence of insular populations then, 8 9 at least for some species, realized insular distributions may be substantially 9 10 less than their fundamental distributions. 10 11 For now, this exercise can be simplified by focusing on just two hypo- 11 12 thetical species. To predict the effects of interspecific interactions it first is 12 13 necessary to define the fundamental insular distributions of these species. 13 14 Specifically, how do these species differ in vagilities and resource require- 14 15 ments? Second, it is necessary to consider the nature of the interaction; pre- 15 16 dation, amensalism, commensalism, parasitism, mutualism or competition? In 16 17 the latter case, is the competitive interaction symmetrical, or does one species 17 18 tend to dominate in interference or exploitative interactions? 18 19 Rather than consider all possible permutations of IDFs and interspecific 19 20 interactions, this exercise will be further simplified by limiting it to the poten- 20 21 tial effects of asymmetrical competition. Figure 10.15 illustrates the IDFs of 21 22 two hypothetical species with asymmetrical competition. Assume that these 22 23 interactions are strong enough to result in competitive exclusion from islands 23 24 where the fundamental IDFs overlap. Also assume that species B is a better 24 25 immigrator than A. Regardless of which species dominates, we expect exclu- 25 26 sive distributions on all but the largest, mainland-like islands, which should 26 27 afford ecological refugia for inferior competitors. If B dominates, then the 27 28 competitively inferior species (A) should be limited to the very small, near 28 29 islands and the very large islands. On the other hand, if A is the dominant 29 30 competitor, then species B, the more vagile species, should be limited to the 30 31 distant islands (and again, the very large, mainland-like islands as well). 31 32 32 33 33 34 34 35 dicted by the species-based model, species richness of non-volant mammals on islands 35 36 of the Great Lakes Region of North America declines more rapidly for islands in arch- 36 37 ipelagoes with more severe (less hospitable) immigration filters (for these mammals 37 38 which either swim or colonize islands by traveling across ice, filter severity is high- 38 est for the lotic archipelagos, intermediate for the coastal marine archipelago, and low- 39 est for the lacustrine archipelago, which is characterized by relatively weak currents 39 40 and more prolonged periods of ice-cover; see Lomolino, 1994). 40 298 M.V. Lomolino 1 ()a Large island refugia 1 for species A A 2 B 2 3 3 4 4 nly 5 B, o 5 6 Species 6 7 7 8 8 9 9 Species A, only

10 Island area (refugia on small islands) 10 11 11 12 12 13 13 14 14 15 Island isolation 15 16 16 17 ()b Large island refugia 17 18 for species B A B 18 19 19 20 ly 20 n 21 o 21 , 22 A 22

23 s 23 24 ie 24 c Species B, only 25 e 25 Sp (refugia on distant islands) 26 26 Island area 27 27 28 28 29 29 30 30 31 Island isolation 31 32 Fig. 10.15. Effects of interspecific interactions on realized insular distribution func- 32 33 tions of two hypothetical species. (a) Here the better immigrator, species B, tends to 33 34 dominate interspecific interactions, but requires larger islands to maintain its popula- 34 35 tions. As a result, species B is widely distributed across the archipelago, while pop- 35 ulations of species A (the ecologically subordinate and less vagile species) should be 36 restricted to relatively small islands (i.e., those too small to maintain populations of 36 37 species B) or the relatively large islands (i.e., those large enough so that they afford 37 38 ecological refugia for species A). (b) Here the better immigrator, species B, is eco- 38 logically subordinant (i.e., A is the dominant competitor or predator) and requires 39 larger islands to maintain its populations. As a result, species A (the relatively poor 39 40 immigrator) is restricted to the less-isolated islands, while populations of species B 40 A biogeographical model 299 1 1 2 Case studies: effects of competition and predation on IDFs 2 3 Although examination of the archipelago scale effects of interspecific inter- 3 4 actions has only just begun, there is some evidence that isolated habitats often 4 5 serve as ecological refugia. In his recent analysis of distributions of fresh- 5 6 water fish among pools of the Red River in Oklahoma, Taylor (1997) reports 6 7 that distributions of small mouth bass (Micropterus dolomieu) and creek chub 7 8 (Semotilus atromaculatus) appear to be strongly affected by asymmetrical 8 9 competition (with bass being competitively dominant). The distributions of these 9 10 two species are nearly exclusive, with the exception of one co-occurrence in 10 11 a relatively large, near pool (Fig. 10.16). 11 12 The effects of predation on insular distributions may be similar to that of 12 13 intense asymmetrical competition, and here again there is some empirical 13 14 evidence that distant islands serve as refugia. Meadow voles and deermice 14 15 on islands of Lakes Huron and Michigan are largely restricted to the more 15 16 isolated islands, apparently as a result of predation by shrews, which are 16 17 smaller and less vagile than their vertebrate prey (Fig. 10.17; see also 17 18 Lomolino, 1984, 1989). These species also co-occur on near islands if they 18 19 are relatively large. Experimental introductions of shrews (predators on young 19 20 mice) onto islands of another archipelago of the Great Lakes Region demon- 20 21 strated that predation by short-tailed shrews (Blarina brevicauda) may 21 22 increase extinction rates of at least some insular rodents (Lomolino, 1984). 22 23 Taking this a step further, after these prey are extirpated, predators may go 23 24 extinct, especially on small islands where alternative resources are scarce. 24 25 Thus, communities on small islands within the immigration range of shrews 25 26 may undergo cycles of colonization by prey, colonization by predator, extinc- 26 27 tion of prey followed by extinction of predator and reinitiation of the cycle. 27 28 Unless there is some factor causing synchronization of this ecological cycle 28 29 among islands, a temporally dynamic checkerboard pattern may be expected 29 30 on islands of intermediate isolation – each in different phases of the cycle 30 31 (e.g., see Fig. 10.17, islands from 0.5 to 5 km isolation). More generally, 31 32 where fundamental ranges overlap, competition and predation may alter 32 33 assembly sequences of insular biotas, increasing the number of ‘holes’ or 33 34 departures from perfect nestedness. 34 35 As a final case study, consider Jared Diamond’s (1975) seminal paper on 35 36 assembly rules and bird communities of New Guinea satellite islands. The 36 37 37 38 38 should be restricted to the more isolated islands (i.e., those lacking A) or the rela- 39 tively large islands (i.e., those large enough so that they afford ecological refugia for 39 40 species B). 40 300 M.V. Lomolino 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 Fig. 10.16. Distribution functions of two species of freshwater fish among pools of 16 17 the Red River in Oklahoma (Taylor, 1997). Small mouth bass (Micropterus dolomieu) 17 appear to dominate in interspecific interactions with the creek chub (Semotilus 18 atromaculatus), but appear more limited in its dispersal abilities. As a result, the 18 19 distributions of these species is nearly exclusive, with populations of small mouth bass 19 20 being restricted to the less-isolated pools, while all but one population of the creek 20 chub are restricted to the most-isolated pools. The one exception is the occurrence of 21 both species on a relatively large, less-isolated pool (i.e., one large and near enough to 21 22 maintain populations of creek chubs despite the effects of competition and possible 22 23 predation from small mouth bass). 23 24 24 25 25 26 purpose of that paper was to develop the concept of alternative stable com- 26 27 munities, which as Diamond observed, ‘was brilliantly explored by Robert 27 28 MacArthur (1972) in Geographical Ecology’. Building on MacArthur’s 28 29 theoretical discussion, Diamond developed a set of empirical procedures, 29 30 including incidence functions and analysis of checkerboard patterns, to inves- 30 31 tigate how community structure is influenced by characteristics of the islands, 31 32 by species interactions or by chance (see also Connor & Simberloff, 1979, 32 33 1983; Simberloff & Connor, 1981; Diamond & Gilpin, 1982; Gilpin & 33 34 Diamond, 1982, 1984). The purpose here is to complement Diamond’s work 34 35 by providing an explicit spatial reference for checkerboard patterns. This may 35 36 be accomplished by plotting distributions of guild members across biogeo- 36 37 graphical space (i.e., area–isolation plots). 37 38 There are at least two, fundamentally different checkerboard patterns. On 38 39 the one hand, if distributions of ecologically similar species were influenced 39 40 by intense competition, but not by island characteristics (e.g., area and 40 A biogeographical model 301 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 Fig. 10.17. Insular distribution patterns of non-volant mammals on islands of two 28 29 lacustrine archipelagoes of Lake Huron and Lake Michigan (S = shrews [Blarina bre- 29 30 vicauda and Sorex cinereus], P = deermice (Peromyscus maniculatus) and M = 30 meadow voles (Microtus pennsylvanicus)). Because shrews tend to be relatively poor 31 immigrators, their populations are generally restricted to the less-isolated islands. By 31 32 preying on meadow voles and deermice (especially juveniles), shrews may be able to 32 33 exclude populations of these rodents from relatively small, less-isolated islands. On 33 34 the other hand, the ecologically subordinant but more vagile deermice and meadow 34 voles may find refugia on the more isolated or larger islands. 35 35 36 36 37 isolation), then we would expect species to be exclusively, but uniformly dis- 37 38 tributed across biogeographical space (Fig. 10.18(a)). On the other hand, if 38 39 distributions are influenced by competition as well as by differences in vagili- 39 40 ties and resource requirements, then a complex set of insular distributions is 40 302 M.V. Lomolino 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 A biogeographical model 303 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 Fig. 10.18. (a) A spatial checkboard distribution of two hypothetical species. In this 14 case, both species are uniformly distributed across the archipelago (i.e., their distrib- 14 15 utions are independent of isolation and area), but they never occupy the same island. 15 16 This pattern would be expected where competition is intense, but species are essen- 16 tially equivalent with respect to immigration abilities and resource requirements. 17 (b) thru (d) In contrast to the above, hypothetical pattern, analysis of insular distrib- 17 18 ution patterns for three of Diamond’s (1975) avian guilds reveals that exclusive 18 19 distribuitons are often achieved by species segregating their realized distributions over 19 20 different ranges of isolation and area (Perault and Lomolino, unpublished report, 1996; 20 see also Figs. 10.16 and 10.17). Regression analysis was used to estimate linear insu- 21 lar distribution functions of these species (see Lomolino, 1986). Insular distributions 21 22 were significantly (P < 0.05) associated with island area and isolation for Ptillinopus 22 23 rivoli, P. solomenensis, Macropygia nigrirostris, and Pachcephala pectoralis. Insular 23 distributions were significantly associated with just area for P. melanura, and 24 marginally (P < 0.09) associated with area for M. mackinlayi. 24 25 25 26 26 27 expected similar to those illustrated in Fig. 10.18(b) to (d). In such cases, 27 28 competitors may be restricted to only a portion of their fundamental distrib- 28 29 ution (i.e., biogeographic refugia). 29 30 Examination of insular distributions of three of Diamond’s avian guilds 30 31 has taken place: the fruit-pigeon, cuckoo-dove and gleaning flycatcher guilds 31 32 (D. Perault and M.V. Lomolino, unpublished data). At first glance, distribu- 32 33 tions of members of the fruit-pigeon guild (Ptillinopus solomenensis and P. 33 34 rivoli; Fig. 10.18(b) may seem consistent with non-interactive, species-based 34 35 model. However, as the arrows in Fig. 10.18(b) indicate, the distribution of 35 36 P. rivoli is biased toward large, near islands, while that of P. solomenensis 36 37 is biased toward small, distant islands. For the cuckoo-dove guild (Fig. 37 38 10.18(c)), Macropygia nigrirostris occurs on near and large islands, while M. 38 39 mackinlayi tends to be restricted to smaller islands. Finally, Pachycephala 39 40 pectoralis of the gleaning flycatcher guild (Fig. 10.18(d)) occurs on the largest 40 304 M.V. Lomolino 1 islands while P. melanura is generally restricted to near, but small islands. 1 2 Therefore, at least for these guilds, insular distributions appear to be influ- 2 3 enced by the combined effects of interspecific interactions and interspecific 3 4 differences in resource requirements and vagilities. 4 5 Including the effects of interspecific interactions will certainly add a sub- 5 6 stantial degree of complexity to the model, but it seems to be a more accu- 6 7 rate account of the complexity of nature. Moreover, this exercise identifies 7 8 another potentially illuminating area of future research. Do vagilities or 8 9 resource demands covary with ecological dominance? In the examples dis- 9 10 cussed above, immigration abilities appeared to be lower for the competi- 10 11 tively dominant species (see also MacArthur, 1972; Tilman & Wedin, 1991). 11 12 Is this a general pattern in nature? Given that there may emerge some pat- 12 13 tern of covariation in vagilities and ‘interspecific dominance’ among species, 13 14 does this pattern vary with the nature of interspecific interactions (e.g., 14 15 negative covariation for competitors, positive covariation for predators)? 15 16 Research directed at these questions may provide illuminating linkages 16 17 between the fields of biogeography and community ecology. 17 18 18 19 19 20 Summary of the hierarchical model 20 21 The species-based, hierarchical model developed in the previous sections can 21 22 account for patterns spanning three biogeographical scales. These include the 22 23 following. 23 24 24 (a) Species level patterns: distribution of a focal species as a function of area 25 25 and isolation (i.e., insular distribution functions). 26 26 (b) Archipelago (inter-community) level patterns: differences in community 27 27 structure (especially species richness and species composition) among 28 28 islands. 29 29 (c) Inter-archipelago patterns: differences in intercommunity patterns (e.g., 30 30 differences in species–area relationships, species–isolation relationships 31 31 and nestedness) among archipelagoes. Alternatively, these patterns may 32 32 also be compared among taxa (e.g., comparing species–isolation rela- 33 33 tionships of bats vs. birds). 34 34 35 Thus, the model can be used to study a diversity of ecological patterns of 35 36 isolated systems. The model, however, remains relatively simple. Because the 36 37 empirical cases discussed here are limited to animals, the model remains 37 38 untested for plants and other taxa. In addition, the model is limited to events 38 39 occurring in ecological time and it is entirely deterministic, focusing on char- 39 40 acteristics of species or islands that affect immigration and extinction. 40 A biogeographical model 305 1 MacArthur and Wilson’s equilibrium theory also assumed immigration and 1 2 extinction were deterministic with respect to island characteristics. On the 2 3 other hand, their equilibrium theory assumed that these processes were 3 4 stochastic with respect to species characteristics. That is, all species were 4 5 assumed equivalent. For this reason, the equilibrium theory remains neutral 5 6 with respect to patterns in species composition. 6 7 7 8 8 9 Linkages: species to archipelago levels 9 10 The true challenge of the hierarchical model presented here lies in identify- 10 11 ing and understanding the key linkages among biogeographical scales. At a 11 12 fundamental level, the linkages among components of insular diversity, in 12 13 ecological time, are immigration and extinction. Linkages among the three bio- 13 14 geographic levels listed above will therefore concern the scale-apppropriate 14 15 characteristics of species and ecosystems that influence these fundamental 15 16 biogeographic processes. Four such linkages are requisite for moving from 16 17 the species to archipelago levels. 17 18 18 (a) The frequency distribution of resource requirements and immigration 19 19 abilities among species; 20 20 (b) Patterns of covariation of resource requirements and immigration abili- 21 21 ties among species; 22 22 (c) Effects of interspecific interactions on establishment and persistence of 23 23 insular populations; 24 24 (d) Correlations between interspecific dominance and immigration abilities 25 25 or resource requirements. 26 26 27 27 28 Linkages: archipelago to inter-archipelago 28 29 29 Moving from within to between archipelago levels involves at least the fol- 30 30 lowing linkages. 31 31 32 (a) Differences and possible covariation in strength of immigration filters 32 33 and productivity among archipelagoes; 33 34 (b) Differences and possible covariation in representative vagilities and 34 35 resource requirements among species pools. 35 36 36 37 37 38 Conclusions 38 39 The requisite information to explore many of these linkages seems to be avail- 39 40 able. On the other hand, for those linkages where adequate data is lacking, 40 306 M.V. Lomolino 1 areas of potentially important research have now been identified. This infor- 1 2 mation may enable comparisons among biotas at regional to global scales, 2 3 scales where biogeographers are seldom blessed with the luxury of having 3 4 identical ecosystems and species pools. By acknowledging the importance of 4 5 these linkages, some otherwise perplexing ambiguities may be resolved. 5 6 Rather than ignoring differences among archipelagoes and faunal groups, our 6 7 questions may actually focus on these differences. In a characteristically 7 8 intriguing paper, Janzen (1967) asked whether ‘mountain passes are higher 8 9 in the tropics’, arguing that, in comparison to species inhabiting temperate 9 10 ecosystems, those of the tropics have lower vagilities, or at least lower propen- 10 11 sities for immigration across mountains. If this is a genuine tendency, how 11 12 should patterns in richness and nestedness vary between faunal groups of 12 13 tropical and temperate regions? To adequately address such questions, differ- 13 14 ences among archipelagoes as well as differences among regional biotas need 14 15 to be studied. 15 16 As biogeography and ecology continue to develop as scientific disciplines, 16 17 the scope of our studies needs to be broadened. The interface between bio- 17 18 geography and local-scale ecology is the domain of an exciting new discipline, 18 19 macroecology, which focuses on variation of population and community level 19 20 parameters at geographical scales (Brown, 1995). The diverse, multi-scale 20 21 nature of this new discipline, and biodiversity research in general, calls for 21 22 hierarchical approaches and others that can capitalize on natural experiments, 22 23 especially those ‘designed’ at landscape to regional scales (see Edwards et 23 24 al., 1994; Brown, 1995). 24 25 Although not an equilibrium theory, the model presented here, like much 25 26 of metapopulation theory (see Gilpin & Hanski, 1991), was derived from a 26 27 fundamental premise of MacArthur and Wilson’s theory – that insular 27 28 community structure is dynamic and results from the combined effects of 28 29 immigration and extinction. Unlike the equilibrium theory which assumes 29 30 species are equivalent, the current model assumes that many patterns in insu- 30 31 lar community structure result from, not despite, differences among species. 31 32 The hierarchical model, while encompassing three biogeographic scales, 32 33 distills down to a relatively simple, deterministic model: one focusing on the 33 34 species and ecosystem characteristics affecting immigration and extinction. 34 35 Patterns in insular community structure derive from the non-random varia- 35 36 tion among species and islands that affect these fundamental biogeographic 36 37 processes. It is not being denied that immigration and extinction also are influ- 37 38 enced by stochastic factors, but such factors cannot account for the patterns 38 39 addressed here. 39 40 Finally, note that my focus has been limited to just a subset of the earth’s 40 A biogeographical model 307 1 ecosystems – islands and other ecosystems distributed as isolated patches in 1 2 ‘seas’ of other habitats. On the other hand, studies of these systems have had 2 3 a disproportionately important influence on the development of ecology, evo- 3 4 lutionary biology and conservation biology. Hopefully, the model presented 4 5 here, in combination with insights from metapopulation theory and other 5 6 disciplines will contribute to our ability to understand the forces structuring 6 7 isolated communities and, ultimately, conserve endangered biotas, many of 7 8 these persisting on true islands or ever shrinking fragments of native habitats. 8 9 9 10 10 11 Acknowledgments 11 12 12 This work was partially supported by a grant from the National Science Foun- 13 13 dation (DEB-9322699). David Perault, James H. Brown, Russell Davis, Rob 14 14 Channell, Gregory A. Smith, and Evan Weiher provided useful comments on 15 15 an earlier version of this manuscript. 16 16 17 17 18 18 19 References 19 20 Aberg, J., Jansson, G. Swenson, J.E, & Anglestam, P. (1995). The effect of matrix 20 21 on the occurrence of hazel grouse (Bonasa bonasia) in isolated habitat 21 fragments. Oecologia 103: 265–269. 22 Alatalo, R.V. (1982). Bird species distributions in the Galapagos and other 22 23 archipelagoes: competition or chance? Ecology 63: 881–887. 23 24 Allen, T.F.H. (1987). Hierarchical complexity in ecology: a non-euclidean 24 25 conception of the data space. Vegetatio 69: 17–26. 25 Auerbach, M.J. & Schmida, A. (1987). Spatial scale and the determinants of plant 26 species richness. Trends in Ecology and Evolution 2: 238–242. 26 27 Brown, J.H. (1978). The theory of insular biogeography and the distribution of 27 28 boreal birds and mammals. Great Basin Naturalist Mem. 2: 209–227. 28 Brown, J.H. (1995). Macroecology. Chicago, IL: University of Chicago Press. 29 Brown, J.H. & Gibson, A.C. (1983). Biogeography. Mosby Company, St Louis. 29 30 Brown, J.H. & Kodric-Brown, A. (1977). Turnover rates in island biogeography: 30 31 effect of immigration on extinction. Ecology 58: 445–449. 31 Brown, J.H. & Lomolino, M.V. (1989). On the nature of scientific revolutions: 32 independent discovery of the equilibrium theory of island biogeography. 32 33 Ecology 70: 1954–1957. 33 34 Brown, J.H. & Nicolleto, P.F. (1991). Spatial scaling of species composition: body 34 35 masses of North American land mammals. American Naturalist 138: 35 1478–1512. 36 Calder, W.A., III. (1984). Size, Function and Life History. Cambridge, MA: 36 37 Harvard University Press. 37 38 Connell, J.H. (1961). The influence of interspecific competition and other factors on 38 the distribution of the barnacle Chthalamus stellatus. Ecology 42: 710–723. 39 Connor, E.H. & Simberloff, D. (1979). The assembly of species communities: 39 40 chance or competition? Ecology 60: 1132–1140. 40 308 M.V. Lomolino

1 Connor, E.H. & Simberloff, D. (1983). Interspecific competition and species co- 1 2 occurrence patterns on islands: null models and the evaluation of evidence. 2 Oikos 41: 455–465. 3 Darlington, P.J. (1957). Zoogeography: The Geographical Distribution of Animals. 3 4 New York: John Wiley. 4 5 Dexter, R.W. (1978). Some historical notes on Louis Agassiz’s lecutres on 5 6 zoogeography. Journal of Biogeography 5: 207–209. 6 Diamond, J.M. (1975). Assembly of species communities. In Ecology and 7 Evolution of Communities, ed. M.L. Cody & J.M. Diamond, pp. 342–444. 7 8 Cambridge, MA: Harvard University Press. 8 9 Diamond, J.M. & Gilpin, M. (1982). Examination of the ‘null’ model of Connor 9 and Simberloff for species co-occurrences on islands. Oecologia 52: 64–74. 10 Dunn, C.P. & Loehle, C. (1988). Species–area parameter estimation: testing the null 10 11 model of lack of relationships. Journal of Biogeography 15: 721–728. 11 12 Edwards, P.J., May, R.M. & Webb N.R. (1994). Large-scale Ecology and 12 Conservation Biology. London: Blackwell Scientific Publications. 13 Gilbert, F.S. (1980). The equilibrium theory of island biogeography: fact or fiction? 13 14 Journal of Biogeography 7: 209–235. 14 15 Gilpin, M. & Diamond, J.M. (1982). Factors contributing to non-radomness in 15 16 species co-occurrence on islands. Oecologia 52: 75–84. 16 Gilpin, M. & Diamond, J.M. (1984). Are species co-occurrences on islands non- 17 random, and are null hypotheses useful in community ecology? In Ecological 17 18 Communities: Conceptual Issues and the Evidence. ed. D.R. Strong, D. 18 19 Simberloff, L.G. Abele, & A.B. Thistle, pp. 297–315. NJ: Princeton 19 University Press. 20 Gilpin, M. & Hanski, I. (1991). Metapopulation Dynamics: Empirical and 20 21 Theoretical Investigations. London: Academic Press. 21 22 Goel, N.S. & Richter-Dynn, N. (1974). Stochastic Models in Biology. NJ, USA: 22 Academic Press. 23 Gould, S.J. (1979). An allometric interpretation of species-area curves: the meaning 23 24 of the coefficient. American Naturalist 114: 335–343. 24 25 Grant, P.R. (1972). Interspecific competition among rodents. Annual Review of 25 26 Ecology and Systematics 3: 79–106. 26 Harestad, E.L. & Bunnell, E.L. (1979). Home range and body weight – a re- 27 evaluation. Ecology 60: 389–402. 27 28 Hengeveld, R. (1987). Scales of variation: their distinction and ecological 28 29 importance. Annales Zoologica Fennici 24: 195–202. 29 Hope, J.H. (1973). Mammals of Bass Strat Islands. Proceedings of the Royal 30 Society of Victoria 85: 163–195. 30 31 Janzen, D.H. (1967). Why mountain passes are higher in the tropics. American 31 32 Naturalist 101: 233–249. 32 Jurgens, K.D. & Prothero, J. (1991). Lifetime energy budgets in mammals and 33 birds. Comparative Biochemistry and Physiology 100A: 703–709. 33 34 Lindstedt, S.L., Miller, B.J. & Buskirk, S.W. (1986). Home range, time, and body 34 35 size in mammals. Ecology 67: 413–418. 35 36 Lomolino, M.V. (1984). Immigrant selection, predatory exclusion and the 36 distributions of Microtus pennsylvanicus and Blarina brevicauda on islands. 37 American Naturalist 123: 468–483. 37 38 Lomolino, M.V. (1986). Mammalian community structure on islands: immigration, 38 39 extinction and interactive effects. Biological Journal of the Linnoean Society 39 28: 1–121, and in Island Biogeography of Mammals, Chapter 1. NY: 40 Academic Press. 40 A biogeographical model 309

1 Lomolino, M.V. (1988). Winter immigration abilities and insular community 1 2 structure of mammals in temperate archipelagoes. In Biogeography of the 2 Island Region of Western Lake Erie. Ohio University Press. 3 Lomolino, M.V. (1989). Bioenergetics of cross-ice movements of Microtus 3 4 pennsylvanicus, Peromyscus leucopus and Blarina brevicauda. Holarctic 4 5 Ecology 12: 213–218. 5 6 Lomolino, M.V. (1993). Winter filtering, immigrant selection and species 6 composition of insular mammals of Lake Huron. Ecography 16: 24–30. 7 Lomolino, M.V. (1994). Species richness patterns of mammals inhabiting nearshore 7 8 archipelagoes: area, isolation, and immigration filters. Journal of Mammalology 8 9 75: 39–49. 9 Lomolino, M.V. & Davis, R. (1997). Biogeographic scale and biodiversity of 10 mountain forest mammals of western North America. Global Ecology and 10 11 Biogeography Letters 6: 57–76. 11 12 Lomolino, M.V., Brown, J.H. & Davis, R. (1989). Island biogeography of montane 12 forest mammals in the American Southwest. Ecology 70: 180–194. 13 Lubchenco, J. (1978). Plant species diversity in a marine intertidal community: 13 14 importance of herbivore food preference and algal competitive abilities. 14 15 American Naturalist 112: 23–39. 15 16 MacArthur, R.H. (1972). Geographical Ecology: Patterns in the Distribution of 16 Species. NJ, USA: Princeton University Press. 17 MacArthur, R.H. & Wilson, E.O. (1963). An equilibrium theory of insular 17 18 zoogeography. Evolution 17: 373–387. 18 19 MacArthur, R.H. & Wilson, E.O. (1967). The Theory of Island Biogeography. NJ: 19 Princeton University Press. 20 Manly, B.F.J. (1991). Randomization and Monte-Carlo Methods in Biology. NY: 20 21 Chapman and Hall. 21 22 Menge, B.A. (1972). Competition for food between two intertidal starfish species 22 and its effect on body size and feeding. Ecology 53: 635–644. 23 Michaud, J.M. (1995). Biogeography of the bog copper butterfly (Lycaena 23 24 epixanthe) in southern Rhode Island Peatlands: a metapopulation perspecitve. 24 25 MS thesis, Department of Natural Resources, University of Rhode Island, USA. 25 26 Munger, J.C. & Brown, J.H. (1981). Competition in desert rodents: an experiment 26 with semipermeable exclosures. Science 211: 510–512. 27 Munroe, E.G. (1948). The geographical distribution of butterflies in the West 27 28 Indies. PhD dissertation, Cornell University, New York, USA. 28 29 Munroe, E.G. (1953). The size of island faunas. In Proceedings of the 7th Pacific 29 Science Congress of the Pacific Science Association. volume IV, Zoology. pp. 30 52–53. Auckland, New Zealand: Whitcome and Tombs. 30 31 Niering, W.A. (1963). Terrestrial ecology of Kapingamarangi Atoll, Caroline 31 32 Islands. Ecological Monographs 33: 131–160. 32 O’Neill, R.V., DeAngelis D.L., Waide, J.B. & Allen, T.F.H. (1986). A Hierarchical 33 Concept of Ecosystems. Princeton, NJ: Princeton University Press. 254 pp. 33 34 Paine, R.T. (1966). Intertidal community structure: experimental studies between a 34 35 dominant competitor and its principal predator. Oecologia 15: 93–120. 35 36 Patterson, B.D. & Atmar, W. (1986). Nested subsets and the structure of insular 36 mammalian faunas and archipelagoes. In Island Biogeography of Mammals, 37 ed. L. Heaney & B.D. Patterson, pp. 65–82. Biological Journal of the 37 38 Linnaean Society 28 (1&2) and London: Academic Press. 38 39 Patterson, B.D. & Brown, J.H. (1991). Regionally nested patterns of species 39 composition in granivorous rodent assemblages. Journal of Biogeography 18, 40 395–402. 40 310 M.V. Lomolino

1 Peters, R.H. (1983). The Ecological Implications of Body Size. Cambridge: 1 2 Cambridge University Press. 2 Pielou, E.C. (1979). Biogeography. NY: John Wiley. 3 Preston, F.W. (1962). The canonical distribution of commonness and rarity: Part I. 3 4 Ecology 43: 185–215. 4 5 Rosenzweig, M.L. (1995). Species – Diversity in Space and Time. NY: Cambridge 5 6 University Press. 6 Sauer, J. (1969). Oceanic islands and biogeographic theory: a review. Geographical 7 Reviews 59: 582–593. 7 8 Schmida, A. & Wilson M.V. (1985). Biological determinants of species diversity. 8 9 Journal of Biogeography 12: 1–20. 9 Schoener, T.W. (1968). Sizes of feeding territories among birds. Ecology 49: 10 123–141. 10 11 Schoener, T.W. (1983). Field experiments on interspecific competition. American 11 12 Naturalist 122: 240–285. 12 Schoener, T.W. & Schoener, A. (1983). Distribution of vertebrates on some very 13 small islands. I. Occurrence sequences of individual species. Journal of Animal 13 14 Ecology 52: 209–235. 14 15 Simberloff, D. & Connor, E.F. (1981). Missing species combinations. American 15 16 Naturalist 118: 215–239. 16 Simberloff, D & Martin, J. (1991). Nestedness of insular avifaunas: simple 17 summary statistics masking complex species patterns. Ornis Fennica 68: 17 18 178–192. 18 19 Simpson, G.G. (1940). Mammals and landbridges. Journal of the Washington 19 Academy of Science 30: 137–163. 20 Taylor, C.M. (1997). Fish species richness and incidence patterns in isolated and 20 21 connected stream pools: effects of pool volume and spatial position. Oecologia 21 22 (in press). 22 Tilman, D. & Wedin, D. (1991). Plant traits and resource reduction for five grasses 23 growing on a nitrogen gradient. Ecology 72: 685–700. 23 24 Vagvolgyi, J. (1975). Body size, aerial dispersal, and origin of Pacific Island snail 24 25 fauna. Systematic Zoology 24: 465–488. 25 26 Werner, E.E. (1979). Foraging efficiency and habitat switching in competing 26 sunfishes. Ecology 60: 256–264. 27 Whitehead, D.R. & Jones, C.E. (1969). Small islands and the equilibrium theory of 27 28 insular biogeography. Evolution 23: 171–179. 28 29 Wiens, H.J. (1962). Atoll Environment and Ecology. New Haven, CT: Yale 29 University Press. 30 Wiens, J.A., Addicott, J.F., Case, T.J. & Diamond, J.M. (1986). The importance of 30 31 spatial and temporal scale in ecological investigations. In Community Ecology, 31 32 ed. J.M. Diamond & T.J. Case, pp. 145–153. NY: Harper and Row. 32 Woodroffe, C.D. (1986). Vascular plant species-area relationship on Nui Atoll, 33 Tuvalu, Central Pacific: a reassessment of the small island effect. Australian 33 34 Journal of Ecology 11: 21–31. 34 35 Wright, D.H. (1983). Species energy theory: an extension of species-area theory. 35 36 Oikos 41: 496–506. 36 Wylie, J.L. & Currie, D.J. (1993). Species-energy theory and patterns of species 37 richness: I. Patterns of bird, angiosperm, and mammals species richness on 37 38 islands. Biological Conservation 63: 137–144. 38 39 39 40 40 1 11 1 2 2 3 Interaction of physical and biological processes 3 4 in the assembly of stream fish communities 4 5 5 6 6 7 Elizabeth M. Strange and Theodore C. Foin 7 8 8 9 9 10 10 11 11 12 12 13 13 14 Introduction 14 15 The concept of assembly rules originally was developed by Diamond (1975) 15 16 to describe regular patterns of species occurrence on islands, and has since 16 17 been generalized to include all processes influencing community develop- 17 18 ment, the emphasis on colonization and extinction having lessened somewhat 18 19 (Drake, 1990a; Nee, 1990). The contemporary meaning of community assem- 19 20 bly also de-emphasizes the term ‘rule’ in favor of less-fixed combinations of 20 21 interactions and outcomes that are known to occur within different commu- 21 22 nities (Post & Pimm, 1983; Wilbur & Alford, 1985; Gilpin et al. 1986; Robin- 22 23 son & Dickerson, 1987; Robinson & Edgemon, 1988; Barkai & McQuaid, 23 24 1988; Drake, 1988, 1990b, 1991; Drake et al., 1993; Case, 1991; Keddy, 24 25 1992; Weiher & Keddy, 1995a, b). Community assembly now encompasses 25 26 the ecology and short-term, observable evolution of the community from 26 27 initial colonization to species saturation, and logically includes a number of 27 28 topics that are referred to by other names, such as trophic cascades (Carpen- 28 29 ter & Kitchell, 1993), top-down vs. bottom-up regulation of trophic webs 29 30 (Matson & Hunter, 1992), and the regulation of biodiversity (Rosenzweig, 30 31 1995). The scope of the processes involved and the applicability of assembly 31 32 rules to all communities makes this one of the central problems in ecology. 32 33 Although the physical environment is important in a number of ways, there 33 34 has been more focus upon biological interactions in assembly processes. The 34 35 traditional view is that environmental gradients act principally to filter some 35 36 species out of the pool of potential immigrants (Drake, 1990b). In most nat- 36 37 ural communities, however, physical conditions may do much more than act 37 38 as a coarse filter of unsuitable species; they may exert subtle, quantitative 38 39 effects that play a continuing role in shaping community composition and 39 40 dynamics (Chesson, 1986). 40

311 312 E.M. Strange and T.C. Foin 1 There is a need to incorporate physical variation into the picture of com- 1 2 munity assembly to account for interactions between physical variation and 2 3 biotic processes within specific systems. Because such interactions will 3 4 depend on both the nature and rates of interaction, if general rules of com- 4 5 munity assembly are to emerge, they will likely do so by comparative analy- 5 6 sis of a number of different kinds of communities. Stream fish communities 6 7 are particularly appropriate for the study of how abiotic factors influence 7 8 community assembly because they have been studied intensively and their 8 9 biological interactions defined, and because flow conditions are known to be 9 10 critically important and are readily quantified. For example, native topminnows 10 11 in the American southwest can survive predation by introduced mosquito fish 11 12 (Gambusia spp.) if floods are frequent enough to reduce populations of Gam- 12 13 busia periodically. In the absence of floods, mosquito fish typically eliminate 13 14 topminnows within 1–3 years. Gambusia is known to be poorly adapted to 14 15 flood conditions, and observations in streams suggest that floods are a key 15 16 factor in coexistence (Meffe, 1984, 1985; Minckley & Meffe, 1987). 16 17 In this chapter, models of interactions are developed for stream fishes in 17 18 what is known as the Trout Zone Assemblage (Moyle, 1976), typical of the 18 19 headwaters of cold montane streams throughout North America, and results 19 20 are used to interpret the development of other communities. In northern Cali- 20 21 fornia streams, high scouring discharges can occur in winter and spring as a 21 22 result of rainfall and snowmelt, and cause high mortality of young-of-year 22 23 (Erman et al., 1988). This contributes to differential recruitment success among 23 24 species that spawn in different seasons, and as a consequence hydrology exerts 24 25 a strong control over community structure and organization in these systems 25 26 (Seegrist & Gard, 1972; Gard & Flittner, 1974; Strange et al., 1992; Strange, 26 27 1995). 27 28 Analysis of fluctuations in fish abundance over a 15-year period in Martis 28 29 Creek, a small coldwater stream in the Truckee River drainage of northern 29 30 California, shows that the impact of invasion by the piscivorous European 30 31 brown trout (Salmo trutta) on native fishes is mediated by the timing of floods 31 32 and droughts (Strange et al., 1992; Strange, 1995). The winter-spawning 32 33 brown trout is favored in years when winter floods are absent but spring floods 33 34 reduce recruitment of native fishes, all of which are spring spawners. On the 34 35 other hand, when floods occur in winter but not in spring, the spring-spawn- 35 36 ing native-species are favored and the winter-spawning brown trout shows 36 37 poor recruitment. During times of drought, relative abundances shift in favor 37 38 of native species that show greater physiological adaptation to low flow 38 39 conditions than brown trout, which is a native of Europe and is presumably 39 40 not adapted to drought conditions characteristic of northern California. 40 Predicting fish communities 313 1 This shifting pattern of relative abundances is well known, but is almost 1 2 impossible to investigate experimentally. In order to examine the influence 2 3 of varying flows on relative abundance, to determine what conditions would 3 4 cause local extinction, and to define conditions which would favor establish- 4 5 ment of invading species, an age-structured population model was developed 5 6 for the fish community of these types of streams. The model was used to set 6 7 up specific strengths and types of interactions between species and, through 7 8 variation in inputs, investigated how stream flow can drive variation in stream 8 9 fish densities and assemblage structure. 9 10 Assembly was modeled by introducing different stream fish types, one at 10 11 a time, under a variety of physical and biological conditions. Although each 11 12 species is represented by a Leslie matrix formulation, the model has two 12 13 important additional features. First, the community is represented by linking 13 14 the individual matrices through the survival vectors rather than through a 14 15 community matrix. Second, the linkage of survival vectors is explicitly related 15 16 to flow regimes, competitive interactions, and predation. By expressing 16 17 species survival rates as a function of density, greater biological realism is 17 18 achieved, and the model is capable of a range of dynamical behavior not pos- 18 19 sible with the linear formulations of the community matrix. Using an exper- 19 20 imental format that permitted variation of single factors and realistic combi- 20 21 nations, we were able to examine how flow-driven variations in stream fish 21 22 densities combine with ongoing competition and predation in the assembly 22 23 of stream fish assemblage structure. Validation of the model was accom- 23 24 plished by comparing our predictions with results of field studies of stream 24 25 fish assemblages in two coldwater streams in the Sierra Nevada region of 25 26 northern California. 26 27 Together, model analysis and field observations allowed us to analyze how 27 28 stream fish assembly dynamics are influenced by interactions between phys- 28 29 ical and biological conditions and to determine what general assembly rules 29 30 might be operating. Our findings also suggest some general conclusions 30 31 regarding the interactive roles of physical variation and biotic processes in 31 32 the assembly of communities in other physically variable environments. 32 33 33 34 34 The simulation model 35 35 36 Stream fish life history types and species 36 37 In order to maximize the generality of the model, a system was established 37 38 using four common stream fish life history types. These combine theoretical 38 39 studies of fish life histories (Winemiller, 1992; Winemiller & Rose, 1992) 39 40 and demographic properties of fish species typical of northern California 40 314 E.M. Strange and T.C. Foin

1 Table 11.1. Summary of demographic characteristics of model life history types 1 2 2 3 Life history type Demographic characteristics 3 4 4 Catostomid matures age 4 5 r = 0.2475 under 25 cfs 5 6 maximum annual fecundity = 6000 eggs 6 7 mean generation time = 6.2 years 7 maximum age = 7 8 8 Cyprinid matures age 1 9 r = 0.8082 under 25 cfs 9 10 maximum annual fecundity = 400 eggs 10 11 mean generation time = 2.15 years 11 12 maximum age = 2 12 13 Cottid matures age 1 13 r = 0.6279 under 25 cfs 14 maximum annual fecundity = 350 eggs 14 15 mean generation time = 2.3 years 15 16 maximum age = 4 16 17 Salmonid matures age 2 17 18 r = 0.4109 under 25 cfs 18 maximum annual fecundity = 1600 eggs 19 mean generation time = 3.7 years 19 20 maximum age = 6 20 21 21 22 22 23 coldwater streams (Moyle, 1976). The four life history types can be identi- 23 24 fied with real species: (a) a salmonid, characterized by maturity at 2 years, 24 25 moderate fecundity, and a reproductive span of 5 years; (b) a cyprinid, char- 25 26 acterized by maturity after 1 year, low fecundity, and a reproductive span of 26 27 2 years; (c) a catostomid, characterized by late maturity, high fecundity, and 27 28 a reproductive span of 4 years; and (d) a cottid, characterized by maturation 28 29 after 1 year, low fecundity, and a reproductive span of 4 years. The salmonid– 29 30 catostomid–cyprinid–cottid fish assemblage is characteristic of coldwater 30 31 streams throughout North America (Moyle & Herbold, 1987). Demographic 31 32 characteristics of model life histories are summarized in Table 11.1. 32 33 By changing parameter values, we could make each life history correspond 33 34 to a specific species for purposes of validation against field data. Parameter 34 35 values for six species were supplied for field validation. The species are the 35 36 introduced European piscivore, brown trout (Salmo trutta), rainbow trout 36 37 (Onchorynchus mykiss), which was introduced to California’s Sierra Nevada 37 38 region, and several species native to the Truckee River drainage, including 38 39 two cyprinids, Lahontan redside (Richardsonius egregius) and Lahontan 39 40 speckled dace (Rhinichthys osculus robustus), the cottid Paiute sculpin (Cottus 40 Predicting fish communities 315 1 beldingi), and the catostomid Tahoe sucker (Catostomus tahoensis). The 1 2 model assumes that the four basic life history types are spring-spawners that 2 3 show weak year-classes following spring floods. A salmonid predator, based 3 4 on the European brown trout, is assumed to spawn in the winter and show 4 5 poor recruitment following winter floods. 5 6 6 7 7 8 Overall structure of the simulation model 8 9 The model consists of linked, age-structured Leslie matrices, one for each 9 10 life history, that simulate annual changes in the population sizes of interact- 10 11 ing fish species as survival is modified by stream discharge, competitive inter- 11 12 actions, and predation. For each yearly integration, the model calculates the 12 13 net change in the number of individuals in each age/size class of each species 13 14 as a function of age-specific growth and mortality rates. Recruitment varies 14 15 as a function of female fecundity and the survival of early life stages. Spec- 15 16 ified patterns of high stream discharge during early development, interspecific 16 17 competition, and predation all contribute to the rate of first year mortality. 17 18 Adult survival is reduced principally by low flow conditions. 18 19 19 20 20 21 Relationship of first year survival to stream discharge 21 22 Curves fit to field data by regression analysis show a negative exponential 22 23 relationship between recruitment of a year-class and mean monthly discharge 23 24 during May for spring-spawners and during January for winter-spawners 24 25 (Strange et al., 1992). The model varies first year survival as a continuous 25 26 negative exponential function of mean monthly May discharges above 25 cfs 26 27 for spring-spawners and as a continuous function of mean January discharges 27 28 above 25 cfs for winter-spawners. Differential sensitivity to extremely high 28 29 or low stream discharge was estimated by differences in egg size and rates 29 30 of first year growth (Winemiller, 1992; Winemiller & Rose, 1992). It was 30 31 assumed that 25 cfs was the critical threshold for changes in survival rate; 31 32 the larger the egg size and the higher the potential growth rate, the less the 32 33 reduction in survival above 25 cfs. Because salmonids have relatively large 33 34 eggs and rapid first year growth, they were assumed to show 25% higher sur- 34 35 vival as stream discharge increased than the moderate-sized catostomid life 35 36 history. The small-bodied cyprinid and cottid life histories were assumed to 36 37 show 25% lower survival than the catostomid. 37 38 Optimal temperatures for salmonids and cottids are generally lower than 38 39 for catostomids and cyprinids, and salmonids and cottids are less tolerant of 39 40 higher water temperatures resulting from low flows (Moyle, 1976). Thus, the 40 316 E.M. Strange and T.C. Foin 1 model assumes that first-year survival under prerecruitment discharges less 1 2 than 25 cfs is highest for the catostomid and cyprinid life histories and 25% 2 3 less for the salmonid and cottid life histories based on North American species. 3 4 4 5 5 6 Relationship of adult survival in response to droughts 6 7 Adult survivorship is also reduced under low flows. In the model, survival 7 8 rates of fish older than one year are reduced as a continuous function of mean 8 9 annual flows less than 25 cfs. Mean annual flow is calculated as the average 9 10 of the mean January and mean May discharges at each yearly time step. As 10 11 was the case for first-year survival under low flows, the model assumes that 11 12 percentage decrease in adult survival under discharges less than 25 cfs is least 12 13 for the catostomid and cyprinid life histories and 25% more for the salmonid 13 14 and cottid life histories. Decreases in adult survival were assumed to be 50% 14 15 more for the European piscivorous salmonid based on brown trout. 15 16 16 17 17 18 Competition during the first year of life 18 19 Density-dependent mortality is characteristic of young-of-year fish (Wootton, 19 20 1990), and models of fish populations generally assume that only this age 20 21 group contributes to density effects (e.g., Levin & Goodyear, 1980; DeAn- 21 22 gelis et al., 1980). This was the basis for using an inverse logistic function 22 23 to represent competitive interactions that modify survival to age 1 based on 23 24 stream discharge. The densities of each species and its competitors were used 24 25 as inputs to reduce survival below that due to streamflow alone. The percent 25 26 reduction ranges from 0 (under no competition at low densities) to 90% (under 26 27 maximum competition at highest densities). Species densities at each time 27 28 step are given by the number of eggs spawned by each species plus the num- 28 29 ber of eggs spawned by competing species modified by weighting factors 29 30 representing the strength of the competitive effect. 30 31 31 32 32 33 Predation intensities 33 34 The model contains only one piscivorous species. Adults of the piscivorous 34 35 salmonid prey on young-of-year of other species based on experimental stud- 35 36 ies of brown trout (Garman & Nielsen, 1982). To model predator functional 36 37 response, first year survival of prey species based on stream discharge was 37 38 modified as a function of the ratio of predator to prey species for each prey 38 39 species. We assumed that the piscivorous salmonid consumes prey species 39 40 according to a type III functional response, and we used an inverted logistic 40 Predicting fish communities 317 1 function to model the relationship between prey survival and predator con- 1 2 sumption. The percentage reduction in prey survival ranges from 0 (under no 2 3 predation) to 90% (under maximum predation). The numerical response of 3 4 the predator was determined to be an asymptotically increasing function of 4 5 the total number of prey consumed per predator up to the maximum con- 5 6 sumption rate. This function gives a percent increase in survival of piscivorous 6 7 salmonids > 150 mm standard length (age 2 and over in our model) based 7 8 on data from Garman and Nielsen (1982). The survival rate for each predator 8 9 size class is the baseline survival for that age/size class in the model in the 9 10 absence of predation; it increases up to a maximum of 90% under maximum 10 11 consumption. 11 12 12 13 13 14 Parameter estimation 14 15 Pre-recruitment stream flows were sampled randomly from discharge distri- 15 16 butions based on the historical record of discharges in our reference stream 16 17 (USGS gaging station no. 10339400, Martis Creek, California, 1972–1994). 17 18 The time series included a range of discharges considered representative of 18 19 the natural regime and included the most extreme high and low discharges 19 20 on record. Rates of survival and reproduction for each life history type were 20 21 estimated from field data (Strange, 1995) and information in the general 21 22 stream fish literature for representative species (Moyle, 1976). Survival rates 22 23 associated with the function relating early survival to stream discharge are 23 24 based on estimates derived from our field data (Strange, 1995) and estimates 24 25 in the stream fish literature for early survival of stream fishes under a range 25 26 of stream discharges (Allen, 1951; Cooper, 1953; Shetter, 1961; Latta, 1962; 26 27 McFadden et al., 1967; McFadden, 1969; Mortensen, 1977; Nehring, 1988). 27 28 Baseline survival was estimated as 20% for age 1, 30% for age 2, 50% for 28 29 age 3, and 70% for ages up to the maximum age/size class, for which sur- 29 30 vival is 0 (Needham et al., 1945; McFadden & Cooper, 1962; McFadden et 30 31 al., 1967; Hunt, 1969). Age/size classes are based on length-frequency histo- 31 32 grams for representative species in our study stream and the general litera- 32 33 ture (Moyle, 1976). Fecundity of each life history type is the mean for the 33 34 average-sized mature female in each mature age/size class and based on 34 35 fecundity estimates for representative species (Moyle, 1976). 35 36 36 37 37 38 Simulation experiments 38 39 Our simulation experiments were designed to reveal how interactions between 39 40 physical and biological conditions vary species’ relative success and outcomes 40 318 E.M. Strange and T.C. Foin 1 of assembly processes. Model life history types were introduced one at a time 1 2 under low abundance (50 individuals), with invasions occurring every ten 2 3 years. Experimental treatments varied invasion sequence, species’ interac- 3 4 tions, and the physical regime. Simulations examined outcomes for invasion 4 5 first by early-maturing life history types with short generation times com- 5 6 pared to entry first by the late-maturing catostomid life history, which has the 6 7 longest generation time. Results were compared for the different sequences 7 8 in the case of invasion by competing species that spawn in the spring and for 8 9 a predator–prey system in which the predator spawns in winter and the prey 9 10 species spawn in spring. For the competition experiments, the cyprinid and 10 11 cottid life history types were modeled as strong competitors of each other but 11 12 weak competitors of the catostomid and salmonid life histories. Similarly, the 12 13 salmonid and the catostomid were assumed to be strong competitors of each 13 14 other and weak competitors of the cyprinid and cottid. Symmetric competi- 14 15 tion was assumed between the salmonid and the catostomid. Based on a study 15 16 of competitive interactions between a cottid and cyprinid in a northern Cal- 16 17 ifornia stream (riffle sculpin, Cottus gulosus, and speckled dace, Rhinichthys 17 18 osculus, Baltz et al., 1982), the cottid is a better competitor. For the preda- 18 19 tor–prey system, the predator life history was a winter-spawning piscivorous 19 20 salmonid, and prey species included the spring-spawning cyprinid, cottid, and 20 21 catostomid. Under predation, prey species were assumed to show intraspecific 21 22 competition only. Details of the simulation experiments are presented below. 22 23 23 24 24 25 Relative success of invader life histories under a range of physical and 25 26 biological conditions 26 27 To define conditions that will alter relative success, we examined how the 27 28 success of different invader life histories varied for contrasting invasion 28 29 sequences and under different physical and biological conditions. Physical 29 30 conditions were varied by alternating floods and droughts at specified fre- 30 31 quencies (# of floods or droughts over 100 years from the time of the first 31 32 invasion) with a moderate streamflow (50 cfs) based on average discharges 32 33 observed in our study stream. For each invasion sequence, the percentage 33 34 change in relative abundances was determined under the different physical 34 35 conditions for a competition community and a predator-prey system com- 35 36 pared to a baseline model without competition or predation. Floods occurred 36 37 during a species’ month of hatching and reduced survival to age 1. For the 37 38 competition community, all invading species were assumed to spawn in spring 38 39 and show poor recruitment in response to spring floods. For the predator–prey 39 40 system, floods occurred only in winter and therefore only affected recruit- 40 Predicting fish communities 319 Table 11.2. Outline of simulation experiments 1 1 2 2 Invasion sequence Biological regime Physical regime 3 3 4 COT-CYP-SAL-CAT CAT-SAL strong competitors, no droughts, 4 5 COT-CYP strong competitors, 10% droughts, 5 and COT,CYP weak 20% droughts, 6 competitors of CAT,SAL 30% droughts 6 7 vs. 7 8 no spring floods, 8 10% spring floods, 9 20% spring floods, 9 10 30% spring floods 10 11 vs. 11 50 cfs 12 12 CAT-SAL-CYP-COT same same 13 13 14 COT-CYP-SAL-CAT SAL predator of no droughts, 14 15 COT, CYP, CAT; 10% droughts, 15 prey show intraspecific 20% droughts, 16 competition only 30% droughts 16 17 vs. 17 18 no winter floods 18 19 10% winter floods 19 20% winter floods 20 30% winter floods 20 21 vs. 21 22 50 cfs 22 23 CAT-SAL-CYP-COT same same 23 24 24 25 COT = COTtidae, CYP = CYPrinidae, SAL = SALmonidae, CAT = CATosto- 25 midae. 26 26 27 27 28 ment of the winter-spawning predator. A flood was defined as 200 cfs, based 28 29 on maximum discharges observed in our study stream. A drought was defined 29 30 as a streamflow during the month of hatching of 10 cfs and an average annual 30 31 flow of 10 cfs. Experimental treatments are outlined in Table 11.2. 31 32 32 33 33 34 Simulation of invasion into the stream fish community 34 35 To illustrate how specific interactions between physical and biological 35 36 conditions can influence assembly outcomes, assembly of the competition 36 37 community and the predator–prey system was examined under regimes that 37 38 differed in the timing of floods and droughts. Floods or droughts were intro- 38 39 duced at the time particular life histories invaded and population trajectories 39 40 followed for up to 50 years after the last invasion. Streamflow regimes were 40 320 E.M. Strange and T.C. Foin 1 based on a lognormal distribution of flows based on the historical record of 1 2 discharges in our study stream. For the competition community, a decade of 2 3 drought (May and January mean = 15 cfs, cv = 0.3) during the period of 3 4 cottid and salmonid invasion was first simulated. Next, a decade of high flows 4 5 (May mean = 150 cfs, cv = 0.1) was simulated during the period of cyprinid 5 6 and catostomid invasion. The same treatment was applied to each invasion 6 7 sequence. For the predator–prey system, a period of high flows in winter (Jan- 7 8 uary mean = 150 cfs, cv = 0.1) was introduced at the time of invasion of the 8 9 piscivorous salmonid under each invasion sequence. 9 10 10 11 11 12 Model validation 12 13 Model predictions were compared with field data for two coldwater streams 13 14 in the Truckee River drainage of northern California to examine how the 14 15 physical regime may mediate processes of community assembly in the case 15 16 of invasion by the piscivorous European brown trout. One stream is our ref- 16 17 erence stream, Martis Creek (Moyle & Vondracek, 1985; Strange et al., 1992; 17 18 Strange, 1995), and the other is nearby Sagehen Creek (Seegrist & Gard, 18 19 1972; Gard & Seegrist, 1972; Gard & Flittner, 1974). Modeled species 19 20 included two salmonid species which are not native to the Truckee drainage, 20 21 brown trout (Salmo trutta) and rainbow trout (Oncorhynchus mykiss), two 21 22 native cyprinids, Lahontan redside (Richardsonius egregius) and Lahontan 22 23 speckled dace (Rhinichthys osculus robustus), a native cottid, Paiute sculpin 23 24 (Cottus beldingi), and a native catostomid, Tahoe sucker (Catostomus tahoen- 24 25 sis). Native Truckee drainage species and rainbow trout spawn in spring, 25 26 whereas brown trout is a winter-spawner. Census data for each of the six 26 27 species and stream discharges for each stream were used as model inputs. 27 28 Stream discharge data were obtained for Martis Creek from records for USGS 28 29 stream gage no. 10339400 and for Sagehen Creek from records for USGS 29 30 stream gage no. 10343500. Initial population sizes were the actual censused 30 31 population sizes for the first year of each census. 31 32 Population limits for the competition and predation functions were adjusted 32 33 according to the range of population sizes for the different species observed 33 34 in each stream. Predation rates and other parameter values were the same as 34 35 for the base model used in the simulation experiments. 35 36 36 37 37 38 38 39 39 40 40 Predicting fish communities 321 1 1 Results 2 2 3 Changes in relative success under different physical and biological 3 4 conditions 4 5 Simulation results show how the relative success of interacting stream fish 5 6 species can vary during community assembly depending on order of inva- 6 7 sion, type of biotic interaction, and differences in life history response to 7 8 streamflow conditions (Fig. 11.1, 11.2). 8 9 9 10 Competition 10 11 In the case of competing salmonid and catostomid life history types, the 11 12 salmonid shows increased advantage as floods increase because of its earlier 12 13 maturity and shorter generation time. However, as droughts become more fre- 13 14 quent, dominance shifts in favor of the catostomid because of the salmonid’s 14 15 lower tolerance of drought conditions. Thus, although the salmonid has an 15 16 advantage if it invades before the catostomid, advantage is lost under frequent 16 17 droughts (Fig. 11.1). If the salmonid invades after its competitor, advantage 17 18 will be lost even under moderate drought frequencies (Fig. 11.2). 18 19 For the competing cyprinid and cottid life histories, relative advantage shifts 19 20 to the weaker competitor, the cyprinid, when either floods or droughts become 20 21 more frequent. The cyprinid is favored under floods because of a higher pop- 21 22 ulation growth rate and is favored under droughts because the cottid shows 22 23 greater sensitivity to drought conditions. Thus, the advantage of invading first 23 24 is lost by the cottid as floods and droughts increase (Fig. 11.1). When the 24 25 cyprinid invades first, it increases in relative abundance as floods and droughts 25 26 increase and the cottid declines (Fig. 11.2). 26 27 Such shifts in the relative success of competing species under changing 27 28 physical conditions strongly influence assembly outcomes. The cyprinid and 28 29 catostomid dominate assemblage structure under frequent droughts because 29 30 of their higher tolerance of drought conditions, whereas the salmonid and 30 31 cyprinid are favored under frequent floods. Advantage is gained by the 31 32 cyprinid under floods because it matures earlier and shows a shorter genera- 32 33 tion time so it can maximize reproductive success in good years. The longer- 33 34 lived salmonid shows greater advantage as flood frequency increases because 34 35 it spreads reproductive value over many age classes and is therefore better 35 36 able to withstand many years of poor recruitment. In contrast, the cottid shows 36 37 advantage only if floods and droughts are infrequent and if it invades before 37 38 its competitor. 38 39 39 40 40 322 E.M. Strange and T.C. Foin 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 Fig. 11.1. Percentage change in relative abundances of model life history types under 36 37 competition and predation for different flood and drought frequencies alternated with 37 38 50 cfs for invasion sequence cottid (invades year 0), cyprinid (invades year 10), 38 salmonid (invades year 20), and catostomid (invades year 30). For the competition 39 experiments, the catostomid and salmonid show symmetric competition and the 39 40 cottid is a stronger competitor than the cyprinid. For the predation experiments, the 40 Predicting fish communities 323 1 1 2 Predation 2 3 Results for the predator-prey system indicate how storm events that impact 3 4 predator recruitment can function to regulate prey populations. When a winter- 4 5 spawning salmonid predator invades late in the invasion sequence, early invad- 5 6 ing prey species that spawn in spring can persist and increase in relative abun- 6 7 dance as long as frequent droughts or winter floods limit increase of the predator 7 8 population (Fig. 11.1). Even a species that invades after the predator can per- 8 9 sist if storm events that limit predator recruitment become more frequent. By 9 10 contrast, if the salmonid predator invades when droughts are infrequent, extinc- 10 11 tion of prey species that invade later will occur, and a prey species that invades 11 12 first will show advantage only as both droughts and floods become more fre- 12 13 quent (Fig. 11.2). Results for both invasion sequences show that all species can 13 14 coexist under droughts if floods occur that limit only the predator population. 14 15 15 16 16 17 Assembly scenarios 17 18 Results of sample assembly scenarios show that when floods or droughts occur 18 19 at the time particular species invade there are significant consequences for 19 20 assembly dynamics that can overwhelm the effects of invasion order (Fig. 20 21 11.3–11.5). If the cottid and salmonid life histories invade under droughts, 21 22 the system is dominated by the catostomid and cyprinid life histories, irre- 22 23 spective of the order in which the salmonid or cottid invade, because the 23 24 catostomid and cyprinid strategies are advantageous under such conditions 24 25 (Fig. 11.3). Similarly, if increases of the catostomid and cyprinid are limited 25 26 by high flows at the time of their invasion, the cottid and salmonid are able 26 27 to benefit (Fig. 11.4). If the cottid and salmonid invade under such condi- 27 28 tions, and before their catostomid and cyprinid competitors, they will be able 28 29 to dominate the assemblage (Fig. 11.4, top). Even when the catostomid and 29 30 cyprinid invade first, the cottid and salmonid are able to increase (Fig. 11.4, 30 31 bottom). When the short-lived cyprinid invades under extreme high flows, 31 32 invasion fails irrespective of the order of invasion relative to its cottid com- 32 33 petitor. By contrast, the late-maturing, long-lived catostomid, which spreads 33 34 reproductive value over many age classes, can invade successfully even 34 35 though high flows occur at the time of invasion. 35 36 36 37 Fig. 11.1. (continued) 37 38 salmonid is a winter-spawning predator and the prey species spawn in spring and show 38 intraspecific competition only. Percentage change in relative abundance is the change 39 in relative percent based on mean populations over 100 years after the first invasion 39 40 calculated as (experimental treatment-baseline)/baseline × 100. 40 324 E.M. Strange and T.C. Foin 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 Fig. 11.2. Percentage change in relative abundances of model life history types under 37 competition and predation for different flood and drought frequencies alternated with 37 38 50 cfs for invasion sequence catostomid (invades year 0), salmonid (invades year 10), 38 39 cyprinid (invades year 20), and cottid (invades year 30). Relative competitive abilities, 39 predator dynamics, and abscissa are as in Fig. 11.1. 40 40 Predicting fish communities 325 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 Fig. 11.3. Simulated community assembly of competing species for invasion sequence 30 31 cottid–cyprinid–salmonid–catostomid (top) and invasion sequence catostomid–salmonid– 31 32 cyprinid–cottid (bottom) under the average regime for Martis Creek, California with 32 periods of drought. A decade of extreme low flows (May and January mean = 15 cfs, 33 cv = 0.3) was introduced at the time of invasion of the cottid and salmonid. Relative 33 34 competitive abilities as in Fig. 11.1. 34 35 35 36 36 37 In the case of the predator–prey system, the flow regime at the time of 37 38 predator invasion also strongly influences the resulting assemblage (Fig. 11.5). 38 39 When the winter-spawning piscivorous salmonid invades under extreme high 39 40 flows during its spawning season, predator and prey species can coexist if 40 326 E.M. Strange and T.C. Foin 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 Fig. 11.4. Simulated community assembly of competing species for invasion sequence 30 31 cottid–cyprinid–salmonid–catostomid (top) and invasion sequence catostomid–salmonid– 31 32 cottid–cyprinid (bottom) under the average regime for Martis Creek, California with 32 periods of high flows. A decade of spring high flows (May mean = 150 cfs, cv = 0.1) 33 was introduced at the time of invasion of the cyprinid and catostomid. Relative 33 34 competitive abilities as in Fig. 11.1. 34 35 35 36 prey species show comparatively high numbers at the time of predator inva- 36 37 sion, either because they invade before the predator or before predator num- 37 38 bers increase (Fig. 11.5, top). However, if the predator is able to recover from 38 39 deleterious high flows before prey species invade, predator invasion can lead 39 40 to prey extinction (Fig. 11.5, bottom). 40 Predicting fish communities 327 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 Fig. 11.5. Simulated community assembly of a predator–prey system for invasion 31 sequence cottid–cyprinid–salmonid–catostomid (top) and catostomid–salmonid– 31 32 cyprinid–cottid (bottom) under the average regime for Martis Creek, California with 32 33 a decade of high flows in winter (January mean = 150 cfs, cv = 0.1) at the time of 33 34 invasion of the salmonid predator. The cottid, cyprinid, and catostomid spawn in spring 34 and are prey of the winter-spawning salmonid. Prey species show intraspecific 35 competition only. 35 36 36 37 37 38 Model predictions and field observations: Martis Creek 38 39 Model predictions generally show a good fit to census data for both Martis 39 40 Creek and Sagehen Creek. In the case of Martis Creek (Fig. 11.6), the model 40 328 E.M. Strange and T.C. Foin 1 satisfactorily predicts the major trends in species’ abundances over 15 years. 1 2 From 1979–1983, species native to Martis Creek (Tahoe sucker, Lahontan 2 3 redside, Paiute sculpin, and Lahontan speckled dace) dominated the assem- 3 4 blage and showed relatively stable ranked abundances. During this time winter 4 5 and spring discharges were dissimilar and varied from year to year, and brown 5 6 trout and spring-spawning species coexisted, with brown trout showing rela- 6 7 tively low numbers. In 1983, severe spring floods decimated populations of 7 8 spring-spawning species while the winter-spawning brown trout increased 8 9 under favorable winter flows. For several years following the spring floods 9 10 of 1983, winter and spring flows were favorable but native species failed to 10 11 recover from flood-related declines while brown trout showed a dramatic 11 12 increase. Paiute sculpin remained rare and Lahontan redside became locally 12 13 extinct. From 1987 to 1992, drought conditions prevailed in Martis Creek, 13 14 and brown trout declined while remnant populations of native species 14 15 increased. The model also accurately predicts this decline of brown trout and 15 16 recovery of Tahoe sucker and speckled dace during the last years of our study. 16 17 The model does not predict the dramatic reversal in rainbow trout and 17 18 brown trout abundance in 1987, which followed severe winter floods in 1986. 18 19 This is probably a census error. We didn’t census fish populations in Martis 19 20 Creek in 1986, and there may have been important changes in that year which 20 21 are not included in our model. The model also underestimates the decline in 21 22 Tahoe sucker from 1983–1990. This may be a true underestimation, or we 22 23 may have failed to adequately census Tahoe sucker during those years. 23 24 Despite these shortcomings, the accurate prediction of overall trends suggests 24 25 that the model captures key dynamics. 25 26 26 27 27 28 Model predictions and field observations – Sagehen Creek 28 29 The model provides an excellent fit to census data for Sagehen Creek (Fig. 29 30 11.7). Both model predictions and field data show relative stability of the 30 31 Sagehen Creek fish populations over the census period (1952–1961). This 31 32 period included two major floods, one in May 1952 and another in Decem- 32 33 ber 1955. Average annual flows remained favorable and stable. Differences 33 34 in the seasonal timing of floods combined with overall stability of annual 34 35 flows appears to have prevented any major shift in the relative abundances 35 36 of brown trout and prey or competing species in Sagehen Creek. 36 37 37 38 38 39 39 40 40 Predicting fish communities 329 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 Fig 11.6. Model predictions (above) compared to census data (below) for the fish 39 species of Martis Creek, California, 1979–1994. No census was taken in 1986. Data 39 40 are relative percentages based on total numbers. 40 330 E.M. Strange and T.C. Foin 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 Fig 11.7. Model predictions (above) compared to census data (below) for the fish 38 species of Sagehen Creek, California, 1952–1961. Census data are from Gard and Flitt- 39 ner (1974) and Gard and Seegrist (1972). Data are relative percentages based on total 39 40 numbers. 40 Predicting fish communities 331 1 1 Discussion 2 2 3 Assembly and invasion of stream fish communities 3 4 Our results underscore the important role of physical variation in community 4 5 organization and assembly processes. Results suggest that invader success 5 6 will vary depending on how physical conditions at the time of invasion affect 6 7 the ratio of invader to resident species and the outcomes of their interactions. 7 8 As a result, the outcome of assembly from the same species pool will vary 8 9 depending on specific interactions between the physical regime and underly- 9 10 ing biotic factors. Invasion sequence is important, but the historical context 10 11 of changing physical conditions affecting rates of biotic processes ultimately 11 12 determines assemblage structure. 12 13 In the case of the Trout Zone Fish Assemblage of the Sierra Nevada, our 13 14 evidence indicates that the community is readily invaded by brown trout when 14 15 this species has strong recruitment years coupled with low recruitment of 15 16 native stream fishes. Such conditions shift relative abundances in favor of 16 17 brown trout, whose predatory activity can affect recruitment of the other fishes 17 18 significantly. For the same reason, the native community can resist brown 18 19 trout invasion if brown trout recruitment is weak and their own recruitment 19 20 is strong. Thus, the influence of changing physical conditions on population 20 21 densities is critical in determining any particular assembly outcome. 21 22 The ability of our model to predict outcomes, not only in our study stream 22 23 but in one that has the same species pool but is otherwise dynamically inde- 23 24 pendent, is evidence that the model is a reasonable simulation of the dynam- 24 25 ics of the real system. The one exception that is not accounted for very well 25 26 is rainbow trout in Martis Creek, but even in this case it is not clear whether 26 27 the differences between actual and predicted abundance are flaws in the model 27 28 or problems with the census. Otherwise, we are satisfied that the model ver- 28 29 ifies our understanding of assembly processes in such stream fish systems. 29 30 If top-down regulation and predator–prey ratios were the dominant influ- 30 31 ences on community assembly and invasion success, the order of entry into 31 32 the community would have been expected to be more important than it was 32 33 in our simulations. Although there are some effects of invasion order (e.g., 33 34 the success or failure of the short-lived cottid and cyprinid forms), the model 34 35 suggests that variation in stream flows (especially in the form of floods 35 36 and droughts) is the dominant factor determining the relative advantage and 36 37 invasion potential of each of the fishes in the system. 37 38 This has important implications for studies of biological invasions. Previ- 38 39 ous studies have suggested that if the habitat is unfavorable for invading 39 40 species, communities will show ‘environmental resistance’ to invasion (Baltz 40 332 E.M. Strange and T.C. Foin 1 & Moyle 1993), whereas highly interactive, species-rich communities will 1 2 show ‘biotic resistance’ (Post & Pimm, 1983; Pimm, 1989; Case, 1991). How- 2 3 ever, our findings make clear that ongoing physical variation will continually 3 4 alter ratios of invader and resident species and outcomes of their interactions 4 5 if species respond in different ways to changing physical conditions. As a 5 6 result, community susceptibility to invasion will vary, and no community will 6 7 be ‘invasion-resistant’ under all conditions. 7 8 8 9 9 10 Life history–environment interactions 10 11 Our findings also suggest that studies of life history–environment interactions 11 12 could benefit by shifting focus from defining optimal traits or strategies based 12 13 on average physical conditions to predicting how interactions between phys- 13 14 ical and biotic conditions will alter relative success as the environment varies. 14 15 Previous models of life history evolution have focused on average environ- 15 16 ments or habitats and attempted to predict a single optimal strategy in stable 16 17 vs. variable environments. For example, the theory of r/K selection proposes 17 18 that high equilibrium population size (high K) is favored in stable environ- 18 19 ments whereas a high intrinsic growth rate (high r) is favored in variable 19 20 environments (MacArthur, 1962; MacArthur & Wilson, 1967; Pianka, 1970). 20 21 Habitat-based models suggest that habitat characteristics serve as a ‘template’ 21 22 for determining life history success, favoring certain traits and eliminating 22 23 others (Grime, 1974, 1977; Southwood, 1977; Begon, 1985; Silby & Calow, 23 24 1985). However, Stearns (1992) has argued that a major weakness of these 24 25 models is that they attempt to tie life history directly to habitat, whereas what 25 26 is needed is an understanding of the link between habitat, mortality regimes, 26 27 and life history – that is, the mechanisms linking habitat and life history. In 27 28 terms of community assembly, our results suggest that, if different life his- 28 29 tories are favored at different times as physical conditions change, no single 29 30 strategy will show overall advantage in a variable environment, or uniform 30 31 advantage across environmental gradients. 31 32 This suggests that models of community assembly in variable environments 32 33 should seek to identify not only those conditions which favor particular life 33 34 histories, but also conditions that will alter relative success. Following van 34 35 der Valk (1981), Keddy (1992) suggests that ‘the environment acts as a fil- 35 36 ter removing species which lack traits for persisting under a particular set of 36 37 conditions’. In a study of experimental wetlands, Weiher and Keddy (1995a) 37 38 found that environmental filters can constrain assembly and result in a com- 38 39 mon end state even though pathways may differ. Our results suggest that, in 39 40 systems in which the environmental filter continually changes, assembly from 40 Predicting fish communities 333 1 the same species pool is likely to result in multiple or alternate states as the 1 2 physical regime alters rates of biotic processes. Rather than a single subset 2 3 of species or succession to a single endpoint, many outcomes are possible 3 4 depending on how physical and biotic processes interact. Although the 4 5 sequence of states may appear random, knowledge of how specific interactions 5 6 between physical and biological conditions result in particular states can allow 6 7 prediction of critical transitions. 7 8 8 9 9 10 Implications of the model for the concept of assembly rules 10 11 Our study suggests that assembly of a stream fish system can be understood 11 12 by quantifying a small number of factors in realistic combinations. Knowing 12 13 how recruitment is affected by flow conditions and integrating these rela- 13 14 tionships across a series of years was the key to understanding community 14 15 assembly. This is not to claim, however, that the system is simple and eas- 15 16 ily understood; our results also underline how important particular interactions 16 17 between physical and biotic factors are in determining the outcome from a 17 18 fixed pool of interacting species. Precise details matter a great deal. 18 19 How can this study be reconciled with others (e.g., Drake et al., 1993) 19 20 which suggest more dominant, higher-order interactions and greater com- 20 21 plexity? One possibility is that the fishes of stream systems are not organized 21 22 as patches which are readily and routinely invaded, as in Drake’s model or 22 23 in the case of the birds of the Bismarck Archipelago (Diamond, 1975). In 23 24 streams, alien species can only be transplanted or invade from upstream or 24 25 downstream. Thus, it is possible that the system itself has a major role in 25 26 restricting and governing the success of an invader, whereas in terrestrial 26 27 systems invasion is easier and biotic interactions are more problematic. On 27 28 the other hand, Case (1996) has recently shown that simple systems like ours 28 29 can exhibit very complex dynamical behavior because the outcomes of biotic 29 30 interactions vary depending upon their magnitude – exactly what we have 30 31 postulated for the role of seasonally variable stream flows. 31 32 Intensified study of community assembly is a very promising avenue in the 32 33 study of ecological complexity, one of the central unsolved problems in ecol- 33 34 ogy (Weiner, 1995). Our study shows that population interactions with the 34 35 physical environment can be sufficient to understand patterns of relative abun- 35 36 dance, local extinction, and invasion into a community. This implies that a 36 37 limited number of variables can explain fluctuations in relative abundance in 37 38 this system. Perhaps this is a very simple system, but the fact remains that 38 39 we did not need to posit greater complexity due to emergent properties to 39 40 understand the essential features of stream fish assembly dynamics. We 40 334 E.M. Strange and T.C. Foin 1 concede that this does not mean that emergent properties are unimportant in 1 2 the assembly of other ecosystems. One may question whether there are general 2 3 assembly or response rules (Drake, 1990a, Keddy, 1992; Weiher & Keddy, 3 4 1995b) that apply across many ecosystems, or whether it is premature to speak 4 5 of rules. Nevertheless, comparative study of assembly in a number of repre- 5 6 sentative ecosystems is likely to be helpful in illuminating the study of 6 7 complexity in ecology. 7 8 8 9 9 10 Acknowledgments 10 11 We are indebted to Dr Peter Moyle, director of the Martis Creek field stud- 11 12 ies, and the many students of the Department of Wildlife, Fish, and Conser- 12 13 vation Biology at the University of California, Davis who assisted with field 13 14 sampling. This work was supported in part by a Graduate Fellowship to Dr 14 15 Elizabeth Strange from the Electric Power Research Institute. We appreciate 15 16 the comments of Dr Christopher Dewees, Dr Peter Moyle, Dr Evan Weiher, 16 17 and an anonymous reviewer on earlier drafts. 17 18 18 19 19 20 References 20 21 Allen, K.R. (1951). The Horokiwi stream: a study of a trout population. New 21 22 Zealand Marine Department Fisheries Bulletin 10. 238 pp. 22 Baltz, D.M. & Moyle, P.B. (1993). Invasion resistance to introduced species by a 23 native assemblage of California stream fishes. Ecological Applications 23 24 32:246–255. 24 25 Barkai, A. & McQuaid, C. (1988). Predator–prey role reversal in a marine benthic 25 26 ecosystem. Science 242:62–64. 26 Baltz, D.M., Moyle, P.B. & Knight N.J. (1982). Competitive interactions between 27 benthic stream fishes, riffle sculpin, Cottus gulosus, and speckled dace, 27 28 Rhinichthys osculus. Canadian Journal Fisheries and Aquatic Sciences 28 29 39:1502–1511. 29 Begon, M. (1985). A general theory of life-history variation. In Behavioural 30 Ecology: Ecological Consequences of Adaptive Behaviour. ed. R.M. Silby & 30 31 R.H. Smith, pp. 91–97. London: Blackwell Sci. Pub. 31 32 Carpenter, S.R. & Kitchell, J.F. (ed.) (1993). The Trophic Cascade in Lakes. 32 Cambridge: Cambridge University Press. 33 Case, T.J. (1991). Invasion resistance, species build-up and community collapse in 33 34 metapopulation models with interspecies competition. Biological Journal of the 34 35 Linnaean Society 42:239–266. 35 36 Case, T.J. (1996). Surprising behavior from a familiar model and implications for 36 competition theory. American Naturalist 146:961–966. 37 Chesson, P.L. (1986). Environmental variation and the coexistence of species. In 37 38 Community Ecology, ed. J.M. Diamond & T.J. Case, pp. 240–256. NY, USA: 38 39 Harper and Row. 39 Cole, L.C. (1954). The population consequences of life history phenomena. 40 Quarterly Review Biology 29:103–137. 40 Predicting fish communities 335

1 Cooper, E.L. (1953). Mortality rates of brook trout and brown trout in the Pigeon 1 2 River, Ostego County, Michigan. Progressive Fish Culturist 15:163–169. 2 DeAngelis, D.L., Svoboda, L.J., Christensen, S.W. & Vaughan, D.S. Stability and 3 return times of Leslie matrices with density-dependent survival: applications to 3 4 fish populations. Ecological Modelling 8:149–163. 4 5 Diamond, J.M. (1975). Assembly of species communities. In Ecology and 5 6 Evolution of Communities, ed. M.L. Cody and J.M. Diamond, pp. 342–444. 6 Cambridge: Harvard University Press. 7 Drake, J.A. (1988). Models of community assembly and the structure of ecological 7 8 landscapes. In Proceedings of the International Conference on Mathematical 8 9 Ecology, ed. L. Gross, T. Hallam, & S. Levin, pp. 585–604. Singapore: World 9 Press. 10 Drake, J.A. (1990a). Communities as assembled structures: do rules govern pattern? 10 11 Trends in Ecology and Evolution 5: 159–163. 11 12 Drake, J.A. (1990b). The mechanics of community assembly and succession. 12 Journal of Theoretical Biology 147: 213–233. 13 Drake, J.A. (1991). Community-assembly mechanics and the structure of an 13 14 experimental species ensemble. American Naturalist 137: 1–26. 14 15 Drake, J.A., Flum, T.E., Witteman, G.J., Voskuil, T., Hoylman, A.M., Creson, C., 15 16 Kenny, D.A., Huxel, G.R., Larue, C.S. & Duncan, J.R. (1993). The 16 construction and assembly of an ecological landscape. Journal of Animal 17 Ecology 62: 117–130. 17 18 Erman, D.C., Andrews, E.D., & Yoder-Williams, M. (1988). Effects of winter 18 19 floods on fishes in the Sierra Nevada. Canadian Journal of Fisheries and 19 Aquatics Science 45: 2195–2200. 20 Gard, R. & Seegrist, D.W. (1972). Abundance and harvest of trout in Sagehen 20 21 Creek, California. Transactions of the American Fisheries Society 101: 21 22 463–477. 22 Gard, R. & Flittner, G.A. (1974). Distribution and abundance of fishes in Sagehen 23 Creek, California. Journal of Wildlife Management 38: 347–358. 23 24 Garman, G.C. & Nielsen, L.A. (1982). Piscivority by stocked brown trout (Salmo 24 25 trutta) and its impact on the nongame fish community of Bottom Creek, 25 26 Virginia. Canadian Journal of Fisheries and Aquatic Sciences 39: 862–869. 26 Gilpin, M., Carpenter, M.P., & Pomerantz, M.J. (1986). The assembly of a 27 laboratory community: multispecies competition in Drosophila. In Community 27 28 Ecology, ed. J.M. Diamond and T.J. Case, pp. 23–40. NY: Harper and Row. 28 29 Grime, J.P. (1974). Vegetation classification by reference to strategies. Nature 250: 29 26–31. 30 Grime, J.P. (1977). Evidence for the existence of three primary strategies in plants 30 31 and its relevance to ecological and evolutionary theory. American Naturalist 31 32 111: 1169–1194. 32 Hunt, R.L. (1969). Overwinter survival of wild fingerling brook trout in Lawrence 33 Creek, Wisconsin. Journal of the Fisheries Research Board of Canada 26: 33 34 1473–1483. 34 35 Keddy, P.A. (1992). Assembly and response rules: two goals for predictive 35 36 community ecology. Journal of Vegetation Science 3: 157–164. 36 Latta, W.C. (1962). Periodicity of mortality of brook trout during first summer of 37 life. Transactions of the American Fisheries Society 91: 408–411. 37 38 Levin, S.A. & Goodyear, C.P. (1980). Analysis of an age-structured fishery model. 38 39 Journal of Mathematical Biology 9: 245–274. 39 MacArthur, R.H. (1962). Some generalized theorems of natural selection. 40 Proceedings of the National Academy of Sciences, USA 48: 1893–1897. 40 336 E.M. Strange and T.C. Foin

1 MacArthur, R.H. & Wilson, E.O. (1967). Theory of Island Biogeography. 1 2 Princeton: Princeton University Press. 203 pp. 2 McFadden, J.T. (1969). Dynamics and regulation of salmonid populations in 3 streams. In Symposium on Salmon and Trout in Streams, ed. T.G. Northcote, 3 4 pp. 313–329. Institute of Fisheries, University of British Colombia, Vancouver. 4 5 McFadden, J.T. Cooper, E.L. (1962). An ecological comparison of six populations 5 6 of brown trout (Salmo trutta). Transactions of the American Fisheries Society 6 81: 202–217. 7 McFadden, J.T., Alexander, G.R. & Shetter, D.S. (1967). Numerical changes and 7 8 population regulation in brook trout, Salvelinus fontinalis. Journal of the 8 9 Fisheries Research Board of Canada 24: 1425–1459. 9 Matson, P.A. & Hunter, M.D. (ed.) (1992). Top-down and bottom-up forces in 10 population and community ecology. Ecology 73:723–765. 10 11 Meffe, G.K. (1984). Effects of abiotic disturbance on coexistence of predator–prey 11 12 fish species. Ecology 65: 1525–1534. 12 Meffe, G.K. (1985). Predation and species replacement in American southwestern 13 fishes: a case study. The Southwestern Naturalist 30: 173–187. 13 14 Minckley, W.L., & Meffe G.F. (1987). Differential selection by flooding in stream- 14 15 fish communities of the arid American southwest. In Community and 15 16 Evolutionary Ecology of North American Stream Fishes, ed. W.J. Matthews & 16 D.C. Heins, pp. 93–104. University Oklahoma Press, Norman. 17 Mortensen, E. (1977). Population, survival, growth and production of trout Salmo 17 18 trutta in a small Danish stream. Oikos 28: 9–15. 18 19 Moyle, P.B. (1976). Inland Fishes of California. Berkeley, CA, USA: University 19 of California Press. 405 pp. 20 Moyle, P.B. & Vondracek, B. (1985). Persistence and structure of the fish 20 21 assemblage of a small California stream. Ecology 66: 1–13. 21 22 Moyle, P.B. & Herbold, B. (1987). Life-history patterns and community structure in 22 stream fishes of western North America: comparisons with eastern North 23 America and Europe. In Community and Evolutionary Ecology of North 23 24 American Stream Fishes, ed. W.J. Matthews & D.C. Heins, pp. 25–34. 24 25 Norman: University of Oklahoma Press. 25 26 Murphy, G.L. (1968). Pattern in life history and the environment. American 26 Naturalist 102: 391–403. 27 Needham, P.R., Moffett, J.W. & Slater, D.W. (1945). Fluctuations in wild brown 27 28 trout populations in Convict Creek, California. Journal of Wildlife 28 29 Management 9: 9–25. 29 Nee, S. (1990). Community construction. Trends in Ecology and Evolution 5: 30 337–340. 30 31 Nehring, R.B. (1988). Stream fisheries investigations. Federal aid project F-51-R, 31 32 Job No. 1, Fish flow investigations, Federal Aid in Fish and Wildlife 32 Restoration, Job Final Report F-51-R, Colorado Div. Wildl., Fish Res. Sec., 33 Fort Collins. 34 pp. 33 34 Pianka, E.R. (1970). On ‘r’ and ‘K’ selection. American Naturalist 104: 592–597. 34 35 Pimm, S.L. (1989). Theories of predicting success and impact of introduced 35 36 species. In SCOPE 37, Biological Invasions: A Global Perspective, ed. J.A. 36 Drake, H.A. Mooney, F. Di Castri, F. Kruger, R. Groves, M. Rejmanek, & M. 37 Williamson, pp. 351–368. Chichester, UK: John Wiley. 37 38 Post, W.M. & Pimm, S.L. (1983). Community assembly and food web stability. 38 39 Mathematical Biosciences 64: 169–192. 39 Robinson, J.V. & Dickerson, Jr. J.E. (1987). Does invasion sequence affect 40 community structure? Ecology 68: 587–595. 40 Predicting fish communities 337

1 Robinson, J.V. & Edgemon, M.A. (1988). An experimental evaluation of the effect 1 2 of invasion history on community structure. Ecology 69: 1410–1417. 2 Rosenzweig, M.L. (1995). Species Diversity in Space and Time. Cambridge: 3 Cambridge University Press. 3 4 Schaffer, W.M. (1974). Optimal reproductive effort in fluctuating environments. 4 5 American Naturalist 108: 783–790. 5 6 Seegrist, D.W. & Gard, R. (1972). Effects of floods on trout in Sagehen Creek, 6 California. Transactions of the American Fisheries Society 102: 478–482. 7 Shetter, D.S. (1961). Survival of brook trout from egg to fingerling stage in two 7 8 Michigan trout streams. Transactions of the American Fisheries Society 90: 8 9 252–258. 9 Silby, R.M. & Calow, P. (1985). Classification of habitats by selection pressures: a 10 synthesis of life-cycle and r/K theory. In Behavioural Ecology: Ecological 10 11 Consequences of Adaptive Behavior, ed. R.M. Silby & R.H. Smith, pp. 75–90. 11 12 London: Blackwell Science. 12 Southwood, T.R.E. (1977). Habitat, the templet for ecological strategies? Journal of 13 Animal Ecology 46: 337–366. 13 14 Stearns, S.C. (1992). The Evolution of Life Histories. Oxford: Oxford University 14 15 Press. 15 16 Strange, E.M. (1995). Pattern and process in stream fish community organization: 16 field study and simulation modeling. Unpublished doctoral dissertation, 17 University of California, Davis. 17 18 Strange, E.M., Moyle, P.B. & Foin, T.C. (1992). Interactions between stochastic 18 19 and deterministic processes in stream fish community assembly. Environmental 19 Biology of Fishes 36: 1–15. 20 van der Valk, A.G. (1981). Succession in wetlands: a Gleasonian approach. 20 21 Ecology 62: 688–696. 21 22 Weiher, E. & Keddy, P.A. (1995a). The assembly of experimental wetland plant 22 communities. Oikos 73: 323–335. 23 Weiher, E. & Keddy, P.A. (1995b). Assembly rules, null models, and trait 23 24 dispersion: new questions from old patterns. Oikos 74: 159–164. 24 25 Weiner, J. (1995). On the practice of ecology. Journal of Ecology 83: 153–158. 25 26 Wilbur, H. & Alford, R.A. (1985). Priority effects in experimental pond 26 communities: responses of Hyla to Bufo and Rana. Ecology 66: 1106–1114. 27 Winemiller, K.O. (1992). Life-history strategies and the effectiveness of sexual 27 28 selection. Oikos 63: 318–327. 28 29 Winemiller, K.O. & Rose, K.A. (1992). Patterns of life-history diversification in 29 North American fishes: implications for population regulation. Canadian 30 Journal of Fisheries and Aquatic Sciences 49: 2196–2218. 30 31 Wootton, R.J. (1990). Ecology of Teleost Fishes. London: Chapman and Hall. 404 31 32 pp. 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 12 1 2 2 3 Functional implications of trait–environment 3 4 linkages in plant communities 4 5 5 6 6 7 Sandra Díaz, Marcelo Cabido, and Fernando Casanoves 7 8 8 9 9 10 10 11 11 12 12 13 13 14 Introduction 14 15 In this chapter, three fundamental concepts in community ecology are dealt 15 16 with: assembly rules, trait–environment linkages, and plant functional types. 16 17 Although all of them have received considerable attention in the literature, 17 18 they have been rarely discussed in an integrated way. The detection of general 18 19 rules underlying observed patterns has been a major aim of community ecol- 19 20 ogy as it grows into an integrated, predictive science (Keddy, 1989; Drake, 20 21 1990; Barbault & Stearns, 1991; Grime, 1993). The interest in identifying 21 22 consistent and predictable associations between plant traits, types of plants, 22 23 and environmental conditions is an integral part of this search for general- 23 24 ization. These aspects have received renewed interest in the last few years. 24 25 In the face of the challenges of massive loss of biodiversity and global climate 25 26 change, accurate predictions are not anymore simply desirable for the sake 26 27 of ‘good science’. They have become an urgent need. 27 28 This chapter is aimed at presenting an approach in which plant traits are 28 29 used to construct functional types and to identify consistent trait–environment 29 30 linkages. On this basis, present fundamental community/ecosystem processes 30 31 can be predicted, as well as their likely shifts under changing climatic 31 32 conditions. Conceptual issues are first analyzed and empirical studies in the 32 33 literature summarized concerning assembly rules (in a broad sense), trait– 33 34 environment linkages, and functional types. As an illustration of the approach, 34 35 an example of the operation of environmental conditions as filters/assembly 35 36 rules on a regional pool of plant traits along a steep climatic gradient is pre- 36 37 sented. 37 38 38 39 39 40 40

338 Trait–environment linkages in plant communities 339 1 1 2 Assembly rules: interactions and filters 2 3 The issue of assembly rules is central to community ecology. Assembly rules 3 4 are generalized restrictions to coexistence, and represent constraints on how 4 5 communities are selected as subsets of a species pool (Diamond, 1975; Keddy, 5 6 1989; Wilson & Gitay, 1995). The concept of assembly rules was born in the 6 7 context of animal ecology (Diamond, 1975) and most of the theory and empir- 7 8 ical examples to date have remained in that context. Assembly rules have 8 9 been suggested for plants as well (Lawton, 1987; Cody, 1989; Drake, 1990; 9 10 Watkins & Wilson, 1992; Wilson et al., 1995a). However, most of those 10 11 authors emphasize interactions between organisms, rather than with other 11 12 selective forces. Wilson & Gitay (1995) explicitly define an assembly rule as 12 13 ‘a restriction on species presence or abundance that is based on the presence 13 14 or abundance of one or several other species, or types of species (not simply 14 15 the response of individual species to the environment)’. 15 16 Other authors have defined assembly rules in a broader – or looser – sense 16 17 (Fig. 12.1). According to Keddy (1989, 1992), filters of any kind imposed to 17 18 the regional species pool can be regarded as assembly rules. The objective 18 19 of assembly rules should then be the prediction of which subset of the total 19 20 species pool for a given region will occur in a specified habitat. While 20 21 Diamond (1975) associates the idea of assembly rules to ‘forbidden combi- 21 22 nations’, thus emphasizing the importance of biotic interactions, Keddy (1992) 22 23 links them to the idea of deletion, reinforcing his broader focus. He explic- 23 24 itly mentions climatic conditions, disturbance regime, and biotic interactions 24 25 as examples of filters. Antecedents of these ideas may be found in the work 25 26 of Woodward and Diament (1991), although these authors do not explicitly 26 27 mention assembly rules. They proposed a conceptual model in which climate, 27 28 fire (disturbance), and site productivity (interactions) act as successive 28 29 ‘filters’, selecting certain traits and functions out of the regional species pool. 29 30 These ‘filters’ fit into the definition of assembly rules in the broad sense 30 31 mentioned above and adopted hereafter in this chapter. 31 32 32 33 33 34 Filters and trait–environment linkages at different scales 34 35 The concept of trait–environment linkages refers to sets of plant attributes 35 36 consistently associated with certain environmental conditions, irrespective of 36 37 the species involved (Keddy, 1992). It is plant traits (and therefore plant func- 37 38 tion) which is the subject of ‘filtering’ processes (Woodward & Diament, 38 39 1991; Keddy, 1992). Climatic, disturbance, and interaction filters tend to act 39 40 at decreasing spatial (and to some degree temporal) scales (Fig. 12.1). At any 40 340 S. Díaz et al. 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 Fig. 12.1. Assembly rules in a broad sense are filters at different scales imposed to 15 the regional pool of species, traits, or functional types (adapted from Woodward & 16 Diament, 1991 and Keddy, 1992). 16 17 17 18 18 19 particular site, a hierarchy of filters can be found: only those traits/functions 19 20 which can survive under the prevailing climatic conditions, and then under 20 21 the predominant disturbance regime, have the opportunity to be ‘filtered out’ 21 22 (or not) by the interactions with other organisms. 22 23 Most of the work on assembly rules for plants has concentrated on the 23 24 interactions filter (e.g., Wilson & Roxburgh, 1994; Wilson et al., 1995b). On 24 25 the other hand, there are uncountable examples documenting the selective 25 26 action of climate and disturbance on plant communities, but they have not 26 27 been presented within the context of assembly rules. However, the concept 27 28 of ‘filters’ at different scales is present in some cases – either implicitly or 28 29 explicitly. The works of van der Valk (1981) focused in topographic–edaphic 29 30 conditions and interactions between plants, of Noble and Slatyer (1980) on 30 31 disturbance and vital attributes, and of Box (1981) and Woodward (1987) on 31 32 relations between vegetation and climate at regional to global scales are 32 33 examples. 33 34 34 35 35 36 Climate change and changing filters 36 37 There is growing evidence that the Earth’s climate and atmospheric compo- 37 38 sition may change to an unprecedented degree in human history during the next 38 39 century (Houghton et al., 1996). As a consequence, many research groups 39 40 around the world are involved in the prediction of likely responses of natural 40 Trait–environment linkages in plant communities 341 1 and semi-natural plant communities. The consideration of a hierarchy of envi- 1 2 ronmental ‘sieves’ seems a useful approach to this problem. Climate change 2 3 is expected to modify filters at all scales. Although it is regarded mainly as 3 4 a change in climatic filters, it is also likely to modify disturbance filters 4 5 through altered frequency of fires, flammability, flooding regime, or land use 5 6 patterns (Houghton et al., 1996). It may also change the biotic interactions 6 7 filters. This is because species are expected to respond individualistically, as 7 8 they have done in the past (Davis, 1981; Huntley, 1992; Bradshaw & McNeilly, 8 9 1991), and therefore communities will likely disassemble and reassemble in 9 10 different ways, and thus plants may experience different neighborhoods. 10 11 11 12 12 13 Spatial gradients as proxies for temporal change 13 14 Present distribution of organisms in space – especially along steep environ- 14 15 mental gradients – may be used as a model of possible changes in time. 15 16 Paleoecological studies tend to support this view (Solomon & West, 1986; 16 17 Delcourt & Delcourt, 1987, 1991). Richardson and Bond (1991) proposed a 17 18 hierarchy of factors controlling pine invasion: climate, disturbance and biotic 18 19 interactions with the resident biota. They concluded that the effects of pre- 19 20 dicted global warming on the distribution of pines are unlikely to be simple 20 21 functions of temperature and precipitation, except at climatic extremes. 21 22 Using present distribution of vegetation in space as a proxy for temporal 22 23 changes is liable to be inaccurate when important features of the expected 23 24 environmental change have no clear equivalent at present. Examples of these 24

25 are increased levels of atmospheric CO2 or altered neighborhoods, including 25 26 inﬂux of potentially dominant alien species (Woodward & Diament, 1991). 26 27 However, the analysis of present spatial patterns, together with paleoecolog- 27 28 ical data, are arguably the only empirical support in investigating likely 28 29 vegetation changes at a broad scale (region to globe). 29 30 30 31 31 32 Plant traits and ecosystem function 32 33 Within a framework of abiotic constraints and prevailing disturbance regime, 33 34 the individual traits of the dominant living organisms strongly inﬂuence 34 35 community/ecosystem processes (Hobbie, 1992; Schulze & Mooney, 1994; 35 36 Jones et al., 1994; Schulze & Zwölfer, 1994; Schulze, 1995). This is implicit 36 37 in the concept of positive-feedback switches in plant communities, i.e., 37 38 processes in which the dominant organisms modify the environment, making 38 39 it more suitable for themselves (Wilson & Agnew, 1992). Whereas vegeta- 39 40 tive traits (e.g., size, leaf turnover, longevity, chemical composition) tend to 40 1 2 ., 1992; ., 1985; 3 ., 1997

4 et al et al et al 5

6 ., 1997

7 et al

8 ., 1993; Sala ., 1993 ., 1993 ., 1991; Reich 9 ., 1993 et al et al et al 10 et al 11 et al ., 1997 ., 1990 12 ., 1990 et al et al 13 et al 14 15 16 ., 1989; Harris, 1991; Shaver ., 1992 ., 1993 17 et al ., 1997 ., 1987; Gange ., 1988; Shaver ., 1992 ., 1979; Aber & Melillo, 1982; McClaugherty et al

18 et al et al et al et al et al 19 et al 20 Schulze & Chapin, 1987; Hobbie, 1992 Kellliher Leishman 21 Shaver 22 23 24 25 26 27 28 29 30 31 Carrying capacity for herbivores McNaughton Interception, water relations, runoff Calder, 1990; Woodward & Diament, 1991; Holling, 1992: Nutrient cycling McNaughton & Oesterheld, 1990 Production efficiency Reich Roughness/albedoTemperature bufferingSoil stability Schulze, 1982; Holling, 1992 Schulze & Zwölfer, 1994 Holling, 1992 Rate of succession Amaranthus & Perry, 1994 32 33 34 35 36 37 Examples of individual plant traits which strongly influence processes the community/ecosystems in they are dominant 38 39 Table 12.1. BiomassLifespan Flammability Inertia Christensen, 1985; Dublin Richardson & Bond, 1991; Chapin Canopy structure Aerodynamic conductance Jarvis & McNaughton, 1986; Kelliher Nutrient content Nutrient cycling Swift Seed numberDispersal modePresence of root symbionts Diversity Expansion over landscape Expansion over landscape Noble, 1989; Hodgson & Grime, 1990 Howe & Smallwood, 1982; Noble, 1989; Hodgson Grime, 1990; Grime Reserve organsPollination modePersistent seed bank Resilience Resilience Expansion over landscape Faegri & Van der Pijl, 1979; Schulze Zwölfer, 1994 Grime, 1979; Noble & Slatyer, 1980 Thompson & Grime, 1979; Leaf turnover rate Nutrient cycling Schulze & Chapin, 1987; Nadelhoffer Secondary growthRamificationRoot architecture Carbon sequestration Water uptake Structural complexity Larcher, 1995; Schulze, 1982 Lawton, 1983, 1987; Brown, 1991; Marone, 1991 Woodward & Diament, 1991; Kelliher 40 Individual traitsRelative growth rate Community/ecosystem processes Productivity Source Grime Trait–environment linkages in plant communities 343 1 be associated to ecosystem processes in situ (productivity, nutrient cycling, 1 2 carrying capacity), regeneration traits (e.g., seed output, dispersal mode, seed 2 3 persistence) tend to determine stability, recolonization after major distur- 3 4 bances, and migration over the landscape (Table 12.1). This means that, given 4 5 a certain set of abiotic and disturbance constraints, it may be possible to pre- 5 6 dict major community/ecosystem processes on the basis of consistent trait– 6 7 environment linkages. 7 8 8 9 9 10 Plant functional types 10 11 As mentioned earlier, given a regional species pool, local conditions filter 11 12 traits, rather than taxa (Woodward & Diament, 1991; Keddy, 1992). Traits 12 13 are not filtered independently from each other, since selective pressures act 13 14 on integrated individuals (Gould & Lewontin, 1979). There is evidence that 14 15 plant traits tend to be associated in recurrent, predictable patterns. Plant design 15 16 constraints (Grime, 1977; Grime et al., 1988), tight integration among phys- 16 17 iological processes (Chapin, 1980; Chapin et al., 1993), allocational trade- 17 18 offs and codependency between traits (Lambers & Poorter, 1992; Reich et 18 19 al., 1992) widely overlap with phylogenetic constraints in determining these 19 20 consistent specialization patterns (Harvey & Pagel, 1991; Westoby et al., 20 21 1995). 21 22 Because of the existence of these recurrent specialization patterns, the 22 23 apparently unmanageable diversity of plant species in natural ecosystems can 23 24 be summarized into fewer functional types (FTs), namely sets of plants 24 25 exhibiting similar responses to environmental conditions and having similar 25 26 effects on the dominant ecosystem processes (Walker, 1992; Gitay & Noble, 26 27 1997). The attempts to reduce communities to a reasonably low number of 27 28 ‘building blocks’ are as old as ecology itself. They have given origin to a 28 29 myriad of related, and not always clearly defined, concepts, such as guilds, 29 30 functional types, growth forms, and strategies (see Simberloff & Dayan, 1991; 30 31 Gitay & Noble, 1997 for critical review). More recently, and as accurate and 31 32 swift predictions are urgently called for, these issues seem to have come back 32 33 into the limelight of ecological research (e.g., Steffen et al., 1992). The term 33 34 ‘functional types’ is usually preferred, at least within the context of plant 34 35 community ecology. This may be because its connotations are somewhat 35 36 looser than the ones of ‘guilds’ or ‘strategies’. 36 37 In the last years, several research groups have tried to construct plant func- 37 38 tional types in a a posteriori way, on the basis of the screening of numerous 38 39 traits organized into comparative databases. These initiatives vary in aims, geo- 39 40 graphical scope, and the range of species and traits considered. For example, 40 344 S. Díaz et al. 1 based on vegetative and regeneration traits, Grime et al. (1988) classified 1 2 more than 400 herbaceous and woody species within the British flora into 2 3 strategies, with resource use and response to disturbance being the main cri- 3 4 teria. Leishman and Westoby (1992) took a similar approach for 300 species 4 5 of Australian semi-arid woodlands. Montalvo et al. (1991) and Fernández- 5 6 Alés et al. (1993) analyzed vegetative and regeneration traits of 179 and 42 6 7 Mediterranean grassland species, respectively, classifying them into FTs with 7 8 different responses to stress, grazing and ploughing, and discussing them 8 9 within an evolutionary context. Diaz et al. (1992) classified herbaceous plant 9 10 species and populations from central Argentina on the basis of morphologi- 10 11 cal and life-history traits, and analyzed their role in community-level 11 12 processes. McIntyre et al. (1995) classified herbaceous plants from Australia 12 13 according to their responses to disturbance. Golluscio and Sala (1993) clas- 13 14 sified 24 forbs from the Patagonian steppe into groups with different strate- 14 15 gies of soil water uptake. Boutin and Keddy (1993) classified 43 herbaceous 15 16 species from wetlands across eastern North America into guilds, mostly on 16 17 the basis of vegetative traits measured in the field and under controlled con- 17 18 ditions. In all the cases in which both vegetative and regeneration traits were 18 19 considered, there was only weak coupling between the two sets of traits. This 19 20 suggests that a satisfactory classification of FTs on the basis of a single 20 21 criterion is unlikely. 21 22 22 23 23 A case study: trait–environment linkages and FTs in central-western 24 24 Argentina 25 25 26 Climatic variations along a regional gradient 26 27 Diaz and Cabido (1995, 1997) studied dominant plant traits across a steep 27 28 environmental gradient in central-western Argentina, with a difference in 28 29 annual rainfall of more than 800 mm, and a difference in altitude of more 29 30 than 1500 m between extreme points (Table 12.2). The dominant vegetation 30 31 types ranged between montane grasslands (highest and wettest extreme), 31 32 woodlands (intermediate points), and open xerophytic shrublands (driest 32 33 extreme). Open halophytic shrublands on lowlands with saline soils are scat- 33 34 tered along the gradient, and were considered as points representing extreme 34 35 and permanent water deficit. On the basis of published vegetation relevés and 35 36 meteorological data, the gradients were divided into 13 climatically homoge- 36 37 nous sectors (Table 12.2). 37 38 38 39 39 40 40 Trait–environment linkages in plant communities 345

1 Table 12.2. Climatically homogeneous sectors identified along a regional gradient 1 2 in central-western Argentina (ca. 31° 25′–32°S, 64° 10′–68° 37′W). 2 3 3 4 Overall Mean Annual 4 cover Altitude temperature rainfall 5 Sector Dominant vegetation (%) (m asl) (°C) (mm) 5 6 6 7 1 Montane grasslands 100 2155 8.1 911.5 7 2 Montane grasslands 100 1850 8.9 840.4 8 3 Montane grasslands 100 1450 11.4 887.3 8 9 4 Montane grasslands 100 1000 13.1 996 9 10 5 Montane woodlands 100 900 13.1 996 10 6 Montane woodlands 90 750 15.6 826.4 11 7 Montane woodlands 90 600 17.5 662 11 12 8 Xerophytic woodlands 80 350 19.6 520 12 13 9 Xerophytic woodlands 70 368 19.6 520 13 14 10 Xerophytic woodlands 50 652 18.3 381 14 11 Xerophytic open shrublands 50 500 18.2 260 15 12 Xerophytic open shrublands 30 641 18 85 15 16 16 17 Cold dry winters and rainfall heavily concentrated to the warm season are charac- 17 18 teristic of the climate over the whole region. Sector 13 (not shown in table) is 18 19 composed of open halophytic shrublands under conditions corresponding to Sectors 19 9–12. See Fig. 12.2(a) for further details on selected Sectors. 20 20 21 21 22 The regional pool: plant species and functional types 22 23 Diaz and Cabido (1995, 1997) then selected the 100 most abundant plant 23 24 species out of the regional pool. The species set comprised 30 families and 24 25 numerous growth forms. There seems to be no obvious limitation to the 25 26 dispersal of propagules of these species across the entire region. There is no 26 27 substantial geographical barrier, the degree of habitat fragmentation is very 27 28 low, and there are abundant animal dispersers. On the basis of the multi- 28 29 variate analysis of key plant traits, they classified the dominant species into 29 30 eight functional types (Table 12.3). The traits that best discriminated between 30 31 FTs were vegetative traits, which were also strongly correlated with each other. 31 32 There was no consistent association between these attributes and regeneration 32 33 traits (Table 12.3). Different FTs were dominant in different Sectors along 33 34 the gradient (Fig. 12.2(b), for further details see also Diaz & Cabido, 1997). 34 35 35 36 36 37 Climatic conditions as filters 37 38 On the basis of these results, we formally tested whether different climatic 38 39 conditions consistently ‘filtered’ certain categories of plant traits out of the 39 40 regional pool, i.e., what values of specific leaf area, longevity, seed size, etc. 40 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 Neobouteloua lophostachya Juncus uruguensis Briza subaristata Nierembergia hippomanica Plantago myosurus Bidens triplinervia 12 Poa stuckertii Stipa tenuissima Deyeuxia hieronymi 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 and DCA on a set of the 100 most abundant species along two 20 21 21 22 22 23 23 TWINSPAN various seed sizes and shapes, number low to intermediate (<1000), various dispersal agents, wind pollination, reproductive peak in spring to early autumn. seed number low to intermediate (<1000), various seed sizes, shapes, and dispersal agents, animal unspecialized and specialized pollination, reproductive peak highly variable, uncommon in mid summer. 24 various seed sizes, shapes and numbers (<100 to >1000), various dispersal agents, wind pollination, reproductive peak in spring to early autumn, concentrated to mid summer. 24 25 25 26 26 50

27 ≤ 27 3 28 28 29 29 pathway, high 3

30 50 cm) herba- 30 climatic gradients in central-western Argentina (Diaz & Cabido, 1997) ≤ 100 cm). C 31 ≤ 31

32 pathway, high SLA, 32 3 33 33

34 and C 34 4 35 35 36 Plant functional types (FTs) produced by 36 Growth form: tussock grasses and large-leaved forbs ( Growth form: short ( ceous and semi-woody erect, creeping, or rosette-like dicots. C and LWR, annual or short lived, very low carbohydrate storage and C immobilization into xylem, low resistance to drought, shoot biomass peak spring to early autumn. Growth form: short graminoids ( LWR, intermediate SLA, lifespan short to moderate, moderate carbohydrate storage in reserve organs, very pathway, high SLA and LWR, moderately long lived, moderate carbohydrate storage in reserve organs, and some short-term C immobilization into standing dead matter, moderate resistance to drought, shoot biomass peak spring to early autumn, concentrated to mid summer. cm). C 37 37 38 38

39 Table 12.3. 39

40 Functional type FT 1 Vegetative traits Regeneration traits FT 2. FT 3 Examples 40 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 Larrea cuneifolia Zucagnia punctata Heterothalamus alienus Acacia caven Prosopis flexuosa Schinopsis haenkeana 12 Tillandsia bryoides Deinacanthon urbanianum Dyckia floribunda 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 numerous large seeds (4–10 cm), various seed shapes and dispersal agents, animal specialized pollination, reproductive peak in spring to early autumn. seed size intermediate to large (2–10 mm), irregular seed shape, moderate seed number (100–1000), animal- and wind-assisted dispersal, animal specialised pollination, reproductive peak in spring. 24 very numerous large flattened seeds (4–10 mm), various dispersal agents (vertebrate dispersal common), animal specialized pollination, reproductive peak in early spring to autumn. 24 25 25 26 26 path-

27 3 27 28 28 29 29 30 30 pathway, low SLA

31 3 31 32 32 33 33 100 cm). CAM pathway, high

34 ≤ 34 300 cm). C 35 ≤ 35 36 36 and LWR, very long lived, evergreen, high carbohydrate storage and C immobilization, ramiﬁed at the ground level, high resistance to drought, leaf biomass constant or very small peak in spring low C immobilization into xylem and bark, moderate resistance to drought, shoot biomass peak highly variable, uncommon in mid summer. Growth form: saxicolous or epiphytic rosettes ( Growth form: trees (>300 cm). C LWR, very low SLA, evergreen, moderate carbohydrate storage in reserve organs, some short-term C immobilization into standing dead material, moderate resistance to drought, shoot biomass constant along the annual cycle or with a peak in spring. way, high to moderate SLA, low LWR, very long lived, deciduous or semi-deciduous, very high carbohydrate storage and C immobilization, very low ramiﬁcation at ground level, high drought resistance, leaf biomass peak in late spring to early summer. Growth form: evergreen shrubs and small trees ( 37 37 38 38 39 39

40 FT 4 FT 5 FT 6 40 1 1 2 2 3 3 4 4 5 5 6 6

7 spp. 7 8 8 9 9 10 10 11 11 Heterostachys ritteriana Senna aphylla Suaeda divaricata 12 Opuntia sulphurea Cereus validus Gymnocalycium 12 13 13 14 14 15 15 16 16

17 ) 17 18 18 2–4 mm), dispersed 2 mm) and spher- ≤

19 ≤ 19

20 continued 20 21 21 22 22 23 23 Table 12.3. ( numerous small ( oidal seeds, various dispersal and pollination agents, reproductive peak in spring to early autumn. numerous small- to intermediate-sized, spheroidal seeds ( 24 by highly mobile animals, animal specialized pollination, reproductive peak in spring to early autumn. 24 25 25 26 26 27 27 28 28 29 29

30 pathway, 30 31 3 31 32 32 33 33 200 cm). C

34 ≤ 34 35 35 36 36 Growth form: aphyllous or scale-leafed shrubs ( extremely low SLA and LWR, long lived, evergreen, high carbohydrate storage, some C immobilization, highly ramiﬁed at the ground level, high resistance to drought and salinity, green biomass constant throughout the year. Growth form: globular, cylindrical, and columnar-branched stem succulents of various sizes. CAM pathway, aphyllous, with green succulent stems, long to very long lived, evergreen, some carbohydrate storage and C immobilization, very low ramiﬁcation at the ground level, very high resistance to drought, green biomass constant throughout the year. 37 37 38 38 39 39

40 Functional type FT 7 Vegetative traits FT 8 Regeneration traits Examples See Table 12.4 for further explanation of traits and definition scales measurement. 40 Trait–environment linkages in plant communities 349 1 were associated to different sectors along the gradient (see Table 12.4 for 1 2 description of traits and categories). The null hypothesis of trait frequency 2 3 distribution in local sectors being indistinguishable from the regional pool 3 4 was rejected in 71% of the pair-wise comparisons (Table 12.5), strongly sug- 4 5 gesting the existence of environmental filters. Vegetative traits were ‘filtered’ 5 6 more often than regeneration traits (74% and 63% of the individual compar- 6 7 isons, respectively). Specific leaf area (SLA), lifespan, ramification, C immo- 7 8 bilization into support tissue, canopy height, and pollination mode were the 8 9 traits showing differences in the largest number of pair-wise comparisons. 9 10 The climatic factors which appear to have the strongest ‘filtering effect’ at 10 11 the regional level were those related with the low temperatures (both means 11 12 and extremes) predominating at high altitude (>750 m asl). The Sector most 12 13 different from the regional pool were high mountain grasslands, with the 13 14 differences becoming stronger as altitude increased (Sectors 1–4 in Table 14 15 12.5, Sector 1 in Fig. 12.2). The ‘filtering effect’ of water deficit became 15 16 important at lower altitudes (<750 m asl). It was very strong in sitituations 16 17 where water deficit is permanent throughout the year, as in the case of halo- 17 18 phytic shrublands (Sector 13 in Table 12.5 and Fig. 12.2). 18 19 Semiarid to arid woodlands, woodland–shrublands, and shrublands (Sec- 19 20 tors 8–12 in Table 12.5, Sectors 10 and 12 in Fig. 12.2) showed differences 20 21 with the regional pool for a comparatively smaller number of traits. This is 21 22 probably because water deficit disappears or is ameliorated during some 22 23 months, allowing the survival of a relatively wide variety of trait combina- 23 24 tions during the favorable season. There was a set of plants which remain 24 25 active during the whole year, and tend to be evergreen, sometimes succulent, 25

26 with CAM or more commonly C3 photosynthesis. There is also a less 26 27 persistent set of species which is only present or active during the growing 27

28 season. This is represented by annual herbaceous plants (mainly C4 grasses), 28 29 and deciduous shrubs and trees. Finally, montane woodlands (Sector 6 in Fig. 29 30 12.2 and Sectors 5–7 in Table 12.5) were transitional between vegetation and 30 31 climatic types. Accordingly, they presented a variety of categories for each 31 32 plant trait, and thus showed comparatively the smallest differences with the 32 33 regional pool. 33 34 The results summarized in Table 12.5 and Fig. 12.2 strongly suggest a 34 35 ﬁltering effect exerted by climatic conditions at the regional scale. This 35 36 resulted in consistent trait–climate linkages, with possible implications for 36 37 ecosystem processes (discussed in following sections). Moderately to highly 37 38 disturbed areas were excluded from this study, and local biotic interactions 38 39 were not analyzed. However, a similar trait-based multivariate approach can 39 40 be taken to investigate their possible roles as ﬁlters. 40 1 1 2 2

3 = 2 3 4 . 4 5 5 300 = 4; >300–

6 10 = 2; >10 = 3 6 ≤ 7 − 7 8 8 9 9 10 10 = 1; 10–<100 = 2; >100 = 3

11 1 11 − g

12 2 12 13 13 1 = 1; LWR>1 = 2 = 3 3 14 ≅ 14 15 15 600 = 6 = 2; C ≥ 16 4 16 17 17 18 18 20 cm =1; 20–60=2; >60–<100 = 3; 100– zoophyllous = 2 ≤ <600 = 5, 19 succulent stem = 1 late spring–summer, summer = 4 mobility (ants, rodents) = 1; highly mobile animals (large mammals, bats, birds) = 2; wind 3 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 ., 1992) 29 29 30 et al 30 31 ., 1994) 31

32 et al 32 33 33 34 Traits recorded on the 100 most abundant species along a climatic gradient in central-western Argentina 34 35 35 36 36

37 Table 12.4. 37 38 38 39 39 Poorter & Bergkotte, 1992; Reich which cannot be used in further biosynthesis (xylem and bark) woody dicots = 2; with trunk and bark 3 photosynthetic tissue) spring, spring-summer, late summer–early autumn = 3; Carlquist, 1988, Hargrave Specific leaf area (SLA) (surrogate for relative growth rate; aphyllous = 0; >0–<10 cm TraitPhotosynthetic pathway CAM = 1; C Description of classes Ten randomly chosen, healthy-looking adult individuals were considered for traits measured in the field or laboratory. Leaf weight ratio (LWR) LWR<1 = 0; LWR Canopy height Leaf succulence (indicator of resistance to drought and/or salinity) non-succulent = 0; slightly succulent 1; highly LifespanCarbon immobilization into support tissue in the form of compounds herbaceous monocots = 0; dicots 1; semi- Ramification at the ground level (adaptive under severe drought; non-woody species = 0; 1 single trunk 1; 2 annual = 1; biennial 2; 3–10 years 3; 11–50 4; >50 = 5 40 Drought resistance (taproot and/or highly succulent stem)ThorninessShoot phenology (seasonality of maximum production no evident drought-resisting organs = 0; taproot or highly Seed sizeNumber of seeds per plant no evident peak = 1; winter, autumn, early spring 2; late Seed dispersal modePollination mode40 <100 seeds = 1; 100–999 2; 1000–5000 3; >5000 = 4 no thorns = 0; slightly thorny 1; very (e.g., Cactaceae) 2 no obvious dispersal agent = 0; animals with relatively low <2 = 1; 2–<4 = 2; 4–10 3; >10 = 4 anemophyllous = 0; unspecialized zoophyllous 1; specialized Trait–environment linkages in plant communities 351

1 Table 12.5. Summary results of pair-wise comparisons (Chi-squared statistic) 1 2 between trait frequency distributions in local sectors and in the region considered 2 as a whole (regional pool) 3 3 4 4 Sectors 5 5 6 Traits 12345678910111213 6 7 7 Vegetative traits 8 Specific leaf area ****–******** 8 9 Lifespan *****–******* 9 10 Ramification *****–******* 10 Leaf weight ratio *****––****** 11 Carbon immobilization ****–**–***** 11 12 Canopy height ***–*–******* 12 13 Leaf succulence ********––––* 13 Thorniness *****–***–––– 14 Drought resistance ****–––––**** 14 15 Photosynthetic pathway ****–**–––––* 15 16 Shoot phenology ***––*––––––* 16 17 Regeneration traits 17 18 Pollination mode ********–**** 18 Seed size ***––––**–*** 19 Seed number ******––––––* 19 20 Dispersal mode **–––*––***–– 20 21 21 22 * = difference with regional pool significant at P < 0.05; simultaneous inference; 22 23 Bonferroni’s correction (Agresti, 1990); – = no significant difference (P ≥ 0.05). 23 See Tables 12.2 and 12.4 for definition of sectors and traits, respectively. 24 24 25 25 26 Patterns in vegetative traits 26 27 The major patterns of association between different categories of traits and 27 28 Sectors along the gradient are summarized in Fig. 12.2 (c)) (only five repre- 28 29 sentative Sectors included for the sake of simplicity). Different sets of traits 29 30 tended to be associated with each other and with particular Sectors. This is 30 31 particularly obvious on comparison of the extreme situations (Sector 1 vs. 31 32 Sectors 12 and 13 in Fig. 12.2 (c)). High relative growth rate (high specific 32 33 leaf area), high investment in photosynthetic tissue (high leaf weight ratio), 33 34 small stature, and short lifespan were distinctive plant traits under comparatively 34 35 cold and moist conditions. In contrast, low relative growth rate, high invest- 35 36 ment in support tissue, intermediate stature, and high persistence in time were 36 37 associated with severe water deficit. These patterns were reflected in the dom- 37 38 inant FTs in each Sector (Fig. 12.2 (b)), and were consistent with trade-offs 38 39 repeatedly mentioned in the literature (Grime, 1977; Chapin, 1980; Lambers 39 40 & Poorter, 1992; Reich et al., 1992). Woodlands and woodland–shrublands 40 1 2 3

4 he rest of the 5 6 7 8 9 10 11 12 13 ), and expected community/ecosystem processes

14 c

15 ). Frequency distribution of different trait categories c 16 17 18 19 20 ), dominant plant traits (

21 b 22 23 24 25 26 27 ), vegetation structure (

28 a 29 30 31 32 33 34 35 36 37 38 39 ) in representative Sectors along a regional gradient central-western Argentina. See Table 12.2 for climatic conditions t d Sectors, and Table 12.4 for definition scales of measurement traits mentioned in ( Fig. 12.2. Summary of climatic conditions ( ( 40 in all the Sectors are available upon request. Trait–environment linkages in plant communities 353 1 (Sectors 5–11 in Table 12.5, Sectors 6 and 10 in Fig. 12.2) showed not only 1 2 intermediate values for plant traits, but also the widest variety of values. This, 2 3 again, was in agreement with the relative abundance of different FTs in them 3 4 (Fig. 12.2 (b)). 4 5 5 6 Patterns in regeneration traits 6 7 In the case of regeneration traits, anemophylly was the most frequent polli- 7 8 nation syndrome in grasslands and halophytic shrublands (Sectors 1 and 13 8 9 in Fig. 12.2 and Sectors 1–4 and 13 in Table 12.5), whereas specialized 9 10 zoophyllous syndromes (mammal-, bird-, and hymenopteran-pollination) 10 11 were the most frequent in woodlands and xerophytic shrublands (Sectors 6 11 12 to 12 in Fig. 12.2, Sectors 5 to 12 in Table 12.5). Small seeds predominated 12 13 in grasslands and halophytic shrublands. In the first case, this was associated 13 14 with low seed production per plant. In the second case, in contrast, seeds 14 15 were produced in large numbers. The predominant pattern in woodlands and 15 16 xerophytic shrublands was the production of numerous big seeds. As in the 16 17 case of analysis by FTs (Díaz & Cabido, 1997), these traits were only loosely 17 18 associated with each other and with vegetative traits. 18 19 19 20 20 21 Trait–environment linkages, FTs, and ecosystem function 21 22 Assuming that (a) different local environmental conditions are filtering cer- 22 23 tain traits (and therefore FTs) out of the regional pool; and (b) predominant 23 24 plant traits have a considerable influence at the community and ecosystem 24 25 levels (Table 12.1), then the next step in this approach is the prediction of 25 26 the relative magnitude of major community/ecosystem processes on the basis 26 27 of consistent trait–environment linkages (Fig. 12.2 (d)). 27 28 Woodlands and woodland–shrublands (Sectors 6 and 10 in Fig. 12.2, Sec- 28 29 tors 5–10 in Table 12.5) should exhibit maximal plant biomass accumulation, 29 30 carbon sequestration in biomass, temperature buffering, and soil water reten- 30 31 tion, as compared with high-altitude grasslands and xerophytic and halophytic 31 32 vegetation (Sectors 1 and 12, and 13 in Fig. 12.2, Sectors 1–4 and 11–13 in 32 33 Table 12.5). They should also have maximum structural complexity (spatial 33 34 arrangement of plant structures, sensu Brown, 1991), given both by the con- 34 35 vergence of several FTs and by the complex architecture of most trees and 35 36 shrubs. Structural complexity has strong influence on higher trophic levels 36 37 (Lawton, 1983, 1987; Marone, 1991) and its positive association with animal 37 38 diversity in these systems, in particular, has been documented by Gardner et 38 39 al. (1995). The predominance of specialized animal pollination also suggests 39 40 that specialized plant–animal interactions may play a more important role (or 40 354 S. Díaz et al. 1 at least may be more common) in woodlands and woodland–shrublands than 1 2 in other systems. 2 3 In the case of montane grasslands (Sector 1 in Fig. 12.2, Sectors 1–4 in 3 4 Table 12.5), maximum biomass turnover, productivity, and nutrient cycling, 4 5 and only moderate capacity for C sequestration in biomass and water uptake 5 6 by the rhizosphere are some of the expected community/ecosystem processes. 6 7 These systems should also exhibit maximum carrying capacity for ungulates. 7 8 Their dominant functional types have high leaf/support tissue ratio (McNaugton 8 9 et al., 1989). They should also show comparatively high nutritional quality: 9 10 RGR tends to be associated positively with N concentration in plant tissue 10 11 (Lambers & Poorter, 1992; Poorter & Bergkotte, 1992; Reich et al., 1992) 11 12 and negatively with the concentration of defensive compounds (Bryant et al., 12 13 1983). From the structural viewpoint, these communities are far less complex 13 14 than woodlands and shrublands (Díaz et al., 1992, 1994; Gardner et al., 1995). 14 15 Minimum biomass accumulation, biomass turnover, productivity, nutrient 15 16 cycling, C sequestration in biomass, and carrying capacity for ungulates were 16 17 predicted for open shrublands (Sectors 12 and 13 in Fig. 12.2, Sectors 11–13 17 18 in Table 12.5). The low overall vegetation cover per unit area (Table 12.2) 18 19 also supports these predictions. The dominant FTs have various root archi- 19 20 tectures, but the overall root web is less developed than those predominating 20 21 in woodlands and woodland–shrublands. This, together with the low vegeta- 21 22 tion cover, should lead to a comparatively low water uptake by vegetation. 22 23 Structural complexity is higher than in grasslands, but lower than in wood- 23 24 lands and woodland–shrublands. 24 25 Woodlands, woodlands–shrublands, and short open shrublands should 25 26 show maximum persistence if environmental conditions have become unsuit- 26 27 able for regeneration. This is because the dominant FTs are likely to survive 27 28 as adults for very long periods under climatic conditions unfavourable for 28 29 regeneration. On the other hand, their capacity for expansion over the land- 29 30 scape is expected to be moderate to low, mainly because their dominant FTs 30 31 take a long time to establish and reach maturity. Grasslands should show a 31 32 very low persistence in situ under unfavourable conditions, but very fast 32 33 expansion rates, mainly due to the faster establishment and development of 33 34 their constituent FTs, and the predominance of wind pollination. 34 35 Further studies of the actual persistence of seeds in soil banks are clearly 35 36 needed in order to improve these predictions. At first inspection, however, 36 37 the predominance of large seeds in woodlands, woodland–shrublands, and 37 38 open xerophytic shrublands, in contrast to smaller seeds in grasslands and 38 39 especially in halophytic shrublands (associated with high seed numbers) (Fig 39 40 12.2 (c)), suggests differences among Sectors in capacity for regeneration in 40 Trait–environment linkages in plant communities 355 1 situ following disturbances, including extreme climatic events, such as untimely 1 2 or unusually severe droughts or frosts. 2 3 3 4 4 5 Predicting ecosystem function under changing climatic conditions 5 6 It is possible to apply the same rationale explained above to the prediction 6 7 of major community/ecosystem processes in the face of climate change. Under 7 8 changing environmental conditions, there would be shifts in predominant 8 9 plant traits and thus ecosystem functions along regional gradients. One thing 9 10 that should be considered is the persistence in situ (inertia) of the established 10 11 vegetation and the establishment rate of the invading populations. The inter- 11 12 play between these two aspects could lead to very different time lags between 12 13 environmental and vegetation changes. Although it is widely accepted that 13 14 species will show individualistic responses as they have done in the past, it 14 15 is reasonable to expect that those species that show similarities in essential 15 16 functional characteristics will behave in relatively similar ways in their mig- 16 17 ration over the landscape. Therefore, on the assumption of consistent 17 18 trait–environment linkages, one should be able to predict expansions and 18 19 retreats of different functional types, future predominant traits, and future 19 20 ecosystem processes, at different points along regional gradients. This ratio- 20 21 nale is valid provided at least one member of each functional type has the 21 22 capacity to migrate over the landscape fast enough to keep pace with chang- 22 23 ing environmental conditions (Schulze & Swölfer, 1994; Chapin et al., 1996). 23 24 This requisite seems to be met by the plants we analyzed, since there was no 24 25 strong correlation between FTs and dispersal ability. 25 26 We used the climatic predictions given by the GISS global circulation 26 27 model (Hansen et al., 1983) for central-western Argentina, in order to hypoth- 27 28 esize changes in plant distribution over the study area. Current GCMs have 28 29 many inadequacies and significant errors on regional scales (Rochefort & 29 30 Woodward, 1992). Therefore, we stress the fact that this is simply an exercise 30 31 to illustrate the approach. 31

32 By the time atmospheric CO2 concentration reaches 700 ppm, mean monthly 32 33 temperature and rainfall values obtained from GISS suggest a shift towards 33 34 hotter and drier conditions in the case of xerophytic woodlands, shrublands 34 35 and halophytic ecosystems (Sectors 10–13 in Fig. 12.2, Sectors 8–13 in Table 35 36 12.5). Therefore, a progressive predominance of plant traits may be expected 36 37 typical of very dry conditions (Sectors 12 and 13 in Fig. 12.2, Sectors 11–13 37 38 in Table 12.5) over more mesic shrubland–woodlands and woodlands (Sec- 38 39 tor 10 in Fig. 12.2, Sectors 9 and 10 in Table 12.5). A direct consequence 39 40 would be a shift from ecosystem processes listed in the central column of 40 356 S. Díaz et al. 1 Fig. 12.2(d) towards those in the two columns to the right. But these changes 1 2 are likely to be very slow, since the present vegetation is expected to be highly 2 3 persistent in the established phase once environmental conditions have 3 4 become unfavorable, and the invading FTs need a long time to establish and 4 5 reach maturity. The time lag between climate and vegetation changes may be 5 6 of the order of centuries. 6 7 Hotter and wetter conditions are predicted in the case of montane grass- 7 8 lands (Sector 1 in Fig. 12.2, Sectors 1–4 in Table 12.5). A shift in altitudi- 8 9 nal vegetation belts may thus be expected, with plant traits now typical of 9 10 lower montane vegetation (Fig. 12.2(c), second column to the left) predom- 10 11 inating over those now typical of high montane grasslands (Fig. 12.2(c), left 11 12 column). Ecosystem processes listed in Fig. 12.2(d) should change accord- 12 13 ingly. The rates of change should be faster in this case. FTs predominating 13 14 in high-mountain grasslands show low persistence as adults once environ- 14 15 mental conditions have become unfavourable for establishment, and the 15 16 potentially invading FTs can establish and reach maturity at fast to moder- 16 17 ate rates. Changes in plant distribution along this gradient might be evident 17 18 within decades. 18 19 19 20 20 21 Concluding remarks 21 22 Efforts aimed at identifying assembly rules (Keddy, 1992) and at predicting 22 23 ecosystem function (Schulze & Zwölfer, 1994) at the level of functional types 23 24 are seemingly more fruitful than those focused on plant species. Functional 24 25 diversity reduces the number of variables to be dealt with, and it is probably 25 26 a better predictor of ecosystem processes. This is not to underestimate the 26 27 importance of species richness within FTs. This may provide redundancy to 27 28 the system, and therefore enhance stability/resilience (Schulze & Zwölfer, 28 29 1994). 29 30 Because of the occurrence of recurrent patterns of specialization among 30 31 plants, it is possible to identify FTs and trait–environment linkages on the 31 32 basis of selected individual traits. However, vegetative traits influencing in 32 33 situ resource acquisition and storage, and regeneration traits influencing 33 34 recolonization after disturbance and expansion over the landscape are only 34 35 loosely coupled and cannot be predicted from each other. Therefore, predic- 35 36 tions of vegetation responses under changing climatic conditions (Nemani & 36 37 Running, 1989; Chapin, 1993) or disturbance regime (McIntyre et al., 1995) 37 38 simply on the basis of a unique, or a highly reduced set of either vegetative 38 39 or regeneration traits are likely to be inaccurate. 39 40 Using plant traits and information about the fundamental environmental 40 Trait–environment linkages in plant communities 357 1 conditions operating in an area as inputs allows the identification of plant 1 2 functional types and of consistent trait–environment linkages both at the local 2 3 and regional scale. On these bases, predictions of present and future com- 3 4 munity/ecosystem processes can be made. In taking this last step, two essen- 4 5 tial points should be taken into account, in order to minimize the risks 5 6 involved in scaling-up. First, when choosing individual plant traits, not only 6 7 should short-term physiological attributes, such as relative growth rate, be 7 8 considered, but morphogenetic traits, reproduction and dispersion in time and 8 9 space, and relationships with other trophic levels (e.g., root symbionts, her- 9 10 bivores, pollinators) should also be taken into account. Although the predic- 10 11 tion of ecosystem function on the basis of individual plant traits involves a 11 12 scaling-up process, this approach is also a top-down one, since the commu- 12 13 nity/ecosystem level perspective guides the selection of variables to be mea- 13 14 sured at the individual level, rather than the reverse (Keddy, 1992). Secondly, 14 15 empirical testing of predictions at the relevant scale (by means of ecosystem 15 16 experiments and medium to long-term monitoring across the landscape) is 16 17 the next obvious and indispensable step. 17 18 18 19 19 20 Acknowledgments 20 21 21 We are grateful to M. Balzarini, E. Pucheta, N. Fernández, and R. Rodríguez 22 22 for useful comments, to G. Funes and V. Falczuk, for their help in data col- 23 23 lection, and to D. Abal-Solís and G. Milano for preparing graphs and figures. 24 24 This work was supported by Universidad Nacional de Córdoba, Fundación 25 25 Antorchas – The British Council and the International Scientific Cooperation 26 26 Program of the European Union. 27 27 28 28 29 29 30 References 30 31 Aber, J.D. & Melillo, J.M. (1982). Nitrogen immobilization in decaying hardwood 31 leaf litter as a function of initial nitrogen and lignin content. Canadian Journal 32 of Botany 60: 2263–2269. 32 33 Agresti, A. (1990). Categorical data analysis. Chichester, UK: John Wiley. 33 34 Amaranthus, M.P. & Perry, D.A. (1994). The functioning of ectomycorrhizal fungi 34 35 in the field: linkages in space and time. Plant and Soil 159: 133–140. 35 Barbault, R. & Stearns, S. (1991). Towards an evolutionary ecology linking species 36 interactions, life-history strategies and community dynamics: an introduction. 36 37 Acta Oecologica 12: 3–10. 37 38 Boutin, C. & Keddy, P.A. (1993). A functional classification of wetland plants. 38 Journal of Vegetation Science 4: 591–600. 39 Box, E.O. (1981). Macroclimate and Plant forms: An Introduction to Predictive 39 40 Modelling in Phytogeography. The Hague: Junk. 40 358 S. Díaz et al.

1 Bradshaw, A.D. & McNeilly, T. (1991). Evolutionary response to global climatic 1 2 change. Annals of Botany 67: 5–14. 2 Brown, V.K. (1991). The effects of changes in habitat structure during succession 3 in terrestrial communities. In Habitat Structure, The Physical Arrangement of 3 4 Objects in Space, ed. S.S. Bell, E.D. McCoy & H.R. Mushinsky, pp. 141–168. 4 5 London: Chapman Hall. 5 6 Bryant, J.P., Chapin III, F.S., & Klein, D.R. (1983). Carbon/nutrient balance of 6 boreal plants in relation to vertebrate herbivory. Oikos 40: 357–368. 7 Calder, I.R. (1990). Evaporation in the Uplands. Chichester, UK: John Wiley. 7 8 Carlquist, S. (1988). Comparative Wood Anatomy. Berlin: Springer-Verlag. 8 9 Chapin, F.S.III (1980). The mineral nutrition of wild plants. Annual Review of 9 Ecology and Systematics 11: 233–260. 10 Chapin, F.S.III (1993). Functional role of growth forms in ecosystem and global 10 11 processes. In Scaling Physiological Proceses: Leaf to Globe, ed. J.R. 11 12 Ehleringer & C.B. Field, pp. 287–312. San Diego: Academic Press. 12 Chapin, F.S.III, Autumn, K., & Pugnaire, F. (1993). Evolution of suites of traits in 13 response to environmental stress. The American Naturalist 142: S78–S92. 13 14 Chapin, F.S.III, Bret-Harte, M.S., Hobbie, S. & Zhong, H. (1996). Plant functional 14 15 types as predictors of the transient response of arctic vegetation to global 15 16 change. Journal of Vegetaion Science 7: 347–357. 16 Christensen, N.L. (1985). Shurbland ﬁre regimes and their evolutionary 17 consequences. In The Ecology of Natural Disturbance and Patch Dynamics, 17 18 ed. S.T.A.S. Pickett & P.S. White, pp. 85–100. Orlando, FL: Academic Press. 18 19 Cody, M.L. (1989). Discussion: structure and assembly of communities. In 19 Perspectives in Ecological Theory, J. Roughgarden, R.M. May, & S.A. Levin, 20 pp. 227–241. Princeton, NJ: Princeton University Press. 20 21 Davis, M.B. (1981). Quaternary history and the stability of forest communities. In 21 22 Forest Succession: Concepts and Applications, ed. D.C. West, H. Shugart & 22 D.B. Botkin, pp. 132–153. NY: Springer-Verlag. 23 Delcourt, H.R. & Delcourt, P.A. (1991). Quaternary Ecology – A Paleoecological 23 24 Perspective. London: Chapman & Hall. 24 25 Delcourt, P.A. & Delcourt, H.R. (1987). Long Term Forest Dynamics of the 25 26 Temperate Zone. Ecological Studies 63. New York: Springer-Verlag. 26 Diamond, J.M. (1975). Assembly of species communities. In Ecology and 27 Evolution of Communities, ed. M.L. Cody & J.M. Diamond pp. 342–444. 27 28 Cambridge, MA, USA: Hardvard University Press. 28 29 Díaz, S. & Cabido, M. (1995). Plant functional types and trait–environment 29 linkages: a multiscale approach. Bulletin of the ESA (Suppl.) 76: 64. 30 Díaz, S. & Cabido, M. (1997). Plant functional types and ecosystem function in 30 31 relation to global change: a multiscale approach. Journal of Vegetation 31 32 Science 8: 463–474. 32 Díaz, S., Acosta, A., & Cabido, M. (1992). Morphological analysis of herbaceous 33 communities under different grazing regimes. Journal of Vegetation Science 3: 33 34 689–696. 34 35 Díaz, S., Acosta, A., & Cabido, M. (1994). Community structure in montane 35 36 grasslands of central Argentina in relation to land use. Journal of Vegetation 36 Science 5: 483–488. 37 Drake, J.A. (1990). Communities as assembled structures: do rules govern pattern? 37 38 Trends in Ecology and Evolution 5: 159–164. 38 39 Dublin, H.T., Slinclair, A.R.E., & McGlade, J. (1990). Elephants and ﬁre as causes 39 of multiple states in the Srengeti-Mara woodlands. Journal of Animal Ecology: 40 1147–1164. 40 Trait–environment linkages in plant communities 359

1 Faegri, K. & van der Pijl, L. (1979). The Principles of Pollination Ecology. Oxford: 1 2 Pergamon. 2 Fernández-Alés, R., Laffarga, J.M., & Ortega, F. (1993). Strategies in 3 Mediterranean grassland annuals in relation to stress and disturbance. Journal 3 4 of Vegetation Science 3: 313–322. 4 5 Gange, A.C., Brown, B.V., & Farmer, L.M. (1990). A test of mycorrhizal benefit in 5 6 an early successional plant community. New Phytologist 115: 85–91. 6 Gardner, S.M., Cabido, M., Valladares, G.R., & Díaz, S. (1995). The influence of 7 habitat structure on arthropod diversity in Argentine semi-arid Chaco. Journal 7 8 of Vegetation Science 6: 349–356. 8 9 Gitay, H. & Noble, I.R. (1997). What are plant functional types and how should we 9 seek them? In Plant Functional Types, ed. T.M. Smith, H.H. Shugart, & F.I. 10 Woodward, pp. 3–19. Cambridge: Cambridge University Press. 10 11 Golluscio, R. & Sala, O.E. (1993). Plant functional types and ecological strategies 11 12 in Patagonian forbs. Journal of Vegetation Science 4: 839–846. 12 Gould, S.J. & Lewontin, R.C. (1979). The spandrels of San Marco and the 13 Panglossian paradigm: a critique of the adaptationist programme. Proceedings 13 14 of the Royal Society of London 205: 581–598. 14 15 Grime, J.P. (1977). Evidence for the existence of three primary strategies in plants 15 16 and its relevance to ecological and evolutionary theory. The American 16 Naturalist 111: 1169. 17 Grime, J.P. (1979). Plant Strategies and Vegetation Processes. Chichester, UK: 17 18 John Wiley. 18 19 Grime, J.P. (1993). Ecology sans frontières. Oikos 68: 385–392. 19 Grime, J.P., Hodgson, J.G., & Hunt, R. (1988). Comparative Plant Ecology: A 20 Functional Approach to Common British Species. London: Unwin Hyman. 20 21 Grime, J.P., Mackey J.M.L., Hillier, S.H., & Read, D.J. (1987). Floristic diversity 21 22 in a model system using experimental microcosms. Nature 328: 420–422. 22 Hansen, J., Russell, G., Rind, D., Stone, P., Lacis, A., Lebedef, S., Ruedy, R., & 23 Travis, L. (1983). Efficient three-dimensional global models for climate 23 24 studies. Models I and II. Monthly Weather Review 111: 609–662. 24 25 Hargrave, K.R., Kolb, K.J., Ewers, F.W., & Davis, S.D. (1994). Conduit diameter 25 26 and drouht-induced embolism in Salvia mellifera Greene (Labiatae). New 26 Phytologist 126: 695–705. 27 Harris, P. (1991). Classical biocontrol of weeds: its definition, selection of effective 27 28 agents, and administrative-political problems. Canadian Entomologist 123: 28 29 827–849. 29 Harvey, P.H. & Pagel, M.D. (1991). The Comparative Methods in Evolutionary 30 Biology. Oxford: Oxford University Press. 30 31 Hobbie, S.E. (1992). Effects of plant species on nutrient cycling. Trends in Ecology 31 32 and Evolution 7: 336–339. 32 Hodgson, J.G. & Grime, J.P. (1990). The role of dispersal mechanisms, 33 regenerative strategies and seed banks in the vegetation dynamics of the 33 34 British landscape. In Species Dispersal in Agricultural Habitats, R.G.H. Bunce 34 35 & D.C. Howard, pp. 61–81. London: Belhaven. 35 36 Holling, C.S. (1991). Cross-scale morphology, geometry, and dynamics of 36 ecosystems. Ecological Monographs 62: 447–502. 37 Houghton, J.J., Meiro Filho, L.G., Callander, B.A., Harris, N., Kattenberg, A. & 37 38 Maskell, K. (1996). Climate Change 1995: The Science of Climate Change. 38 39 Contribution of Working Group I to the Second Assessment Report of the 39 Intergovernmental Panel on Climate Change. Cambridge: Cambridge 40 University Press. 40 360 S. Díaz et al.

1 Howe, H.F. & Smallwood, J. (1982). Ecology of seed dispersal. Annual Review of 1 2 Ecology and Systematics 13: 201–228. 2 Huntley, B. (1992). Rates of change in the European palynological record of the 3 last 13 000 and their climatic interpretation. Climate Dynamics 6: 185–191. 3 4 Jarvis, P.G. & McNaughton, K.G. (1986). Stomatal control of transpiration: scaling 4 5 up from leaf to region. Advances in Ecological Research 15: 1–49. 5 6 Jones, C.G., Lawton, J.H., & Shachak, M. (1994). Organisms as ecosystem 6 engineers. Oikos 69: 373–386. 7 Keddy, P.A. (1989). Competition. London: Chapman & Hall. 7 8 Keddy, P.A. (1992). Assembly and response rules: two goals for predictive 8 9 community ecology. Journal of Vegetation Science 3: 157–164. 9 Kelliher, F.M., Leuning, R., & Schulze, E-D. (1993). Evaporation and canopy 10 characteristics of coniferous forests and grasslands. Oecologia 95: 153–163. 10 11 Lambers, H. & Poorter, H. (1992). Inherent variation in growth rate between higher 11 12 plants: a searchj for physiological causes and ecological consequences. 12 Advances in Ecological Research 23: 187–261. 13 Larcher, W. (1995). Physiological Plant Ecology. Berlin: Springer-Verlag. 13 14 Lawton, J.H. (1983). Plant architecture and the diversity of phytophagous insects. 14 15 Proceedings Sixth Annual Tall Timbers Fire Ecology Conference. 28: 23–39. 15 16 Lawton, J.H. (1987). Are there assembly rules for successional communities? In 16 Colonization, Succession and Stability, ed. A.J. Gray, M.J. Crawley, & P.J. 17 Edwards, pp. 225–244. Oxford: Blackwell Scientiﬁc Publications. 17 18 Leishman, M.R. & Westoby, M. (1992). Classifying plants into groups on the basis 18 19 of associations of individual traits–evidence from Australian semi-arid 19 woodlands. Journal of Ecology 80: 417–424. 20 Leishman, M.R., Hughes, L., French, K., Amstrong, D., & Westoby, M. (1992). 20 21 Seed and seedling biology in relation to modelling vegetation dynamics under 21 22 global climate change. Australian Journal of Botany 40: 599–613. 22 McClaugherty, C.A., Pastor, J., Aber, J.D. & Melillo, J.M. (1985). Forest litter 23 decomposition in relation to soil nitrogen dynamics and litter quality. Ecology 23 24 66: 266–275. 24 25 McIntyre, S., Lavorel, S., & Tremont, R.M. (1995). Plant life-history attributes: 25 26 their relationship to disturbance response in herbaceous vegetation. Journal of 26 Ecology 83: 31–44. 27 McNaughton, S.J. & Oesterheld, M. (1990). Extramatrical abundance and grass 27 28 nutrition in a tropical grazing ecosystem, the Serengeti National Park, 28 29 Tanzania. Oikos 59, 92–96. 29 McNaughton, S.J., Oesterheld, M., Frank, D.A., & Williams, K.J. (1989). 30 Ecosystem-level patterns of primary productivity and herbivory in terrestrial 30 31 habitats. Nature 341: 142–144. 31 32 Marone, L. (1991). Habitat features affecting bird spatial distribution in the Monte 32 desert, Argentina. Ecologia Australia 1: 77–86. 33 Montalvo, J., Casado, M.A., Levassor, C., & Pineda, F.D. (1991). Adaptation of 33 34 ecological systems: compositional patterns of species and morphological and 34 35 functional traits. Journal of Vegetation Science 2: 655–666. 35 36 Nadelhoffer, K.J., Giblin, A.E., Shaver, G.R., & Laundre, J.A. (1991). Effects of 36 temperature and organic matter quality on C, N, and P mineralization in soils 37 from six arctic ecosystems. Ecology 72: 242–253. 37 38 Nemani, R.R. & Running, S.W. (1989). Testing a theoretical climate–soil–leaf area 38 39 hydrologic equilibrium of forests using satellite data and ecosystem 39 simulations. Agricultural and Forest Meteorology 44: 245–260. 40 Noble, I.R. (1989). Attributes of invaders and the invading process: terrestrial and 40 Trait–environment linkages in plant communities 361

1 vascular plants. In Biological Invasions: A Global Perspective, ed. J.D. Drake 1 2 et al., pp. 301–313. Chichester, UK: John Wiley. 2 Noble, I.R. & Slatyer, R.O. (1980). The use of vital attributes to predict 3 successional changes in plant communities subject to recurrent distrubances. 3 4 Vegetatio 43: 5–21. 4 5 Poorter, H. & Bergkotte, A. (1992). Chemical composition of 24 wild species 5 6 differing in relative growth rate. Plant, Cell and Environment 15: 221–229. 6 Reich, P.B., Walters, M.B., & Ellsworth, D.S. (1992). Leaf life-span in relation to 7 leaf, plant and stand characteristics among diverse ecosystems. Ecological 7 8 Monographs 62: 365–392. 8 9 Richardson, D.M. & Bond, W.J. (1991). Determinants of plant distribution: 9 evidence from pine invasions. The American Naturalist 137: 639–668. 10 Rochefort, L. & Woodward, F.I. (1992). Effects of climate change and a doubling 10

11 of CO2 on vegetation diversity. Journal of Experimental Botany 43: 11 12 1169–1180. 12 Sala, O.E., Lauenroth, W.K., & Golluscio, R.A. (1997). Plant functional types in 13 temperate arid regions. In Plant Functional Types, ed. T.M. Smith, H.H. 13 14 Shugart, & F.I. Woodward, pp. 217–233. Cambridge: Cambridge University 14 15 Press. 15 16 Schulze, E-D. & Chapin, F.S. (1987). Plant specialization to environments of 16 different resource availability. Ecological Studies 61: 120–148. 17 Schulze, E-D. & Mooney, H.A. (1994). Biodiversity and Ecosystem Function. 17 18 Berlin: Springer-Verlag. 18 19 Schulze, E-D. & Zwölfer, H. (1994). Fluxes in ecosystems. In Flux Control in 19 Biological Systems, ed. E-D. Schulze pp. 421–445. New York: Academic 20 Press. 20 21 Schulze, E-D. (1982). Plant life forms and their carbon, water and nutrient 21 22 relations. Encyclopedia of Plant Physiology, New Series 12B: 615–676. 22 Schulze, E-D. (1995). Flux control at the ecosystem level. Trends in Ecology and 23 Evolution 10: 40–43. 23 24 Shaver, G.R., Giblin, A.E., Nadelhoffer, K.J., & Rastetter, E.B. (1997). Plant 24 25 functional types and ecosystem change in Arctic tundras. In Plant Functional 25 26 Types, ed. T.M. Smith, H.H. Shugart, & F.I. Woodward, pp. 153–173. 26 Cambridge: Cambridge University Press. 27 Simberloff, D. & Dayan, T. (1991). The guild concept and the structure of 27 28 ecological communities. Annual Review of Ecology and Systematics 22: 28 29 115–143. 29 Solomon, A.E. & West, D.C. (1986). Atmospheric carbon dioxide change: agent of 30 future forest growth or decline. In Effects of Changes in Stratospheric Ozone 30 31 and Global Climate, ed. J.G. Titus, pp. 23–38. Washington DC: US 31 32 Environmental Protection Agency. 32 Steffen, W.L., Walker, B.H., Ingram, J.S., & Koch, G.W. (1992). Global Change 33 and Terrestrial Ecosystems: The Operational Plan. The Royal Swedish 33 34 Academy of Sciences: IGBP-Secretariat. 34 35 Swift, M.J., Heal, O.W., & Anderson J.M. (1979). Decomposition in Terrestrial 35 36 Ecosystems. Berkeley, CA: University of California Press. 36 Thompson, K. & Grime, J.P. (1979). Seasonal variations in the seed banks of 37 herbceous species in ten contrasting habitats. Journal of Ecology 67: 893–921. 37 38 Thompson, K., Band, S.R., & Hodgson, J.G. (1993). Seed size and shape predict 38 39 persistence in soil. Functional Ecology 7: 236–241. 39 van der Valk, A.G. (1981). Succession in wetlands: a Gleasonian approach. 40 Ecology 62: 688–696. 40 362 S. Díaz et al.

1 Walker, B.H. (1992). Biodiversity and ecological redundancy. Conservation Biology 1 2 6: 18–23. 2 Watkins, A.J. & Wilson, J.B. (1992). Fine-scale community structure of lawns. 3 Journal of Ecology 80: 15–24. 3 4 Westoby, M., Leishman, M.R., & Lord, J.M. (1995). On misinterpreting the 4 5 ‘phylogenetic correction’. Journal of Ecology 83: 531–534. 5 6 Wilson, J.B. & Agnew, A.D.Q. (1992). Positive-feedback switches in plant 6 communities. Advances in Ecological Research 23: 263–336. 7 Wilson, J.B. & Roxburgh, S.H. (1994). A demonstration of guild-based assembly 7 8 rules for a plant community and determination of intrinsic guilds. Oikos 69: 8 9 267–276. 9 Wilson, J.B. & Gitay, H. (1995). Community structure and assembly rules in a 10 dune slack: Variance in richness, guild proportionality, biomass constancy and 10 11 dominance/diversity relations. Vegetatio, 116: 93–106. 11 12 Wilson, J.B., Allen, R.B., & Lee, W.G. (1995a). An assembly rule in the ground 12 and herbaceous strata of a New Zealand rain forest. Functional Ecology 9: 13 61–64. 13 14 Wilson, J.B., Sykes, M.T., & Peet, R.K. (1995b). Time and space in the community 14 15 structure of a species-rich limestone grassland. Journal of Vegetation Science 15 16 6: 729–740. 16 Woodward, F.I. (1987). Climate and Plant Distribution. Cambridge: Cambridge 17 University Press. 17 18 Woodward, F.I. & Diament, A.D. (1991). Functional approaches to predicting the 18 19 ecological effects of global change. Functional Ecology 5: 202–212. 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 13 1 2 2 3 When does restoration succeed? 3 4 4 5 Julie L. Lockwood and Stuart L. Pimm 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 Introduction 14 15 How effective have efforts been to restore nature? Has anyone ever success- 15 16 fully restored a degraded site, species for species, or function for function? 16 17 We attempt to answer these key questions of restoration ecology by review- 17 18 ing 87 published studies. We also try to relate our results to ecological theory, 18 19 which may strike the reader as overly ambitious. Our ambition stems from 19 20 prior experience of assembling other scattered and highly subjective descrip- 20 21 tions of nature – food webs. On reading these early compilations, many field 21 22 workers felt (correctly) that they could collect much better data. As a conse- 22 23 quence, the quality and quantity of new studies improved (Pimm et al., 1991) 23 24 and served to refine food web theory considerably. 24 25 Obviously, all restoration ecologists wish for more and better data. It is 25 26 hoped our ideas will suggest specific choices from the bewildering array of 26 27 observations for them to report. The ecological theories presented here will 27 28 surely need improvement; indeed, it may be necessary to abandon them com- 28 29 pletely. Theories improve most when confronted with data, even data as ill- 29 30 defined as subjective accounts of restoration’s successes and failures. 30 31 The simplest division in the expressed goals of restoration is between func- 31 32 tion and structure. By ‘function’ we mean ecosystem processes, such as pri- 32 33 mary productivity, water purification, soil erosion, and the loss or retention 33 34 of nutrients. By ‘structure’ we mean measures of species composition. These 34 35 measures range from a complete list of an area’s original inhabitants, to such 35 36 qualitative terms as ‘wildlife habitat’. Function and structure are unlikely to 36 37 be independent, of course. What does ecological theory suggest about the 37 38 likelihood of restoring function or structure? 38 39 39 40 40

363 364 J.L. Lockwood and S.L. Pimm 1 1 2 On getting function 2 3 Large-scale surveys of ecosystem processes reveal consistent patterns. It is 3 4 possible to predict terrestrial primary productivity from evapotranspiration 4 5 (Rosenzweig, 1968), secondary productivity from primary (McNaughton et 5 6 al., 1989; Cyr & Pace, 1993), the stocks of soil carbon and nitrogen from a 6 7 site’s Holdridge life-zone (Post & Pasteur, 1985) and many other features 7 8 (see Schulze & Mooney, 1993). These patterns are broadly independent of 8 9 the species the ecosystems contain. This does not mean that species compo- 9 10 sition does not matter, for much scatter remains about these statistical rela- 10 11 tionships. It does suggest, however, that nature achieves these relationships 11 12 using a vast array of possible species combinations. Thus, in order to restore 12 13 ecosystem functions, it may not be necessary to restore any special species 13 14 combination. 14 15 To restore ecosystem functions, it may not even be necessary to have the 15 16 original number of species. The connections between the number of species 16 17 an area supports and its ecosystem processes are tenuous. Clearly, with no 17 18 species there can be no functions. Yet, except for systems with unrealistically 18 19 few species, there is, as yet, little evidence for a relationship between struc- 19 20 ture and function (Naeem et al., 1994; Lockwood & Pimm, 1994). Finally, 20 21 restoration of ecosystem function does not even require native species (Aron- 21 22 son et al., 1993). 22 23 Particular ecosystem functions simply appear to be consistent with a range 23 24 of species numbers and broad arrays of species compositions including native 24 25 and introduced species. Metaphorically, the restoration of function should be 25 26 an easy target to hit. 26 27 27 28 28 29 On getting structure 29 30 In contrast, the restoration of the original species composition, or any par- 30 31 ticular species composition, must surely require a close approximation to the 31 32 original ecosystem processes. For example, without fire some prairies will 32 33 become forests. Although restoration of this ecosystem function may be nec- 33 34 essary to restore structure, surely it is not sufficient. Simply burning an area 34 35 may still not restore the original prairie’s composition. Differences in species 35 36 composition in prairies may be determined in part by biotic interactions super- 36 37 imposed on the template of ecosystem processes. Metaphorically, structure 37 38 will be a hard target to hit, because more constraints must be satisfied. 38 39 Critics will argue that this case has been overstated: subtle differences in 39 40 environmental condition may drive all differences in species composition. 40 When does restoration succeed 365 1 After all, not all prairie fires are the same. Restore these exact conditions and 1 2 perhaps the original species composition will return. 2 3 The experience of restoration provides a test of these competing hypothe- 3 4 ses. If nature is so sloppy as to accommodate a wide variety of species com- 4 5 positions within a narrow range of ecosystem conditions, then function should 5 6 be much easier to restore than structure. There will be many pathways to the 6 7 former and few to the latter. 7 8 8 9 9 10 Partial or complete structure 10 11 Critics will now (rightly) notice that we have applied a double standard. 11 12 ‘Structure’ is meant to be some precise measure of species composition – 12 13 often the original! ‘Function’ is intended to mean some rough measures of 13 14 productivity, nutrient retention, and so on. Society often imposes these dif- 14 15 ferent standards upon restoration. There are circumstances when it does not. 15 16 Some partial structure may be the goal; by this we mean any loose grouping 16 17 of native species but not a particular sub-set of that group – for example, 17 18 hardwood trees, prairie plants, ground cover, ‘habitat for wildlife’ and so on 18 19 are considered partial structure. 19 20 Following earlier arguments, function and partial structure should be 20 21 equally easy to restore: there are multiple ways to achieve both. There are a 21 22 variety of species composition that will provide suitable ‘habitat for wildlife’ 22 23 or include a group of prairie plants. One might define, say, a prairie as easily 23 24 by its partial structure (grasses, forbs, but no trees) as by its processes (short 24 25 hydroperiod, episodic fires). Ease of definition may also equate to ease of 25 26 restoration: the more precise an object’s definition, the more difficult it will 26 27 be to create. 27 28 28 29 29 30 Secondary succession or community assembly? 30 31 So far, a crude theory of the likely patterns in the success of a restoration 31 32 has been derived. It is based on the number of acceptable endpoints and thus 32 33 the number of pathways that will lead us there. Our analogy begs the ques- 33 34 tion: what if there are few pathways, but they are easy, well-worn ones? Per- 34 35 haps familiarity is more important than number. Long-held ideas on secondary 35 36 succession suggest that community composition moves along pathways 36 37 towards a final end-point with goal-driven determination. 37 38 Communities assemble by the addition of species through invasions and 38 39 by the loss of species through extinctions. In this sense, a pathway is the 39 40 sequence of species invasions and extinctions that leads to a particular species 40 366 J.L. Lockwood and S.L. Pimm 1 composition. Succession, by its very name, implies that the pathways are tra- 1 2 ditional and commonly traveled. Some restorations may be able to mimic 2 3 these traditional roles, and so incorporate natural processes of disturbance and 3 4 recovery. 4 5 Unfortunately, there is no guarantee that desired species A need follow the 5 6 introduction of species B. As a society, neither ‘every cog and wheel’ 6 7 (Leopold, 1948), – all the species, nor the ‘assembly instructions’ – the eco- 7 8 logical history of the site have been kept. In ecological succession, species 8 9 follow a time-worn path through community assembly. In many restorations, 9 10 these pathways may be lost due to excessive habitat fragmentation, lingering 10 11 pollution, and so on. 11 12 12 13 13 14 What does the theory of community assembly tell us? 14 15 If the familiar pathway toward community regeneration is lost, what does 15 16 ecological theory tell us to expect? Computer simulations and small-scale 16 17 experiments in community assembly now recognize two endpoints along a 17 18 continuum of possible behaviors (Drake, 1988; Luh & Pimm, 1993; Lock- 18 19 wood et al. 1997). The rate at which species are allowed to attempt colo- 19 20 nization determines the rate at which the species composition of that com- 20 21 munity turns over. Communities in which species are allowed to attempt 21 22 invasion often, churn through those species that establish regularly and 22 23 quickly. Virtually all species in the species pool (i.e., the group of species 23 24 that potentially could be community members) become members of the 24 25 community at some time during the assembly process. Each species persists 25 26 for only a very short period of time; no species come to dominate (Lock- 26 27 wood et al., 1997). 27 28 Communities in which species are allowed to attempt invasion much less 28 29 often will show a greater persistence, and may, in fact, become completely 29 30 invasion resistant. By ‘persistent’ it is typically meant that no (or, in prac- 30 31 tice, little) turnover in species composition occurs (Drake, 1990; Pimm, 1991). 31 32 A small set of all species in the pool will become community members and 32 33 they will persist for long periods. Given that different places will likely expe- 33 34 rience different histories, each species has its chance somewhere, rather than 34 35 sometime (Drake et al., 1993). 35 36 Interestingly, theory also suggests that from any initial set of species, 36 37 several states (each with different species compositions) are possible (Drake, 37 38 1988, 1990; Luh & Pimm, 1993). The presence of a species early in the devel- 38 39 opment of the community may have profound effects on the species compo- 39 40 sition at any selected point in the assembly history. Communities are not just 40 When does restoration succeed 367 1 the result of processes that operate today, but are also the products of a long, 1 2 capricious history. The historical sequence in which species are added to and 2 3 removed from a community determine its extant structure. 3 4 Within the practical realm of restoration, assembly theory then suggests 4 5 that it may be possible to re-establish a set of species, but not always the 5 6 desired set. If species attempt invasion often, any resultant community will 6 7 likely be ephemeral. Indeed, the repeat of any exact mix of species may be 7 8 statistically improbable. If species attempt invasion occasionally, when com- 8 9 position is highly dependent on history, restoring a community from its extant, 9 10 constituent species may be impossible (Drake, 1988, 1990; Pimm, 1991; Luh 10 11 & Pimm, 1993). This Humpty-Dumpty effect (Pimm, 1991) holds that species 11 12 important in determining structure may not be present in the community com- 12 13 position we see today. Thus, there is no guarantee that practical restoration 13 14 can exactly or sufficiently repeat a community’s history, even if all the species 14 15 of the extant community are available for reintroduction, if the reintroduc- 15 16 tion attempts are sufficiently separated in time and abiotic conditions are 16 17 sufficiently replicated. 17 18 In sum, and based on results from community assembly experiments, it is 18 19 first argued that restoring ecosystem functions or partial structure should be 19 20 relatively easy to achieve. The restoration of a particular species composition 20 21 will not be easy or it may be impossible. Second, it is argued that commu- 21 22 nities in which functional or structural goals are successfully restored may be 22 23 transitory and may quickly revert back to an ‘unsuccessful’ state. The only 23 24 exceptions to these two ‘rules’ come when the restorationist can employ the 24 25 natural patterns of secondary succession. 25 26 26 27 27 28 Restoration as ecological experimentation 28 29 The most difficult bridge to span between the disciplines of ecological the- 29 30 ory and restoration practice will likely remain the similarity of their terms 30 31 and the contrariety in their meanings (Pimm, 1984). We now try to translate 31 32 our theoreticians’ lexicon (e.g., persistence and community composition) into 32 33 the vocabulary of practical restoration. In so doing, our method of evaluat- 33 34 ing restoration projects will be outlined within the context of community 34 35 assembly theory. 35 36 Restoration often involves the manipulation of the physical and chemical 36 37 environment. These may involve major changes and we do not underrate their 37 38 difficulty. A restoration attempt is also an uncontrolled experiment in assem- 38 39 bling a community (Pimm, 1991; Luh & Pimm, 1993). Species are added 39 40 (‘seeding’) to the community either by planting or inoculating the restoration 40 368 J.L. Lockwood and S.L. Pimm 1 site with a desired set of species or by natural colonization of species from 1 2 adjacent sources. Species are removed (‘weeding’) either through active 2 3 removal (often actual weeding) of unwanted species or from the unintended 3 4 loss of all individuals of a desired species that is purposefully introduced. 4 5 Recall that assembly theory suggests two ‘rules’, one concerning the 5 6 achievement of persistence and the other the achievement of a particular 6 7 species composition. Our translations between the theoretical and practical 7 8 meanings of these two terms, given below, provide our two criteria for eval- 8 9 uating success and failure. A restored community is considered to be persis- 9 10 tent when the turnover in species composition falls to a sufficiently low level 10 11 (i.e., something resembling the natural turnover rate for a particular system). 11 12 Ideally, those responsible for management of the restoration will cease to 12 13 ‘weed and seed’ when this happens. Thus, when management stops (no more 13 14 weeding and seeding), we consider persistence achieved. 14 15 The appropriate species composition is considered to be successfully 15 16 restored when those responsible for evaluating a project judge that a bio- 16 17 logical goal has been successfully attained. Thus, it is assumed that, if the 17 18 restorationist has been successful at restoring a goal, it is because he has re- 18 19 established an appropriate species composition. From our theory derived 19 20 above, function and partial structure goals are expected to be successfully 20 21 attained using a variety of ‘correct’ species compositions. Structural goals 21 22 can only be attained using a limited set of species compositions. By further 22 23 dividing goals into function, full structure and partial structure categories, 23 24 the rate at which each is successfully attained can be assessed and thus the 24 25 applicability of our theory can be tested. 25 26 Persistence may not be tightly linked with the restoration of species com- 26 27 position. Thus, management may cease even though several goals may not 27 28 have been attained (e.g., successfully establishing a persistent set of prairie 28 29 plants which do not necessarily include those rare species most desired). Sim- 29 30 ilarly, goals may be achieved but management may not stop (e.g., improving 30 31 lake water clarity through continued ‘weeding and seeding’ of desired species). 31 32 Thus, these two criteria are considered relatively independent: one criteria 32 33 may be met without having to necessarily accomplish the other. 33 34 Restoration projects can be reviewed and identified as either ‘successes’ or 34 35 ‘failures’ in terms of these two major features: achievement of persistence 35 36 and an appropriate species composition. First, restorationists are either ‘weed- 36 37 ing and seeding’ or else they have achieved a sufficiently low level of species 37 38 turnover that this is not deemed necessary. Second, restorationists either 38 39 achieve or fail to achieve their other goals. A project is deemed completely 39 40 successful only if it satisfies all projected biological goals and achieves 40 persistence as defined by the cessation of management. When does restoration succeed 369 1 By using these translations of practical to theoretical terms, a project can 1 2 be judged as successful or failed based simply on knowledge gained from 2 3 published reports. Some of these reports appear in peer-reviewed journals. 3 4 Others do not, and may not be as carefully scrutinized. Still others are sec- 4 5 ondhand reports from compilations of case studies. Wherever possible, the 5 6 primary source was obtained along with other corroborating publications. It 6 7 is recognized, however, that not all publications have passed through equally 7 8 stringent publication guidelines and that relying on secondhand interpreta- 8 9 tions may misrepresent results. The results from peer-reviewed and not-peer- 9 10 reviewed plus compilation publications will be compared to assay the effects 10 11 of data quality. 11 12 It is understood that the terms ‘success’ and ‘failure’ are loaded with mean- 12 13 ing. Here it is assumed that practitioners hope to restore some attribute(s) of 13 14 the original community and have these attributes persist for long periods of 14 15 time. Under other assumptions, a project classified as ‘failed’ may be quite 15 16 successful (e.g., the project successfully revitalizes a fishing industry or 16 17 improves the quality of life for nearby residents). 17 18 Our criteria of success and failure incorporate several biases associated 18 19 with the sociological setting of restoration efforts. The authors’ judgment of 19 20 whether or not a specific goal was met is relied upon. There is likely a bias 20 21 toward optimistic judgments on the part of some authors. This bias is diffi- 21 22 cult to assess, but will likely inflate the ease with which goals were attained. 22 23 When management ceases, it will be assumed this is because changes in 23 24 species composition are minimal enough to require no further intervention. 24 25 This is likely not always true. Individuals responsible for an expensive, legally 25 26 mandated restoration efforts, for example, may ‘declare victory and go home’, 26 27 ceasing to assume responsibility regardless of the site’s changing species com- 27 28 position. Although some projects that potentially fit this class can be identi- 28 29 fied, it is not possible to reliably estimate the size of this bias nor correct for 29 30 it. Thus, it is also likely that the ease at which persistence is achieved is over- 30 31 estimated. It is acknowledged that our criteria will consistently err on the side 31 32 of optimism and therefore our conclusions must be carefully tempered. 32 33 33 34 34 35 The literature base 35 36 Literature was gathered using computer databases, citation indexes, and bib- 36 37 liographies. To include a project, it had to satisfy three constraints. First, each 37 38 project must have intended goals. Second, projects in which the damaged 38 39 ecosystem was assigned a new use, through exotic species introduction or 39 40 physical manipulation were excluded (Aronson et al., 1993; Bradshaw 1988). 40 By doing this, the scope of the review has been narrowed to include only 370 J.L. Lockwood and S.L. Pimm 1 those projects that seek to restore some portion of the native ecosystem. Third, 1 2 each project was subjected to at least some initial management intervention. 2 3 This may include simply the cessation of pollution or some other source of 3 4 degradation, but most often there was some physical manipulation of the site. 4 5 Eighty-seven restoration projects met our criteria. Projects were incorpo- 5 6 rated as we found them, so our sample represents how easily they could be 6 7 attained. Thus, if projects describing wetland restorations in the continental 7 8 United States were easier to find than those for forests in central Siberia, they 8 9 were more likely to be included in our analyses. Our review does include a 9 10 variety of community types, however, and they are located around the world. 10 11 The compilation of restoration efforts is representative, not exhaustive. 11 12 All the projects are listed in the Appendix. It includes the author(s) of the 12 13 publications, the location of the project, the success category (see below), and 13 14 the author(s) judgment of whether or not each goal was met. Three descrip- 14 15 tive variables are also included: duration, system type, and size. The dura- 15 16 tion of each project was calculated by subtracting the date of initial man- 16 17 agement from the date of latest publication. Thus, each project is judged 17 18 during the time period for which we have information and not beyond. For 18 19 many recent projects, this information is necessarily preliminary. Projects 19 20 range in duration from 1 to 53 yr with the average being 6.3 yr. Duration is 20 21 categorized as short-term (1–5 yr), medium-term (6–20 yr), and long-term 21 22 (> 20 yr). Each project is classified broadly according to system type. ‘Fresh- 22 23 water’ includes rivers, streams and lakes (15 studies), ‘marine’ includes sea- 23 24 grass beds, rocky subtidal areas, and artificial reefs (18 studies), ‘terrestrial’ 24 25 includes prairies and woodlands (23 studies) and ‘wetland’ includes any area 25 26 which is partially or periodically flooded (31 studies). The size of the site 26 27 where the restoration takes place is grouped into three categories: small (1–10 27 28 ha), medium (11–50 ha), and large (> 51 ha). There are 59 small, 20 medium, 28 29 and only 8 large sites. 29 30 30 31 31 32 Classifying the goals 32 33 Structural goals include anything that pertains to the return of species. The 33 34 success rates of returning a community species-for-species (full structure) v. 34 35 limited sets of appropriate species (partial structure) is of interest, so we report 35 36 these goals separately. 36 37 For example, consider two projects intended to restore a freshwater wet- 37 38 land. One may list the establishment of a ‘sufficient diversity of freshwater 38 39 plants’ as its goal. The second lists the return of each of the species observed 39 40 at the site prior to disturbance. The first desires the return of diversity per se 40 When does restoration succeed 371 1 and thus pays less attention to the original species composition nor even to 1 2 the use of only original species (although not exotics). Some of the original 2 3 species may never return, nor are they encouraged to do so, but the finished 3 4 product still resembles a native freshwater wetland. The return of partial struc- 4 5 ture is sufficient for a successful judgment. The second project demands the 5 6 return not only of native species but most of the original species composition. 6 7 Here, those responsible are very concerned about which species are present 7 8 and which are not. Full structure is desired and required for a successful judg- 8 9 ment. 9 10 Functional goals comprise variables that ecologists consider under the 10 11 broad umbrella of ecosystem processes. Functional goals, like partial struc- 11 12 ture, do not require the return of a specific set of species. Functional goals 12 13 may call for similar intervention as for structural goals. For example, a func- 13 14 tional goal might be achieved by the planting of Spartina to control erosion 14 15 but the goal is that process and not Spartina and its associated biota. 15 16 16 17 17 Classifying the results 18 18 19 Unsuccessful 19 20 In 17 projects, management neither ceased nor were all the biological goals 20 21 achieved. Commonly, projects under this category faced physical restraints. 21 22 For example, improper turbidity in a seagrass restoration prevented the long- 22 23 term establishment of underwater vegetation (Teas, 1977), or previously 23 24 unknown contaminants uncovered during restoration operations caused the 24 25 die-off of large portions of the newly planted marsh vegetation (Josselyn et 25 26 al., 1990). 26 27 27 28 Partially successful 28 29 This class includes the majority of the projects considered (53). There are 29 30 two subclasses of partial success: 30 31 31 32 All goals met but management continues 32 33 In 11 studies, management continues in efforts to control for unwanted species 33 34 or species are continually added to the site, but the other goals are met. 34 35 For example, in many prairie restorations when other than a desired species 35 36 enters, it is actively removed. If a desired species fails or becomes only poorly 36 37 established it is repeatedly introduced. It is possible to achieve intended goals 37 38 while this management continues. In one case, the goals of establishing a few 38 39 dominant prairie grasses and controlling erosion were both met (Kramer, 1980). 39 40 Although those responsible were concerned about species composition and 40 372 J.L. Lockwood and S.L. Pimm 1 continued to manage the site to direct which species established, the goals 1 2 stated at the outset of the project were apparently met by the current, if chang- 2 3 ing, species composition. 3 4 Also included in this class are restorations where management continues 4 5 in efforts to correct problems with physical characteristics of the site but not 5 6 necessarily to control for unwanted species. These projects often face the dif- 6 7 ficulties associated with restoring water quality, soil properties or elevation 7 8 characteristics. For example, when large river systems or lakes are consid- 8 9 ered for restoration those responsible must improve water quality by stopping 9 10 non-point pollution sources (Berger, 1992; Markarewicz & Bertram, 1991). 10 11 Often, such improvements are achieved allowing for the return of many native 11 12 fish. However, the continuation of this function (i.e., providing fish habitat) 12 13 is dependent on the continued improvements in water quality through con- 13 14 stant management. Here functions are restored without much regard to the 14 15 exact species composition that supports that function (i.e., as long as fish have 15 16 something to feed on they will survive; it is not necessary to return the set 16 17 of species that the fish originally fed on). 17 18 18 19 Management stops but not all goals are met 19 20 This type of project is the most common in our review, with 42 case studies. 20 21 Typically, some of the desired species persist and some, but not all, func- 21 22 tions are restored. For example, the Salmon River salt marshes were restored 22 23 through a series of projects initiated in the early 1980s (Frenckel & Morlan, 23 24 1990). The project was successful in restoring a diversity of plants typical of 24 25 salt marshes and biomass production. The native species richness of the site 25 26 was not restored. Since all three goals were mentioned and only two achieved, 26 27 we classify this project as partially successful. 27 28 There is a subset of 13 projects in which management ceased but the reasons 28 29 for the cessation are not clearly detailed or expected. Most often, manage- 29 30 ment ceased despite none of the stated goals having been met. An unusual 30 31 example involves a wetland that, after years of neglect, was set for restora- 31 32 tion in mitigation for construction on a nearby site (Josselyn et al., 1990). As 32 33 their goals, the authors listed restoring a diversity of wetland plants and the 33 34 use of the area by birds. Neither goal was met. Inadvertently, the area did 34 35 provide habitat for an endangered plant, the Humboldt Bay Owl’s clover 35 36 (Orthocarpus castillejoides var. humboldtiensis) causing those responsible to 36 37 cease management (at least temporarily). 37 38 38 39 39 40 40 When does restoration succeed 373

1 Table 13.1. Under our success criterion, the achievement of persistence and a 1 2 species composition that will meet the projected goals of the project determine 2 success 3 3 4 4 Total number of 5 Characteristics of projects that ﬁt Success 5 6 the project characterization category Type of goals listed 6 7 7 Persistent and met all 17 Successful 13 functional goals 8 biological goal 18 partial structure 8 9 1 complete structure 9 10 Persistent and not 42 Partially 41 functional goals 10 all goals met Successful 32 partial structure 11 25 complete structure 11 12 Not persistent and all 11 Partially 13 functional goals 12 13 goals met Successful 10 partial structure 13 14 0 complete structure 14 Not persistent and not 17 Unsuccessful 13 functional goals 15 all goals met 10 partial structure 15 16 7 complete structure 16 17 17 18 Using this criterion, only 17 (20%) of the projects in our literature base we consid- 18 19 ered completely successful. Within those 17 projects, 32 goals were given. Only 19 one of those was the restoration of the native community species-for-species. 20 20 21 21 22 Complete success 22 23 In this class are 17 projects that satisﬁed all prescribed goals and manage- 23 24 ment ceased. For example, a project in Florida to mitigate marsh habitat 24 25 destruction due to phosphate mining listed as its goals the establishment of 25 26 a diversity of trees and shrubs common to freshwater wetlands in the area 26 27 (Erwin, 1985). Both of these goals were met and management was determined 27 28 no longer necessary, thus we consider this project completely successful. 28 29 29 30 30 31 Results 31 32 (a) An equal percentage (20%) of all projects are completely successful or 32 33 unsuccessful; most projects are only partially successful (60%: Table 13.1). 33 34 (b) There is no difference in the success rates reported from publications from 34 35 peer reviewed journals and publications from non-reviewed or compiled 35 36 sources. In 24 peer reviewed publication, 4 projects were judged com- 36 37 plete successes, 4 failed, and 16 achieved partial success. In 63 non- 37 38 reviewed sources, 13 were considered complete successes, 13 failed and 38 39 37 were partially successful. These proportions are almost identical (χ2 39 40 = 0.46, 2df, P = 0.79) and we henceforth discount biases between sources. 40 374 J.L. Lockwood and S.L. Pimm

1 Table 13.2. Biological goals listed by all projects 1 2 2 3 Function Successful Failed Total 3 4 4 Water quality 11 (92%) 1 12 5 Sustaining habitat 16 (73%) 6 22 5 6 Productivity 13 (38%) 21 34 6 7 Erosion control 9 (75%) 3 12 7 Total 49 (61%) 31 180 8 8 9 Structure 9 Partial 46 (66%) 24 70 10 Complete 2 (6%) 32 34 10 11 Total 48 (46%) 56 104 11 12 12 13 There are two types of goals: functional and structural. We divide the structural 13 14 goals into those that are restoring only partial structure (e.g., diversity per se) and 14 those that are restoring complete structure (i.e., portions of the community species- 15 for-species). The number of successes and failures are listed beside each goal type. 15 16 Function and partial structure are considered restored more often than complete 16 17 structure. 17 18 18 19 (c) Of the 184 goals given, 97 (52%) were achieved (Table 13.2). Functional 19 20 goals were listed a total of 80 times and considered restored by the 20 21 author(s) in 49 of those attempts (61%). Structural goals were listed 104 21 22 times with success occurring in 48 of those attempts (46%). 22 23 (d) Most of the successes under structure, however, came from restoring div- 23 24 ersity per se or establishing the dominant species (46 out of the 48 24 25 successes) rather than a speciﬁc species composition (Table 13.2). Restor- 25 26 ing a speciﬁc assemblage species-for-species was listed as a goal 34 times 26 27 but only two attempts were considered successful (6%). Thus, function 27 28 and partial structure are restored at least 10 times more frequently than 28 29 species-for-species restoration (χ2 = 17.76, P < 0.001). 29 30 (e) When we consider the subset of projects deemed ‘completely successful’ 30 31 only one listed among its goals the return of full structure (Southward, 31 32 1979; Hawkins & Southward, 1992). The other goals, function and partial 32 33 structure, were mentioned 13 and 18 times, respectively. The restoration 33 34 of complete structure is considerably underrepresented among this group 34 35 of completely successful projects (χ2 = 7.1, P < 0.05). Those projects that 35 36 wished to restore a diversity of the appropriate species are at least as 36 37 likely to be included in our complete success category as those who 37 38 wished to restore function. However, those projects that attempted to 38 39 restore the original community species-for-species are far less frequently 39 40 considered completely successful. 40 When does restoration succeed 375

1 Table 13.3. No association existed between complete success and such descriptive 1 2 variables as duration, size or type of system 2 3 3 4 Duration Size System type 4 5 Complete χ2 = 1.565 χ2 = 2.346 χ2 = 0.153 5 6 success df = 2 df = 2 df = 3 6 7 P = 0.457 P = 0.3094 P = 0.9849 7 8 8 9 9 10 (f) We found no χ2 association between projects deemed completely suc- 10 11 cessful and any descriptive variables (Table 13.3). This is likely the result 11 12 of combining a large number of projects that incorporate many temporal 12 13 and spatial scales. 13 14 (g) It is considered that 59 of 87 attempts (61%) to have achieved persis- 14 15 tence. This rate falls to 48% if we exclude a set of 17 projects in which 15 16 it is unclear why management ceased. 16 17 17 18 18 19 Discussion 19 20 A broad-scale view of the practice of restoration has purposely been taken to 20 21 search for patterns. Two have been found. First, those who practise restora- 21 22 tion (or who evaluate restoration projects) are frequently pleased with efforts 22 23 to restore such features as water quality, habitat structure suitable for deer, a 23 24 substantial biomass of grasses or trees, or reduce the rate of soil loss. In con- 24 25 trast, only a very few are pleased with efforts to restore a particular species 25 26 composition. Thus, function and partial structure are indeed considered 26 27 restored more often than species composition. 27 28 Second, in a total of 34 deliberate attempts, only two projects restored 28 29 original species composition. In these projects, those responsible were likely 29 30 utilizing secondary successional pathways and the spread of adjacent extant 30 31 communities (Gore & Johnson, 1981; Southward, 1979; Hawkins & South- 31 32 ward, 1992). In one project a river was diverted along a 2.5 km stretch while 32 33 coal was mined from under the original bed. Restoration then proceeded by 33 34 re-inforcing the natural bed, controlling erosion and forcing water flow back 34 35 along its original course. The upstream portion of the river served as the 35 36 source pool for colonizing invertebrates and eventually that set of character- 36 37 istic species established (Gore & Johnson, 1981). The second successful 37 38 attempt was comparable. After an oilspill and its clean-up along a rocky tidal 38 39 coastline, the invertebrate sessile community was destroyed. Eventually, the 39 40 species typical of the area returned after the remaining oil and solvent 40 376 J.L. Lockwood and S.L. Pimm 1 disappeared. Again, the source for the colonizing species was presumably an 1 2 undisturbed adjacent coastline since no reintroduction of species was 2 3 attempted (Southward, 1979; Hawkins & Southward, 1992). Thus, it may be 3 4 that original species richness will only be restored when secondary succes- 4 5 sional pathways are still available for exploitation. 5 6 Two possible theoretical explanations have been outlined above for why 6 7 these patterns exist. The first emphasizes the necessity of exactly replicating 7 8 original physical conditions. The second is the nature of community assem- 8 9 bly dynamics: even given identical physical conditions, restoration of a par- 9 10 ticular species composition may be impossible, at best improbable, and even 10 11 then, only transitory. When restorationists can use the historically familiar, 11 12 frequently explored pathways of secondary succession, it may be possible to 12 13 escape such a pessimistic fate. 13 14 We report a relatively high rate of persistence, which we equate with the 14 15 cessation of management. This may be the product of the relatively short 15 16 average duration of the projects (6.3 yr). It remains to be seen whether per- 16 17 sistence is indeed achieved despite the cessation of management. The level 17 18 of persistence achieved by most restoration projects is important beyond sat- 18 19 isfying political and economic requirements. Collecting long-term community 19 20 turnover data from restoration projects may be the only way to distinguish 20 21 which of the two assembly scenarios described above (i.e., constant species 21 22 turnover or achieving alternative persistent states) best describe most restora- 22 23 tion attempts. 23 24 No matter which of these two assembly models proves the best descriptor 24 25 of most restoration attempts, incorporating several separate restoration 25 26 protocols within any one site may be most efficacious at restoring lost species. 26 27 Restoring original composition is a worthy goal, of course. Those who practise 27 28 restoration should accept that the communities they build might be functional 28 29 replicates, but – as our models predict – they will not be structural replicates. 29 30 Any attempt at hitting that one ‘target’ will almost surely fail for the reasons 30 31 given above. 31 32 Varying the restoration protocol across the restoration site will increase the 32 33 numbers and variety of species the site will support. For example, some areas 33 34 may be devoted to repeated and frequent attempts at re-establishing a set of 34 35 desired native species. In adjacent areas, however, these same species are 35 36 given only a few scattered opportunities at establishment. These variations in 36 37 protocol effectively provide different colonization rates and sequences across 37 38 the restoration site. These differences will likely produce a series of variant 38 39 communities which, taken together, may hold all of the desired native species 39 40 while also providing a complete set of original functions. In their totality, 40 When does restoration succeed 377 1 these variant communities (together called a meta-communities by Wilson, 1 2 1992) may be the best way to ensure that most desired species are restored. 2 3 To our knowledge, this suggestion has not yet been explored by any rest- 3 4 oration. 4 5 5 6 6 7 References 7 8 Aronson, J, Floret, C. Le Floch E. Ovalle C., & Pontanier, R. (1993). Restoration 8 and rehabilitation of degraded ecosystems in arid and semi-arid lands. I. A 9 view from the south. Restoration Ecology 1: 8–17. 9 10 Berger, J.J. (1992). Restoring attributes of the Willamette River. In Restoration of 10 11 Aquatic Ecosystems. National Resource Council, Washington, DC: National 11 Academic Press. 12 Bradshaw, A.D. (1998). Alternative endpoints for reclamation. In Rehabilitating 12 13 Damaged Ecosystems vol II, ed. John Cairns Jr. Boca Raton, FA: CRC Press, 13 14 Inc. 14 15 Cyr, H. & Pace, M.L. (1993). Magnitude and patterns of herbivory in aquatic and 15 terrestrial ecosystems. Nature 361 (6488): 148–149. 16 Drake, J.A. (1988). Models of community assembly and the structure of ecological 16 17 landscapes. In Mathematical Ecology, ed. T. Hallam, L. Gross, & S. Levin, 17 18 Singapore: World Press. 18 Drake, J.A. (1990). The mechanics of community assembly and succession. Journal 19 of Theoretical Biology. 147: 213–233. 19 20 Drake, J.A., Flum, T.E., Witteman, G.J., Voskuil, T., Hoylman, A.M., Creson, C., 20 21 Kenny, D.A., Huxel, G.R., LaRue, C.S. & Duncan, J.R. (1993). The 21 construction and assembly of an ecological landscape. Journal of Animal 22 Ecology 62: 117–130. 22 23 Erwin, K.L., Best, G.R., Dunn, W.J. & Wallace, P.M. (1985). Marsh and forested 23 24 wetland reclamation of a central Florida phosphate mine. Wetlands 4: 87–104. 24 25 Frenkel, R.E. & Morlan, J.C. (1990). Restoration of the Salmon River salt marshes: 25 retrospect and prospect. Final Report to the US Environmental Protection 26 Agency, Seattle, Washington. 26 27 Gore, J.A. & Johnson, L.S. (1981). Strip-mined river restoration. Water Spectrum 27 28 13(1): 31–38. 28 Hawkins, S.J. & Southward, A.J. (1992). The Torrey Canyon oil spill: recovery of 29 rocky shore communities. In Restoring the Nation’s Marine Environment, ed. 29 30 G.W. Thayer, Maryland Sea Grant College, College Park, MD. 30 31 Josselyn, M., Zedler, J., & Griswold, T. (1990). Lincoln street marsh wetland 31 mitigation (Appendix II). In Wetland Creation and Restoration: Status of the 32 Science, ed. J.A. Kusler & M.E. Kentula. Washington, DC: Island Press. 32 33 Kramer, G.L. (1980). Prairie studies at caterpillar tractor co. Peoria proving ground. 33 34 In Proceedings of the Seventh North American Prairie Conference, ed. C.L. 34 35 Kucera. University of Missouri, Columbia, MO. 35 Leopold, A. (1948). A Sand County Almanac, Oxford: Oxford University Press. 36 Lockwood, J.L. & Pimm, S.L. (1994). Species: would any of them be missed? 36 37 Current Biology 4(5): 455–457. 37 38 Lockwood, J.L., Powell, R., Nott, M.P., & Pimm, S.L. (1997). Assembling 38 ecological communities in time and space. Oikos 80: 549–553. 39 Luh, H-K. & Pimm, S.L. (1993). The assembly of ecological communities: a 39 40 minimalist approach. Journal of Animal Ecology 62: 749–765. 40 378 J.L. Lockwood and S.L. Pimm

1 McNaughton, S.J., Oesterheld, M., Frank, D.A., & Williams, K.J. (1989). 1 2 Ecosystem productivity and herbivory in terrestrial habitats. Nature 341: 2 142–144. 3 Makarewicz, J.C. & Bertram, P. (1991). Evidence for the restoration of the Lake 3 4 Erie ecosystem. BioScience 41(4): 216–223. 4 5 Naeem, S., Thompson, L.J., Lawlor, S.P., Lawton, J.H., & Woodﬁn, R.M. (1994). 5 6 Declining biodiversity can alter the performance of ecosystems. Nature 368: 6 734–737. 7 Pimm, S.L. (1984). The complexity and stability of ecosystems. Nature 307: 7 8 321–326. 8 9 Pimm, S.L. (1991). The Balance of Nature? Ecological issues In the Conservation 9 of Species and Communities. Chicago, IL: University of Chicago Press. 10 Pimm, S.L., Lawton, J.H., & Cohen, J.E. (1991). Food web patterns and their 10 11 consequences. Nature 350: 669–674. 11 12 Post, W.M. & Pasteur. A. (1985). Global patterns in carbon cycling. Nature 317: 12 613–616. 13 Rosenzweig, M.L. (1968). Net primary productivity of terrestrial environments: 13 14 predictions from climatological data. American Naturalist 102: 67–84. 14 15 Schulze, E-D. & Mooney, H.A. (1993). Biodiversity and ecosystem function. 15 16 Berlin: Springer-Verlag. 16 Southward, A.J. (1979). Cyclic ﬂuctuations in population density during eleven 17 years recolonization of rocky shores in West Cornwall following the ‘Torrey 17 18 Canyon’ oil-spill in 1967. In Cyclic Phenomena in Marine Plants and 18 19 Animals, ed. E. Naylor & A.G. Harnoll. Oxford: Pergamon Press. 19 Teas, H.J. (1977). Ecology and restoration of mangrove shorelines in Florida. 20 Environmental Conservation 4(1): 51–58. 20 21 Wilson, D.S. (1992). Complex interactions in metacommunities, with implications 21 22 for biodiversity and higher levels of selection. Ecology 73: 1984–2000. 22 23 23 24 24 Appendix 25 25 26 All projects used in this study are listed by author. Also included are location, 26 27 achievement of persistence and all goals, success according to both criteria, 27 28 and the intended goals. Each project is categorized under three descriptive 28 29 variables: duration, system type, and size. 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 Functional (2) noPartial 2 (3) no 2 Functional (3) no 1 Complete (2) no 1 15 Partial (2) no 1 15 16 16 17 17 18 18 19 19

20 of Goal Type All Complete System 20

21 Project attributes 21 22 22 23 23 24 24 Table 13.4 25 25 26 26 27 27 28 28 29 29 30 30 31 31 Homestead Monument Santa Barbara, CA Diego County, CA 32 Watershed, CAmine restoration Functional (2) no 2 Partial32 (2) yes 1 33 33 34 34 35 35

36 . No Vernal Pools in San Yes No36 Partial Complete (1) no 1 Freshwater 1 37 37 et al . . Shoreline, CA Partial (2) yes 1 38 . 38 39 39 et al et al Wheeler, R.L.et al Front Range, CO Functional (2) yes 1 Josselyn, M. No Bracut Marsh, CA Yes No Partial Partial (1) no 2 Wetland 1 Hutchinson, D.E. NoPritchett, D.A. Prairie restoration at Josselyn NoJosselyn, M. No Vernal Pools in No Yes Hayward Regional Yes Partial Partial Yes No (1) yes Partial 3 No Complete (1) no Partial Terrestrial 1 2 Functional (1) no 1 Freshwater 1 Wetland 1 Berger, J.J.Berger, J.J.Buckner, D.L. & No NoErwin, K.L. Mattole River No Freshwater marsh, Fonseca, M.S. Blanco River, CO Yes Yes Yes Agrico phosphate , FL No Eelgrass bed, NC Yes Yes Yes No Yes Yes Yes No No Yes Partial Partial No Partial Partial Functional (1) yes (1) yes Partial 1 2 (1) yes (1) no Complete 1 1 (1) no Wetland Freshwater 1 3 1 Freshwater Wetland 2 2 Marine 40 Author(s)Berger, J.J. Peer Location of project No Williamette River, OR Persistence? goals? No success? goal No attained? Duration No type Functional Size (1) yes 1 Freshwater 2 40 1 1 2 2 3 3 4 4

5 System 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 Functional (3) yesFunctional (4) yes 1 Functional (5) yes 1 Partial 1 (6) yes 1 Functional (3) yesFunctional (4) no 1 1 15 Functional (2) noFunctional (3) noPartial 2 2 (4) no 2 15 16 16

17 ) 17 18 18 19 19 continued 20 of Goal Type All Complete 20 21 21 22 22 23 23 24 Table 13.4 ( 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 bed, FL second site, FL Partial (2) yes 1 32 Lakes, UK32 Partial (2) yes 1 33 33 34 34 35 35

36 . Yes Norfolk Broadland Yes Yes Yes36 Functional (1) yes 1 Freshwater 37 37 et al . CA Functional (2) yes 1 . bed, FL Complete (2) no 2 38 . . & 38 39 39 Willard, D.E.Willard, D.E. Sandusky Bay, OH Functional (2) yes 2 Partial (2) yes 1 et al et al et al et al Levin D.A. & Levin D.A. & No The Harbor Project, NoMoss, B. Lack Puckaway, WI Yes Yes Yes Yes No Partial Partial Functional (1) yes (1) yes 2 1 Wetland Freshwater 2 1 Author(s)Josselyn, M. Peer No Location of project Lincoln Street Marsh, Persistence? No goals? success? goal Yes Partial attained? Duration type Functional (1) yes 1 Size Wetland 1 Roberts, L. Thorhaug, A.Erwin, K.L. YesMclaughlin, P.A. Florida Keys seagrass Yes Yes Yes Turkey Point seagrass Agrico phosphate mine Yes Yes Yes Yes Yes No Yes Partial Partial Functional Partial (1) yes (1) yes 1 2 (1) yes 1 Marine Marine 1 1 Wetland 1 40 Butler, R.S. Josselyn, M. Yes Lake Tohopekaliga, FL No Sweetwater Marsh, CA Yes Yes No No Partial40 Partial Complete (1) yes Partial 1 (1) yes 2 Freshwater 2 Wetland 1 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1

Asuquo Obot, E. Yes Nigerian freshwater Yes No Partial Complete (1) no 2 Wetland 2 wetland Smith, I. et al. Yes Seagrass bed, NC Yes No Partial Functional (1) no 1 Marine 1 Anderson, B.W. No Lower Colorado River Yes No Partial Partial (1) yes 1 Terrestrial 1 & Ohmart, R.D. Test plots Complete (2) no 1 Childress, D.A. No MT Yes Yes Yes Partial (1) yes 1 Wetland 2 & Eng, R.L. Hueckel, G.J. Yes Elliott Bay, Puget Yes No Partial Functional (1) yes 1 Marine 1 et al. Sound rocky intertidal Jessee, W.N. Yes Pendelton Artiﬁcial Yes No Partial Functional (1) no 1 Marine 1 et al., Carter Reef, CA Functional (2) yes 1 et al. Teas, H.J. Yes Mangrove shorelines, No No No Functional (1) no 1 Wetland 1 FL Partial (2) no 1 Cammen, L. Yes Salt Marsh, NC Yes No Partial Functional (1) no 1 Wetland 2 Functional (2) no 1 Functional (3) yes 1 Gore, J.A. & . Yes Tongue River, WY Yes No Partial Complete (1) yes 1 Freshwater 1 Johnson, L.S Complete (2) no 1 Functional (3) no 1 Brunori, C. No Freshwater marsh along No Yes Partial Functional (1) yes 1 Wetland 1 Elk River, MA Brunori, C. No Freshwater marsh on No Yes Partial Functional (1) yes 1 Wetland 1 Bohemia River, MA Bontje, M.P. No Salt Marsh, NJ Yes No Partial Partial (1) no 1 Wetland 2 Partial (2) yes 1 Partial (3) yes 1 Partial (4) no 1 Partial (5) no 1 Brown, V.K. & Yes Grassland in UK Yes No Partial Complete (1) no 2 Terrestrial 2 Gibson, C.W.D. 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1

Table 13.4 (continued)

All Complete Type of Goal System Author(s) Peer Location of project Persistence? goals? success? goal attained? Duration type Size

Broome, S.W. No Saltwater Marsh at No No No Functional (1) no 1 Wetland 1 Pamlico estuarary, NC Carangelo, P.D. No Hog Island mitigation, TX No No No Partial (1) no 1 Marine 1 Carangelo P.D. No Huffco Project, TX Yes No Partial Functional (1) no 1 Marine 1 Functional (2) no 1 Carangelo, P.D. No TX No No No Functional (1) no 1 Marine 1 Carangelo, P.D. No Clark Island Project, TX No No No Functional (1) no 1 Marine 1 Bragg, T.B. No Allwine Prairie No No No Complete (1) no 1 Terrestrial 1 Transplants, NE Bragg, T.B. No NE No Yes Partial Partial (1) yes 2 Terrestrial 2 Steigman, K.L. No Heard Wildlife No No No Complete (1) no 1 Terrestrial 1 & Ovenden, L. Sancturary transplants, Nichols, O. et al. No Jarrah forest of Yes No Partial Functional (1) no 1 Terrestrial 3 Western Australia Complete (2) no 1 Functional (3) yes 1 Hayes, T.D. No Rolling plains prairie No Yes Partial Partial (1) yes 2 Terrestrial 2 et al. of north-cental Texas Partial (2) yes 2 Southward, A.J. No Rocky shores of West Yes Yes Yes Complete (1) yes 2 Marine 2 Cornwall, UK Pickart, A.J. No Dune revegetation at Yes No Partial Complete (1) no 1 Terrestrial 1 Buhne Point, CA Functional (2) yes 1 Functional (3) no 1 Kramer, G.L. No Prairie establishment, IA No Yes Partial Functional (1) yes 1 Terrestrial 1 Partial (2) yes 1 Drake, L.D. No Prairie establishment, IA Yes Yes Yes Functional (1) yes 1 Terrestrial 1 Partial (2) yes 1 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 PartialFunctional (4) yes (3) yes 1 1 15 Functional (3) no 1 Partial (3) noFunctional 2 (3) yes 1 PartialPartial (2) yesFunctional 1 (3) yes (2) no 1 1 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31

32 in Prairie, WIof DuPage, IL Functional (2) no 1 32 33 33 34 34 35 35 36 36 37 37

38 .38 Apalachicola Bay, F Functional (2) yes 1 39 39 & K.C. Haines Croix, Virgin Islands Partial (2) no 1 & Dinsmore, J.JCharette, D.J. wetlands, IA& Williams M shorelines, FL Functional (2) yes 2 Partial Partial (2) no 1 (2) no 1 & Martin, M.A. Newling, C.J. Galveston Bay, TX Functional (2) yes 1 Lewis, R.R. III No Mangrove restoration St Yes No Partial Functional (1) no 1 Wetland 2 LaGrange, T.G. YesShisler, J.K. & Freshwater NoBeeman, S. Salt Marshes, N.J.Gillo, J.L. Yes No YesGoforth, H.W. Jr. Salt Marsh, FL No NoNewling, C.J. al No Mangrove Marshland, FLet No Partial No Partial No Partial Saltwater marsh in L Partial No (1) yes Yes Yes (1) no 2 Yes 1 Partial No Functional Yes Wetland No (1) yes No Wetland Yes 1 1 Partial 1 Functional Functional Functional (1) yes (1) yes Wetland (1) yes 1 1 1 1 Wetland Wetland Wetland 1 1 1 Holler J.R. No Microfungla community No No No Complete (1) no 1 Terrestrial 1 40 Woehler, E.E. Dale, E.E. Jr. No& Smith, T.C.Kirt, R.R. Prairie, WI NoWebb, J.W. and Prairie at Pea Ridge Yes National Military Park, AR No Salt Marshes No Prairie at College Yes No Yes No Yes No Partial No Complete Partial No (1) no Partial Partial (1) no 1 Complete 1 Functional (1) no (1) yes 1 Terrestrial 1 1 Terrestrial 1 Terrestrial Wetland 1 1 40 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1

Table 13.4 (continued)

All Complete Type of Goal System Author(s) Peer Location of project Persistence? goals? success? goal attained? Duration type Size

Lewis, R.R. III No Seagrass bed at Yes No Partial Partial (1) no 1 Marine 1 & R.C. Phillips Craig Key, FL Functional (2) no 1 Kenworthy, W.J. No Seagrass bed in Back Yes No Partial Functional (1) yes 1 Marine 1 et al. Sound, NC Partial (2) yes 1 Functional (3) no 1 George, D.H. No Lagoon at Cape Yes Yes Yes Partial (1) yes 1 Wetland 1 Canaveral, FL Partial (2) yes 1 Hoffman, R.S. No Eelgrass bed at Yes No Partial Complete (1) no 1 Marine 1 Mission Bay, CA Functional (2) no 1 Harrison, P.G. No Eelgrass bed Yes Yes Yes Functional (1) yes 1 Marine 1 southwestern British Partial (2) yes 1 Nitsos, R. No Eelgrass transplants in Yes Yes Yes Partial (1) yes 1 Marine 1 Morro Bay, CA Functional (2) yes 1 Merkel, K.W. No Eelgrass bed in Chula , Yes No Partial Partial (1) no 1 Marine 1 Vista Wildlife Reserve Frenkel, R.E. & No Salmon River Salt Yes No Partial Complete (1) no 2 Wetland 2 Morian, J.C. Marshes, OR Functional (2) yes 2 Partial (3) yes 2 Allen, H.H. & Yes Saltmarsh in No No No Partial (1) no 1 Wetland 1 Webb J.W. Mobile Bay, AL Functional (2) no 1 Lewis, R.R. III No Needlerush marsh, FL Yes No Partial Partial (1) yes 1 Wetland 1 Functional (2) no 1 Berger, J.J. & No Kissimmee River Yes No Partial Complete (1) no 1 Wetland 3 Kent, M. et al. demonstration project, FL Complete (2) no 1 Complete (3) no 1 Complete (4) no 1 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1

Sperry, T.M./ Yes University of Wisconsion Yes No Partial Complete (1) no 3 Terrestrial 2 Zimmerman J.H. Arboretum Prairie, WI J.A. Schwarzmeier/ Partial (2) yes 3 Greene, H.C. & J.T. Curtis Woehler, E.E. & No Prairies on public Yes No Partial Parital (1) no 1 Terrestrial 1 Martin, M.A. land, WI Partial (2) yes 1 Wombacher, J. No Prairie at Knox No No No Partial (1) no 3 Terrestrial 1 & R. Garay/ College, IL Complete (2) no 3 Burton, P.J. et al. Complete (3) no 3 Burton, P.J. No Morton Aboretum, IL No No No Partial (1) no 3 Terrestrial 1 et al./ Complete (2) no 3 Schulenberg, R. Partial (3) no 3 Makarewicz, Yes Lake Erie No No No Functional (1) no2 Freshwater 3 J.C. & Bertram P. Partial (2) yes 2 Charlton M.N. et al. Urabe, J. Yes Japanese Pond Yes Yes Yes Functional (1) yes 2 Freshwater 1 Functional (2) yes 2 Lutz, K.A. No Prairie, OH Yes No Partial Complete (1) no 2 Terrestrial 1 Partial (2) yes 2 Edwards, R.D. No Saltwater Marsh in Yes No Partial Complete (1) no 1 Wetland 1 & Woodhouse, Pamlico Esturary, NC Partial (2) yes 1 W.W. Functional (3) yes 1 Haynes, R.J. & No Bottomland forest in Yes Yes Yes Partial (1) yes 2 Terrestrial 2 Moore, L. the Southeastern US Functional (2) yes 2 Partial (3) yes 2 Vassar, J.W. No Prairie in Mark Twain Yes Yes Yes Partial (1) yes 1 Terrestrial 2 et al. National Forest, MO Kilburn, P.D. Prairies, IL Yes Yes Yes Partial (1) yes 1 Terrestrial 1 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1

Table 13.4 (continued)

All Complete Type of Goal System Author(s) Peer Location of project Persistence? goals? success? goal attained? Duration type Size

Keller, W. & . Yes Sudbury Lakes, No No No Functional (1) yes 2 Freshwater 3 Yan, N.D Ontario, Canada Complete (2) no 2 Gameson, A.L.H. . No Thames Estuary, UK No Yes Partial Functional (1) yes 3 Wetland 3 & Wheeler, A Partial (2) yes 3 Sparks, R. No Illinois River No Yes Partial Functional (1) yes 3 Freshwater 3 Functional (2) no 3 Partial (3) yes 3 Edmondson, W.T. No Lake Washington, WA Yes No Partial Functional (1) yes 2 Freshwater 3 Functional (2) no 2 Partial (3) no 2 Partial (4) yes 2 Moore. D.C. & Yes Garroch Head, Isle of No Yes Partial Partial (1) yes 2 Marine 2 Rodger, G.K. Bute,Scotland Functional (2) yes 2 Olmstead, L.I. & . Yes Mud Creek, AR Yes No Partial Complete (1) no 1 Freshwater 1 Cloutman, D.G Functional (2) yes 1 Functional (3) no 1 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 When does restoration succeed 387

1 References 1 2 Allen, H.H. & Webb, J.W. (1983). Influence of breakwaters on artificial saltmarsh 2 3 establishment on dredged material. In Proceedings of the Ninth Annual 3 4 Conference on Wetlands Restoration and Creation, ed. F.J. Webb Jr. 4 5 Hillsborough Community College, Tampa, FA. 5 Anderson, M.R. & Cottam, G. (1970). Vegetational change on the Greene prairie in 6 relation to soil characteristics. In Proceedings of a Symposium on Prairies and 6 7 Prairie Restoration, ed. P. Schramm, Knox College, Galesburg, IL. 7 8 Anderson, B.W. & Ohmart, R.D. (1979). In The Mitigation Symposium: a national 8 workshop in mitigating losses of fish and wildlife habitats. General technical 9 report RM-65. USDA Forest Service. 9 10 Asuquo Obot, E., Chinda, A., & Braid, S. (1992). Vegetation recovery and 10 11 herbaceous production in a freshwater wetland 19 years after a major oil spill. 11 African Journal of Ecology 30: 149–156. 12 Beeman, S. (1983). Techniques for the creation and maintenance of intertidal 12 13 saltmarsh wetlands for landscaping and shoreline protection. In Proceedings of 13 14 the Tenth Annual Conference on Wetland Restoration and Creation, ed. F.J. 14 15 Webb, Hillsborough Community College, Tampa, FA. 15 Berger, J.J. (1992a). Restoring attributes of the Willamette River. In Restoration of 16 Aquatic Ecosystems. National Resource Council, Washington, DC: National 16 17 Academic Press. 17 18 Berger, J.J. (1992b). Citizen restoration efforts in the Mattole river watershed. In 18 Restoration of Aquatic Ecosystems. National Resource Council. Washington, 19 DC: National Academic Press. 19 20 Berger, J.J. (1992c). The Blanco River. In Restoration of Aquatic Ecosystems. 20 21 National Resource Council. Washington, DC: National Academic Press. 21 Bontje, M.P. (1987). The application of science and engineering to restore a salt 22 marsh, 1987. In Proceedings of a Conference: Increasing Our Wetland 22 23 Resources, ed. J. Zelany & J.S. Feierbend. Washington, DC: National Wildlife 23 24 Federation. 24 25 Bragg, T.B. (1976). Allwine prairie preserve: a reestablished bluestem grassland 25 research area. In Proceedings of the 5th North American Prairie Conference, 26 Iowa State University, IA. 26 27 Bragg, T.B. (1988). Prairie transplants: preserving ecological diversity. In The 27 28 Prairie: Roots of Our culture; Foundation of our economy. Proceedings of the 28 Tenth North American Prairie Conference, ed. A. Davis & G. Stanford. 29 Dallas, TX: Native Prairie Association of Texas. 29 30 Broome, S.W. (1987). Creation and development of brackish-water marsh habitat. 30 31 In Proceedings of a Conference: Increasing our Wetland Resources, ed. J. 31 Zelany & J.S. Feierbend. Washington, DC: National Wildlife Federation. 32 Brown, V.K. & Gibson, C.W.D. (1994). Re-creation of species-rich calcichlorous 32 33 grassland communities. In Grassland Management and Nature Conservation, 33 34 ed. R. Haggar. 34 35 Brunori, C.R. (1987). Examples of wetland creation and enhancement in Maryland. 35 In Proceedings of a Conference: Increasing our Wetland Resources, ed. J. 36 Zelany & J.S. Feierbend. Washington, DC: National Wildlife Federation. 36 37 Buckner, D.L. & Wheeler, R.L. (1990). Construction of cattail wetlands along the 37 38 east slope of the Front Range in Colorado. In Environmental Restoration, ed. 38 J.J. Berger. Washington, DC: Island Press. 39 Burton, P.J., Robertson, K.R., Iverson, L.R. & Risser, P.G. (1988). Use of resource 39 40 partitioning and disturbance regimes in the designing and management of 40 388 J.L. Lockwood and S.L. Pimm

1 restored prairies. In Reconstruction of Disturbed Arid Lands: An Ecological 1 2 Approach, ed. E.B. Allen. Boulder, CO: Westview Press, Inc. 2 Butler, R.S., Hulon, M.W., Moyer, E.J. & Williams, V.P. (1992). Littoral zone 3 invertebrate communities as affected by a habitat restoration project on Lake 3 4 Tohopekaliga, Florida. Journal of Freshwater Ecology 7(3): 317–328. 4 5 Cammen, L.M. (1976a). Macroinvertebrate colonization of Spartina marshes 5 6 artificially established on dredge spoil. Estuarine and Coastal Marine Science 6 4: 357–372. 7 Cammen, L.M. (1976b). Abundance and production of macroinvertebrates from 7 8 natural and artificially established salt marshes in North Carolina. The 8 9 American Midland Naturalists 96(2): 487–493. 9 Carangelo, P.D. (1987). Creation of seagrass habitat in Texas: results of research 10 investigations and applied programs. In Proceedings of a Conference: 10 11 Increasing our Wetland Resources, ed. J. Zelany & J.S. Feierbend, 11 12 Washington, DC: National Wildlife Federation. 12 Carter, J.W., Jessee, W.N. Foster, M.S. & Carpenter, A.L. (1985). Management of 13 artificial reefs designed to support natural communities. Bulletin of Marine 13 14 Science 37(1): 114–128. 14 15 Charlton, M.N., Milne J.E. Booth, W.G. & Chiocchio, F. (1993). Lake Erie 15 16 offshore in 1990: restoration and resilience in the Central Basin. Journal of 16 Great Lakes Research 19(2): 291–309. 17 Childress, D.A. & Eng, R.L. (1979). Dust abatement project with wildlife 17 18 enhancement on Canyon Ferry Reservoir, Montana. In The mitigation 18 19 symposium: a national workshop in mitigating losses of fish and wildlife 19 habitats. General technical report RM–65. USDA Forest Service. 20 Dale, E.E. Jr. & Smith, T.C. (1980). Changes in vegetation on a restored prairie at 20 21 Pea Ridge National Military Park, Arkansas. In Proceedings of the Seventh 21 22 North American Prairie Conference, ed. C.L. Kucera. University of Missouri, 22 Columbia. 23 Edmondson, W.T. (1977). Recovery of Lake Washington from eutrophication. In 23 24 Recovery and Restoration of Damaged Ecosystems, ed. J. Cairns Jr., K.L. 24 25 Dickson, & E.E. Herricks. Charlottesville, VA: University of Virginia Press. 25 26 Edwards R.D. & Woodhouse Jr. W.W. (1982). Brackish marsh development. In 26 Proceedings of the Ninth Annual Conference on Wetlands Restoration and 27 Creation, ed. F.J. Webb Jr. Hillsborough Community College, Tampa, FL. 27 28 Erwin, K.L. (1990). Agrico 8.4 acre wetland (Appendix II). In: Wetland Creation 28 29 and Restoration: Status of the Science ed. J.A. Kusler, & M.E. Kentula. 29 Washington, DC: Island Press. 30 Erwin, K.L., Best, G.R. Dunn, W.J. & Wallace, P.M. (1985). Marsh and forested 30 31 wetland reclamation of a central Florida phosphate mine. Wetlands 4: 87–104. 31 32 Fonseca, M.S., Kenworthy, W.J. Colby D.R. Rittmaster, K.A. & Thayer, G. (1990). 32 Comparison of fauna among natural and transplanted eelgrass Zostera marina 33 meadows: criteria for mitigation. Marine Ecology Progress Series 65: 33 34 251–264. 34 35 Frenkel, R.E. & Morland, J.C. (1990). Restoration of the Salmon River salt 35 36 marshes: retrospect and prospect. Final Report to the US Environmental 36 Protection Agency, Seattle, WA. 37 Gameson, A.L.H. & Wheeler, A. (1977). Restoration and recovery of the Thames 37 38 Estuary. In: Recovery and Restoration of Damaged Ecosystems, ed. J. Cairns 38 39 Jr., K.L. Dickson, & E.E. Herricks. Charlottesville, VA: University of Virginia 39 Press. 40 George, D.H. (1982). Lagoon restoration at Cape Canaveral Air Force Station, 40 When does restoration succeed 389

1 Florida. In Proceedings of the Ninth Annual Conference on Wetlands Restoration 1 2 and Creation, ed. F.J. Webb Jr. Hillsborough Community College, Tampa, FL. 2 Gilio, J.L. (1983). Conversion of an impacted freshwater wet prairie into a 3 functional aesthetic marshland. In Proceedings of the Tenth Annual Conference 3 4 on Wetland Restoration and Creation, ed. F.J. Webb. Hillsborough 4 5 Community College, Tampa, FL. 5 6 Goforth, H.W., Jr. & Williams, M. (1983). Survival and growth of red mangroves 6 (Rhizophora manle L.) planted upon marl shorelines in the Florida Keys (A 7 five-year study). In Proceedings of the Tenth Annual Conference on Wetland 7 8 Restoration and Creation, ed. F.J. Webb, Hillsborough Community College, 8 9 Tampa, FL. 9 Gore, J.A. & Johnson, L.S. (1981). Strip-mined river restoration. Water Spectrum 10 13(1): 31–38. 10 11 Green, H.C & Curtis, J.T. (1953). The re-establishment of prairie in the University 11 12 of Wisconsin arboretum. Wildflower 29: 77–88. 12 Harrison, P.G. (1988). Experimental eelgrass transplants in southwestern British 13 Columbia, Canada. In Proceedings of the California Eelgrass Symposium: 13 14 Chula Vista, California, ed. K.W. Merckel & J.L. Stuckrath, National City, 14 15 CA: Sweetwater River Press. 15 16 Hayes, T.D., Riskind, D.H. & Pace III, W.L. (1987). Patch-within-patch restoration 16 of man-modified landscapes within Texas state parks. In Landscape 17 Heterogeneity and Disturbance, ed. M.G. Turner. NY: Springer-Verlag. 17 18 Haynes, R.J. & Moore, L. (1987). Re-establishment of bottomland hardwoods 18 19 within National Wildlife Refuges in the southeast. In Proceedings of a 19 Conference: Increasing our Wetland Resources, ed. J. Zelany & J.S. Feierbend 20 Washington, DC: National Wildlife Federation. 20 21 Hoffman, R.S. (1988a). Fishery utilization of natural versus transplanted eelgrass 21 22 beds in Mission Bay, San Diego, California. In Proceeding of the California 22 Eelgrass Symposium: Chula Vista, California, ed. K.W. Merckel & J.L. 23 Stuckrath. National City, Sweetwater River Press. 23 24 Hoffman, R.S. (1988b) Recovery of eelgrass beds in Mission Bay, San Diego, 24 25 California following beach restoration work. In Proceedings of the California 25 26 Eelgrass Symposium: Chula Vista, California, ed. K.W. Merckel & J.L. 26 Stuckrath. National City, CA: Sweetwater River Press. 27 Holler, J.R. (1981). Studies of soil microfungal populations of a prairie restoration 27 28 project. Proceedings of the Seventh North American Prairie Conference, ed., 28 29 C.L. Kucera. University of Missouri, Columbia. 29 Hueckel, G.J., Buckley, R.M., & Benson, B.L. (1989). Mitigating rocky habitat loss 30 using artificial reefs. Bulletin of Marine Science 44(2): 913–922. 30 31 Hutchison, D.E. (1992). Restoration of the Tall Grass prairie at Homestead 31 32 Monument, Beatrice, Nebraska. Rangelands 14(3): 174. 32 Jessee, W.N., Carpenter, A.L., & Carter, J.W. (1985). Distribution patterns and 33 density estimates of fishes on a southern California artificial reef with 33 34 comparisons to natural reef-kelp habitats. Bulletin of Marine Science 37(1): 34 35 214–226. 35 36 Josselyn, M. Zedler, J. & Griswold, T. (1990a). Vernal pool project in San Diego 36 (Appendix II). In Wetland Creation and Restoration: Status of the Science, ed. 37 J.A. Kusler & M.E. Kentula. Washington, DC: Island Press. 37 38 Josselyn, M. Zedler, J. & Griswold, T. (1990b). Sweetwater Marsh mitigation in 38 39 San Diego County (Appendix II). In Wetland Creation and Restoration: Status 39 of the Science, ed. J.A. Kusler & M.E. Kentula. Washington, DC: Island 40 Press. 40 390 J.L. Lockwood and S.L. Pimm

1 Josselyn, M. Zedler, J. & Griswold, T. (1990c). Hayward Regional Shoreline, 1 2 Alameda county (Appendix II). In Wetland Creation and Restoration: Status 2 of the Science, ed. J.A. Kusler & M.E. Kentula. Washington, DC: Island 3 Press. 3 4 Josselyn, M. Zedler, J. & Griswold, T. (1990d). Bracut marsh mitigation bank. In 4 5 Wetland Creation and Restoration: Status of the Science, ed. J.A. Kusler & 5 6 M.E. Kentula. Washington, DC: Island Press. 6 Josselyn, M. Zedler, J. & Griswold, T. (1990e). Lincoln street marsh wetland 7 mitigation (Appendix II). In Wetland Creation and Restoration: Status of the 7 8 Science, ed. J.A. Kusler & M.E. Kentula. Washington, DC: Island Press. 8 9 Keller W. & Yan, N.D. (1991). Recovery of crustacean zooplankton species 9 richness in Sudbury area lakes following water quality improvements. 10 Canadian Journal of Fisheries and Aquatic Sciences 48: 1635–1644. 10 11 Kenworthy, W.J. Fonseca, M.S. Homziak, J., & Thayer, G.W. (1980). Development 11 12 of a transplanted seagreas (Zostera marina L.) meadow in Back Sound, 12 Carteret county, North Carolina. In Proceedings of the Seventh Annual 13 Conference on Wetland Restoration and Creation, ed. D.P. Cole. Hillsborough 13 14 Community College, Tampa, FA. 14 15 Kilburn, P.D. (1970). Hill prairie restoration. In Proceedings of a Symposium on 15 16 Prairies and Prairie Restoration, ed. P. Schramm. Knox College, Galesburg, IL. 16 Kirt, R.R. (1990). Quantitative trends in the progression toward a prairie state by 17 seed broadcasting and seedling transplant methods. In Proceedings of the 12th 17 18 North American Prairie Conference. 18 19 Kramer, G.L. (1980). Prairie studies at caterpillar tractor co. Peoria proving ground. 19 In Proceedings of the Seventh North American Prairie Conference, ed. C.L. 20 Kucera. University of Missouri, DC. 20 21 LaGrange, T.G. & Dinsmore, J.J. (1989). Plant and animal community responses to 21 22 restored Iowa wetlands. Prairie Naturalist 21(1): 39–48. 22 Levin, D.A. & Willard, D.E. (1990a). Lake Puckaway, Wisconsin (Appendix II). In 23 Wetland Creation and Restoration: Status of the Science, ed. J.A. Kusler & 23 24 M.E. Kentula. Washington, DC: Island Press. 24 25 Levin, D.A. & Willard, D.E. (1990b). The Harbour Project, Sandusky Bay, Ohio 25 26 (Appendix II). In Wetland Creation and Restoration: Status of the Science, ed. 26 J.A. Kusler & M.E. Kentula. Washington, DC: Island Press. 27 Lewis R.R. III. (1982). Restoration of a needlerush (Juncus roemerianus Scheele) 27 28 following interstate highway construction II. Results after 22 months. In 28 29 Proceedings of the Ninth Annual Conference on Wetlands Restoration and 29 Creation, ed. F.J. Webb Jr. Hillsborough Community College, Tampa, FA. 30 Lewis R.R. III. & Haines, K.C. (1980a). Large scale mangrove restoration on St 30 31 Croix, US Virgin Islands – II. Second Year. In Proceedings of the Seventh 31 32 Annual Conference on Wetland Restoration and Creation, ed. D.P. Cole, 32 editor. Hillsborough Community College, Tampa, FL. 33 Lewis R.R. III. & Phillips, R.C. (1980b). Experimental seagrass mitigation in the 33 34 Florida Keys. In Proceedings of the Seventh Annual Conference on Wetland 34 35 Restoration and Creation, ed. D.P. Cole. Hillsborough Community College, 35 36 Tampa, FL. 36 Loftin, K.M., Toth, L.A., & Obeysekeka, J.T.B. (1988). Proceedings: Kissimmee 37 River Restoration Symposium, Orlando, FL. 37 38 Lutz, K.A. (1988). Prairie establishment in southwestern Ohio. In Proceedings of 38 39 the Eleventh North American Prairie Conference: Prairie Pioneers: Ecology, 39 History and Culture, ed. T.B. Bragg & J. Stubbendreck. Lincoln, NE: 40 University of Nebraska Printing. 40 When does restoration succeed 391

1 McLaughlin, P.A., Treat, S.F., & Thorhaug, A. (1983). A restored seagrass 1 2 (Thalassia) bed and its animal community. Environmental Conservation 10(3): 2 247–254. 3 Makarewicz, J.C. & Bertram, P. (1991). Evidence for the restoration of the Lake 3 4 Erie ecosystem. BioScience 41(4): 216–223. 4 5 Merkel, K.W. (1988). Eelgrass transplanting in South San Diego Bay, California. In 5 6 Proceedings of the California Eelgrass Symposium: Chula Vista, California, 6 ed. K.W. Merckel & J.L. Stuckrath. National City, CA: Sweetwater River 7 Press. 7 8 Moore, D.C. & Rodger, G.K. (1991). Recovery of a sewage sludge dumping 8 9 ground. II. Macrobenthic community. Marine Ecology Progress Series 75: 9 301–308. 10 Morrison, D. (1987). Landscape restoration in response to previous disturbance. In 10 11 Landscape Heterogeneity and Disturbance, ed. M.G. Turner. NY: Springer- 11 12 Verlag. 12 Moss, B., Balls, H., Irvine, K., & Stansfield, J. (1986). Restoration of two lowland 13 lakes by isolation from nutrient-rich water sources with and without removal 13 14 of sediment. Journal of Applied Ecology 23: 391–414. 14 15 Newling, C.J., Landin, M.C. & Parris, S.D. (1983). Long-term monitoring of the 15 16 Apalachicola Bay wetland habitat development site. In Proceedings of the 16 Tenth Annual Conference on Wetland Restoration and Creation, ed. F.J. 17 Webb. Hillsborough Community College, Tampa, FL. 17 18 Nichols, O., Wykes, B.J. & Majer, J.D. (1988). The return of vertebrate fauna and 18 19 invertebrate fauna to bauxite mined areas in south-western Australia. In 19 Animals in Primary Succession: The Role of Fauna in Reclaimed Lands, ed. 20 J.D. Majer. Cambridge: Cambridge University Press. 20 21 Nitsos, R. (1988). Morro Bay eelgrass transplant. In Proceedings of the California 21 22 Eelgrass Symposium: Chula Vista, California, ed. K.W. Merckel & J.L. 22 Stuckrath. National City, CA: Sweetwater River Press. 23 Olmstead, L.I. & Cloutman, D.G. (1974). Repopulation after a fish kill in Mud 23 24 Creek, Washington county, Arkansas following pesticide pollution. 24 25 Transactions of the American Fisheries Society 1: 79–87. 25 26 Pickart A.J. (1990). Dune revegetation at Buhne Point, King Salmon, California. In 26 Environmental Restoration, ed. J.J. Berger. Washington, DC: Island Press. 27 Pritchett, D.A. (1990). Creation and monitoring of vernal pools at Santa Barbara, 27 28 California. In Environmental Restoration, ed. J.J. Berger Washington, DC: 28 29 Island Press. 29 Schulenberg R. (1970). Summary of Morton Arboretum prairie restoration work, 30 1963 to 1968. In Proceedings of a Symposium on Prairies and Prairie 30 31 Restoration, ed. P. Schramm. Knox College, Galesburg, IL. 31 32 Shisler, J.K. & Charette, D.J. (1984). Evaluation of artificial salt marshes in New 32 Jersey. New Jersey Agricultural Experiment Station publication number 33 P–40502–01–84. 33 34 Smith, I. Fonseca, M.S. Rivera, J.A. & Rittmaster, K.A. (1988). Habitat value of 34 35 natural versus recently transplanted eelgrass, Zostera marina, for the Bay 35 36 scallop, Argopecten irradians. US Fishery Bulletin 87: 189–196. 36 Sparks, R. (1992). The Illinois River–floodplain ecosystem. In Restoration of 37 Aquatic Ecosystems. National Resource Council. Washington, DC: National 37 38 Academic Press. 38 39 Sperry, T.M. (1983). Analysis of the University of Wisconsin–Madison prairie 39 restoration project. In Proceedings of the Eighth North American Prairie 40 Conference, ed. R. Brewer. Western Michigan University, Kalamazoo, MI: 40 392 J.L. Lockwood and S.L. Pimm

1 Southward, A.J. (1979). Cyclic fluctuations in population density during eleven 1 2 years recolonization of rocky shores in West Cornwall following the ‘Torrey 2 Canyon’ oil-spill in 1967. In Cyclic Phenomena in Marine Plants and 3 Animals, ed. E. Naylor & A.G. Harnoll. Oxford: Pergamon Press. 3 4 Steigman, K.L. & Ovenden, L. (1988). Transplanting tallgrass prairie with a 4 5 sodcutter. In The Prairie: Roots of our Culture; Foundation of our Economy. 5 6 Proceedings of the Tenth North American Prairie Conference. ed. A. Davis & 6 G. Stanford. Native Prairie Association of Texas, Dallas, TX. 7 Teas, H.J. (1977). Ecology and restoration of mangrove shorelines in Florida. 7 8 Environmental Conservation 4(1): 51–58. 8 9 Thorhaug, A. (1983). Habitat restoration after pipeline construction in a tropical 9 estuary: seagrasses. Marine Pollution Bulletin 14(11): 422–425. 10 Toth, L.A. (1991). Environmental responses to the Kissimmee River demonstration 10 11 project. Environmental Sciences Division, South Florida Water Management 11 12 District Technical Publication 91–02. 12 Urabe, J. (1994). Effect of zooplankton community on seston elimination in a 13 restored pond in Japan. Restoration Ecology 2(1): 61–70. 13 14 Vassar, J.W., Henke, G.A. & Blakely, C. (1978). Prairie restoration in North- 14 15 Central Missouri. In Proceedings of the Sixth North American Prairie 15 16 Conference, The Prairie Peninsula – In the Shadow of Transition, ed. R.L. 16 Stuckey & K.J. Reese. College of Biological Sciences, The Ohio State 17 University, OH. 17 18 Webb, J.W. & Newling, C.J. (1985). Comparison of natural and man-made salt 18 19 marshes in Galveston Bay Complex, Texas. Wetlands 4: 75–86. 19 Woehler, E.E. & Martin, M.A. (1980). Annual vegetation changes in a 20 reconstructed prairie. In Proceedings of the Seventh North American Prairie 20 21 Conference, ed. C.L. Kucera. University of Missouri, MO. 21 22 Wombacher, J. & Garay R. (1973). Insect diversity and associations in a restored 22 prairie. In Proceedings of the Third Midwest Prairie Conference, Kansas State 23 University, Manhattan, KS. 23 24 Zimmerman, J.H., & Schwartz, J.A. (1976). Experimental prairie restoration at 24 25 Wingra overlook. In Proceedings of the Seventh North American Prairie 25 26 Conference, ed. C.L. Kucera. University of Missouri, MO. 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 14 1 2 2 3 Epilogue: From global exploration to 3 4 community assembly 4 5 5 6 6 7 Paul Keddy 7 8 8 9 9 10 10 11 11 12 12 13 In accordance with the plan laid down, we proceed to the consideration of follies into 13 14 which men have been led by their eager desire to pierce the thick darkness of futurity. 14 15 Charles Mackay (1841) 15 16 16 17 17 18 Introduction 18 19 The editors of a symposium volume are allowed to frame the question, choose 19 20 the participants, contribute their own chapter, and offer opinions on drafts of 20 21 each chapter. In this sense, they are like hosts at a party: they already own 21 22 the house, and so should attempt to remain in the background, introduce the 22 23 guests as they arrive, serve drinks and snacks unobtrusively, occasionally 23 24 intervene if there are awkward periods of silence, and ensure that, in the end, 24 25 the guests find their own coats and car keys and leave in a good humor. In 25 26 general, then, it is probably unwise for editors to also presume to add 26 27 commentaries, conclusions, or, as in some cases, multiple chapters. They risk 27 28 becoming like the unfortunate host, who, having indulged too freely in his 28 29 own liquor and become intoxicated with his own importance, proceeds to 29 30 bore everyone with loud opinions on every topic that arises. In assuming the 30 31 role of hosts, our intention was to remain sober, restricting ourselves to an 31 32 introduction and a single contributed chapter. The purpose of the volume, 32 33 after all, was to present a variety of views, not, like the over-refreshed host, 33 34 to force readers to accept ours. Indeed, if our own views could not be ade- 34 35 quately expressed in two sections, a third or fourth would probably be equally 35 36 unsatisfactory, resembling the swaying host, who assumes that pushing his 36 37 face closer and increasing the volume of his voice will make him more com- 37 38 pelling. The summary and conclusions of this volume were therefore to be 38 39 contributed by a guest, Jared Diamond, the author whose work so obviously 39 40 inspired many of the chapters. Alas, Dr Diamond’s field work conflicted with 40

393 394 P. Keddy 1 publication deadlines, and so the task of summing up has fallen back upon 1 2 editors. I am, to shift analogies, in the unenviable position of the opening act 2 3 at a rock concert, who, finding the concert star delayed in traffic, must try to 3 4 entertain (or at least distract) the audience to avoid damage to the stadium. 4 5 Since this volume will appear near the end to the twentieth century, it seems 5 6 reasonable to begin by putting the topic of assembly rules within the context 6 7 of broader developments in ecology, beginning with voyages of exploration 7 8 and passing through the concept of habitat templates for communities. I then 8 9 propose to discuss several aspects of assembly rules that challenged all par- 9 10 ticipants, and offer a few brief observations on the chapters themselves. 10 11 11 12 12 13 The end of a great era: global exploration 13 14 We have now reached the end of a great era in biology: the era of global 14 15 exploration, map-making, large collections of new species, and classification 15 16 (e.g., Morris, 1973; Morison, 1978; Edmonds, 1997). This was an essential 16 17 first step in ecology, revealing, as it did, the diversity of life forms on earth, 17 18 and documenting the pool of species from which communities are assembled. 18 19 While there remain new discoveries to be made, particularly in poorly known 19 20 groups such as the arthropods and fungi, and under the oceans or on tropi- 20 21 cal tepui, the great period of explorers and collectors in sailing ships and 21 22 steamers has passed. Morris (1973, vol. 1, pp. 232–249) recounts how only 22 23 a little over a hundred years ago, 16 September 1864, the British Association 23 24 for the Advancement of Science met in Bath, England. Among the celebri- 24 25 ties were the two most controversial figures of African exploration, Richard 25 26 Burton and John Speke. The Times called their impending confrontation a 26 27 gladiatorial exhibition. The topic of debate? The source of the Nile. 27 28 From the comfort of our offices near the end of the twentieth century, it 28 29 may be difficult to imagine the trials and tribulations of early explorers and 29 30 collectors. In the above case, the joint expedition to Africa in 1858 ended 30 31 one phase at Lake Tanganyika with Speke nearly blind from trachoma and 31 32 Burton half-paralyzed by malaria; Speke’s solitary reconnaissance trip 25 days 32 33 later brought him to the shore of Lake Victoria. On Livingstone’s last trip, 33 34 still on the hunt for the Nile headwaters, he was ‘delayed by tribal wars, con- 34 35 stantly sick, losing his teeth one by one’ when he reached the Arab slaver’s 35 36 village of Ujiji and languished near death. On 10 November 1871, he was 36 37 discovered by Henry Stanley of the New York Herald, and greeted with the 37 38 now famous ‘Dr Livingstone, I presume?’ Such stories of the era of explor- 38 39 ation are delightfully recounted by James Morris (1973) in his three-volume 39 40 history of the British Empire, Pax Britannica. Other examples include: 40 Epilogue 395 1 Ferdinand Magellan (c. 1480–1521) who endured conspiracy, mutiny, 1 2 cheating by provisioners, and scurvy, eventually dying in the Philippine 2 3 Islands while his ship, the Victoria, went on to be the first to circum- 3 4 navigate the world. 4 5 James Cook (1728–1779) who charted much of the Pacific ocean includ- 5 6 ing New Zealand and Australia, but who was killed by natives in Hawaii 6 7 in a dispute over a stolen boat. 7 8 8 9 Alexander von Humboldt (1769–1859) who explored the northern areas 9 10 of South America, set the world altitude limit for mountain climbing while 10 11 ascending Mount Chimborazo, prepared a treatise on the political econ- 11 12 omy of Mexico, encouraged young scientists including Charles Darwin 12 13 and Louis Agassiz, and died at 90 while working on the fifth volume of 13 14 Kosmos, an overview of the structure and behaviour of the known universe. 14 15 Augustin de Candolle (1778–1841) and his son Alphonse (1806–1893) 15 16 who laid many of the foundations of plant geography and ecology. 16 17 17 Alfred Russel Wallace (1823–1913), explorer of the Amazon Basin and 18 18 Malay Archipelago, co-discoverer of the evolution by natural selection, 19 19 author of more than ten books ranging in topics from Palm Trees of the 20 20 Amazon (1853) to Man’s Place in the Universe (1903), and founding 21 21 father of zoogeography. 22 22 23 One of the most remarkable discoveries of the early explorers was the sheer 23 24 diversity of life forms on Earth. New climates and new lands yielded a 24 25 fantastic, even unbelievable, array of new species. Within each of the major 25 26 zoogeographic realms, biomes and ecosystem types that they discovered, there 26 27 were further scales of variation. Moisture gradients, elevation gradients, fires 27 28 and herds of wandering herbivores, all generated patterns at more local scales. 28 29 As the results of map-making, systematics, phytogeography and zoogeography 29 30 coalesced, attention naturally turned to more fine-scale descriptions and the 30 31 search for causation. This we may regard as the beginning of the modern era 31 32 of ecology. 32 33 33 34 34 35 The beginning of a new era: the search for causation 35 36 In the current era, ecologists have increasingly turned their attention from 36 37 tabulating the species pool, towards understanding the environmental factors 37 38 that cause these hierarchically nested ecological patterns. The outcome of this 38 39 last century of effort might be roughly summarized with three principles. 39 40 The first principle might state that any particular community or ecosystem 40 396 P. Keddy

1 Flooding 1 impeding natural drainage 2 2 3 3 4 4 5 Burning 5 6 Eutrophication Reservoir 6 siltation construction 7 Increased Off-road 7 Increased nutrients water 8 vehicles 8 Increased 9 disturbance 9 10 Natural 10 11 wetland 11 Decreased 12 nutrients 12 13 Fire Decreased 13 disturbance 14 suppression Flood control 14 Flood control leading to reduced spring 15 Water level siltation 15 16 stabilization 16 Decreased water 17 17 18 18

19 Drainage 19 20 infilling 20 21 Fig. 14.1. Any specified community, in this case a freshwater wetland, arises from an 21 22 array of opposing factors that together produce the local environmental conditions. 22 23 23 24 is produced by multiple environmental factors acting simultaneously. Any 24 25 specified community (and that includes its species and functions) could be 25 26 viewed as being the product of the pushing and pulling by opposing environ- 26 27 mental factors. Since the concluding chapter is being written in November 27 28 1998 while my co-editor Dr Weiher drives us through wetlands along the 28 29 Gulf Coast of Mississippi, I will use a wetland example (Fig. 14.1), but clearly 29 30 the same principles will apply to every other ecosystem, be it birds on islands 30 31 or fish in streams. It may be useful to consider this set of physical factors as 31 32 a kind of habitat template (Southwood, 1977) which both guides and con- 32 33 strains the evolution of species, the formation of biological communities, and 33 34 the functions that the community performs. For example, along most water- 34 35 courses, the movement of water, and particularly the movement of floodwaters, 35 36 creates gravel bars, eroding banks, sand bars, ox bows and deltas (Fig. 14.2). 36 37 In these circumstances, the three most important processes producing the habi- 37 38 tat template are (i) flooding, (ii) erosion and (iii) deposition. These are not 38 39 entirely independent, of course, but this co-variation of factors is one of the 39 40 realities of community ecology. The characteristics of species, the composition 40 Epilogue 397 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 Fig. 14.2. Dynamic processes can create a wide variety of environmental conditions 24 25 that act as a habitat template. In this case, a river creates habitats including terrace 25 26 edges, sand bars, oxbows, sloughs and levees (from Mitsch & Gosselink, 1986). 26 27 27 28 of the communities, and the processes of ecosystems are all related to this 28 29 template. 29 30 Nested within the first principle are a series of more specific relationships 30 31 between communities and environmental factors. The second principle might 31 32 therefore state that there are quantitative relationships between environmental 32 33 factors and the properties of communities. The study of each factor in the 33 34 habitat template provides an opportunity for exploration of such quantitative 34 35 relationships. Examples might include (i) the production of tropical fish as 35 36 determined by floodplain area, (ii) the diversity of plants as controlled by 36 37 substrate fertility, or (iii) the zonation of invertebrates produced by different 37 38 rates of burial by sediment. These relationships summarize the state of human 38 39 knowledge about the factors that create and control ecological communities. 39 40 The challenge for the ecologist is to unravel these factors, discover their 40 398 P. Keddy 1 consequences for communities, and determine their relative importance. The 1 2 challenge for managers and conservationists is to understand these relation- 2 3 ships, and then, if necessary, manipulate or regulate one or more of them to 3 4 maintain or produce the desired characteristics of a landscape. 4 5 These challenges are made difficult by the many kinds of communities, 5 6 and the many factors at work in them. The difficulty is compounded by a 6 7 third principle, the multiple factors that produce a community or ecosystem 7 8 will change through time. Disturbance from factors such as fire, storms, land- 8 9 slides and floods can influence the communities and species found at any site 9 10 (Sousa, 1984; Pickett & White, 1985; Botkin, 1990). The habitat template 10 11 along water courses in Fig. 14.2 is constantly changing as moving water 11 12 reshapes the environment. Similarly, if humans change the factors acting on 12 13 the community in Fig. 14.1, say, by decreasing spring flooding or increasing 13 14 fertility, the balance of forces will shift and the composition or function of 14 15 the community will begin to shift as well. It is far too easy, and therefore far 15 16 too common, for humans to study small fragments of this habitat complex 16 17 (say one species or one vegetation type in one oxbow lake), losing track of 17 18 the fact that the particular species and community types are but transitory 18 19 occurrences at any location. To understand communities, and to manage them 19 20 wisely, it is essential to appreciate their multifactorial and dynamic nature. 20 21 Even if the species pool is determined by evolution and biogeography, and 21 22 the type of species have been set by the habitat template, an important 22 23 question remains: how much of the residual variance in composition can be 23 24 accounted for by other constraints upon species composition? Is the site-to- 24 25 site variation within one habitat type merely a result of stochastic factors, is 25 26 it a consequence of subtle changes in the template, or are there other bio- 26 27 logical constraints upon the combinations of species that can occur at any 27 28 location? From this perspective, assembly rules are nested within the habitat 28 29 template and biogeographic realms as a further set of constraints upon species 29 30 composition. All of the foregoing chapters in this book could be treated 30 31 as elaborations of these principles, with varying degrees of emphasis upon 31 32 biogeography, habitat templates and biological constraints on composition. 32 33 33 34 34 35 The search for assembly rules 35 36 The intractable number of components still remains a central problem in the 36 37 causal analysis of ecological communities. To borrow from realm of general 37 38 systems theory (Weinberg, 1975), one can recognize small, medium and large 38 39 number systems. These have very different properties, and therefore require 39 40 different approaches in analysis and modelling. Small number systems have 40 Epilogue 399 1 few components and few interactions, and these systems are amenable to 1 2 precise mathematical description; the trajectory of a bullet can be predicted 2 3 with reasonable confidence. While science has in general been successful in 3 4 analyzing small number systems, in ecology this can usually be done only 4 5 by artificially isolating a set of populations from the many connections they 5 6 have with other populations, an isolation achieved only by careful judgement 6 7 (Starfield & Bleloch, 1991) or by use of containers (Fraser & Keddy, 1997). 7 8 Desert rodents (Chapter 3) and stream fish (Chapter 12) might be workable 8 9 examples of small number systems. At the other extreme are large number 9 10 systems where there are so many components that the average behavior 10 11 becomes a useful description of the system. The ideal gas laws provide one 11 12 example; the position and velocity of a particular gas molecule are not of 12 13 interest, but the properties of volume, temperature and pressure are. Large 13 14 number systems may not occur in ecology. Regional floras having several 14 15 thousand species (Chapter 5, Chapter 6) may be as large as we normally 15 16 encounter, and even here the traits of the individuals probably differ enough 16 17 that estimates of average behavior are misleading. Even congeneric species 17 18 of animals such as frogs in the genus Rana cannot be easily combined: diet, 18 19 size, physiology and behavior change with life history stage, and even adults 19 20 of species differ in trait such as body size, temperature preferences and 20 21 desiccation tolerance (Moore, 1949; Goin & Goin, 1971). A frog is neither 21 22 a molecule nor a billiard ball. 22 23 Ecology is therefore inordinately difficult, precisely because communities 23 24 and ecosystems are neither large nor small number systems (Lane, 1985). As 24 25 medium number systems, they contain too many components to be treated 25 26 analytically, and too few for statistical analysis. This may be why some 26 27 ecologists are attracted to the study of complexity itself (Chapters 8 and 9). A 27 28 rhinoceros, a frog and a grass plant cannot be averaged like ideal gas mole- 28 29 cules, nor are their behaviors and population dynamics equivalent to random 29 30 events. As the number of components increase arithmetically, the number of 30 31 interactions increase geometrically. Some method of simplification is there- 31 32 fore necessary in order to solve problems involving medium number systems 32 33 (Lane, 1985; Starfield & Bleloch, 1991). Functional groups, guilds and traits 33 34 all have this potential to simplify medium number systems to small number 34 35 systems, which may explain the popularity of such approaches (e.g., Chapters 35 36 1, 5, 9 and 12). Simplification requires carefully preserving critical inter- 36 37 actions and components while excising or ignoring others. The inherent 37 38 difficulty in doing this probably explains why, at present, medium number 38 39 systems require analytical approaches that are as much an art as a science. 39 40 40 400 P. Keddy 1 1 2 Rising to the challenge 2 3 Our introductory chapter ended the same way as our opening remarks at the 3 4 symposium, with the challenge of the maze in Fig. 9 on page 18. At this 4 5 point, it might be worthwhile, although hardly tactful, to evaluate the authors’ 5 6 responses to our challenge. How many authors suggested useful tactics to tra- 6 7 verse the maze? How many were successful in reaching the goal? Answer- 7 8 ing these two questions for each chapter, and summarizing the results in a 8 9 table, might be a useful challenge for graduate courses. Students could read 9 10 chapters, and, for each one, an assessor could have the task of evaluating both 10 11 the tactics offered and the success demonstrated in attaining the goal. The 11 12 graduate class as a whole could then discuss each assessor’s judgement. By 12 13 the end of the course, a chapter by chapter table rating of (i) value of tactics 13 14 and (ii) degree of achievement of goals, could be constructed. The only fur- 14 15 ther guideline that might assist such discussions is a reminder of the open- 15 16 ing view on page 8 that: 16 17 17 18 Contrary to common practice, merely documenting a pattern is not the study of com- 18 munity assembly . . . Asking if there is pattern in nature is akin to asking if bears shit 19 in the woods. Null models provide a valuable and more rigorous way of demonstrating 19 20 pattern, but they still do not specify assembly rules. 20 21 21 22 Booth and Larson (Chapter 7) may have provided the most controversial con- 22 23 tribution. In reading their chapter, one is reminded of Andy Warhol’s asser- 23 24 tion that in the future everyone will have their 15 minutes of fame. For this 24 25 to be possible, collective amnesia, a fascination with the ephemeral, and self- 25 26 delusion are all required. How many ecological ideas trumpeted as new 26 27 discoveries, they ask us, are simply reworded versions of ideas from the turn 27 28 of the century? The lack of computer access to early texts, and the degree to 28 29 which they are physically deteriorating in storage, suggests that it will be 29 30 increasingly easy to pretend that re-statements of old ideas are novel contri- 30 31 butions. Larson and Booth have challenged us all to be aware of the history 31 32 of our discipline and to give credit where credit is due. 32 33 The final chapter in the book, Lockwood and Pimm’s assessment of 87 33 34 published reports on restoration projects provides a realistic assessment of 34 35 the practical limitations to re-assembling natural communities. Seventeen 35 36 projects were rated unsuccessful, and 53 were rated partially successful, leav- 36 37 ing only 17 (20%) as successful in attaining stated goals. These included 13 37 38 functional goals and 19 structural goals. It therefore appears that much of the 38 39 rhetoric around assembly rules cannot yet be translated into practical success 39 40 on the ground. 40 Epilogue 401 1 And so, corrected proofs in hand, we now excuse ourselves from our duties 1 2 as hosts, and return to our rainy trip through the Longleaf Pine forests and 2 3 Pitcher Plant savannas of the Gulf coast, a region with several dozen species 3 4 of carnivorous plants and orchids, and a seemingly endless array of Aster- 4 5 aceae, Cyperaceae and Poaceae. The existence of this Longleaf Pine eco- 5 6 system (Peet & Allard, 1993) can only be explained by invoking processes 6 7 over very different time scales. At the geological scale, over tens of millions 7 8 of years, there were the factors of erosion, deposition, sea level change and 8 9 leaching that produced these rolling hills and sand plains. At the historical 9 10 scale of centuries, there were the recurring effects of fire, and the increas- 10 11 ingly strong effects of human settlement. Current factors appear to include 11 12 fire, seepage, infertility, summer drought, competition, and disturbance by 12 13 burrowing crawfish. The coexistence of up to 40 species in a 50 by 50 cm 13 14 quadrat requires consideration of all these different time scales. By invoking 14 15 these processes, one can offer an entertaining explanation for why some areas 15 16 have so many plants. Superimposed upon such factors, however, remains the 16 17 tantalizing possibility that only some mixtures are biologically possible, or 17 18 ecologically favored. The finer the scale becomes, the more difficult expla- 18 19 nation seems to become. Sitting here at the Western Inn, in Stone County, 19 20 southern Mississippi at 9.30 on a Saturday night, filled with catfish and 20 21 chicken-fried steak (but no beer), the original questions posed in the intro- 21 22 duction remain to tantalize the naturalist and the reader: just how do these 22 23 multivariate factors select the flora of a pitcher plant savannah from the avail- 23 24 able pool, how deterministic is the habitat template, and why are there five 24 25 species of pitcher plant in a savanna as opposed to any other number? 25 26 26 27 27 28 Acknowledgments 28 29 I thank my co-editor Evan Weiher for his many hours of hard work and end- 29 30 less attention to detail. We are both grateful to the contributors for their effort, 30 31 enthusiasm, patience and general good humor, as well as to Maria Murphy 31 32 and Alan Crowden at Cambridge University Press for piloting us though the 32 33 publishing process. Dennis Whigham was helpful in the early stages of organ- 33 34 ising the symposium through the Ecological Society of America. NSERC pro- 34 35 vided the continuity of funding that made the project possible. 35 36 36 37 37 38 References 38 39 Botkin, D.B. (1990). Discordant Harmonies. A New Ecology for the Twenty-first 39 40 Century. New York: Oxford University Press. 40 402 P. Keddy

1 Edmonds, J. ed. (1997). Oxford Atlas of Exploration. New York: Oxford University 1 2 Press. 2 Fraser, L.H. & Keddy, P. (1997). The role of experimental microcosms in 3 ecological research. Trends in Ecology and Evolution 12: 478–481. 3 4 Goin, C.J. & Goin, O.B. (1971). Introduction to Herpetology. 2nd edn. San 4 5 Francisco: W. H.Freeman. 5 6 Lane, P.A. (1985). A food web approach to mutualism in lake communities. In The 6 Biology of Mutualism, Ecology and Evolution, ed. D.H. Boucher, pp. 344–374. 7 New York: Oxford University Press. 7 8 Mackay, C. (1841). Memoirs of Extraordinary Popular Delusions. Reprinted in 8 9 1980 as Extraordinary Popular Delusions and the Madness of Crowds, with a 9 foreword by A. Tobias. NY: Harmony Books. 10 Mitsch, W.J. & Gosselink, J.G. (1986). Wetlands. NY: Van Nostrand Reinhold. 10 11 Moore, J.A. (1949). Patterns of evolution in the genus Rana. In Genetics, 11 12 Paleontology and Evolution, ed. G.L. Jepsen, G.G. Simpson, & E. Mayr, pp. 12 315–338. Princeton: Princeton University Press. Reprinted in 1963 by NY: 13 Atheneum. 13 14 Morison, S.E. (1978). The Great Explorers. The European Discovery of America. 14 15 NY: Oxford University Press. 15 16 Morris, J. (1973). Pax Britannica. 3 Vols. Faber & Faber Ltd. Reprinted 1992, 16 London: Folio Society. 17 Peet, R.K. & Allard, D.J. (1993). Longleaf pine vegetation of the southern Atlantic 17 18 and eastern Gulf Coast regions: a preliminary analysis. In Proceedings of the 18 19 Tall Timbers Fire Ecology Conference No. 18, The Longleaf Pine Ecosystem: 19 Ecology, Restoration and Management, ed. S.M. Hermann, pp. 45–104. 20 Tallahassee, FL: Tall Timbers Research Station. 20 21 Pickett, S.T.A & White, P.S. (1985). The Ecology of Natural Disturbance and 21 22 Patch Dynamics. Orlando, FL: Academic Press. 22 Sousa, W.P. (1984). The role of disturbance in natural communities. Annual Review 23 of Ecology and Systematics 15: 353–391. 23 24 Southwood, T.R.E. 1977. Habitat, the template for ecological strategies? Journal of 24 25 Animal Ecology 46: 337–365. 25 26 Starfield, A.M. & Bleloch, A.L. (1991). Building Models for Conservation and 26 Wildlife Management, 2nd edn. Edina, MN: Burgers International Group. 27 Weinberg, G.M. (1975). An Introduction of General Systems Thinking. New York: 27 28 John Wiley. 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 1 1 2 2 3 Index 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 Page numbers in italics refer to figures and local minima 264 13 14 tables local of species 76 14 measurable properties 8, 9 15 Acacia aneura 178 morphological space occupation 257 15 16 adaptive landscape 262–3 non-random in local community 84–5, 86 16 topography 263, 266, 268 observed 25, 26 17 aerial dispersion of micro-snails 283 optimization approach 262 17 18 aggression patterns 8 18 desert rodents 94 predictions of change 261 19 interspecific in sparrows 198 random 132, 265 19 20 Allen’s rule 209 sandy shoreline 265, 266 20 allocation rules 254 simulated 265 21 allopatric speciation unfavored 31, 32, 33, 34 21 22 desert rodents 92–3 assembly 22 23 interspecific competition 93 attractor constraints 243, 244 23 allowed states 25 categorical level 236, 237 24 anemochorous species, island populations communities from pools 4 24 25 168 constraints 207, 243, 244 25 Anolis lizards, Caribbean 76 deterministic 243, 244 26 anti-intellectualism 13 functions 243–4 26 27 archipelagos 293–5 general theory 245–6 27 inter-archipelago links 305 insular communities 283–5 28 patterns 296, 298, 299, 304 levels 236–7 28 29 differences 306 mechanics 235–7, 247 29 species level links 305 mechanism 246 30 area–isolation plots 282 non-random 76 30 31 Argentina, central-western 344–56 operator 242–3 31 32 arid region productivity 79 patterns 32 assemblage consistent 253–6 33 adaptive landscape 262–3 species membership discontinuity 241 33 34 animal 253–4 probabilistic elements 242–5 34 deviation 262, 265 probability density of state space 244 35 scores 266–7 process 233, 236–7 35 36 energy state 262 regulation 234 36 environmental factors as filters 254 random walk 244 37 evolutionary of desert rodents 99 state space predictions 244–5 37 38 extant on Oahu island 117 stream fish community 331–2 38 extirpation events 264 topological level 236, 237 39 favored 31, 32, 33, 34 see also community assembly 39 40 imaginary 25, 26 assembly rule 3 40

403 404 Index

1 assembly rule (continued) directionality 246 1 2 abundance-based 153–5, 156, 157 generic communities 234 2 architectural 237 propositions 245–6 3 categories 165 self-organization 238–42 3 4 co-occurring species 145–6 Asteraceae, Barkley Sound islands 173 4 community atmospheric composition change 340–1 5 composition within sites 189–202 see also carbon dioxide, atmospheric; 5 6 constraints on composition 251–3 climate change 6 patterns 130 attractors 238–9 7 competitive 58 assembly rule solutions 239 7 8 concepts 211–13, 339 chaotic 244 8 definitions 7–9, 77, 131–2, 207, 208, 209, phase transitions 239 9 236 authority 12 9 10 descriptions 207, 208, 209 avifauna 10 11 development 52–3 Hawaiian extinction patterns 242 11 Diamond’s 23 see also birds 12 empirical patterns 76 avifauna, introduced 12 13 environmental noise 148 rejection criteria 109 13 environmental patchiness 134–5, 136 selection criteria for community 14 environmental variation 135 membership 108 14 15 finding 8 avifauna of New Guinea islands 6, 23, 300 15 functional group 53, 67 insular distribution 302, 303–4 16 generality 99 16 17 goals 9–10, 16 babblers 180 17 history of concept 211–13 Barkely Sound (Vancouver Island; British 18 insular fauna 284 Colombia) 168 18 19 interpretation of term 252 forest species 169, 170, 171, 172, 177 19 20 island systems 168–9 island classes 171, 172 20 limiting similarity 143–6 Barton–David statistical test 96 21 local 72 bass, small mouth 299 21 22 local coexistence of species 75 Bass Straits (Australia) 277, 281 22 mechanistic processes 76 isolation effects 278, 281 23 null hypothesis 60 Bénard cell spontaneous organization 260 23 24 obstacles 9–10, 16–17 Bergmann’s rule 209 24 patterns of community structure 77–88, 89 Bermuda 108, 109 25 plant characters 143–53 great kiskadee release 113 25 26 presence/absence 137–9, 140–1, 142–3 introductions 113 26 primary evidence 206 success 111 27 spatial autocorrelation 135, 136–7 morphological overdispersion 117, 118, 27 28 spatial heterogeneity 138 119 28 29 species interactions 132 passeriforms 113, 117 29 standard statistical tests 133 priority effect 114, 115, 116 30 strategies for finding 253–9 species introduced 126-9 30 31 subtle effects 137 Betula papyrifera 219, 220, 221–2 31 systems used in search 213–15 biodiversity, geographical basis 278, 281 32 terminology 210–11 biogeographic functions 278 32 33 trade-offs 263 biogeographic refugia 303 33 trait-based 266, 267, 268 biogeographic space 282, 297 34 trait–environment patterns 268 guild member distribution plotting 300 34 35 type A 165, 166, 168 biogeographic studies, IDF 29 35 type B 165, 166 biological conditions for stream fish 36 types 131 communities 321, 322, 323 36 37 validity 94 simulation 319–20 37 38 what to test for 134 biomass 38 see also Fox and Brown, assembly rule; constancy 153 39 Fox’s assembly rule evenness relationship 155, 156 39 40 assembly trajectory 233, 236–7 importance in biosphere 4 40 Index 405

1 biota, local on cliffs 225 cheek pouch volume, desert rodents 80, 94 1 2 biotic filtering 130 chub, creek 299 2 biotic processes cliff face vegetation 224–5 3 outcome variation 333 climate change 3 4 physical variation 312 ecosystem function prediction 355–6 4 birds filters 340–1, 345 5 density on protead heathland 179 spatial gradients 341 5 6 insectivorous community composition climatic factors 6 189–91 filters 349, 350, 351 7 island 76 trait–environment linkages in plant 7 8 mulga bushland 178–9 communities 349, 350, 351, 352 8 nectarivorous 179 clover, Humboldt Bay Owl’s 372 9 protead heathland 178–9 co-occurrence 9 10 scrub/woodland/forest 173, 177–8 assembly rules 145–6 10 11 species distribution in New Guinea 23 desert rodents 81, 82 11 see also avifauna; oak woodland birds multiple 97 12 Blarina brevicauda 300, 301 patterns 66–8, 69 12 13 bluebird, eastern 113 species characteristics 94 13 body size coal mining restoration 376 14 desert rodents 94, 99 coexistence 23, 48, 77, 251 14 15 frequency distribution 287–8 allopatric speciation 92 15 immigration ability 290–1 bird species 190, 191, 193, 196, 197, 203 16 intraspecific geographic variation 93 constraints of assembly rules 251 16 17 local community role 94 facilitated 91 17 non-random pattern of distribution 96 functional groups of desert rodents 82–7 18 statistical power of tests 96 restrictions 339 18 19 Bonasa bonasia 293–4, 296 coldwater streams, North America 313–14 19 20 Boschniakia 169 collaboration 17 20 Bose–Einstein statistics 64, 65 colonization 21 bottom-up processes 87, 237 long-distance 283 21 22 butcherbird 180 random independent 71 22 butterfly, bog copper 279 sequential of empty sites 72 23 Colwell and Winkler’s null models 27–9 23 24 canopy interactions 159 communities/community 24 canopy-forming trees 213 alternative composition 166 25 conifer forest 216 amplification 166 25 26 capacity rules 254 building from pools 2, 3, 4 26 carbon dioxide, atmospheric 341 co-adapted/co-evolved 142 27 ecosystem function prediction 352, 355–6 competition 158 27 28 carrs, textures 147, 148 conceptual model of assembly 252 28 29 Catastomus tahoensis 314, 320 constraints on composition 213 29 catbird, gray 113 depauperate 284–5 30 catostomid fish development 209 30 31 demographic characteristics 314 constraints 206, 207 31 invasion 323, 325–7 disturbance of state 239–40 32 causal mechanisms, desert rodents 91–4 ecological compatibility of members 165 32 33 cedar, eastern white 216, 220, 221 ecological substitutes 166 33 seedling survival 220, 223, 224 energy dissipating systems 259–61 34 Chaetodipus intermedius 78 entity aggregations 234 34 35 chaparral 177, 178 generic 234 35 character displacement guilds intrinsic to 151 36 diffuse 93 highly stressed 215 36 37 evolutionary 93–4 invasion susceptibility variation 332 37 38 checkerboard longevity 215 38 distribution 295 maintenance 209 39 patterns 300–1, 302, 303 membership selection criteria 108 39 40 spatial distribution 302–3 non-competitive 215 40 406 Index

1 communities/community (continued) differences between species 79–81 1 2 organization 75 role 94 2 patterns 130 cells 65 3 persistence 366 composition 72 3 4 prey/predator species proportions in favored state 59, 61, 62, 64, 65, 68, 71 4 community 51 sites 105 5 RAD differences 157 southwestern US data 83–4, 85, 86 5 6 random 63 functional groups 105 6 relative abundance distribution of species interactions 98 7 154 strength 84, 85 7 8 replication across habitat range 166 interspecific competition 91–3 8 replication in wide-ranging habitats 173, structure 96 9 177–81, 182, 183, 184, 185–7, 188–9 measures of structure 97 9 10 self-organized 143 morphological differences between species 10 11 sequences of states 234 79–81 11 simulated 62 Narcissus effect 83 12 size 165 nested subset structure 88, 89, 90 12 13 species turnover 166 non-random assemblages 84–5, 86 13 structure constraints on Niagara random assignment 64–5 14 Escarpment 216, 217, 218, 219–20, simulated 63, 105–6 14 15 221–3, 224–5 species 15 community assembly 311 coexistence 79 16 diversity 266 entry 63 16 17 extinction 233 with negative/positive associations 88 17 invasion 233 state, sites 105 18 island paradigm 5–6 structure 86 18 19 models in variable environments 332–3 in desert rodents 100 19 20 null hypothesis 105–7 existence 94 20 optimization model 263–7 unfavored state 59 21 paradigms 5–7 community structure 21 22 partial deviations 264 comparative geographic studies 77 22 predictions 265–6 desert rodents 76 23 recruitment 333 diversity indices 78 23 24 restoration 365–7 ecological similarity patterns 79–87 24 tolerance values 263–4 favored state 71 25 trait–environment paradigm 5, 7 island 272–3 25 26 trait–environment patterns 263–4 mineral resource increase 259–60 26 see also assembly morphological similarity patterns 79–87 27 community composition 165, 166 null models 253–4 27 28 competition 58, 59 patterns 77–88, 89 28 29 constraints 251–3 desert rodents 90, 91 29 determinants 166 identities of species 87–8, 89, 90 30 Niagara Escarpment 216 phylogenetic similarity patterns 79–87 30 31 null models 255–6 species richness 77–9 31 predictability 208 competition 32 protead heathland 178–9 above-ground intensity in wetlands 255 32 33 trait dispersion 256 assembly rule 58 33 within sites asymmetrical 297, 300 34 birds in grass-sagebrush 191, 193 coefficients 48 34 35 insectivorous bird communities 189–91 community composition 58, 59 35 community ecology 1–2, 24 consumer resource 46, 47 36 common goals 2 dynamical character 234 36 37 difference from evolutionary ecology 3 extent and morphological overdispersion 37 38 framework 3 120 38 community, local favored states 69 39 assembly 97 IDF 299–301, 302, 303–4 39 40 body size intraspecific and rejection patterns 109 40 Index 407

1 natural communities 158 cyprinid fish 1 2 non-random consequences 207 demographic characteristics 314 2 site sharing 66–7 invasion 323, 325–7 3 stream fish 3 4 community 321, 322, 324 dace, Lahontan speckled 314, 318, 320 4 in first year 316 Dalechampia 146 5 trait overdispersion 146 deermouse 78, 300, 301 5 6 competition, interspecific 24, 27, 28, 29, 45, habitat generalist 97 6 58, 71 desert basins, isolated 79 7 allopatric speciation 93 Diamond’s assembly rules 23–4 7 8 community structure 96 Dipodomys 78 8 desert rodents 91–3 Dipodomys deserti 85 9 ecological/evolutionary consequences 75 Dipodomys merriami 63, 66, 93, 97, 98 9 10 habitat component assessment 49, 50, 51 Dipodomys ordii 98 10 11 introductions 122–3 discriminant function analysis (DFA) 193, 11 overdispersion 116 194, 199, 200 12 morphological 120 dispersion, morphological for wetland plants 12 13 process in desert rodents 98–9 254–5, 256 13 competitive exclusion 153 dissipative structures 238 14 competitive interactions 213 distribution 14 15 complex systems approach 238 checkerboard patterns 295, 300–1, 302, 15 complexity 11 303 16 congeners, floral trait overdispersion 254 exclusive of species 295, 297 16 17 conifer forest fundamental 297 17 canopy-forming tree 216 patterns 277, 279, 280 18 community longevity 215 insular of non-volant mammals 281 18 19 hierarchy of factors controlling invasion 341 random 95 19 20 see also Niagara Escarpment realized 295, 297, 298, 299–301, 302, 20 Connor and Simberloff’s null models 27 303–4 21 constraints 206, 207 see also insular distribution function (IDF) 21 22 field situations 209 disturbance, community state/cyclical 239–40 22 laboratory situations 207–8 divergence 77 23 trait–environment patterns 260 diversity 23 24 consumer-resource competition 46, 47 community assembly 266 24 consumer-resource model indices 78 25 MacArthur–Tilman 46–8, 49 mulga 180 25 26 multiple regression technique 49, 50 protead heathland 178–9 26 Nevada test site data 48–9 types 166 27 consumption vectors 46–7 dominance, aggressive in desert rodents 94 27 28 convergence 77 Drosophila, interspecific interactions 29 28 29 ecological 149 drought 29 sorting 146, 147 fish survival 316, 319 30 evolutionary 146–7 stream fish community 321, 322, 323, 324 30 31 texture 147–9 Dumetella carolinensis 113 31 core habitat, sparrow occupancy 195, 196, 197 dynamical behavior 246 32 core niches, mulga bushland 180 32 33 core-satellite community structure see nested ecesis 211, 212, 213, 214 33 subset community structure ecological communities 34 Corvus 180 compatibility of members 165 34 35 cottid fish composition prediction 2 35 demographic characteristics 314 ecological complexity 333 36 invasion 323, 325–7 ecological dominance 304 36 37 Cottus beldingi 314, 320 ecological replacements, mulga bushland 180 37 38 Cottus gulosus 318 ecological segregation, desert rodents 99 38 covariation of species, IDF 285 ecological sorting 39 Cracticus 180 convergence 146, 147 39 40 cuckoo-dove 302, 303 trait overdispersion 146 40 408 Index

1 ecological variables, non-linear relationships Fox’s assembly rule 95 1 2 259 local community 59, 61, 62, 64, 68, 71 2 ecosystem function sites 105 3 plant traits 341, 342, 343 southwestern US data 83–4, 85, 86 3 4 restoration 364 null distribution 60 4 trait–environment linkages in plant random assignment 68–9 5 communities 353–5 see also unfavored state 5 6 ecosystem function prediction, with fecundity variance 235 6 trait–environment linkages 355–6 feeding guilds 7 edge, morphological 257 Australian studies 29, 30, 31, 32, 35, 36, 7 8 edge species in island systems 172–3, 176, 37 8 177 Madagascar 39 9 emergent structure 238 North American studies 32, 35, 36, 37–8 9 10 endurance of larger species 283 South American studies 38–9 10 11 energy ferns 169 11 critical point 260 filters 12 dissipating systems 259–61 assembly rules 339, 340 12 13 gradient 260 biotic interactions 341 13 levels 260 climate change 340–1 14 state of assemblage 262 climatic conditions 345 14 15 entropy 238, 259 climatic factors 349, 350, 351 15 environment trait–environment linkages in plant 16 controlled 207 communities 339–40 16 17 variable 332–3 finches 17 environmental factors as filters for emberizid 191, 193 18 assemblages 254 Saint Helena 112, 117 18 19 environmental filtering 130 see also sparrows, emberizid 19 20 environmental patchiness 134–5, 136, 137 fir, Douglas 169 20 environmental productivity 226 fire-climax communities 239–40 21 environmental spatial heterogeneity 138 fish 21 22 environmental variation, assembly rules 135 abundance fluctuations 312–13 22 environmentally mediated patterns 132 assembly modeling 313 23 equilibrium theory of MacArthur and Wilson community species from Australian 23 24 272 springs 226 24 erosion control 371 fish, stream communities 311–13 25 Estrilda astrild 110 adult survival 316 25 26 evenness assembly 331–2 26 RAD 154–5, 156, 157 processes 318 27 succession 157 competition 318, 321, 322, 324 27 28 evolution demographic characteristics 314 28 29 fitness peaks 262 first year survival 315–16 29 within-guild 35, 37 invader relative success 318–19 30 evolutionary convergence 146–7 invasion 331–2 30 31 extinction life history–environment interactions 31 community assembly 233 332–3 32 island community structure 272, 273 modeling 313–17 32 33 oceanic islands 111 order of entry 331 33 patterns in Hawaiian avifauna 242 relative abundance 313 34 rates on islands 274 salmonid predator 315 34 35 extinction–area function 275 stream discharge 315–16, 320 35 extinction–area relationship 278 stream flow 331 36 extirpation events 264 fish, stream community model 313–17 36 37 adult survival response to droughts 316, 37 38 favored state 319 38 community structure 71 assembly scenarios 323–6, 327 39 competition 69 biological conditions 321, 322, 323 39 40 distribution 68, 72 competition in first year 316 40 Index 409

1 competition function population limit 320 fruit-pigeon 302, 303 1 2 implications for assembly rules 333–4 functional groups 59 2 invasion assembly rule 67 3 range of conditions 318–19 coexistence in desert rodents 82–7 3 4 sequences 324, 325–6, 327 computer modeling 107 4 simulation 319–20 desert rodents 99–100, 209 5 Martis Creek observations/predictions ecological attribute 106 5 6 327–8, 329 favored states 59, 68–9 6 outcome prediction 331 interrelationships 257 7 parameter estimation 317 Nevada test site data 64 7 8 physical conditions 321, 322, 323 null hypothesis of community assembly 8 predation function population limit 320 105 9 predation intensity 316–17 randomized 69 9 10 prey extinction 326, 327 randomizing assignments 68–9 10 11 Sagehen Creek observations/predictions fynbos 177 11 328, 330 12 simulation experiments 317–20 Gambusia 312 12 13 stream discharge/first year survival Gloger’s rule 209 13 relationship 315–16 Grand Teton National Park (Wyoming) 14 structure 315 birds 14 15 survival rates 317 in grass-sagebrush 191, 193 15 validation 320 in willows site 192 16 fitness landscape 261–2 sparrows in willows site 193, 194, 195, 16 17 adaptive 262 196, 197–8 17 fitness peaks 261, 262 core habitat 195, 196, 197 18 species evolution 262 DFA 193, 194, 196 18 19 floods and stream fish community 312, 321, factor analysis 193, 194, 196 19 20 322, 323, 324 vegetation height in willows site 193 20 floral traits, overdispersion 254 warblers in willows site 198–9, 200–1, 202 21 flow conditions 312, 331 core habitat 199 21 22 flowering time spacing 144 DFA 199, 200 22 flycatcher factor analysis 199, 200 23 gleaning 303 foraging height 199, 201, 202 23 24 oak woodland 184, 186–7, 188, 189 habitat preference 199, 200 24 focal species model 273–5, 276, 277–8, 279, grassland, montane 352, 354, 355, 356 25 280, 281, 282 Great Basin (North America) 277, 281 25 26 assumptions 274 isolation effects 278, 281 26 empirical distribution patterns 279 Great Lakes (North America), island fauna 27 IDF 274, 275, 277–8, 279, 282 300 27 28 immigration ability 275, 276 grouse, hazel 293–4, 296 28 29 isolation 280, 281, 282 guild 29 food niche 25 classification 150–1 30 food resource, partitioning 37 evolution within 35, 37 30 31 food webs hypothesis 132 31 organization 75 interrelationships 257 32 predator addition 240 intrinsic 151–3, 155, 169, 171 32 33 foraging behaviors 35 plant 343, 344 33 forbidden states 25 structure 75 34 forest gap dynamics 239 random matrix 41, 42–3, 44 34 35 forest species of island systems 169, 170, see also feeding guild 35 171, 172, 177 guild assembly rule 29, 30, 31, 53 36 Fox and Brown applicability 37–40 36 37 assembly rule 68 deterministic components 51 37 38 guild 41, 42-3, 44 development 46–9, 50, 51–2 38 random data set generation 41, 42–3, 44 favored assemblages 31, 32 39 test on Nevada test site data 64–6 Kelt’s development 49, 51 39 40 Fox’s assembly rule 59, 71, 82, 95 methodology 40–1, 42–3, 44–5 40 410 Index

1 guild assembly rule (continued) distance 275, 276 1 2 null hypothesis 40, 53 filters 298 2 operational mechanism 36 island biogeography 289–90 3 probabilistic components 51 island community structure 272, 273 3 4 probability states 53 rates on islands 274 4 process 51–2 resource requirements 284 5 regression technique incorporation 52, 53 sigmoidal nature of function 291 5 6 resource immigration–isolation relationship 275, 278 6 availability 32, 33 immigrator 7 competition model 48 ability 297, 298 7 8 Schoener–Pimm regression technique 53 passive 275 8 site probability states 51 incidence function 23 9 statement 31–2 incisor width in desert rodents 80, 93, 94 9 10 stochastic component 51 information flow 259 10 11 unfavored assemblages 31, 32 insular communities 11 Wilson 41, 44 assembly 283–5 12 guild proportionality 32, 149–51, 254, 256 faunal assemby rules 284 12 13 abundance-based 154 nestedness 283, 283, 284–5, 286 13 analyses 143 see also focal species model 14 species–species interactions 159 insular distribution 293 14 15 gymnosperm/angiosperm forest 149 realized 295, 297, 298, 299–301, 302, 15 303–4 16 habitat insular distribution function (IDF) 273, 274, 16 17 abiotic variation 166 275, 282 17 broadly distributed type 166 archipelago 293 18 component of interspecific competition 49, asymmetrical competition 297, 298 18 19 50, 51 biogeographical studies 29 19 20 exclusivity 68 competition effects 299–301, 302, 303–4 20 heterogeneity covariation of species 285 21 desert rodents 78 focal species model 282 21 22 patch model 153 general form 277 22 islands 178 intercept/slope 287 23 niche 25 predation effects 299–301, 302, 303–4 23 24 patch use 167 resource requirement measure 282 24 suitability 166 shifts 293 25 vegetation structure 166 sigmoidal form 277, 278, 279 25 26 wide-ranging in community replication insular resources 288–9 26 173, 177–81, 182, 183, 184, 185–7, inter-archipelago patterns 304 27 188–9 inter-archipelago scales 293–5 27 28 heath 177 inter-taxa scales 293–5 28 29 see also protead heathland interactions 29 hedgerows, European 213 assembly rules 339 30 historical context 14–16 coefficients for species in local community 30 31 historical events, proneness to 207–8 50 31 historical revisionism 14 direct/indirect 75 32 Holdridge life-zone 364 plant 340 32 33 Humpty Dumpty effect 159, 237, 367 interspecific interactions 27, 28, 297 33 Huron, Lake (North America) 300, 301 Drosophila 29 34 hypergeometric distribution, multivariate 106 realized insular distribution 297, 298 34 35 see also competition, interspecific 35 Icarus effect 28 introductions 36 immigration Bermuda 126–9 36 37 ability 275, 276, 283, 284–5, 289–90, 297, failed 114 37 38 298 human actions 122 38 competitive dominance 304 interspecific competition 122–3 39 shrews 301 modes 113 39 40 Darlington’s pattern 286 novel situation 121 40 Index 411

1 Oahu island 126–9 static theory 272 1 2 oceanic islands 110–11, 112 theory 52 2 St Helena 126–9 vagility 297 3 species-specific attributes 122 island systems 167–9, 170, 171, 174–7 3 4 successful 114 assembly rules 168–9 4 Tahiti 126–9 dispersal driven 168 5 invasion 26 edge species 172–3, 176, 177 5 6 biotic resistance 332 extinction driven 168 6 community forest species 169, 170, 171, 172, 177 7 assembly 233 mid-sized 168–9, 170, 171–3, 174–7 7 8 susceptibility variation 332 nestedness 168, 168–9 8 disrupting 240 plant distributions 168–9, 170, 171–3, 9 environmental resistance 331–2 174-7 9 10 long time period 240 shoreline species 171–2, 174–5, 177 10 11 order 331 species distribution 167–8 11 resistant state 240 species richness variance 173 12 restoration 366, 367 islands, oceanic 12 13 sequence 208 assembly history of avian community 109 13 European hedgerows 213 extinctions 111 14 stream fish community 321, 323–6, 324, geographical isolation 109 14 15 327, 331–2 human manipulation of flora/fauna 109–10 15 island introductions 110–11, 112 16 habitat 178 avifauna 108–9 16 17 paradigm 5–6 modes 113 17 populations morphological overdispersion 116–17, 118, 18 anemochorous species 168 119–20 18 19 resource requirements 288 native range size 120–2 19 20 small island effect 287, 292 passeriform species pool 120 20 species physiographic differences 110 21 co-occurrence 203 priority effect 114–16 21 22 richness 52–3 size 111, 112 22 sets 166 species pool formation 110–13 23 island biogeography 272–4 isolation 278, 280, 281, 282 23 24 assembly of insular communities 283–5 minimal critical 291 24 biogeographic levels 304 see also species–isolation relationship 25 linkages 305 25 26 body size frequency distribution 287–8 Jack Horner effect 133–4 26 community structure 282–5, 286, 287–95, guild proportionality 149–50 27 296 Juniperus virginiana 216 27 28 equilibrium theory 273 28 29 focal species model 273–5, 276, 277–8, Kelt’s development of guild assembly rule 29 279, 280, 281, 282 49, 51 30 geographical basis of biodiversity 278, 281 kiskadee, great 113 30 31 inter-archipelago/inter-taxa scales 293–5 31 nestedness of insular communities 283–5 lawns, species richness 138, 140, 141 32 patterns 211 Leslie matrix 313, 315 32 33 resource requirements 297 Leucadendron leaf size 144 33 richness 292 life forms, importance in biosphere 4 34 small island effect 287, 292 life history–environment interactions in 34 35 species stream fish community 332–3 35 accumulation 288 lily 169 36 distribution 277, 279, 280 limiting similarity model 254 36 37 richness patterns 285, 287–93 linear modeling 261 37 38 species–area relationship 285, 287–9, 293 locomotion mode of desert rodents 94, 99 38 species-based hierarchical model 304–5 Loess curves 261 39 species–isolation relationships 285, 287, low-productivity environments 225–6 39 40 289–91 40 412 Index

1 MacArthur–Tilman consumer-resource model Narcissus effect 28, 44–5, 71 1 2 46–8, 49 desert rodents 86–7 2 macchia 177 local community structure 83 3 macroecology 306 null models 133, 134 3 4 macrohabitat selection 78 native range, size on oceanic islands 120–2 4 Macropygia nigrirostris 303 natural history 72 5 Maianthemum dilatatum 169 nearest-neighbor distances 257, 282 5 6 maquis 177 neighborhoods, altered 341 6 marsh habitat destruction, Florida 373 nested subset community structure 88, 89, 90 7 marsupials 37 nestedness 7 8 Martis Creek (Truckee river; California) 312, immigrator resource requirements 285 8 320, 325–7 insular communities 283, 284–5, 285, 286 9 field observations/model predictions island systems 168, 168–9 9 10 327–8, 329 patterns 91 10 11 mass effect 274 Nevada test site data 32, 36, 38, 41, 48–9 11 matrix randomization method 62–4, 71 Fox and Brown test 64–6 12 Maxwell–Boltzmann statistics 65 mammal 60, 61 12 13 M’Closkey B 25 non-random assemblages 85–6 13 Melanodryas 180 plant communities 67 14 Michigan, Lake (North America) 300, 301 Wright and Biehl approach 70 14 15 micro-snail aerial dispersion 283 Niagara Escarpment 217 15 Microdipodops 78 cliff face 16 Microeca 180 community 216, 218 16 17 microhabitat selection 78 species pool 224–5 17 Micropterus dolomieu 299 community 18 Microtus pennsylvanicus 301 composition 216, 218 18 19 migration, active 283 structure constraints 216, 217, 218, 19 20 migrules 211, 212 219–20, 221–3, 224–5 20 minimal spanning tree (MST) 117 seed 21 model systems for assembly rules 76 bank 219, 220, 222 21 22 module packing effect 138, 143 rain 219, 220, 221 22 Monte-Carlo simulation 27 seedling 23 algorithm of Fox and Brown 64, 65, 66 establishment 219, 220 23 24 favored states 64, 65, 66 survivorship 220, 223, 224 24 Fox’s assembly rule 82 soil samples 219 25 guild assembly rule testing 37, 39 niche 25 26 J.P. Morgan effect 28, 111, 112 convergence 146 26 morphological features of oceanic island limitation 137, 138 27 birds 116 theory 24 27 28 morphological overdispersion criteria 120 niche separation 28 29 morphological ratios 95 mean 25–6, 27 29 Morris D, MacArthur–Tilman consumer- null models 28 30 resource model derivation 46–8, 51 species assembly 30, 31 30 31 mosquito fish 312 non-linearity, ecological consequences 31 mouse 238–41 32 kangaroo 78 Nothofagus forest, convergence 148, 149 32 33 rock pocket 78 null distribution, species ranges 67 33 see also deermouse null hypothesis 34 mulga bushland assembly rule 60 34 35 assembly criteria 180–1 coefficent theta value 107 35 birds 178 community assembly 105–7 36 density 180 computer modeling 106–7 36 37 insectivorous 180–1 Fox’s assembly rule 82 37 38 species diversity 179, 180 guild assembly rule 40, 53 38 core niches 180 testing 37, 38, 39 39 ecological replacements 180 morphological ratios 95 39 40 species turnover 179 random data set generation 41, 42–3, 44 40 Index 413

1 randomness 95 patterns 1 2 shared-site 69–71 assembly 8 2 simulated communities 106 species replacement 257 3 statistical power of tests 96 Perognathus longimembris 66 3 4 null models 6, 8, 14 Perognathus parvus 66, 67 4 assembly rules 133 Peromyscus maniculatus 63, 66, 78 5 Colwell and Winkler’s 27–9 persistence 5 6 community achievement 368 6 composition 255–6 communities 366 7 structure 253–4 restoration 373, 375, 376 7 8 Connor and Simberloff 27 restored community 368 8 difficulty in framing 132–4 phase transitions 240–1 9 explicit ecological process 134 phosphate mining 373 9 10 obvious feature inclusion 133 physical conditions 311 10 11 patch model 135, 136 stream fish community 321, 322, 323 11 safeguards 133–4 simulation 319–20 12 selection 210 variation 312 12 13 species scattering 134 pine 13 strategies for finding assembly rules 253–6 lodgepole 169 14 trait–environment patterns 261 see also conifer forest 14 15 see also Narcissus effect Pinus contorta 169, 171 15 plant characters 16 Oahu island 108–9 texture convergence 146–9 16 17 extant assemblage 117 vegetative 146 17 introduction success 111, 112–13 plant communities, trait–environment 18 morphological overdispersion 117, 118, 119 linkages 338 18 19 priority effect absence 115, 116 plant distribution, Barkley Sound islands 168 19 20 species introduced 126-9 plant functional types (FTs) 343–4 20 oak woodland birds invading 356 21 core species 183, 184, 185–6, 188–9 montane grassland 356 21 22 diversity 203 regional pool 345, 346–8 22 flycatchers 184, 186–7, 188, 189 trait–environment linkage 357 23 species composition/diversity 181, 182, Argentinian study 344–56 23 24 183, 184, 185 woodlands 355–6 24 vireos 184, 186, 188 plant traits 25 of western North America 181, 182, 183, ecosystem function 341, 342, 343 25 26 184, 185–7, 188–9 prediction 357 26 woodpeckers 185, 187 frequency distribution 349, 350, 351 27 wrens 184, 185, 186 FTs 343–4, 345, 346–8, 349 27 28 oil spill restoration 376–7 regeneration 343, 344, 347–8 28 29 Oncorhynchus mykiss 314, 320 regenerative patterns 351, 352, 353 29 Orthocarpus castillejoides var. trait–environment linkages 353–5 30 humboldtiensis 372 vegetative 341, 343, 344, 346-8 30 31 overdispersion, morphological on oceanic patterns 351, 352, 353 31 islands 116–17, 118, 119–20 Poaceae, Barkley Sound islands 173 32 pollination, Stylidium 145 32 33 Pachycephala 303–4 pollinators, species sharing 144 33 passeriforms Polypodium vulgare 169 34 Bermuda 113, 117 Polystichum munitum 169 34 35 morphological overdispersion 117, 118, Pomatostomus 180 35 119 prairies 36 Saint Helena 117 burning 364 36 37 species pool on oceanic islands 120 potholes 7 37 38 Tahiti 117 predation 38 patch model 135, 136–7 IDF 299–301, 302, 303–4 39 habitat heterogeneity 153 intensity in stream fish 316–17 39 40 patchiness, environmental 134–5, 136 stream fish community 322, 323, 324 40 414 Index

1 predator redthroat 181 1 2 brown trout 312, 320, 331 reductionism 247 2 invasion of stream fish community 323, refugia, biogeographic 303 3 325–6, 327 Reithrodontomys megalotis 85 3 4 salmonid 314 rejection criteria 122 4 presence/absence matrices 62 relative abundance distribution (RAD) 154, 5 presence/absence rules 137–9, 140–1, 142–3 156 5 6 environmental scale 143 evenness 154–5, 156, 157 6 large-scale patterns 139, 143 predictivity of shape 157, 158 7 species richness 137–8, 139 shape 157, 158 7 8 prey, extinction in stream fish community relaxation fauna 277, 281 8 326, 327 rescue effect 274 9 prey/predator species porportions in resource availability 9 10 community 151 favored assemblages 33, 34, 37 10 11 priority effect, oceanic islands 114–16 guild assembly 31 11 process, guild assembly rule 51–2 rules 32, 33 12 production, soil fertility 259 partitioning 37 12 13 productivity skewed 32–3, 34, 35, 36, 37 13 arid regions 79 specialization 37 14 relative for archipelagos 293 unfavored assemblages 33, 34, 37 14 15 propagule size on cliff face 220, 224, 225 resource requirements 15 protead heathland covariation 283, 284 16 birds 178–9 distribution influence 301 16 17 density 179 faunal groups 283 17 species diversity 178–9 immigration 284–5 18 community composition 178–9 island populations 288 18 19 diversity 178–9 species differences 297 19 20 species richness 179 resource utilization 20 vegetation structure 179 curves 23 21 Pseudotsuga menziesii 169, 171 functions 24 21 22 Ptilinopus fruit pigeons 6, 303 resources 22 demands 304 23 rainforest canopy gap distribution pattern 242 ecological dominance 304 23 24 random assemblages 265 food 37 24 patterns 132 insular 288–9 25 random assignment 68–9 production curves 23 25 26 local community 64–5 soil 257 26 random data set generation 41, 42–3, 44 use 33 27 Stone, Dayan and Simberloff 45 desert rodents 94 27 28 Wilson 41, 44 wetland plant assemblage dispersal 254–5 28 29 random data set null model testing 133 see also consumer resource competition; 29 random distributions 95 consumer resource model 30 random draws 67 restoration 363–7 30 31 random patterns test 137 assembly theory 367 31 random shifts 137 community persistence 368 32 randomization tests 27 completely successful projects 373 32 33 randomness descriptive variables 375 33 null hypothesis 95 ecological experimentation 367–9 34 species invasion patterns 212 ecosystem functions 364 34 35 rat, kangaroo 78 environmental manipulation 367–9 35 allopatric speciation 92 expressed goals 363 36 geographic variation 93 function 363, 365 36 37 habitat generalist 97 goals 37 38 interspecific competition 98–9 achievement 372, 374 38 recruitment, community assembly 333 classification 370–1 39 Red River (Oklahoma) 299 functional/structural 370–1, 374 39 40 redside, Lahontan 314, 320 invasion 366, 367 40 Index 415

1 literature base 369–73 local community 1 2 management cessation 368, 372, 373 favored states 83–4, 85, 86 2 natural patterns of secondary succession structure 100 3 367 local species richness 79 3 4 partially successful projects 371–2, 373 locomotion mode 94, 99 4 patterns 375–7 macro-/micro-habitat selection 78 5 persistence achievement 373, 375, 376 morphological group classification 83 5 6 projects 385–90 morphological trait non-random ratios 80 6 results 373–5 Narcissus effect 86–7 7 results classification 371–3 non-random assemblages 84–5, 86 7 8 review of literature 369–70 pattern dependence between species 89 8 secondary succession 375 resource use 94 9 species composition 364–5, 375 seed-eating of Sonoran desert 25–6 9 10 achievement 368 spatial scales of assemblages 91 10 11 species weeding 368 species 11 structure 363, 364–5 diversity 78, 89 12 success patterns 365 identity patterns 87–8, 89, 90 12 13 success rates of projects 373 with negative/positive associations 88 13 success/failure judgement 369 strength of interaction in local community 14 succession 365–6, 367 84, 85 14 15 unsuccessful projects 371, 373 trait patterns at different spatial scales 81 15 weeding and seeding 368 root interactions 159 16 Rhinichthys osculus robustus 314, 320 rotation/reflection, spatial autocorrelation 137 16 17 Richardsonius egregius 314, 320 17 robins 180 Sagehen Creek (Truckee river; California) 18 rodents 320, 328, 330 18 19 feeding guilds 37–8 Saint Helena 108, 109 19 20 rainforest–steppe interface 38–9 finches 112, 117 20 rodents, desert 76 introductions 113, 126–9 21 aggressive dominance 94 success 111 21 22 allopatric speciation 92–3 morphological overdispersion 118, 119 22 arid southwest of North America 60, 61 passeriforms 117 23 body size 99 priority effect 114, 115 23 24 analysis 80, 81 species introduced 126–9 24 correlations 94 Salmo trutta 312, 320 25 causal mechanisms 91–4 Salmon River salt marsh restoration 372 25 26 cheek pouch volume 80, 94 salmonid fish 26 co-occurence 81, 82 demographic characteristics 314 27 community structure patterns 87–8, 89, 90 invasion 323, 325-7 27 28 ecological segregation 99 salt marsh 28 29 ecological sorting 93–4 intrinsic guilds 152 29 evolutionary assemblages 99 restoration 372 30 evolutionary character displacement 93–4 sclerophyllous vegetation 177 30 31 facilitated coexistence 91 sculpin 31 Fox’s assembly rule 82–3 Paiute 314, 320 32 functional groups 99–100, 209 riffle 318 32 33 classification 83 seedlings 33 coexistence 82–7 establishment 219, 220 34 granivorous 38, 58, 60–1, 99 survivorship 220, 223, 224 34 35 habitat seeds 35 generalists 97 bank 219, 220, 222 36 heterogeneity 78 persistence in soil banks 354 36 37 incisor width 80, 93, 94 rain 219, 220, 221 37 38 interactions in structure of local size 224, 353 38 community 98 self-indulgence 13 39 interspecific competition 91–3 self-organization 238, 246 39 40 process 98–9 assembly trajectory 238–42 40 416 Index

1 self-organization (continued) abundance 1 2 determinism 239 relative 154 2 pattern production 242 self-organization 240 3 phase transitions 240–1 accumulation in island populations 288 3 4 random fluctuations 239 assembly and guild assembly rule 29, 30, 4 species abundance/diversity 240 31 5 succession 240 combinations 23 5 6 system dynamics variations 240 constraints on success at site 211–12 6 Semotilus atromaculatus 299 distribution 7 shared-site null hypothesis 69–71 island biogeography 277, 279, 280 7 8 shiner, Ouachita Mountain 279 relative abundance 154 8 shoreline diversity 77 9 island systems species 171–2, 174–5, 177 desert rodents 78, 89 9 10 sandy 265, 266 self-organization 240 10 11 shrew(s) frequency inequality 64 11 data hidden combinations 234 12 from North America 61 local assemblages 76 12 13 site occupation 70–1 number importance in biosphere 4 13 Great Lakes islands introductions 300 occurrence 69 14 immigration 301 patterns 304 14 15 masked 279 invasion 212 15 short-tailed 300 regional pool 94 16 shrubland 353–4 role/strategy 23 16 17 Sialia sialis 113 scattering models 134 17 similarity, limiting 143–6, 254 similarity 91 18 theory 137, 138 turnover in community 166 18 19 site weeding 368 19 20 occupation from shrew data 70–1 widespread 69 20 richness 69 see also co-occurrence; coexistence; 21 size dominance 10–11 speciation 21 22 small island effect 287, 292 species composition 22 small-mammal diversity in Australia 29, 30, achievement 368 23 31–2 restoration 375 23 24 soil fertility 259 turnover 366 24 deviation 266 type B assembly rules 165 25 soil resources 257 species interactions 25 26 Sorex cinereus 301 assembly rule 132 26 sorting, ecological 93–4 nature 297 27 sparrows persistence of insular population 297 27 28 core habitat occupation 195, 196, 197 realized insular distribution 295, 297, 298, 28 29 emberizid coexistence 191, 193, 196, 197 299–301, 302, 303–4 29 grass-sagebrush of Grand Teton National see also interspecific interactions 30 Park (Wyoming) 191, 193 species pool 30 31 interspecific aggression 198 communities 2, 3 31 interspecific co-occupancy of habitat formation on oceanic islands 110–13 32 avoidance 197–8 isolated desert areas 79 32 33 willows site in Grand Teton National Park phylogenetic variability 111 33 (Wyoming) 193, 194, 195, 196, 197–8 species range 66–8 34 Spartina 371 biogeographic 69 34 35 spatial autocorrelation 135, 136–7, 159 simulated 62 35 spatial gradients, temporal change 341 site sharing 66–7 36 spatial niche overlap, mean 30, 31 species richness 154–5 36 37 spatial patterns, gradients 341 assembly rules 137, 138 37 38 specialization, resource availability 37 curves 296–7 38 speciation 36, 37 deviation 266 39 allopatric 92–3 distribution across site 40 39 40 species lawns 138, 140, 141 40 Index 417

1 local plant characters 146–9 1 2 desert rodents 79 thermodynamics, non-linear 260, 261 2 vs regional 139 Thinichthys osculus 318 3 patterns 77–9, 285, 287–93 thornbills 180–1 3 4 large-scale 139, 143 Thuja occidentalis 216, 218, 219, 221–3, 224 4 prediction 283, 285 Tokeshi’s principle 133 5 presence/absence rules 137–8, 139, 140–1 top-down processes 87 5 6 protead heathland 179 control 237 6 regional pools 139, 141 topminnow 312 7 trait-based assembly rules 266, 267 topographic–edaphic conditions, plant 7 8 variance 137, 138, 139 interactions 340 8 species–area relationship 285, 287–9, 291 trade-offs in assembly rules 263 9 effect 62 trait 9 10 island biogeography 285, 287–9, 293 dispersion 254, 256 10 11 sigmoidal 291–2 overdispersion 146 11 species-by-site matrix (SSM) 169 plants and edaphic factors 257 12 species–energy group 293 space components 258 12 13 species–isolation relationship 285, 287, variation along gradients 256–7 13 289–91 trait patterns 258, 259 14 negative sigmoidal 292 desert rodents 80–1 14 15 standard area of deviation 117 soil fertility 259 15 Stone, Dayan and Simberloff, Nevada data trait–environment linkages 236–9 16 set analysis 45 FTs 357 16 17 stream flow, stream fish community 312, 331 wetland plant assemblages 257, 258 17 Stylidium 145 trait–environment linkages in plant 18 co-occurring species 145–6 communities 338 18 19 substitutes, ecological 166 climatic factors 349, 350, 351, 352 19 20 succession community/ecosystem process prediction 20 evenness 157 343 21 history of concept 212–13 ecosystem function 353–5 21 22 predictability in unproductive areas 215 prediction 355–6 22 secondary 365–6, 375 filtering processes 339–40 23 self-organization 240 filters 345 23 24 stages 212, 214 FTs 344–5, 346–8, 349, 350, 351, 352, 24 trajectories in highly stressed communities 353–6 25 215 montane grasslands 354 25 26 successional change 213 open shrubland 354 26 sucker, Tahoe 314, 320 plant–animal interactions 353–4 27 supertramp strategy 23 regional pool 345, 346-8 27 28 sycophancy 12 trait frequency distribution 349, 350, 351 28 29 Sylvia warblers 189–91 woodlands/woodland–shrubland 353–4 29 see also plant traits, regenerative; plant 30 Tahiti 108, 109 traits, vegetative 30 31 introductions 113 trait–environment paradigm 5, 7 31 success 111 trait–environment patterns 32 morphological overdispersion 117, 118, assembly rules 261–3, 268 32 33 119 community assembly 263–4 33 passeriforms 117 constraints 260 34 priority effect 114, 115 predictions 261–2 34 35 species introduced 126–9 value 266 35 taxonomy, guild classification 150–1 range 258, 261 36 Taxus 169, 170, 171 trees, canopy composition prediction 213 36 37 Taxus brevifolius 169, 171 tribal units 12 37 38 template-filter model 254 trout 38 temporal change, spatial gradients 341 brown 312, 320, 331 39 texture convergence 147–9 rainbow 314, 320 39 40 analysis 158 Trout Zone Fish Assemblage 312, 331 40 418 Index

1 unfavored state coexistence 190, 203 1 2 assemblage 31, 32, 33, 34 willows site in Grand Teton National Park 2 Fox’s assembly rule 95 198–9, 200–1, 202 3 guild assembly rule 31, 32 water deficit 349 3 4 local community 59 waxbill, common 110 4 sites 105 weeding and seeding, restoration 368 5 resource availability 33, 34, 37 wetlands 5 6 see also favored state above-ground competition intensity 255 6 adaptive landscape 268 7 vagility 304 guild proportionality 256 7 8 distribution influence 301 limiting similarity in species 144–5 8 ecological dominance 304 mineral resource increase 259–60 9 species differences 297 morphological dispersion 254, 254–5, 9 10 vegetation 256 10 11 gradient 136 plant assemblages 254–5 11 height and warbler foraging 190–1 plant invasion 209 12 transition 14, 15 trait–environment linkages 257, 258 12 13 vegetative characters 146 Wilson JB, guild assembly rule 41, 44 13 vireo woodlands/woodland–shrubland 352, 353–4 14 oak woodland 184, 186, 188 woodpeckers 185, 187 14 15 white-eyed 113 wrens 184, 185, 186 15 Vireo griseus 113 Wright and Biehl approach, shared-site null 16 vole, meadow 300, 301 hypothesis 69, 70 16 17 17 warblers yew, Pacific 169 18 paruline 198–9, 200–1, 202 18 19 sylvine 189–91 zero net growth isocline (ZNGI) 46–7 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40