Journal of Vegetation Science September 2013 777 — 974 Issue 5 Volume 24 Journal of Vegetation Science Journal of Vegetation Science Volume 24 • Issue 5 • September 2013 • ISSN 1100-9233 Advances in plant community ecology Volume 24 Issue 5 September 2013 Contents
Special Feature: Functional Diversity
N.W.H. Mason & F. de Bello – Functional diversity: a tool for answering challenging ecological questions 777 S. Pavoine, A. Gasc, M.B. Bonsall & N.W.H. Mason – Correlations between phylogenetic and functional diversity: 781 mathematical artefacts or true ecological and evolutionary processes? N.W.H. Mason, F. de Bello, D. Mouillot, S. Pavoine & S. Dray – A guide for using functional diversity indices to reveal 794 changes in assembly processes along ecological gradients F. de Bello, C.P. Carmona, N.W.H. Mason, M.-T. Sebastià & J. Lepš – Which trait dissimilarity for functional diversity: 807 trait means or trait overlap? N.W.H. Mason & S. Pavoine – Does trait conservatism guarantee that indicators of phylogenetic community structure 820 will reveal niche-based assembly processes along stress gradients? E. Laliberté , D.A. Norton & D. Scott – Contrasting eff ects of productivity and disturbance on plant functional diversity 834 at local and metacommunity scales P. Gerhold, J.N. Price, K. Püssa, R. Kalamees, K. Aher, A. Kaasik & M. Pärtel – Functional and phylogenetic community 843 assembly linked to changes in species diversity in a long-term resource manipulation experiment L. Chalmandrier, T. Münkemüller, L. Gallien, F. de Bello, F. Mazel, S. Lavergne & W. Thuiller – A family of null models 853 to distinguish between environmental fi ltering and biotic interactions in functional diversity patterns R.J. Pakeman & A. Eastwood – Shifts in functional traits and functional diversity between vegetation and seed bank 865 M. Bernard-Verdier, O. Flores, M.-L. Navas & E. Garnier – Partitioning phylogenetic and functional diversity into alpha 877 and beta components along an environmental gradient in a Mediterranean rangeland M. Hejda & F. de Bello – Impact of plant invasions on functional diversity in the vegetation of Central Europe 890 Š. Janeček, F. de Bello, J. Horník, M. Bartoš, T. Černý, J. Doležal, M. Dvorský, K. Fajmon, P. Janečková, Š. Jiráská, 898 O. Mudrák & J. Klimešová – Eff ects of land-use changes on plant functional and taxonomic diversity along a productivity gradient in wet meadows T. Herben, Z. Nováková & J. Klimeš ová – Comparing functional diversity in traits and demography of Central 910 European vegetation C.M. Hulshof, C. Violle, M.J. Spasojevic, B. McGill, E. Damschen, S. Harrison & B.J. Enquist – Intra-specifi c and 921 inter-specifi c variation in specifi c leaf area reveal the importance of abiotic and biotic drivers of species diversity across elevation and latitude M. Carboni, A.T.R. Acosta & C. Ricotta – Are diff erences in functional diversity among plant communities on 932 Mediterranean coastal dunes driven by their phylogenetic history? S. Lavorel, J. Storkey, R.D. Bardgett, F. de Bello, M.P. Berg, X. Le Roux, M. Moretti, C. Mulder, R.J. Pakeman, 942 S. Dí az & R. Harrington – A novel framework for linking functional diversity of plants with other trophic levels for the quantifi cation of ecosystem services M. Moretti, F. de Bello, S. Ibanez, S. Fontana, G.B. Pezzatti, F. Dziock, C. Rixen & S. Lavorel – Linking traits between 949 plants and invertebrate herbivores to track functional eff ects of land-use changes V.D. Pillar, C.C. Blanco, S.C. Müller, E.E. Sosinski, F. Joner & L.D.S. Duarte – Functional redundancy and stability 963 in plant communities
Indexed and abstracted in: Absearch, Biological Abstracts, Elsevier BIOBASE/Current Awareness in Biological Sciences, Current Contents, Environmental Periodicals Bibliography, International Association for Vegetation Science ISI Web of Science
jjvs_v24_i5_oc.inddvs_v24_i5_oc.indd 1 77/27/2013/27/2013 11:57:4611:57:46 AMAM International Association for Vegetation Science JOURNAL OF VEGETATION SCIENCE Governing Board for 2011–2015 Offi cial organ of the International Association for Vegetation Science (www.iavs.org) PRESIDENT VICE PRESIDENT & MEETINGS COMMITTEE CHAIR Martin Diekmann, University of Bremen, Germany; [email protected] Valério De Patta Pillar, Federal University of Rio Grande do Sul, The Journal of Vegetation Science publishes papers on all aspects of plant community ecology, with particular emphasis on papers that develop new SECRETARY Porto Alegre, Brazil concepts or methods, test theory, identify general patterns, or that are otherwise likely to interest a broad readership. Papers may focus on any aspect of Susan Wiser, Landcare Research, Lincoln, New Zealand; [email protected] VICE PRESIDENT & MEMBERSHIP COMMITTEE CHAIR vegetation science including theory, methodology, spatial patterns (including plant geography and landscape ecology), temporal changes (including pal- VICE PRESIDENT & PUBLICATIONS OFFICER Michael Palmer, Oklahoma State University, Stillwater, OK, USA aeoecology and demography), processes (including ecophysiology), and description of ecological communities (by phytosociological or other methods), Robert K. Peet, University of North Carolina, Chapel Hill, NC, USA VICE PRESIDENT & GLOBAL FUND COMMITTEE CHAIR provided the focus is on increasing our understanding of plant communities. Papers that apply ecological concepts, theories and methods to the management VICE PRESIDENT & VEGETATION CLASSIFICATION Alicia Acosta, University of Rome Three, Italy and/or description of vegetation should be directed to our associated journal, Applied Vegetation Science. The Journal accepts original papers (including COMMITTEE CHAIR ADMINISTRATOR (ex offi cio) shorter ones), synthesis papers, forum contributions and reports. It commissions reviews of selected software. Javier Loidi, University of the Basque Country, Bilbao, Spain Nina Smits, Alterra, Wageningen, The Netherlands; [email protected] The Journal of Vegetation Science was founded by Eddy van der Maarel, Robert K. Peet and Robert Neuhäusl in 1990. Goals The International Association for Vegetation Science (IAVS) is a worldwide union of scientists and others interested in theoretical and practical Submissions studies of vegetation: its composition and structure, history, classifi cation, distribution, ecology, dynamics, management and uses in the landscape. Journal of Vegetation Science For submission instructions, subscription and all other information visit: http://wileyonlinelibrary.com/journal/jvs. accepts The main goals of the IAVS are to facilitate personal contacts among vegetation scientists all over the world and to promote research in all aspects articles for Open Access publication. Please visit http://olabout.wiley.com/WileyCDA/Section/id-406241.html for further information about OnlineOpen. of vegetation science and its applications. Please forward any general submission queries to jvsci@editorialoffi ce.co.uk Toward these ends we hold meetings and excursions, sponsor publications, and provide mechanisms for vegetation scientists to communicate with each other and the world at large. CHIEF EDITORS David Ward, University of Kwazulu-Natal, Pietermaritzburg, Contact IAVS South Africa Alessandro Chiarucci Details about IAVS can be found on our website at www.iavs.org. For additional information about membership, contact Nina Smits (admin@ University of Siena, Italy; [email protected] Karsten Wesche, Senckenberg Museum of Natural History, Görlitz, iavs.org). For other information contact President Martin Diekmann ([email protected]) or Secretary Susan Wiser ([email protected]) Valério De Patta Pillar Germany Otto Wildi, Swiss Federal Institute for Forest, Snow and Landscape Membership Federal University of Rio Grande do Sul, Porto Alegre, Brazil; [email protected] € € with Research, Birmensdorf, Switzerland For all membership queries, please contact the Administrator ([email protected]). Membership costs 20 for full members and 10 for student Milan Chytrý Kerry D. Woods, Bennington College, VT, USA members. Free memberships are available for individuals in countries with currency problems (contact Administrator Nina Smits admin@iavs. Masaryk University, Brno, Czech Republic; [email protected] Martin Zobel, University of Tartu, Estonia org for further information). Members qualify for a deep discount in the subscriptions prices of our publications. To join, go to http://iavs.org/ Meelis Pärtel [Chair] MembershipRenew.aspx University of Tartu, Estonia; [email protected] EDITORIAL BOARD Meetings Gudrun Bornette, University Claude Bernard Lyon 1, Villeurbanne, IAVS hosts annual symposia and associated fi eld excursions to facilitate communication and to increase awareness and understanding of the ASSOCIATE EDITORS France diversity of world vegetation. For more information on future meetings, go to http://www.iavs.org/MeetingFuture.aspx Alica T. R. Acosta, University of Rome Three, Italy Zoltán Botta-Dukát, MTA Centre for Ecological Research, Vácrátót, Sándor Bartha, MTA Centre for Ecological Research, Vácrátót, Hungary Hungary Information for subscribers Production Editor: Hans Henrik Bruun, University of Copenhagen, Denmark Helge Bruelheide, University of Halle, Germany Journal of Vegetation Science is published in six issues per year. Hoshia Rose (email: [email protected]). Beverly S. Collins, Western Carolina University, Cullowhee, NC, USA Richard T. Busing, US Geological Survey, Corvallis, OR, USA Institutional subscription prices for 2013 are: Print & Online: US$1852 Wiley’s Corporate Citizenship initiative seeks to address the environmen- € Sara Cousins, Stockholm University, Sweden Paolo Cherubini, Swiss Federal Institute for Forest, Snow and Landscape (US & Rest of World), 1203 (Europe), £946 (UK). Prices are exclusive of tal, social, economic, and ethical challenges faced in our business and Francesco de Bello, University of South Bohemia, Cˇ eské Budeˇjovice, Research, Birmensdorf, Switzerland tax. Asia-Paci fi c GST, Canadian GST and European VAT will be applied which are important to our diverse stakeholder groups. Since launching Czech Republic Guillaume Decocq, Jules Verne University of Picardy, Amiens, France at the appropriate rates. For more information on current tax rates, please the initiative, we have focused on sharing our content with those in need, Philip M. Dixon, Iowa State University, Ames, IA, USA go to www.wileyonlinelibrary.com/tax-vat. The price includes online enhancing community philanthropy, reducing our carbon impact, creating Miquel De Cáceres, Forest Science Centre of Catalonia, Solsona, Spain access to the current and all online back fi les to January 1st 2009, where Roger del Moral, University of Washington, Seattle, WA, USA Leandro Duarte, Federal University of Rio Grande do Sul, Porto Alegre, global guidelines and best practices for paper use, establishing a vendor available. For other pricing options, including access information and code of ethics, and engaging our colleagues and other stakeholders in our Rasmus Ejrnæs, University of Aarhus, Rønde, Denmark Brazil terms and conditions, please visit www.wileyonlinelibrary.com/access. Hannes Feilhauer, University of Erlangen-Nuremberg, Erlangen, efforts. Follow our progress at www.wiley.com/go/citizenship Jason Fridley, Syracuse University, NY, USA Delivery Terms and Legal Title Germany Frank Gilliam, Marshall University, Huntington, WV, USA Where the subscription price includes print issues and delivery is to the Copyright and Photocopying David Gibson, Southern Illinois University, Carbondale, IL, USA Sarah Goslee, USDA Agricultural Research Service, University Park, PA, USA recipient’s address, delivery terms are Delivered at Place (DAP); the Copyright © 2013 International Association for Vegetation Science. All Nicolas Gross, Chizé Centre for Biological Studies, Villiers en Bois, rights reserved. No part of this publication may be reproduced, stored or Rune Halvorsen, University of Oslo, Norway recipient is responsible for paying any import duty or taxes.Title to all France transmitted in any form or by any means without the prior permission in Michael Huston, Texas State University, San Marcos, TX, USA issues transfers FOB our shipping point, freight prepaid. We will endeavour Sabine Güsewell, Swiss Federal Institute of Technology (ETH), Zürich, to fulfi l claims for missing or damaged copies within six months of writing from the copyright holder. Authorization to photocopy items for Zaal Kikvidze, Ilia University, Tbilisi, Georgia Switzerland publication, within our reasonable discretion and subject to availability. internal and personal use is granted by the copyright holder for libraries and Ingolf Kühn, Centre for Environmental Research Leipzig-Halle, Halle, other users registered with their local Reproduction Rights Organisation Jessica Gurevitch, State University of New York, Stony Brook, NY, USA JOURNAL OF VEGETATION SCIENCE (ISSN 1100-9233), is published Germany (RRO), e.g. Copyright Clearance Center (CCC), 222 Rosewood Drive, US mailing Florian Jansen, University of Greifswald, Germany in January, March, May, July, September and November. Danvers, MA 01923, USA (www.copyright.com), provided the appropriate Norman Mason, Landcare Research, Hamilton, New Zealand ˇ agent: Mercury Media Processing, LLC 1634 East Elizabeth Ave, Linden Jan Lepš, University of South Bohemia, Ceské Budeˇjovice, Czech Republic fee is paid directly to the RRO. This consent does not extend to other Richard Michalet, University of Bordeaux, Talence, France NJ 07036 USA. Periodical postage paid at Rahway, NJ. Simona Maccherini, University of Siena, Italy kinds of copying such as copying for general distribution, for advertising John Morgan, La Trobe University, Bundoora, Australia Postmaster: Send all address changes to JOURNAL OF VEGETATION Fraser Mitchell, Trinity College, Dublin, Ireland or promotional purposes, for creating new collective works or for resale. Tohru Nakashizuka, Tohoku University, Sendai, Japan SCIENCE, Journal Customer Services, John Wiley & Sons Inc., 350 Main St., Jesse Nippert, Kansas State University, Manhattan, KS, USA Special requests should be addressed to: [email protected] Mike Palmer, Oklahoma State University, Stillwater, OK, USA [Forum Editor Malden, MA 02148-5020. Jodi Price, University of Western Australia, Perth, Australia and Software Review Editor, [email protected]] Publisher Disclaimer Håkan Rydin, Uppsala University, Sweden Juli G. Pausas, Desertifi cation Research Centre (CIDE-CSIC), Valencia, Spain Journal of Vegetation Science is published by Blackwell Publishing Ltd, The Publisher, International Association for Vegetation Science and Rachel J. Standish, University of Western Australia, Perth, Australia Editors cannot be held responsible for errors or any consequences aris- János Podani, Eötvös University, Budapest, Hungary 9600 Garsington Road, Oxford OX4 2DQ, UK; Tel. +44 (0) 1865 776868; Andrew Tanentzap, York University, Toronto, Canada ing from the use of information contained in this journal; the views Andreas Prinzing, University of Rennes 1, France Email: [email protected]. Cyrille Violle, CNRS, Montpellier, France Blackwell Publishing was acquired by John Wiley & Sons in February and opinions expressed do not necessarily refl ect those of the Publisher, Francisco Pugnaire, Arid Zones Experimental Station (EEZA-CSIC), International Association for Vegetation Science and Editors, neither Almeria, Spain Kerstin Wiegand, University of Göttingen, Germany 2007. Blackwell’s programme has been merged with Wiley’s global Scientifi c, Technical, and Medical business to form Wiley-Blackwell. does the publication of advertisements constitute any endorsement by Gillian Rapson, Massey University, Palmerston North, New Zealand Gail W.T. Wilson, Oklahoma State University, Stillwater, OK, USA the Publisher, International Association for Vegetation Science and Journal Customer Services: For ordering information, claims and any enquiry Duccio Rocchini, Edmund Mach Foundation, Trento, Italy Monika Wulf, Leibniz-Centre for Agricultural Landscape Research, Editors of the products advertised. Müncheberg, Germany concerning your journal subscription please go to www.wileycustomerhelp. Stephen Roxburgh, CSIRO, Canberra, Australia com/ask or contact your nearest offi ce: ISSN 1100-9233 (Print) Sam Scheiner, US National Science Foundation, Arlington, VA, USA CONSULTANT EDITORS Americas: Email: [email protected]; Tel: +1 781 388 8598 or +1 800 835 ISSN 1654-1103 (Online) Sebastian Schmidtlein, University of Bonn, Germany Robert K. Peet, University of North Carolina, Chapel Hill, NC, USA 6770 (Toll free in the USA & Canada). View this journal online at wileyonlinelibrary.com/journal/jvs Lindsay Turnbull, Institute of Environmental Sciences, Zürich, Switzerland Europe, Middle East and Africa: Email: [email protected]; J. Bastow Wilson, Access to this journal is available free online within institutions in the University of Otago, Dunedin, New Zealand Tel: +44 (0) 1865 778315. Vigdis Vandvik, University of Bergen, Norway developing world through the AGORA initiative with the FAO, the Asia Pacifi c: Email: [email protected]; Tel: +65 6511 8000. HINARI initiative with the WHO and the OARE initiative with UNEP. For Japan: For Japanese speaking support, Email: [email protected]; Tel: +65 information, visit www.aginternetwork.org, www.healthinternetwork.org 6511 8010 or Tel (tollfree): 005 316 50 480. Cover photograph: Area de Conservación Guanacaste, Costa Rica, a summit of cloud-forested volcano, looking west towards lowland dry tropical forests and www.oaresciences.org Pacifi c coast. Photo by C. Hulshof. See Hulshof et al., ‘Intra-specifi c and inter-specifi c variation in specifi c leaf area reveal the importance of abiotic and biotic Visit our Online Customer Get-Help available in 6 languages at www. drivers of species diversity across elevation and latitude’, Journal of Vegetation Science 24: 921–931. wileycustomerhelp.com. Printed in Singapore by C.O.S. Printers Pte Ltd
jjvs_24_5_ic.inddvs_24_5_ic.indd 1 77/27/2013/27/2013 11:56:1311:56:13 AMAM Journal of Vegetation Science 24 (2013) 777–780 SPECIAL FEATURE: FUNCTIONAL DIVERSITY Functional diversity: a tool for answering challenging ecological questions Norman W. H. Mason & Francesco de Bello (SPECIAL FEATURE EDITORS)
Keywords Abstract Beta diversity; Community assembly; Ecosystem function; Environmental filtering; Functional trait diversity provides a powerful means of addressing ecology’s per- Niche complementarity; Phylogenetic diversity sistent questions, through its dual role as an indicator of mechanisms driving dif- ferences in species composition between communities and as a predictor of Received 7 May 2013 ecosystem-level processes. Functional traits provide a means of testing mecha- Accepted 7 May 2013 nisms behind species turnover between communities because environmental Co-ordinating Editor: Bastow Wilson heterogeneity, competition and disturbance influence species fitness via their traits. Functional traits also provide a link between species and multiple ecosys- Mason, N.W.H. (corresponding author, tem-level processes, such as primary productivity, nutrient fluxes and resilience, [email protected]): Ecosystem since species influence these processes via their traits. This special issue demon- Processes, Landcare Research, Private Bag strates that functional diversity offers a practical means of investigating ecology’s 3127, Hamilton, New Zealand persistent questions. de Bello, F. ([email protected]): Institute of Botany, Academy of Sciences of the Czech Republic, Dukelsk a 135, 379 82, T rebon, Czech Republic and Department of Botany, University of South Bohemia, Na Zlate stoce 1, 370 05, Cesk e Budejovice, Czech Republic
Quantification of functional diversity and its Introduction components Many studies are emerging using functional traits and Understanding the diversity of functional diversity indices functional diversity (FD) to assess long-standing ecologi- cal questions. However, as pointed out by Cadotte et al. The quantification of FD requires a set of measureable (2011), the body of empirical evidence on the patterns concepts. Many definitions of FD have been provided by of FD is still rather small. This special issue presents a different authors. Most definitions agree that FD is the number of studies demonstrating how FD might help us diversity of the functional traits of the species in a commu- to solve some of the long-standing questions in ecology. nity. However, this general definition gives little specific In this Editorial, we outline how the special issue has guidance on how the concept of diversity should be advanced different aspects of FD research. These aspects applied to functional trait data. Mason et al. (2005) pro- include: (1) the quantification of FD; (2) the use of FD posed a definition of FD as the distribution in functional to reveal mechanisms driving community assembly and trait space of the species presence and abundance in a com- ecosystem functions; and (3) relationships between FD munity, including three components: (1) the amount of and other aspects of biodiversity. We devote a specific functional trait space filled by species in the community section to each of these areas, providing a background (functional richness); (2) the evenness of abundance distri- and synthesis of the different studies comprising this bution in filled trait space (functional evenness); and (3) special issue. We discuss some possible developments the degree to which the distribution of species abundances and uncertainties that still remain to be resolved in the maximizes divergence in functional traits (functional field. divergence).
Journal of Vegetation Science Doi: 10.1111/jvs.12097 © 2013 International Association for Vegetation Science 777 Functional diversity N.W.H. Mason & F. de Bello
The advantage of this definition is that it identifies dif- within-community diversity into within vs. between spe- ferent measurable components of FD. Often the plethora cies components. of available indices is bewildering for ecologists trying to Partitioning the total diversity of a region (gamma diver- decide how to measure FD, but some indices that appear to sity) into within- (alpha diversity) and between-commu- be complementary are actually very closely related mathe- nity (beta diversity; Whittaker 1975) components, is matically (Pavoine & Bonsall 2011). In this special issue, receiving increasing attention as a means of prioritizing two papers (Mason et al. 2013; Pavoine et al. 2013) show conservation efforts. Methods have emerged that permit that categorizing and decomposing FD indices according to direct comparison of the alpha and beta components of the above three FD components not only helps to simplify taxonomic, functional and PD (Bernard-Verdier et al. selection of indices, but also provides a framework for 2013; Carboni et al. 2013; Jane cek et al. 2013). The FD interpreting variation in FD index values along ecological within a community (alpha diversity) can be further gradients. The power of mathematical decomposition is decomposed into between-species and within-species very well illustrated in Pavoine et al. in assessing evidence components. It is becoming evident that taking into for correlations between FD and phylogenetic diversity account this within-species FD may be essential to under- (PD) indices in field and simulated data. In particular, Pav- standing the patterns of species co-existence and of species oine et al. demonstrate that most existing indices are distributions (Hulshof et al. 2013). In particular, trait dis- affected by co-varying factors (usually species richness and similarity measures that take into account within-species evenness in species abundance), and that removal of the trait variability provide a more robust way to assess the effect of these co-varying factors often requires observed joint effects of environmental filtering and niche comple- index values to be expressed relative to an appropriate null mentarity on community assembly (de Bello et al. 2013). model. Community assembly processes Alpha, beta and gamma functional diversity The last few years have seen a rapid increase in the use of The definition of FD proposed in Mason et al. (2005) FD indices to reveal plant community assembly processes. focuses very much on trait diversity between species This is one of the most persistent questions in ecology and within communities (alpha diversity). However, recent one of the main themes of this special issue. In general, years have seen a proliferation of studies examining FD at evidence for niche complementarity (when trait dissimilar- a variety of different scales (e.g. Fig. 1) – within species, ity between co-occurring species increases fitness) is between communities (i.e. beta diversity), regional (i.e. assumed when observed FD is higher than that expected gamma diversity) and even global. This special issue spans under an appropriate null model. By contrast, evidence of all these scales, particularly decomposing regional diversity environmental filtering is assumed when observed FD is into within- and between-community components and, less than expected (Cornwell et al. 2006). In this special issue, Mason et al. (2013) use simple assembly models to provide some guidance on which indices, combined with which null models, can reliably detect changes in assembly Functionaldiversity(FD) processes across a wide range of ecological contexts. partitioning Comparison of results from several field studies within this special issue (Gerhold et al. 2013; Jane cek et al. 2013; FD between communities Laliberte et al. 2013) reveal various possible shifts in FD along productivity and disturbance gradients. There is an obvious gap in the literature for an integrated study, incor- FD between species
Total FD within communities porating a large number of data sets and using standardized methods for calculating FD, to test whether assembly
FD within species processes shift in a consistent direction along ecological within communities gradients – particularly gradients of productivity and dis- REGION1 REGION2 turbance. Three studies in this special issue (Chalmandrier et al. Fig. 1. Different hierarchical levels at which functional diversity (FD) can 2013; Herben et al. 2013; Laliberte et al. 2013) demon- be studied. This hypothetical example shows the case of two regions with strate the importance of scale in using FD indices to reveal a different spatial partitioning of FD. Region 1, compared to Region 2 presents a higher total diversity but a smaller between-species FD within assembly processes. Laliberte et al. found increased envi- communities. Region 1 presents a smaller within-species FD within ronmental filtering with increasing fertilization/productiv- communities. ity at the metacommunity scale, due to exclusion of short
Journal of Vegetation Science 778 Doi: 10.1111/jvs.12097 © 2013 International Association for Vegetation Science N.W.H. Mason & F. de Bello Functional diversity and slow-growing species. At local community scale, Links between functional and phylogenetic diversity where biotic interactions influence species relative abun- dances and co-occurrences, they found evidence for Phylogenetic diversity is increasingly used as a means of increasing niche complementarity with increasing produc- detecting niche-based community assembly processes. tivity. Chalmandrier et al. and Herben et al. show that the This obviously assumes that variation in the traits that spatial scale of study influences the definition of the pool influence assembly processes is phylogenetically con- of species on which null models are based, which strongly served. Another, often unacknowledged, assumption is determines conclusions on community assembly. that trait conservatism at the species pool level is retained A variety of novel uses for FD in the study of community during community assembly. Studies dealing with PD in assembly processes have been revealed in this special issue. this special issue generally suggest that PD indices are Hejda & de Bello (2013) used FD indices to show that inva- poorly suited to the study of plant community assembly sive plants often introduce novel trait combinations to processes. Gerhold et al. (2013) do show some congru- invaded plant communities. Pakeman & Eastwood (2013) ence between the responses of PD and FD to experimen- show contrasts in the relationship between FD and ecologi- tally manipulated productivity in meadow communities, cal gradients for vegetative vs seed bank communities. If but also found that PD and FD were generally poorly cor- future studies find a consistent disconnect between FD in related. Bernard-Verdier et al. (2013) and Pavoine et al. vegetation and seed banks, this might improve our under- (2013) demonstrate that PD very poorly captures varia- standing of how disturbance impacts FD in plant commu- tion in functional structure along gradients of soil resource nities. Hulshof et al. (2013) compare the contribution of availability and salinity. This is true even when there is intra- and inter-specific trait variation in FD. They reveal a strongly significant phylogenetic trait conservatism at the decrease in the relative contribution of intra-specific varia- species pool level. Pavoine et al. also show that when tion as species richness increases. These novel applications strong correlations between FD and PD are obtained, this of FD indices to explore questions around community is almost always attributable to co-varying factors such as assembly demonstrate that this is a young and expanding species richness and evenness in species abundances. field. Using simulated communities Mason & Pavoine (2013) show that PD indices are unable to recover known shifts in assembly processes along ecological gradients. They Ecosystem processes within and across trophic demonstrate that this is because turnover in phylogenetic levels composition and diversity along ecological gradients does Managing ecosystems to ensure the provision of multiple not capture functional turnover. In other words, the phy- ecosystem processes and services is a key challenge for logenetic signal in trait variation at the species pool level is theoretical and applied ecology. Recent syntheses and often not maintained during community assembly. empirical studies have highlighted that functional traits predict the effects of global changes on ecosystem services Conclusion better than species diversity per se (Cadotte et al. 2011). Therefore, functional traits and FD are receiving increas- Collectively, the studies within this special issue provide a ing attention as the main biotic component through good example of the potential power of functional diver- which different organisms and communities can influ- sity to answer ecological questions. They also reveal many ence ecosystems processes. In this special issue, three of the potential pitfalls involved in using functional diver- studies show how to incorporate FD, within and across sity indices as indicators of community assembly and eco- trophic levels, to improve understanding of variation in system function. Many ecological questions will remain ecosystem processes. Lavorel et al. (2013.) propose a new problematic for years to come and inspire, new generations framework for assessing multitrophic linkages via func- of ecologists to develop novel approaches in search of tional traits. Moretti et al. (2013) explore these multi- answers. The present special issue shows that functional trophic trait linkages and their effects on multiple diversity may lead to new perspectives on persistent ecolo- ecosystem processes focusing on the linkages between gical question. plant and grasshopper traits and their interactions on grassland biomass. These studies will help to test hypoth- Acknowledgements eses and identify knowledge and data gaps that limit our We thank all authors, editors and referees that, with their understanding of functional processes in ecosystems. passion and dedicated work, made this special issue in the Another study demonstrates the benefit of functional Journal of Vegetation Science possible. Our most special redundancy for stability and resilience of ecosystems after thanks go to Sally Sellwood, for her invaluable contribu- disturbance (Pillar et al. 2013).
Journal of Vegetation Science Doi: 10.1111/jvs.12097 © 2013 International Association for Vegetation Science 779 Functional diversity N.W.H. Mason & F. de Bello tion as editorial secretary. We also thank Robin Pakeman, Jane cek, S., de Bello, F., Horn ık, J., Barto s, M., Cern y, T., Valerio Pillar, Bastow Wilson and Meelis Partel€ for their Dole zal, J., Dvorsky, M., Fajmon, K., Jane ckova, P., Jirask a, help in improving this editorial. S., Mudrak, O., & Klime sova, J. 2013. Effects of land-use changes on plant functional and taxonomic diversity along a References productivity gradient in wet meadows. Journal of Vegetation Science 24: 898–909. de Bello, F., Carmona, C.P., Mason, N.W.H., Sebastia, M-T., & Laliberte, E., Norton, D.A., & Scott, D. 2013. Contrasting effects Lep s, J. 2013. Which trait dissimilarity for functional of productivity and disturbance on plant functional diversity diversity: trait means or trait overlap? Journal of Vegetation at local and metacommunity scales. Journal of Vegetation Sci- Science 24: 807–819. ence 24: 834–842. Bernard-Verdier, M., Flores, O., Navas, M-L., & Garnier, E. Lavorel, S., Storkey, J., Bardgett, R.D., de Bello, F., Berg, M.P., 2013. Partitioning phylogenetic and functional diversity into Le Roux, X., Moretti, M., Mulder, C., Pakeman, R.J., D ıaz, alpha and beta components along an environmental gradi- S., & Harrington, R. 2013. A novel framework for linking ent in a Mediterranean rangeland. Journal of Vegetation Science functional diversity of plants with other trophic levels for the 24: 877–889. quantification of ecosystem services. Journal of Vegetation Sci- Cadotte, M.W., Carscadden, K., & Mirotchnick, N. 2011. Beyond ence 24: 942–948. species: functional diversity and the maintenance of ecologi- Mason, N.W.H. & Pavoine, S. 2013. Does trait conservatism cal processes and services. Journal of Applied Ecology 48: guarantee that indicators of phylogenetic community struc- 1079–1087. ture will reveal niche-based assembly processes along stress Carboni, M., Acosta, A.T.R., & Ricotta, C. 2013. Are differences gradients? Journal of Vegetation Science 24: 820–833. in functional diversity among plant communities on Medi- Mason, N.W.H., Mouillot, D., Lee, W.G., & Wilson, J.B. 2005. terranean coastal dunes driven by their phylogenetic his- Functional richness, functional evenness and functional tory? Journal of Vegetation Science 24: 932–941. divergence: the primary components of functional diversity. Chalmandrier, L., Munkem€ uller,€ T., Gallien, L., de Bello, F., Ma- Oikos 111: 112–118. zel, F., Lavergne, S., & Thuiller, W. 2013. A family of null Mason, N.W.H., de Bello, F., Dray, S., Mouillot, D., & Pavo- models to distinguish between environmental filtering and ine, S. 2013. A guide for using functional diversity indi- biotic interactions in functional diversity patterns. Journal of ces to reveal changes in assembly processes along Vegetation Science 24: 853–864. ecological gradients. Journal of Vegetation Science 24: 794– Cornwell, W.K., Schwilk, D.W., & Ackerly, D.D. 2006. A trait- 806. based test for habitat filtering: convex hull volume. Ecology Moretti, M., de Bello, F., Ibanez, S., Fontana, S., Pezzatti, G., Dzi- 87: 1465–1471. ock, F., Rixen, C., & Lavorel, S. 2013. Linking traits between Gerhold, P., Price, J.N., Pussa,€ K., Kalamees, R., Aher, K., Kaasik, plants and invertebrate herbivores to track functional effects A., & Partel,€ M. 2013. Functional and phylogenetic commu- of land-use changes. Journal of Vegetation Science 24: 949–962. nity assembly linked to changes in species diversity in a long- Pakeman, R.J. & Eastwood, A. 2013. Shifts in functional traits term resource manipulation experiment. Journal of Vegetation and functional diversity between vegetation and seed bank. Science 24: 843–864. Journal of Vegetation Science 24: 865–876. Hejda, M. & de Bello, F. 2013. Impact of plant invasions on func- Pavoine, S. & Bonsall, M.B. 2011. Measuring biodiversity to tional diversity in the vegetation of Central Europe. Journal explain community assembly: a unified approach. Biological of Vegetation Science 24: 890–897. Reviews 4: 792–812. Herben, T., Novakov a, Z., & Klime sova, J. 2013. Comparing Pavoine, S., Gasc, A., Bonsall, M.B., & Mason, N.W.H. 2013. functional diversity in traits and demography of Central Correlations between phylogenetic and functional diversity: European vegetation. Journal of Vegetation Science 24: 910– mathematical artefacts or true ecological and evolutionary 920. processes? Journal of Vegetation Science 24: 781–793. Hulshof, C.M., Violle, C., Spasojevic, M.J., McGill, B., Dams- Pillar, V.D., Blanco, C.C., Muller,€ S.C., Sosinski, E.E., Joner, F., & chen, E., Harrison, S., & Enquist, B.J. 2013. Intra-specific Duarte, L.D.S. 2013. Functional redundancy and stability in and inter-specific variation in specific leaf area reveal the plant communities. Journal of Vegetation Science 24: 963–974. importance of abiotic and biotic drivers of species diversity Whittaker, R.J. 1975. Communities and ecosystems. Macmillan, across elevation and latitude. Journal of Vegetation Science 24: New York, NY, US. 921–931.
Journal of Vegetation Science 780 Doi: 10.1111/jvs.12097 © 2013 International Association for Vegetation Science Journal of Vegetation Science 24 (2013) 781–793 SPECIAL FEATURE: FUNCTIONAL DIVERSITY Correlations between phylogenetic and functional diversity: mathematical artefacts or true ecological and evolutionary processes? Sandrine Pavoine, Amandine Gasc, Michael B. Bonsall & Norman W.H. Mason
Keywords Abstract Assembly processes; Biodiversity; Diversity indices; Environmental filtering; Niche Questions: Is phylogenetic diversity (PD) an accurate surrogate for functional complementarity; Null model; Phylogenetic diversity (FD)? How are FD:PD correlations affected by the diversity index used, signal; Simpson index; Species richness; covarying factors and/or the strength of the phylogenetic signal in ecological Surrogates traits?
Abbreviations Location: Field study, coastal marsh plain Mekhada, Algeria, complemented by FD = Functional diversity; PD = Phylogenetic simulated data. diversity Methods: FD and PD indices might correlate simply because variation in species Received 31 March 2012 richness and evenness (referred to as co-factors) influences both FD and PD val- Accepted 4 January 2013 ues. We partition FD and PD indices into components influenced by species rich- Co-ordinating Editor: Andreas Prinzing ness, evenness and species’ (functional and phylogenetic) characteristics. When a simple partition was not found, comparison to null models was used to remove Pavoine, S. (corresponding author, the effects of co-factors. We examined correlations between ten FD and PD indi- [email protected]): Gasc, A. ([email protected]): ces, among which several were shown to be connected using our mathematical Museum national d’Histoire naturelle, partitioning approach and several were transformed by comparison with null Departement d’Ecologie et Gestion de la models to control for effects of co-factors. In doing this, FD values were calcu- biodiversite, UMR 7204 CNRS-UPMC, 61 rue lated using simulated trait values with varying phylogenetic signal. We then Buffon, 75005, Paris, France Pavoine, S. & Bonsall, M.B. (michael.bonsall selected a subset of complementary FD and PD indices in exploring the influence @zoo.ox.ac.uk): University of Oxford, of environmental variables on diversity across 75 plant assemblages in Mekh- Department of Zoology, Mathematical Ecology ada. Research Group, South Parks Road, Oxford, OX1 3PS, UK Results: Altogether, mathematical partitioning and the comparison to null Gasc, A.:Museum national d’Histoire models successfully removed the effects of co-factors when comparing FD and naturelle, Departement Systematique et PD. For all indices affected by species richness, FD:PD correlations approached 1, Evolution, UMR 7205 CNRS, 43 rue Buffon, irrespective of the trait evolution model used. In contrast, simulations showed 75005, Paris, France that FD:PD correlations measured with indices unaffected by co-factors Bonsall, M.B.: St. Peter’s College, New Inn decreased when the phylogenetic signal in traits decreased. Applied to plant Hall Street, Oxford, OX1 2DL, UK assemblages in Mekhada, complementary diversity indices showed that, despite Mason, N.W.H. (masonn@landcareresearch. co.nz): Landcare Research, Private Bag, 3127, significant (but moderate) FD:PD correlation, FD but not PD was significantly Hamilton, New Zealand correlated with the main stress gradient (salinity). Conclusions: From both our simulations and analysis of plant community diversity, PD was a poor surrogate for FD. In Mekhada, PD was also less corre- lated with environmental variables than FD. Species richness was found to be a better surrogate for FD than PD in identifying the ecological processes that dis- tribute species along the salinity gradient.
ecosystem function and prioritize areas for conservation. Introduction These studies generally assume that PD is a surrogate for Phylogenetic diversity (PD) is increasingly used to under- functional diversity (FD), with closely related species being stand community assembly processes, predict rates of more similar in their functional traits than distantly related
Journal of Vegetation Science Doi: 10.1111/jvs.12051 © 2013 International Association for Vegetation Science 781 Correlating phylogenetic with functional diversity S. Pavoine et al. ones (Webb et al. 2002; Blomberg et al. 2003). Disparities simply correlating PD with FD is thus not sufficient; it is between FD and PD have also been used to infer that phy- imperative to find an association between PD and some logeny captures more ecologically relevant functional vari- aspect of the ecological process (e.g. along an environmen- ation than the limited number of traits used to estimate FD tal gradient as in our case study) or processes affecting the (Cadotte et al. 2009). However, there has been little criti- community organization, similar to the association cal assessment of whether PD can really act as a general between FD and the same aspect of the ecological process surrogate for FD. Testing this requires appropriate indices (es). For example, several studies in meadow grassland for examining correlation and association between FD and communities have found that increased FD was associated PD. This study explores the relationships between FD and with higher rates of productivity (Petchey et al. 2004). At PD indices. Using real and simulated data, we test whether a broader scale, the correlation between FD of forest tree PD is likely to be a useful surrogate for FD in community assemblages and plant productivity was found to be higher ecological studies. In doing so we aim to identify which in less productive, boreal environments (due to niche com- diversity indices have the power to discriminate relation- plementarity effects) compared to productive, temperate ships between FD and PD, and are not confounded by co- environments, where niche complementarity effects are varying factors. weakened by highly productive species occurring in monospecific stands (Paquette & Messier 2011). As might be expected, the choice of which functional traits to IS PD likely to be a realistic surrogate for FD? include in an analysis is the most important aspect in mea- Our starting point, as with many studies, relies on the suring FD (Petchey & Gaston 2002). For example, PD has hypothesis that (unmeasured) traits are evolutionary con- been found to be a better indicator of ecosystem function served. As shown in Webb et al. (2002) and Kraft et al. (e.g. primary productivity; Flynn et al. 2009) than FD, (2007), patterns in PD depend on whether critical traits are simply as a result of the difficulties associated with finding conserved or convergent. For PD to be a generally useful relevant traits with which to measure FD (Cadotte et al. surrogate of FD, the relationship between FD and PD thus 2009). needs to be robust against phylogenetically independent trait variation. Mismatch between FD and PD may be Identifying non-redundant indices for testing FD and explained by lack of resolution in phylogenies but also by PD correlations phenotypic plasticity in traits of distant species (Kluge & Kessler 2011). For many functional aspects of diversity, in In this study, we focus on the mathematical explanation a range of different phylogenies, phylogenetic distances that affects the relationships between PD and FD. Biodiver- are able to explain a significant amount of variation in trait sity is a multifaceted concept, which no single index can values across taxa (Blomberg et al. 2003). However, varia- hope to quantify in an easily interpretable way (Pavoine & tion in traits across species is the outcome of both environ- Bonsall 2011). Measures of both FD and PD are affected by mental and historical factors (Freckleton & Jetz 2009), so how species are distributed in functional or phylogenetic that the strength of the phylogenetic signal in trait varia- space and also how abundances vary among species. When tion can differ markedly according to the specific traits and the objective of a study is to identify ecological mecha- the phylogeny under investigation (Blomberg et al. 2003). nisms that explain species’ occurrences and abundances, a Indeed, cases where phylogenetic and trait variation are guiding principle in selecting interpretable FD and PD indi- very tightly linked may well be the exception rather than ces is that each index should quantify a single, indepen- the rule (Blomberg et al. 2003). Both conserved and con- dent aspect of biodiversity (Mason et al. 2005). vergent traits have been found to be involved in ecological Interpreting correlations between FD and PD requires indi- processes, such as environmental filtering and competition ces that uphold this principle. (Emerson & Gillepsie 2008). There are many biological rea- The FD and PD indices depend on three metrics: (1) spe- sons to expect no relationships between PD and FD, or cies’ identities; (2) species’ abundances; and (3) species’ even a negative relationship (Emerson & Gillepsie 2008). distribution in functional or phylogenetic space. In this One obvious explanation is that species may evolve while study, our first aim is to provide a definitive means of persisting within an assemblage, thus decreasing the rela- assessing whether indices proposed to measure FD and PD tionships between FD and PD (Kluge & Kessler 2011). are truly independent of diversity measures that depend Overall, trait divergence (Prinzing et al. 2008) and adap- only on species’ identities and abundances, such as species tive diversification (Fukami et al. 2007) can decrease the richness and evenness. Our approach seeks to mathemati- link between FD and PD. cally partition diversity indices into their basic compo- As the ultimate objective for the use of PD is to elucidate nents. This partitioning provides a more definitive the ecological mechanisms that structure communities, approach to assessing whether an index is truly
Journal of Vegetation Science 782 Doi: 10.1111/jvs.12051 © 2013 International Association for Vegetation Science S. Pavoine et al. Correlating phylogenetic with functional diversity independent from, or shares some redundancies with, might arise simply due to some artefact (like variation in other indices. species richness), rather than a real ecological process.
Using null models to remove redundancy between Methods indices We have used real data to evaluate the relevance of statisti- Recently, conceptual arguments have been made for using cal approaches that compare FD and PD. When necessary, null models to explore (and hence remove) redundancy this real data set was compared with simulated data (see between FD or PD indices and existing diversity measures. FD and PD correlations in simulated data). The field study We have explored this approach for those indices for which is located at La Mafragh, Mekhada, (15 000 ha, 36°48′N, a simple mathematical partitioning has not been obtained. 008°00′E), a coastal marsh plain of low elevation (from 1 For example, for indices that are based only on species’ to 4 m a.s.l. for most of the area) in the east of Annaba in characteristics, comparison of observed FD or PD values to Algeria. This region is located in a sub-humid bioclimate those expected from null models that randomize species with warm winters (Emberger 1955). It is furrowed by riv- occurrences (Hardy 2008) have been used to control for the ers, and constitutes a basin filled by alluvial and colluvial trivial effects of species richness. Also, indices that are based deposits. The lowest parts are composed of large and small on species’ characteristics and abundances can be compared marshes. The study area was defined in de Belair (1981). It with a null model that randomizes the abundances across constitutes 10 000 ha within the coastal plain where 102 species within each community (Hardy 2008) to control for sites were defined on a regular grid (average size of the effect of the evenness of species’ abundances. sites = 100 ha). Five of these sites, with very high hetero- geneity, and 22 other sites (predominantly located in the central marsh) with three species or less, were excluded Aims and objectives from our analyses, leaving 75 sites. The plant abundance Our broad aim is to understand the association between data were collected in 1979 based on three releves per site FD and PD in communities. The ultimate objectives of this (separated by 200 m). Details on the sampling design can work are twofold. First, we aim to provide a test of whether be found in de Belair (1981) and Pavoine et al. (2011). (and under what conditions) PD is correlated with FD. Sec- A total of 56 plant species was observed. Nine biological ond, we aim to identify a set of appropriate indices for test- traits are available for the recorded plant species: (1) life ing correlations between FD and PD in real plant cycle (perennial, annual, biennial and/or seasonal), (2) communities, according to the following criteria: pollination (autogamous, entomogamous and/or anemog- amous), (3) median of the potential flowering period, (4) 1. Each index should quantify a single component of FD length of the flowering period, (5) maximum and (6) mini- or PD that is independent of species richness and evenness. mum height of adult individuals, (7) spikiness (absent, 2. Each index should provide good power to detect FD:PD sometimes present, always present), (8) succulent leaves correlations. (absent, sometimes present, always present), (9) hairy 3. Correlation levels between FD and PD should not be leaves (absent, sometimes present, always present). Life dependent on any mathematical artefact. cycle and flowering periods are important to determine We begin by decomposing FD and PD indices into sim- how plants can cope with harsh (xeric) seasonal condi- pler components to identify those that are truly indepen- tions. For instance, most dicots in La Mafragh were annual dent from species richness and evenness. We then use null or biennial, completing their life cycle before the onset of models to remove remaining redundancies between indi- the xeric season (Pavoine et al. 2011). Pollination mode ces and analyse whether this affects the strength of FD:PD may influence species occurrence along the main stress correlations. We simulate trait values to test whether the gradient if biotic pollen vectors are less abundant in high strength of the phylogenetic signal in trait variation affects stress, low productivity sites. Spikiness, succulent leaves FD:PD correlations. Finally, we examine FD:PD correla- and hairy leaves are morphological adaptations that aid tions across real plant assemblages in the Algerian coastal water conservation. Spikiness and hairy leaves also reflect marsh plain using a subset of identified non-overlapping solar radiation, which can be important for avoiding pho- FD and PD indices. Doing this, we intend to examine the toinhibition and photorespiration in high stress environ- potential for FD:PD correlations to reveal underlying eco- ments. Plant height is an indicator of light competition logical processes that structure species communities rather ability, and is usually associated with a K-selected (com- than simply providing a robust test of a range of different petitive) ecological strategy (Westoby et al. 2002). processes. This especially relates to the use of diversity par- A phylogenetic tree was established based on Webb & titioning measures to identify indices where correlations Donoghue (2005) for topology; further details are given in
Journal of Vegetation Science Doi: 10.1111/jvs.12051 © 2013 International Association for Vegetation Science 783 Correlating phylogenetic with functional diversity S. Pavoine et al.
Pavoine et al. (2011). As with any study in phylogenetic Tree size (TS) is the sum of all branch lengths in a phy- community ecology, our results might be affected by logeny (Faith 1992) or a functional dendrogram (Petchey uncertainties associated with the phylogenetic tree (which & Gaston 2002). FEve (Villeger et al. 2008) is an abun- is only an estimation of the unknown real tree). The phy- dance-weighted measure of functional regularity. If a sin- logenetic tree we utilize has been widely used in studies on gle quantitative trait is considered, FEve would have its phylogenetic community analyses (179 citations for Webb maximum value if all species’ values were regularly & Donoghue 2005 according to the ISI Web of Knowledge, spaced in the quantitative trait axis and if all species had August 2012). Using this phylogenetic tree we integrated similar abundances. We have also used this index when the most up-to-date estimates of branch lengths all species have equal abundances and named it FEvep. (Wikstrom€ et al. 2001; Hedges & Kumar 2009) (in contrast HED and EED (Cadotte et al. 2010) measure evenness in many previous studies on phylogenetic community diver- phylogenetic or functional distinctiveness (where distinc- sity have just used Wikstrom€ et al. 2001; or simply tiveness is high for species that are distant to all others in counted nodes ignoring branch lengths, and hence lack a phylogeny or a functional dendrogram). B (Shimatani comprehensive details on evolutionary history). The pedo- 2001) is a measure of the covariance between the abun- logical data were also collected in 1979 from each site (de dances of pairs of species and the functional or phyloge- Belair 1981). Eight soil variables were considered: Clay netic distances between them. We also examined four 2+ 1 (%), silt (%), sand (%), K2O(&), Mg (mEq 100 g ), indices that are very similar: FAD (Walker et al. 1999), Na+ (mEq 100 g 1), K+ (mEq 100 g 1), elevation (m). The which is the sum of pair-wise functional or phylogenetic whole data set is described in detail and available as Sup- distances between species; MFAD = FAD/(S 1) plementary Material in Pavoine et al. (2011). (Schmera et al. 2009), where S is the number of species; As we could not consider the myriad of diversity indices meanD, which is the mean of pair-wise functional or developed so far, we analysed the diversity of the 75 plant phylogenetic distances between species (Weiher et al. = assemblages with the richnessP (S number of species), the 1998); and QE (Rao 1982), which is the mean of pair- – ¼ S 2 Gini Simpson index (G 1 i¼1 pi ,wherepi is the rela- wise functional or phylogenetic distances between tive abundance of a species) and ten distinct diversity indi- species, where species are weighted by their relative ces that represented different aspects of PD and FD. These abundances. indices were originally developed to measure FD, PD or both. As shown in Pavoine & Bonsall (2011), all these indi- Using mathematical partitioning to identify redundancy ces of diversity can be adapted to describe either FD or PD. among diversity measures Several of the ten diversity indices are known to depend The indices on only a single aspect of diversity: meanD, which mea- The ten diversity indices we studied include information sures the functional/phylogenetic divergence among spe- on species richness, species’ abundances and characteris- cies; B, which measures the covariance between species’ tics. Although the indices are based on similar assump- functional/phylogenetic distances and abundances; EED, tions, they differ in which aspect of FD and PD they which measures evenness in species’ functional/phyloge- describe. Basically, when applied to FD each of these indi- netic distinctiveness. The other more integrative indices ces is defined according to one of the following three crite- might depend on the effects of species number, evenness ria: functional distances among species (obtained through in the species abundance, species’ functional and phyloge- an extension of Gower distance; Gower 1971; Electronic netic characteristics and the interaction between these Appendix S1); a functional space (obtained by principal components. We have partitioned these more integrative coordinates analysis applied to functional distances; Gower diversity indices into simple indices, i.e. into indices that 1966); a functional dendrogram (obtained by an depend on a single aspect of diversity. We did this by iden- unweighted pair group method based on the arithmetic tifying, in the equations of these indices, terms that depend mean, UPGMA, algorithm applied to functional distances; only on species richness (and not on species’ abundances Petchey & Gaston 2002). Similarly, when applied to PD, or characteristics), terms that depend only on species even- each of the diversity indices is based on one of the follow- ness, terms that depend only on species’ characteristics and ing criteria: phylogenetic distances (sum of branch lengths terms associated with an interaction between species’ on the shortest path that connects two species on the phy- abundances and characteristics (if, for instance, some char- logenetic tree); a phylogenetic space (obtained by principal acteristics are common in more abundant species in a com- coordinates analysis applied to phylogenetic distances; munity while other characteristics are retained in rare Gower 1966); the phylogenetic tree. A complete descrip- species). Partitioning of indices reveals redundancy tion of the indices is given in Appendix S2. between indices.
Journal of Vegetation Science 784 Doi: 10.1111/jvs.12051 © 2013 International Association for Vegetation Science S. Pavoine et al. Correlating phylogenetic with functional diversity
accelerating rates of evolution (Blomberg et al. 2003). Using null models to provide complementary indices With the height of the phylogenetic tree standardized to We also evaluated the ability of null models (Gotelli & equal one, and the variance–covariance matrix of species’ Graves 1996) to remove the confounding effects of species trait values equal to r²C(h,g), where h is the value at the richness and evenness for estimating correlations of FD root node and g is the rate of trait change, we used the fol- and PD. This approach, based on a transformation of each lowing parameter values for the AC model: r ²=1, h = 0, diversity index, was applied on both simulated and real g = exp( 2), g = exp( 10), g = exp( 20); phylogenetic data. A transformed index was defined as standardized signal increases with g,wheng tends towards 0 the model effect size (SES) (Gotelli & McCabe 2002), such that tends towards BM (see Blomberg et al. 2003 for details).
SES = (obsX meanX)/SDX where the subscript (X) We also used completely random traits drawn from a stan- denotes the original diversity index, obsX is the observed dard normal distribution with mean 0 and variance 1. We value of the index in one of the 75 plant assemblages, and simulated species pools with nine quantitative traits based meanX and SDX are the mean and SD from an underlying on one of the evolutionary models (we chose to simulate null model. nine traits as our real data set contained nine traits). Spe- We generate two different null models. A pool of species cies abundances for species pools were derived from the was defined by the 56 plant species observed in the study real abundances of plants within the 75 plant assemblages area. For the first null model, random assemblages were from Mekhada. The real phylogenetic tree associated with determined with the same number and abundance of spe- plant assemblages from Mekhada was used to describe cies as that observed in the real assemblages. However, the evolutionary relationships between simulated species. identities of the species were replaced with the identity of For diversity indices that required functional distances species randomly drawn from the species pool (model 1p among species, we used the squared Euclidean metric in Hardy 2008). This null model supposes that the presence because, by definition, it is expected to vary linearly with and abundance of species within assemblages is indepen- phylogenetic distances under the Brownian motion model dent of their traits and phylogeny. The second null model (e.g. Harmon & Glor 2010). We simulated 1000 species applies only to diversity indices that include species’ abun- pools per evolutionary model for the raw diversity indices, dances. It assumes that species have random abundances but only 50 species pools under the SES transformations within an assemblage. Using each real assemblage, random (due to computational limits). Each SES transformation assemblages were obtained by permuting abundances was based on 200 theoretical values per null model. among present species (model 1s in Hardy 2008). SES1 will For each simulated data set, we applied all indices to define the SES transformation associated with the first null evaluate PD and simulated FD within the 75 plant assem- model and SES2 the SES transformation associated with blages; we then computed Spearman rank correlation the second null model. between FD and PD with each of the diversity indices. This For all indices that do not include species’ abundances, a non-parametric measure of correlation is more appropriate transformation by SES1 is expected to provide new indices than Pearson product-moment correlation as the high free from variations in species richness. For all other indi- number of simulations prevented us from analysing the ces (QE, B, FEve), the new indices obtained by SES1 and effects of distribution shape and Pearson correlation is sen- SES2 transformations are expected to depend on interac- sitive to extreme values. tions between abundances and species’ characteristics only. All indices and the transformations used are summa- The FD and PD correlations in real data rized in Table 1. We estimated the phylogenetic signal in functional trait variation with the Mantel test since it allows us to analyse The FD and PD correlations in simulated data the strength of the phylogenetic signal across all nine traits We analysed the impact of phylogenetic signal strength on simultaneously. Based on our mathematical partitioning the correlation between FD and PD. We assumed that the (see Using mathematical partitioning to identify redun- same index is used for measuring FD and PD (Appendix dancy among diversity measures) and simulations (see the S3). The FD:PD correlation analyses were done for the ten FD and PD correlations in simulated data), we selected a raw diversity indices, the indices transformed by SES1 and subset of four diversity indices in addition to species rich- the indices transformed by SES2. ness and evenness. The SES transformation was applied For this analysis, we simulated traits with various phylo- with 1000 theoretical values per null model and assem- genetic signals. To do this, we applied the Brownian blage. We analysed the impact of the choice of the diversity motion model (BM) and the accelerating (AC) model of index on the FD:PD Spearman correlation. We then analy- trait evolution to the phylogenetic tree with different sed the associations between FD, PD and environmental
Journal of Vegetation Science Doi: 10.1111/jvs.12051 © 2013 International Association for Vegetation Science 785 Correlating phylogenetic with functional diversity S. Pavoine et al.
Table 1. Summary of the indices used. Details on each index can be found in Appendix S2.
Short definitions Abbreviations References Dependence on
Species Species’ Associated richness abundances simpler indices1
Sum of differences among species FAD Walker et al. (1999) Yes No meanD Sum of differences among MFAD Schmera et al. (2009) Yes No meanD species divided by species richness Average difference between two species meanD Weiher et al. (1998) No No Covariance between species’ B Shimatani (2001) No Yes SES1B, SES2B characteristics and abundances Average difference between two QE Rao (1982) Yes Yes meanD, B, SES1B, individuals SES2B, SES1QE Sum of branch lengths on a TS Faith (1992); Yes No SES1TS2 phylogenetic, functional tree Petchey & Gaston (2002) Diversity in species distinctiveness HED Cadotte et al. (2010) Yes No EED, SES1EED Evenness in species distinctiveness EED Cadotte et al. (2010) No No SES1EED Evenness in differences among individuals FEve Villeger et al. (2008) No3 Yes SES1FEve2, SES2FEve2 Evenness in differences among species FEvep Villeger et al. (2008) No3 No SES1FEvep2 1By simple index we mean that unaffected by species richness and/or species’ abundances. 2As in the main text, SES1 means standardized effect size (SES) transformation by the first null model, and SES2 means SES transformation by the second null model. These transformations are expected to remove the dependence on species richness and to control the dependence on species evenness. 3The ranges of these indices do not depend on species richness; however, the overall impact of species richness in the values of these indices has not been determined by our study. We thus relied on null models to remove any potential effects of species richness. variables by linear additive regressions (nominal type I the estimated FD:PD correlations for both MFAD and FAD error a = 5%) to evaluate whether PD can be used as a can still approach 1 due to the effects of species richness in surrogate for FD. All calculations were completed in R (R MFAD and FAD. In fact, FD:PD correlations are mostly dri- Foundation for Statistical Computing, Vienna, AT, AU) ven by species richness if the coefficient of variation (CV) with the packages listed in Appendix S4. of species richness is much higher than the CV for func- tional and phylogenetic meanD (proof in Appendix S5-A). Results In this case, comparing functional and phylogenetic MFAD Mathematical partitioning or FAD is like comparing species richness with itself, which, obviously, is rather meaningless. The FD:PD correlations for FAD, MFAD and HED can arise The HED is also linked to EED by a function of species simply due to variation in species richness richness: HED = ln(S) EED. Partitioning HED thus shows it Mathematical analysis allows a better understanding of is influenced by species richness, and as for FAD and how species richness influences the correlation between MFAD, FD:PD correlations for HED approach 1 when vari- FD:PD for MFAD and FAD. Because MFAD = (S 1) ation in HED is mostly driven by variation in species rich- meanD, and FAD = S(S 1) meanD, where S is the species ness. richness within an assemblage, the correlations between FD:PD for MFAD and FAD depends on (1) the mean and The TS: a complex function of species richness variance of species richness within assemblages, (2) the mean and variance of meanD for functional traits and phy- The last index based on species’ presence/absence that we logeny, (3) the pair-wise covariances (normed by means) studied, TS, is known to be highly dependent on species between species richness, the phylogenetic meanD and the richness (Petchey & Gaston 2002). Several studies have functional meanD (proof in Appendix S5-A). When com- explored the link between TS and species richness by per- paring FD and PD with MFAD and FAD, it is thus difficult forming simulations with a regularly increasing number of to unravel whether the observed correlation is due to spe- species randomly drawn from a species pool. With these cies richness only or to similarities in the functional and simulations, Petchey & Gaston (2002) found that TS phylogenetic characteristics of the species. To answer this increases roughly linearly with species richness for trees question, MFAD and FAD have to be partitioned into spe- with many nodes close to the root; and that, in contrast, cies richness and meanD. TS first increases drastically and then saturates with Even if the pair-wise correlations between species rich- increasing species richness for trees with many nodes close ness, FD and PD (as measured by meanD) all equal zero, to tips.
Journal of Vegetation Science 786 Doi: 10.1111/jvs.12051 © 2013 International Association for Vegetation Science S. Pavoine et al. Correlating phylogenetic with functional diversity
When species are randomly drawn from a pool, the values is expected only when species richness is low. To connection between TS and species richness depends on understand how and why each component of diversity tree shape. Consider a pool of S species, a tree (functional drives the FD:PD correlation, one has to evaluate the dendrogram or phylogeny) composed of K branches effects of species richness, species evenness, meanD and B with lengths lk, k=1, …,K.Letdk be the number of tips on the strength of the correlation. that descend from branch k. On average, the expected value for TS in an assemblage of n species randomly drawn Interpreting FEve and FEvep from the species pool is (proof in Appendix S5-B): Null models are necessary to analyse FEve and FEvep as TSðÞn there is no simple way of extracting the contributions of " #, S XK X d X S d S species richness, species’ abundances and functional or ¼ k k lk lk lk phylogenetic characteristics from the mathematical n k¼1 ksuchthat n ksuchthat n n dk n S dk n expressions for these diversity indices. For FEve and FEvep a minimum spanning tree must be built but it is dependent TS is thus a function of polygons of S and n of various on which species are in the assemblages. Although the degrees, which depends on the shape of the tree consid- range (0,1) of these indices is not dependent on species ered. We need a null model to remove trivial effects of spe- richness, we used the SES transformations to verify cies richness. whether co-factors could affect their values.
The QE: a function of G, meanD and B indices The FD and PD correlations in simulated data For quadratic entropy (QE), Shimatani (2001) showed that Although EED is, in theory, not expected to be affected by = 9 + – QE meanD G B,whereG is the Gini Simpson species richness, we found in our simulations unexpected index, which depends on species richness (S) and species’ P moderate correlations between functional and phyloge- ¼ S 2 abundances (pi), G 1 i¼1 pi ; B is the covariance- netic EED even when traits were simulated independently like measure between the abundances of pairs of species of the phylogeny. We thus applied the SES1 transforma- ¼ (pipj) and the distances between them (dij) B 2 tion to EED. Under model 1, FD:PD correlations obtained P G with SES1HED and SES1EED are equal (Appendix S5-D). dij meanD pipj SSðÞ 1 and meanD is the average i>j Henceforth, we considered only SES1EED. FD:PD correla- distance among species tions obtained with SES1FAD, SES1MFAD, SES1meanD and simply meanD are identical (Appendix S5-D). Hence- XS XS ¼ 1 forth, we only considered meanD. We applied the SES meanD ð Þ dij S S 1 i¼1 j¼1 transformations to all indices that depended on species’ abundances because they could all potentially be affected QE is thus dependent on the main effects of both species’ by the shape of distribution of species’ abundances within abundances (G) and species characteristics (meanD), and the species pool and the communities. SES1 was thus on the interactions between species’ abundances and char- applied to the following indices: B, EED, FEve, FEvep, QE acteristics (B). and TS. Under null model 2, SES2QE = SES2B (Appendix The FD:PD correlation for QE depends on the pair-wise S5-D), SES2 was thus applied to the following indices that covariance between G, functional and phylogenetic me- depend on species’ abundances: B,FEve. anD, and functional and phylogenetic B (Appendix S5-C). Indices confounded by species richness gave high corre- It can also be shown that, when there is no correlation lations even when the phylogenetic signal in trait data was between functional and phylogenetic distances between weak. Many of the indices (HED, FAD, MFAD, TS, QE and species, a FD:PD correlation for QE approaching 1 can only to a lesser extent EED and FEve) we have considered are occur due to the effects of species abundances (Appendix confounded and were found to be influenced by co-factors S5-C). To extend this further, G can also be decomposed (Fig. 1, Appendix S6). The indices B, FEvep and meanD into the species richness (number of species, S) and an seemed unaffected by co-factors. Given that when using index of species evenness independent of species richness: null models, correlations are strongly influenced by the X strength of the phylogenetic signal, null models correctly S S E ¼ 1 p2 remove the influence of the co-factors on the FD:PD corre- S 1 i¼1 i lation. However, for many indices FD:PD correlations were Given that the component of QE dependent on species moderate or low even when traits followed a Brownian richness is (S 1)/S, the impact of species richness on its model of evolution (i.e. when the phylogenetic signal was
Journal of Vegetation Science Doi: 10.1111/jvs.12051 © 2013 International Association for Vegetation Science 787 Correlating phylogenetic with functional diversity S. Pavoine et al.
Fig. 1. Bar plots of the average Spearman correlation between FD and PD obtained when simulating traits according to the Brownian model (Blomberg et al. 2003), AC model (with varying values for parameter g), and random normal distribution. “SES1” is written before the name of indices transformed by method SES1 based on null model 1, and “SES2” before the name of indices transformed by method SES2 based on null model 2 (see Methods). strong) and strongly decreased with increasing accelerating (FEvep, df=73, P = 0.202), 0.398 (B, P < 0.001), 0.465 rate of evolution in the model AC (Fig. 1). All indices (SES1TS, P < 0.001), to 0.526 (meanD, P < 0.001). transformed by null model 2, where abundances are Despite these correlations, linear additive regression permuted among species within assemblages, were associ- revealed that FD was more correlated with environmental ated with low FD:PD correlations even when traits fol- variables than PD. For FD, we found that all indices except lowed a Brownian model of evolution (Fig. 1). This might B were significantly correlated with the environmental be due to a lack of abundance variability in our data set variables: SES1TS (r²=0.410, P < 0.001), FEvep (r²=0.373, (phytosociological estimations). P < 0.001), meanD (r²=0.250, P = 0.011), B (r²=0.030, Based on this analysis, we selected the following indices, P = 0.978). In contrast, for PD, none of the indices were which collectively quantify each of the primary compo- significantly correlated with the environmental variables: nents, richness, evenness and divergence, as identified by SES1TS (r²=0.184, P = 0.082), meanD (r²=0.169, Mason et al. (2005), to analyse our real plant community P = 0.122), B (r²=0.159, P = 0.153), FEvep (r²=0.011, diversity data set: P = 0.373). Species richness (log-transformed, r²=0.218, P = 0.030), but not species evenness (r²=0.179, P = 0.093), 1. SES1TS and meanD, which both describe the size of was correlated with the environmental gradient. Overall, functional or phylogenetic space occupied by the species of in our case study, PD was thus a poor surrogate for FD. an assemblage (i.e. functional or phylogenetic richness), Species richness was a better surrogate. Log-transformed but SES1TS is more adapted to the tree-shape structure of species richness was correlated with FD (correlations, phylogenetic data and meanD is more adapted to the anal- df = 73, FEvep, r = 0.314, P = 0.006; meanD, r = 0.413, ysis of a table of functional traits which does not have an P < 0.001; SES1TS, r = 0.437, P < 0.001), but not with PD intrinsic tree structure. (FEvep, r = 0.006, P = 0.958; meanD, r = 0.222, 2. FEvep, which describes how regularly spaced species’ P = 0.056; SES1TS, r = 0.100, P = 0.392). A complemen- traits or phylogenetic positions are in a multivariate space tary analysis demonstrated that FD diversity indices (i.e. a special case of functional or phylogenetic evenness SES1TS, FEvep and meanD increased with increasing ele- where all species have the same abundance); vation and with decreasing salinity (Appendix S7). 3. B, which determines whether the most functionally or phylogenetically distant species have the highest abun- dances (i.e. functional or phylogenetic divergence). Discussion
Species richness and evenness completed the list of Throughout this paper we searched for interpretable FD diversity indices. and PD indices that could help tease apart the factors influ- encing the correlations between FD, PD and environmen- tal gradients. All indices confounded by species richness The FD and PD correlations in real data provided high FD:PD correlation even for traits that We found a significant phylogenetic signal in the func- evolved independently of the phylogeny. Mathematical tional distances among species (Mantel test, r = 0.310, partitioning and the comparison to null models led to a set P = 0.001, 1000 permutations). FD:PD correlations of indices independent of species richness and where the depended on the index selected and varied from 0.149 effects of species’ abundances are controlled. We selected
Journal of Vegetation Science 788 Doi: 10.1111/jvs.12051 © 2013 International Association for Vegetation Science S. Pavoine et al. Correlating phylogenetic with functional diversity four of these indices and applied them to the plant com- signal. Further, phylogenetically distant species often co- munity structure on the Mekhada coastal marsh plain. occurred in the most stressful conditions, with salt-tolerant Below we discuss the results of our case study; highlight monocots (i.e. Juncaceae and Cyperaceae) and halophyte di- the problems our analyses exposed in comparing FD and cots (i.e. Amaranthaceae) both common in highly saline PD; discuss the ability of our partitioning and null models environments (Pavoine et al. 2011). The fact that PD is a to avoid these problems; and conclude by considering the poor surrogate of FD, when unbiased measures or null implications of our results for the usefulness of PD as a sur- models are used, is not due to a lack of quality of the phylo- rogate for FD. genetic tree. It is due to a convergence in traits between di- cots (Amaranthaceae) and monocots (Juncaceae, Cyperaceae). The separation between these two clades is acknowledged Salinity gradient in Mekhada marsh plain in all current phylogenies. The lack of association between In Mekhada, the main environmental gradient is a salinity PD and the salinity gradient is thus, at least partly, due to gradient, with low elevation areas being on average more similarities in the traits of distant clades with different bi- saline (Pavoine et al. 2011). We analysed whether environ- ogeographic origins. Both the salinity gradient and the mental variables are similarly correlated with both FD and phylogeny are associated with trait values but are indepen- PD. Plant abundances were not influenced by functional dent of one another. traits or phylogeny. However, for indices based on species occurrences, FD significantly decreased with increasing Issues raised when comparing FD and PD salinity. This was found even though the functional traits available for this data set are expected to be less directly A simple error when comparing FD and PD would be to affected by the environmental gradient than unavailable think that there is only a single way to measure diversity. physiological traits (Pavoine et al. 2011). We found moder- In our simulations, we obtained very different FD:PD cor- ate FD:PD correlations; and while relationships between PD relations according to the diversity index used. This was and the primary stress gradient (salinity) in our study area partly due to co-factors affecting diversity values and also were in the same direction as for FD, they were non-signifi- partly due to the infinite possible ways of measuring diver- cant. Even species richness was a better surrogate for the sity even with the same data and the same objective. connections between FD and the salinity gradient. Over the last 30 years, a huge range of diversity indices The correlation between FD and species richness is an has been developed (Schweiger et al. 2008; Pavoine & ecological result rather than a mathematical artefact. Plots Bonsall 2011). When analysing their data, ecologists are with low species richness in Mekhada are located in thus often confronted with the difficult choice of which flooded, salty areas where only species with particular trait indices are best for their circumstances (data, objective combinations can establish (low meanD and SES1TS val- etc.; Ricotta 2005). A typical reaction is to apply the most ues, indicating reduced functional richness). In contrast, widely used index. However, different disciplines seem to areas with higher species richness are located at higher ele- be biased towards using different indices. One can thus be vations where silt and sand add to clay to provide greater tempted to measure FD with the most widely used index heterogeneity of soil types and where more species with in functional community ecology, and to measure PD with diverse traits can establish (high FEvep values, which may the most widely used index in phylogenetic community be associated with increased influence of niche comple- ecology. However, there should be good biological argu- mentarity). In the flooded, salty areas, all but the salt-toler- ments for doing this, since results produced using two dif- ant and halophyte species are filtered out, which causes ferent indices to measure FD and PD are hardly functional clustering in plant assemblages (where co-exist- interpretable (Appendix S3). The first basic rule when ing species have similar traits). In these locations, species comparing FD and PD is thus to use the same index of were perennial, often with spiky structures and glabrous diversity for the two aspects. We followed this rule leaves, and they were mostly anemogamous. Indeed the throughout our study. most salty areas are likely to be unfavourable habitats for In this choice for an index of diversity, it should be kept pollinators due to the high level of disturbance through in mind that interpreting diversity indices is difficult when regular flooding (Pavoine et al. 2011). More generally, on different aspects influence diversity. Simple mathematical a stress gradient, functional diversity and species richness partitioning of diversity indices allowed us to separate the can be correlated if only a few species with similar traits effects of species richness, species’ abundances and func- can establish in the harshest areas (following the “physio- tional or phylogenetic characteristics. Knowing these rela- logical tolerance” hypothesis of Mason et al. 2008). tive effects is important as, for instance, we showed that Patterns of PD were found to be less clear, which can be multiplying meanD by the species richness (MFAD) or the explained by the moderate, instead of strong, phylogenetic squared species richness (FAD) considerably increases FD:
Journal of Vegetation Science Doi: 10.1111/jvs.12051 © 2013 International Association for Vegetation Science 789 Correlating phylogenetic with functional diversity S. Pavoine et al.
PD correlations. A common error when measuring FD and choose (Hardy 2008). To avoid unnecessary complications, PD is to assume that we are only measuring the functional the SES approach should thus be used exclusively when and phylogenetic characteristics of the species. However, mathematical partitioning is not possible. For instance, the relative influence of species richness and/or species’ mathematical partitioning revealed that the complemen- abundances on FD and PD in comparison with species’ tary index meanD could be extracted from both FAD and functional and phylogenetic characteristics depends on the QE. It also showed that the SES1 approach is useless with index used. Choosing an index of FD and PD is thus adopt- FAD, MFAD and meanD, since it provides similar results to ing a very particular view of biodiversity. those observed using meanD. We have seen that disentangling effects of species’ char- acteristics, species’ abundances and species richness, Removing co-factors to disentangle ecological through partitioning or using null models, is essential mechanisms when correlating FD and PD to avoid misleading and exag- The FD:PD correlations should reflect the degree to which gerated correlations due to co-factors (species richness and phylogenetic relatedness corresponds with functional simi- evenness) that are independent of species’ characteristics. larity. The fact that indices confounded by species richness Even if only FD or only PD is considered, this approach to give high FD:PD correlations even when the phylogenetic disentangling multiple contributions to assessment of signal in trait data is weak is undesirable when the aim is diversity might be also very informative. For example, to reveal assembly mechanisms. Consequently, we require Walker et al. (1999) analysed graminoid species in a lightly methods to remove these co-factors when analysing the grazed site and in a heavily grazed site in an Australian correlations between FD and PD and when analysing the rangeland. They found that within the lightly grazed grass- correlations between either FD or PD and environmental land, dominant species were functionally more dissimilar factors (e.g. Mouillot et al. 2011). to one another, and functionally similar species more sepa- An often-reported strategy to remove the influence of rated in abundance rank, than expected on the basis of co-factors consists of regressing FD (and/or PD) on the co- average ecological distances in the community (index me- factors and working on functional (or phylogenetic) resid- anD). They found that this functional redundancy uals. However, our results (especially for index TS) high- between minor and dominant species led to increased light that we need to know how FD and PD are linked via resilience in biomass production, with minor species co-factors before deciding how to remove these co-factors. increasing in abundance when abundance of the domi- To analyse variation in TS relative to species richness, some nant species was reduced by intense grazing. This study studies have, for instance, used species richness and/or its shows that removing the influence of covarying factors quadratic term as an explanatory variable in a linear from measures of FD or PD can aid in revealing ecological regression and have analysed the residuals of the regres- processes. More generally, Table 2 in Pavoine & Bonsall sion. The equation that links TS with species richness (2011) demonstrated how comparing patterns in FD/PD showed that TS does not have a linear relationship with and patterns in species richness or evenness can help to species richness or its quadratic term. tease apart mechanisms of community assembly. We found that at least one index is, by definition of its mathematical formula, unaffected by co-factors: meanD. The PD is a poor surrogate for FD We can be confident that any observed correlation between this index and a co-factor in field or experimental A critical ecological question when comparing FD with PD studies would be due to ecological mechanisms rather than is whether similarities between phylogenetic distances and to a mathematical artefact. In contrast, measures of diver- functional distances among species translate into similari- sity based on indices such as MFAD, FAD, HED, TS and QE ties between the FD and PD of species assemblages. Using confound several components of diversity. Mathematical simulations, we found that FD:PD correlations are moder- partitioning is sufficient to decompose MFAD, FAD, HED ate or low, depending on the diversity index used, even and QE into simpler indices, necessary to decipher the with the Brownian model of trait evolution. The Brownian mechanisms that lead to observed diversity patterns. How- model assumes a correlation close to 1 between the ever, for other indices such as TS, null models are required. squared Euclidean functional distances and time-scaled In addition, any index using species’ abundances might be phylogenetic distances used here among species. A signifi- affected by the shape of the distribution of abundance cant phylogenetic signal in functions is thus not sufficient (skewed vs. even). Null models can also help to control for for ensuring high FD:PD correlation. Recent simulation these sorts of issue. studies corroborate this and have shown that this can affect The use of null models (SES approach), however, adds a conclusions about ecological processes. Mason et al. critical methodological decision – which null model to (2013) found that FD indices gave high power to detect
Journal of Vegetation Science 790 Doi: 10.1111/jvs.12051 © 2013 International Association for Vegetation Science S. Pavoine et al. Correlating phylogenetic with functional diversity changes in assembly processes along a hypothetical stress questions that were not solved by previous analyses that gradient, while PD indices gave only weak power for the focused on a single aspect of biodiversity. same assembly model (Mason & Pavoine 2013). Kraft et al. Analysing the relative contributions and pair-wise cor- (2007) found that the power of PD to detect community relations of species richness, species’ abundances and char- patterns is higher with environmental filtering, larger spe- acteristics allows identification of the mechanisms cies pool sizes and higher number of traits considered, and controlling species occurrences and abundances. This lower with competition. When communities were struc- approach that details FD and PD patterns can be usefully tured by competition, the power to detect phylogenetic applied at local scales to analyse community and ecosys- community structure was found to be high only if traits tem processes, and also at broader scales, to explore, for were more conserved than expected according to a Brown- example, variation in FD and PD along altitudinal and lati- ian model. Kraft et al. (2007) also found that this power tudinal gradients (Tallents et al. 2005). Finally, simple cal- depends on the index of PD used. The power of detection culation of correlation between FD and PD can be of phylogenetic community structure thus definitely complemented by methods that analyse the detailed func- depends on the evolution of traits, the diversity index used tional and phylogenetic compositions of assemblages in a and the ecological processes underpinning community more descriptive way (Pavoine et al. 2011). These analyses composition. can identify which traits and which clades are responsible A phylogenetic signal in traits at the species pool level for the levels of FD and PD and further our understanding is not sufficient to ensure a FD:PD correlation, as illus- of how ecological communities are structured and func- trated by our case study and simulations. Concrete exam- tion. ples are also given elsewhere, e.g. in Pillar & Duarte (2010). They measure a phylogenetic signal at the species Acknowledgements pool level and a phylogenetic signal at the metacommuni- ty level. The phylogenetic signal at the species pool level The authors thank the co-ordinating editor, Andreas Prin- corresponds simply to the degree of evolutionary conser- zing, the referee, Oliver Schweiger, and an anonymous vatism of trait differentiation in the species pool. The phy- referee for the constructive comments they have made on logenetic signal at the metacommunity level consists in our paper. comparing the phylogenetic structure of a community with its trait-based structure. Contrary to the phyloge- References netic signal at the species pool level, the phylogenetic sig- nal at the metacommunity level includes information on Blomberg, S.P., Garland, T. & Ives, A.R. 2003. Testing for phylo- occurrence or abundance of species within a metacom- genetic signal in comparative data: behavioral traits are more munity. Applying their approach to grassland communi- labile. Evolution 57: 717–745. ties in Brazil, Pillar & Duarte (2010) found that, among Cadotte, M.W., Cavender-Bares, J., Tilman, D. & Oakley, T.H. seven traits examined individually, four of them had a 2009. Using phylogenetic, functional and trait diversity to significant phylogenetic signal at species pool level but understand patterns of plant community productivity. PloS only two had significant phylogenetic signal at the meta- ONE 4: e5695. community level. Pillar & Duarte (2010) provided two Cadotte, M.W., Davies, T.J., Regetz, J., Kembel, S.W., Cleland, models of how phylogeny affects trait values: model E. & Oakley, T.H. 2010. Phylogenetic diversity metrics for 1 = T?P?E, where the correlation between traits and ecological communities: integrating species richness, abun- dance and evolutionary history. Ecology Letters 13: 96–105. environment is mediated by the phylogeny; and model de Belair, G. 1981. Biogeographie et amenagement: la plaine de 2 = P?T E, where both the environment and the phy- La Mafragh (Annaba, Algerie). PhD Thesis, UniversitePaul logeny are associated with trait values but independent of Valery, Montpellier, FR. one another. The latter model necessarily involves mod- Emberger, L. 1955. Une classification biogeographique des erate phylogenetic signal. Only under the former model, climats. Recueil des travaux des laboratoires de botanique, de g eolo- can we expect the relationship between PD and environ- gie et zoologie de la facult e des sciences de Montpellier 7: 3–43. ment to act as a surrogate for the relationship between Emerson, B.C. & Gillepsie, R.G. 2008. Phylogenetic analysis of FD and environment. community assembly and structure over space and time. Trends in Ecology & Evolution 23: 619–630. Conclusions and directions for future research Faith, D.P. 1992. Conservation evaluation and phylogenetic diversity. Biological Conservation 61: 1–10. Comparing two aspects of biodiversity (i.e. functional and Flynn, D.F.B., Gogol-Prokurat, M., Nogeire, T., Molinari, N., phylogenetic diversity), which share common components Richers, B.T., Lin, B.B., Simpson, N., Mayfield, M.M. & (species richness and species’ abundances), raise new DeClerck, F. 2009. Loss of functional diversity under land
Journal of Vegetation Science Doi: 10.1111/jvs.12051 © 2013 International Association for Vegetation Science 791 Correlating phylogenetic with functional diversity S. Pavoine et al.
use intensification across multiple taxa. Ecology Letters 12: J.A., Troussellier, M., Velez, L., Watson, R., Douzery, E.J.P. 22–33. & Mouquet, N. 2011. Protected and threatened components Freckleton, R.P. & Jetz, W. 2009. Space versus phylogeny: disen- of fish biodiversity in the Mediterranean Sea. Current Biology tangling phylogenetic and spatial signals in comparative 21: 1044–1050. data. Proceedings of the Royal Society of London Series B - Biologi- Paquette, A. & Messier, C. 2011. The effect of biodiversity on tree cal Sciences 276: 21–30. productivity: from temperate to boreal forests. Global Ecology Fukami, T., Beaumont, H.J.E., Zhang, X.-X. & Rainey, P.B. 2007. and Biogeography 20: 170–180. Immigration history controls diversification in experimental Pavoine, S. & Bonsall, M.B. 2011. Measuring biodiversity to adaptive radiation. Nature 446: 436–439. explain community assembly: a unified approach. Biological Gotelli, N.J. & Graves, G.R. 1996. Null models in ecology. Smithso- Reviews 86: 792–812. nian Institution Press, Washington, DC, US. Pavoine, S., Vela, E., Gachet, S., De Belair, G. & Bonsall, M.B. Gotelli, N.J. & McCabe, D.J. 2002. Species co-occurrence: a 2011. Linking patterns in phylogeny, traits, abiotic variables meta-analysis of J. M. Diamond’s assembly rules model. Ecol- and space: a novel approach to linking environmental filter- ogy 83: 2091–2096. ing and plant community assembly. Journal of Ecology 99: Gower, J.C. 1966. Some distance properties of latent root and 165–175. vector methods used in multivariate analysis. Biometrika 53: Petchey, O.L. & Gaston, K.J. 2002. Functional diversity (FD), 325–338. species richness and community composition. Ecology Letters Gower, J.C. 1971. A general coefficient of similarity and some of 5: 402–411. its properties. Biometrics 27: 857–874. Petchey, O.L., Hector, A. & Gaston, K.J. 2004. How do different Hardy, O.J. 2008. Testing the spatial phylogenetic structure of measures of functional diversity perform? Ecology 85: 847– local communities: statistical performances of different null 857. models and test statistics on a locally neutral community. Pickett, S.T.A. & Bazzaz, F.A. 1978. Organization of an assem- Journal of Ecology 96: 914–926. blage of early successional species on a soil moisture gradi- Harmon, L.J. & Glor, R.E. 2010. Poor statistical performance of ent. Ecology 59: 1248–1255. the Mantel test in phylogenetic comparative analyses. Evolu- Pillar, V.D. & Duarte, L.D.S. 2010. A framework for metacom- tion 64: 2173–2178. munity analysis of phylogenetic structure. Ecology Letters 13: Hedges, S.B. & Kumar, S., eds. 2009. The timetree of life,pp.551. 587–596. Oxford University Press, New York, NY, US. Prinzing, A., Reiffers, R., Braakhekke, W.G., Hennekens, S.M., Kluge, J. & Kessler, M. 2011. Phylogenetic diversity, trait diver- Tackenberg, O., Ozinga, W.A., Schaminee, J.H.J. & van sity and niches: species assembly of ferns along a tropical ele- Groenendael, J.M. 2008. Less lineages – more trait variation: vational gradient. Journal of Biogeography 38: 394–405. phylogenetically clustered plant communities are function- Kraft, N.J.B., Cornwell, W.K., Webb, C.O. & Ackerly, D.D. 2007. ally more diverse. Ecology Letters 11: 809–819. Trait evolution, community assembly, and the phylogenetic Rao, C.R. 1982. Diversity and dissimilarity coefficients: a unified structure of ecological communities. American Naturalist 170: approach. Theoretical Population Biology 21: 24–43. 271–283. Ricotta, C. 2005. Through the jungle of biological diversity. Acta Mason, N.W.H., de Bello, F., Mouillot, D., Pavoine, S., & Dray, Biotheoretica 53: 29–38. S., 2013. A guide for using functional diversity indices to Schmera, D., Eros, T. & Podani, J. 2009. A measure for assessing reveal changes in assembly processes along ecological gradi- functional diversity in ecological communities. Aquatic Ecol- ents. Journal of Vegetation Science24: 794–806. ogy 43: 157–167. Mason, N.W.H., Irz, P., Lanoiselee, C., Mouillot, D. & Argillier, Schweiger, O., Klotz, S., Durka, W. & Kuehn, I. 2008. A compar- C. 2008. Evidence that niche specialization explains species– ative test of phylogenetic diversity indices. Oecologia 157: 485 energy relationships in lake fish communities. Journal of Ani- –495. mal Ecology 77: 285–296. Shimatani, K. 2001. On the measurement of species diversity Mason, N.W.H., Mouillot, D., Lee, W.G. & Wilson, J.B. 2005. incorporating species differences. Oikos 93: 135–147. Functional richness, functional evenness and functional Tallents, L.A., Lovett, J.C., Hall, J.B. & Hamilton, A.C. 2005. Phy- divergence: the primary components of functional diversity. logenetic diversity of forest trees in the Usambara mountains Oikos 111: 112–118. of Tanzania: correlations with altitude. Botanical Journal of Mason, N.W.H. & Pavoine, S. 2013. Does trait conservatism the Linnean Society 149: 217–228. guarantee that indicators of phylogenetic community Villeger, S., Mason, N.W.H. & Mouillot, D. 2008. New mul- structure will reveal niche-based assembly processes tidimensional functional diversity indices for a multifac- along stress gradients?. Journal of Vegetation Science 24: 820 eted framework in functional ecology. Ecology 89: 2290– –833. 2301. Mouillot, D., Albouy, C., Guilhaumon, F., Ben Rais Lasram, F., Walker, B., Kinzig, A. & Langridge, J. 1999. Plant attribute diver- Coll, M., Devictor, V., Meynard, C.N., Pauly, D., Tomasini, sity, resilience, and ecosystem function: the nature and
Journal of Vegetation Science 792 Doi: 10.1111/jvs.12051 © 2013 International Association for Vegetation Science S. Pavoine et al. Correlating phylogenetic with functional diversity
significance of dominant and minor species. Ecosystems 2: Supporting Information 95–113. Webb, C.O., Ackerly, D.D., McPeek, M.A. & Donoghue, M.J. Additional supporting information may be found in the 2002. Phylogenies and community ecology. Annual Review of online version of this article: Ecology and Systematics 33: 475–505. Webb, C.O. & Donoghue, M.J. 2005. Phylomatic: tree assembly Appendix S1. Calculation of functional distances for applied phylogenetics. Molecular Ecology Notes 5: 181–183. among species. Weiher, E., Clarke, G.D.P. & Keddy, P.A. 1998. Community Appendix S2. Complete description of the diversity assembly rules, morphological dispersion, and the coexis- indices. tence of plant species. Oikos 81: 309–322. Appendix S3. Using different indices for measuring Westoby, M., Falster, D.S., Moles, A.T., Vesk, P.A. & Wright, I.J. FD and PD increases the risk of misinterpretation. 2002. Plant ecological strategies: some leading dimensions of Appendix S4. List of R packages used. variation between species. Annual Review of Ecology and Sys- Appendix S5. Mathematical details. – tematics 33: 125 159. Appendix S6. Average Spearman correlation € Wikstrom, N., Savolainen, V. & Chase, M.W. 2001. Evolution of between FD and PD obtained with trait simulations. the angiosperms: calibrating the family tree. Proceedings of the Appendix S7. Correlations between environmental Royal Society of London Series B – Biological Sciences 268: 2211– variables and diversity indices on the marsh plain Mekhada. 2220.
Journal of Vegetation Science Doi: 10.1111/jvs.12051 © 2013 International Association for Vegetation Science 793 Journal of Vegetation Science 24 (2013) 794–806 SPECIAL FEATURE: FUNCTIONAL DIVERSITY A guide for using functional diversity indices to reveal changes in assembly processes along ecological gradients Norman W.H. Mason, Francesco de Bello, David Mouillot, Sandrine Pavoine & Ste´ phane Dray
Keywords Abstract Co-existence; Environmental filtering; Functional divergence; Functional evenness; Question: Which functional diversity indices have the power to reveal changes in Functional richness; Functional trait; Limiting community assembly processes along abiotic stress gradients? Is their power affected similarity; Niche complementarity; Null models; by stochastic processes and variations in species richness along stress gradients? Species richness Methods: We used a simple community assembly model to explore the power Nomenclature of functional diversity indices across a wide range of ecological contexts. The NZ Plant Names Database http:// model assumes that with declining stress the influence of niche complementar- nzflora.landcareresearch.co.nz/ (accessed 5 ity on species fitness increases while that of environmental filtering decreases. August 2010) We separately incorporated two trait-independent stochastic processes – mass Received 29 March 2012 and priority effects – in simulating species occurrences and abundances along a Accepted 24 September 2012 hypothetical stress gradient. We ran simulations where species richness was Co-ordinating Editor: Martin Zobel constant along the gradient, or increased, decreased or varied randomly with declining stress. We compared observed values for two indices of functional rich- ness – total functional dendrogram length (FD) and convex hull volume (FRic) – Mason, N.W.H. (corresponding author, [email protected]): Landcare with a matrix-swap null model (yielding indices SESFD and SESFRic) to remove Research, Private Bag 3127, Hamilton 3240, any trivial effects of species richness. We also compared two indices that mea- New Zealand sure both functional richness and functional divergence – Rao quadratic entropy de Bello, F. ([email protected]): Institute of (Rao) and functional dispersion (FDis) – with a null model that randomizes Botany, Academy of Sciences of the Czech abundances across species but within communities. This converts them to pure Republic, Dukelska´ 135, CZ-379 82 Trˇebonˇ , measures of functional divergence (SESRao and SESFDis). Czech Republic Mouillot, D. ([email protected]): Results: When mass effects operated, only SESRao and SESFDis gave reason- Laboratoire ECOSYM, UMR 5119 CNRS-UM2- able power, irrespective of how species richness varied along the stress gradient. IRD-IFREMER, Place Euge` ne Bataillon cc 93, FD, FRic, Rao and FDis had low power when species richness was constant, and 34095 Montpellier, France variation in species richness greatly influenced their power. SESFRic and SESFD Pavoine, S. ([email protected]): De´ partement d’Ecologie et Gestion de la biodiversite´ ,UMR were unaffected by variation in species richness. When priority effects operated, 7204 CNRS-UPMC, Muse´ um national d’Histoire FRic, SESFRic, Rao and FDis had good power and were unaffected by variation naturelle, 61 rue Buffon, 75005 Paris, France in species richness. Variation in species richness greatly affected FD and SESFD. University of Oxford, Department of Zoology, SESRao and SESFDis had low power in the priority effects model but were unaf- Mathematical Ecology Research Group, South fected by variation in species richness. Parks Road, Oxford OX1 3PS, UK Dray, S. ([email protected]): Conclusions: Our results demonstrate that a reliable test for changes in assem- Laboratoire BBE-CNRS-UMR-5558, Univ. bly processes along stress gradients requires functional diversity indices measur- C. Bernard – Lyon I 43, Bd du 11 Novembre ing either functional richness or functional divergence. We recommend using 1918, 69622 Villeurbanne Cedex, France SESFRic as a measure of functional richness and either SESRao or SESFDis (which are very closely related mathematically) as a measure of functional divergence. Used together, these indices of functional richness and functional divergence provide good power to test for increasing niche complementarity with declining stress across a broad range of ecological contexts.
Journal of Vegetation Science 794 Doi: 10.1111/jvs.12013 © 2013 International Association for Vegetation Science N.W.H. Mason et al. Functional diversity and stress
allowing subordinate species to evade competition with Introduction superior competitors (Mason et al. 2011b). This should Functional diversity indices have the potential to reveal cause multi-species communities where light competition community assembly processes (Mason et al. 2008b, is intense to have greater functional diversity than those 2012). Vegetation changes along gradients of environmen- where light is not limiting (Mouchet et al. 2010) for traits tal stress are amongst the most widely observed patterns in linked to spatial or temporal differentiation in resource use plant ecology (e.g. Richardson et al. 2004; Peltzer et al. and acquisition. There is recent field evidence for this in 2010). Documenting how community assembly processes plant communities (Mason et al. 2012). change along these gradients is crucial for understanding Where size-symmetric below-ground competition dom- the drivers of observed vegetation change. Functional inates, species with similar niches are more likely to co- diversity is not a single, non-divisible concept, but rather is exist, since small differences in competitive ability do not composed of multiple components (Mason et al. 2005), have disproportionate effects on the outcome of competi- with a large variety of functional diversity indices having tion (Rajaniemi 2003). Co-existence between functionally been proposed to measure these components (Schleuter similar species also results from slower growth rates in et al. 2010; Pavoine et al. 2013). Plant ecologists clearly stressed communities, which reduce the pace of competi- need guidance on which indices to use. Recent theoretical tion after disturbance (following the dynamic equilibrium and practical advances in the measurement of functional theory of Huston (1979) and supported by experimental diversity have led to an increase in the use of functional evidence; Rajaniemi 2003; Wardle & Zackrisson 2005). diversity indices to detect changes in community assembly Consequently, selective pressure for niche differentiation processes along ecological gradients (Mason et al. 2011a,b; between co-occurring species is less intense in stressed Pakeman 2011; Spasojevic & Suding 2012). However, communities. Rather, fitness is enhanced by traits that these studies are usually restricted to a single ecological maximize acquisition and retention of limiting below- context, and so provide little guidance to ecologists on ground resources (e.g. Richardson et al. 2005; Lambers which indices to use. Theoretical models allow us to assess et al. 2008; Holdaway et al. 2011). the power of functional diversity indices to detect changes This is consistent with environmental filtering (i.e. when in assembly processes along gradients across a wide range local fitness is enhanced by possession of traits similar to a of contexts. This study uses simple theoretical models that locally ‘optimal’ trait value; Mouillot et al. 2007; Mason et al. embody clearly defined assembly processes to test which 2011b). The prevalence of environmental filtering in stressed functional diversity indices most reliably reveal changes in communities should cause decreased functional diversity these processes along gradients of abiotic stress. (Mouchet et al. 2010). For some stresses, such as nitrogen limitation, fitness could be enhanced by niche differentiation via facilitation (e.g. Walker & Chapin 1986). The hypothesis Why should functional diversity change along stress that functional diversity should decline with stress assumes gradients? facilitation effects will be minor compared with the influence Stress gradients often represent a shift from below-ground of environmental filtering in stressed communities and is competition for soil nutrients and water to above-ground basedonfieldevidencefromplant communities occurring competition for light with declining stress (e.g. Coomes & along a soil phosphorus gradient (Mason et al. 2012). We Grubb 2000). Competition for light is size-asymmetric (i.e. recognize that factors other than below-ground resource lim- larger individuals are disproportionately advantaged) itation may impose stress on plant communities. We refer to whereas below-ground competition is size-symmetric ‘stress gradients’ in this study, for the sake of brevity. (Schwinning & Weiner 1998; Berntson & Wayne 2000; Cahill & Casper 2000). Competition for light provides more Using functional diversity indices to detect trait-based competitive species with an increasing advantage as they assembly processes outperform (i.e. become taller than) less competitive ones (Grime 1973a,b, 2001; Huston & DeAngelis 1987). Niche Mason et al. (2005) identified three primary components differentiation is required for species to co-exist when of functional diversity – functional richness, functional competition is size-asymmetric, since without niche differ- evenness and functional divergence. Each component pro- entiation even small differences in fitness lead to exclusion vides independent information on the distribution of spe- of all but the fittest species (Aikio 2004; Kohyama & Tak- cies in functional trait space, and a separate index is ada 2009). Thus, niche differences (i.e. spatial or temporal required to quantify each component (Mouchet et al. differentiation in resource use and acquisition) between 2010). Of the three components, functional richness and co-occurring species should increasingly enhance local functional divergence (or indices that combine them) have fitness as light competition becomes more intense, by most often been linked to community assembly processes
Journal of Vegetation Science Doi: 10.1111/jvs.12013 © 2013 International Association for Vegetation Science 795 Functional diversity and stress N.W.H. Mason et al.
(Mouchet et al. 2010; Mason et al. 2012; Spasojevic & strong influence on community assembly. Mass effects Suding 2012) or ecosystem functioning (Petchey et al. occur when source–sink metapopulation dynamics allow 2004; Mouillot et al. 2011). species to occur at low abundance in communities where The various indices of functional richness aim to mea- they are unable to maintain a viable population (Kunin sure the volume of niche space occupied by the species 1998; Leibold et al. 2004). This produces a pattern where within a community (Mouchet et al. 2010). Functional environmental heterogeneity and spatial distance both richness should increase when niche complementarity have significant, independent effects on species turnover enhances probabilities of species occurrence (Mason et al. (Cottenie 2005). Potential for mass effects is highest in 2012). Functional divergence measures the degree to highly heterogeneous landscapes where communities with which the abundance of a community is distributed different abiotic conditions are separated by very small spa- toward the extremities of occupied trait space (Mouchet tial distances (Kunin 1998). Mass effects disrupt the rela- et al. 2010). Functional divergence should increase when tionship between traits and occurrence probabilities of niche complementarity enhances species’ relative abun- species, making it difficult for functional richness indices to dances (Mason et al. 2012). Various indices measure both detect changes in assembly processes (Mason et al. 2013). functional richness and divergence (Mouchet et al. 2010) – Priority effects occur when assembly order influences FDvar (Mason et al. 2003), Rao quadratic entropy the outcome of inter-specific competition (e.g. Ejrnæs (de Bello et al. 2010) and functional dispersion (Laliberte et al. 2006). They generally produce patterns where com- & Legendre 2010). They should increase when niche com- munities in similar environments exhibit persistent differ- plementarity enhances either, or both, species’ occurrence ences in species composition (Fukami et al. 2005; Ejrnæs probabilities and abundances. et al. 2006), and seem to be most evident during succes- An important property of functional diversity indices is sion (e.g. Fukami et al. 2005), especially when competi- that they be independent of existing diversity measures tion for light has a strong influence on community (Mason et al. 2005; Pavoine et al. 2013). Functional rich- composition during succession (e.g. Ejrnæs et al. 2006). ness increases monotonically with species richness. This Priority effects disrupt the relationship between species’ means that observed values of functional richness can traits and both occurrences and abundances, potentially increase in the absence of any change in assembly pro- making it difficult for both functional richness and func- cesses, simply due to corresponding increases in species tional divergence indices to detect changes in assembly richness (Mason et al. 2008b). Comparing observed values processes. for functional richness with those expected from matrix- Reliable detection of changes in trait-based assembly swap null models that randomize species’ occurrences processes along stress gradients requires a set of functional (Manly & Sanderson 2002) can remove any trivial effects diversity indices that are robust against trait-independent of species richness (Mason et al. 2011a; Richardson et al. stochastic assembly processes. Specifically, we need indices 2012). This is important for ensuring that changes in func- that can detect the influence of trait-based processes on tional richness do not lead to spurious conclusions about either occurrence probabilities or abundances, since trait- changes in assembly processes. independent processes might disrupt the link between spe- Recent studies have used a null model that randomizes cies’ traits and either occurrence probability or abundance. abundances across species but within communities to test for trait-based assembly processes in biological communi- Aims and objectives ties (Mason et al. 2008a). Comparison of observed values for indices that measure both functional richness and We aim to explore the power of functional diversity indices divergence with expected values from this null model to detect shifts in assembly processes along gradients under effectively converts these indices into pure measures of a wide range of ecological contexts. To achieve this aim we functional divergence (Mason et al. 2012). This is advanta- sought to develop a simple assembly model that varies trait geous, since there is only one existing measure of func- convergence in communities along a hypothetical stress tional divergence (FDiv; Villeger et al. 2008), and it seems gradient, by altering the relative influence of niche com- to have only moderate power to detect assembly processes plementarity and environmental filtering on species fit- (Mouchet et al. 2010). ness. This model simulates trait values for multiple species pools. It quantifies the effects of environmental filtering and niche complementarity on species fitness, using Functional diversity, assembly processes and trait- explicitly defined functions. It varies the relative influence independent stochasticity of these assembly processes along the gradient, with niche Stochastic processes that act independently of species complementarity becoming more influential as stress traits, particularly mass and priority effects, may have a declines. It incorporates two forms of stochasticity – mass
Journal of Vegetation Science 796 Doi: 10.1111/jvs.12013 © 2013 International Association for Vegetation Science N.W.H. Mason et al. Functional diversity and stress
and priority effects – to test whether indices are robust ensures FitnessEFi is bound between 0 and 100 when | | against trait-independent processes. It also tests whether Ti Topt RangeT. We aimed to avoid negative values variation in species richness affects the power of indices to for fitness since we used fitness to scale species’ abun- detect changes in assembly processes along the stress gradi- dances (see below) and it is not logical for species to be ent. Our ultimate aim is to select a set of functional diver- assigned negative abundances. We used a quadratic func- sity indices that will provide a reliable test of the tion as a simple means of modelling a non-linear decline in hypothesis that niche complementarity, and hence func- fitness with increasing distance of the species’ trait value tional diversity, will increase with declining abiotic stress. from the optimal trait value. To do this we use the following selection criteria: We model the effect of niche complementarity on fit- ness as the competitive effect on potential colonizers of 1. Each index must have reasonable power to detect the species that are already ‘present’ in the community. In the increasing influence of niche complementarity with basic model, species’ local fitness determines their selection declining stress (henceforth power, for brevity) when order. We assume that the species closest to the locally either mass effects or priority effects influence commu- optimal trait value has the highest fitness. Since it colo- nity assembly. nizes first, we assume its fitness is unaffected by competi- 2. Power must be high across a range of species richness tion from species that colonize subsequently. Competition levels. affects the fitness of all other species. We calculate compe- 3. Variation in species richness along the stress gradient tition as a declining cubic function of the distance between must not affect the power of any index. species in trait space: 4. Collectively, the indices must provide reasonable power across all the contexts examined. ¼ ð j jÞ3 ð Þ Cij b RangeT Ti Tj 2
Methods where Cij is the competitive effect of present species j on colonizing species i,andb is a scaling factor that ensures | | We begin by outlining the basic assembly model. We then Cij is bound between 0 and 1 when Ti Topt RangeT. detail how priority and mass effects are built in to incorpo- We used a cubic function since it is a simple non-linear rate stochasticity. Finally, we list the FD indices we use and function and gives a steeper decline in the competitive explain the two null models examined. effect of co-occurring species with increasing trait dissimi- larity than does the quadratic function. This in turn pro- vided greater variation in trait convergence along the Basic assembly model stress gradient. The model context is one where the relative influence of The overall effect of competition from the species pres- niche complementarity and environmental filtering ent on the fitness of potential colonizers is estimated as the change along a stress gradient, with the influence of envi- sum of Cij values weighted by the fitness of the species ronmental filtering being greatest at the most stressed sites. present: The aim of our study was to examine the ability of func- "# tional diversity indices to test the hypothesis that the influ- XS ¼ ; ð Þ ence of niche complementarity on community assembly FitnessNCi 100 min Cij Fitnessj 100 3 ¼ will increase relative to that of environmental filtering j 1 with declining abiotic stress. Consequently, we designed where Fitnessj is the overall fitness of present species j (see our model to produce communities of increasing trait below for calculation of overall fitness). When modelled in divergence with declining stress by varying the relative this way, competitive effects are additive and the intensity influence of environmental filtering and niche comple- of competition exerted by a species is proportional to its fit- mentarity on species fitness. We model the effect of envi- ness. A potential colonizing species will experience intense ronmental filtering on fitness as a declining quadratic competition if it has similar traits to a species that is already function of the difference between a species’ trait value present and has high fitness. Overall fitness is calculated as and the locally optimal trait value for a given site: a combination of environmental filtering and niche com- ¼ ð j jÞ2 ð Þ plementarity effects on fitness: FitnessEFi a RangeT Ti Topt 1
Fitnessi ¼ c FitnessEFi þ d FitnessNCi ð4Þ where: RangeT is the range of values for trait T in the spe- cies pool, Ti is the trait value for species i, Topt is the locally where c + d = 1, so that fitness is always bound between 0 optimal trait value for the site and a is a scaling factor that and 100.
Journal of Vegetation Science Doi: 10.1111/jvs.12013 © 2013 International Association for Vegetation Science 797 Functional diversity and stress N.W.H. Mason et al.
In the basic model, species are selected in order of their abundance value, abundance is largely determined by fitness. The first species selected is the one with the highest having complementary traits to species that are already
FitnessEFi (i.e. with a trait value closest to the local opti- present. mum), and its fitness is assumed to be unaffected by com- petition (i.e. Fitness = 100). Fitness for all remaining NCi Generating regional species pools and locally optimal species is then calculated including both Fitness and the EFi trait values influence of competition from the first species (i.e. as mea- sured by Cij 9 Fitnessj in equation 3, with species j being Values for a single trait were simulated for 130 indigenous the first colonizing species). The species with the highest New Zealand herbaceous and woody angiosperm, gymno- fitness is then selected as the second colonizer. The process sperm and fern species following a Brownian motion is repeated until the number of selected species equals the model of evolution with mean of 0 and SD of 1 (Blomberg desired local species richness. Fitness of the selected species et al. 2003). These species form the pool of potential colo- is then scaled to abundance so that the most abundant spe- nizers for our simulated communities. We chose to exam- cies has 103 times the abundance of the least abundant ine only a single trait, since identifying ecological species, and differences in abundance between species are mechanisms driving variation in functional diversity often proportional to their differences in fitness. This degree of requires each trait to be analysed separately (Mason et al. variation in abundances is typical in plant communities. At 2011b). We used Phylomatic (http://www.phylodiversity. high levels of d, colonization order has increasing influence net/phylomatic/) to construct a hypothesized phylogenetic on fitness, since late-arriving species will, on average, have tree for our species, and used the New Zealand Plant Phy- lower FitnessNCi than early colonizers. logeny Database (http://plantphylogeny.landcareresearch. As demonstrated by de Bello et al. (2011), Albert et al. co.nz/WebForms/Home.aspx) to manually graft species (2012), Laliberte & Legendre (2010) and de Bello et al. that were not available in Phylomatic. Branch lengths (2012), calculating functional diversity for multiple traits were computed following Grafen (1989) using the ‘com- can introduce a variety of complications, and necessitates pute.brlen’ in R package ‘ape’ (Paradis et al. 2004; R Foun- complex methodological decisions. To avoid such compli- dation for Statistical Computing, Vienna, AT, Australia). cations, the model uses a single trait to determine both There may be some potential for characteristics of our phy- niche complementarity and environmental filtering effects logeny or the model of trait evolution to alter results. How- on fitness. There is a good precedent for this based on field ever, it seems that aspects like phylogenetic tree shape studies showing that single traits may simultaneously generally have only subtle effects on the likelihood of influence species’ occurrences along stress gradients as detecting assembly processes (Kraft et al. 2007). Also, a well as influencing competitive interactions between spe- recent study using the same simulation model as used here cies at the local community level (Cornwell & Ackerly has shown that the strength of trait conservatism used in 2009; Mason et al. 2012). However, we recognize that spe- generating species pools via trait evolution models does cies’ responses to environmental heterogeneity and inter- not affect the relationship between species turnover and specific competition may be influenced by different traits functional traits along environmental gradients (Mason & in many instances (Chesson 2000). Pavoine 2012). Locally optimal trait values for each site were spaced evenly along the central 60% of the gradient of total traits Varying assembly processes along the gradient present in the species pool. The intention here was to Values for c and d vary along the hypothetical stress gradi- ensure that functional diversity values at the extremes of ent. We simulated species’ abundances for ten points the stress gradient were not affected by limitations in the (termed sites henceforth) along the stress gradient. At the trait values present in the species pool. There is a risk that most stressed site, c = 0.91 and d = 0.09, so that 91% of functional diversity values might be artificially constrained overall fitness is determined by environmental filtering if the optimal trait values at the ends of the stress gradient and 9% is determined by niche complementarity. Moving are too close to the edge of the trait space occupied by the from the most stressed to least stressed end of the gradient, species pool. Altering locally optimal trait values is consis- we decreased c and increased d by intervals of 0.09, so that tent with multiple observations that the trait values of at the least stressed site c = 0.1 and d = 0.9, with 10% of locally dominant species vary monotonically along stress overall fitness due to environmental filtering and 90% due gradients (Peltzer et al. 2010). to niche complementarity. In this way, fitness, and hence Under this basic model, species’ abundances are a abundance, is largely determined by proximity to the deterministic function of their traits and the traits of locally optimal trait value at the most stressed site. At the co-occurring species. This is quite unrealistic, since least stressed site, for all but the species with the highest stochastic processes that act independently of traits also
Journal of Vegetation Science 798 Doi: 10.1111/jvs.12013 © 2013 International Association for Vegetation Science N.W.H. Mason et al. Functional diversity and stress influence abundance in plant communities. Consequently, fitness is strongest. This is consistent with evidence that we do not use the basic model to test whether functional priority effects are stronger in high productivity environ- diversity indices can detect changes in assembly processes. ments where size-asymmetric competition for light is more Rather, we employ two extensions that incorporate differ- intense (Ejrnæs et al. 2006). This model tests whether ent types of stochasticity. functional diversity indices can detect changes in trait- based assembly process in communities where priority effects disrupt the effect of these processes on both occur- Mass effects model rences and abundances. This is a simple extension of the basic model where, for each site, the species present are randomly selected from Functional diversity indices and null models the species pool (e.g. if we wanted a community of ten spe- cies, we would randomly select ten species from the regio- We examined two indices of functional richness – the FD nal species pool). The basic model then calculates the of Petchey & Gaston (2002) and the FRic of Villeger et al. fitness and abundance of the selected species. Under this (2008), which is derived from the convex hull volume of model, species’ occurrences are completely unrelated to Cornwell et al. (2006). For a single trait, FRic is simply the their traits, but abundance is a deterministic function of range of trait values spanned by the species present. To the traits of the species present. This is an extreme form of ensure values for each simulated community for both mass effects. This model tests whether functional diversity these measures were bound between 0 and 1, they were indices can detect changes in trait-based assembly pro- expressed as a proportion of the values for the entire spe- cesses in communities where these processes have weak cies pool. We used FDiv as a measure of functional diver- effects on species’ occurrences (i.e. neither biotic nor abi- gence. For functional evenness (i.e. evenness in the otic filtering excludes species from local communities), but distribution of abundance in functional trait space; Mason a strong effect on species’ abundances. The assumption et al. 2005) we used the FEve of Villeger et al. (2008). We that mass effects are not influenced by the traits linked to examined two mathematically quite similar indices that environmental filtering and niche complementarity is per- are a combination of both functional evenness and func- haps an oversimplification (Mu¨ nkemu¨ ller et al. 2012). tional divergence – FDis (Laliberte & Legendre 2010) and However, seed production and dispersal are the traits most Rao quadratic entropy (referred to as Rao henceforth; Rao likely to drive mass effects, since high propagule pressure is 1982; Pavoine & Doledec 2005; de Bello et al. 2010). the key factor in maintaining sink populations (Mouquet Although there is some debate about the appropriate dis- et al. 2004). Further, there is evidence that traits linked to similarity measure to use in calculating Rao (de Bello et al. regenerative strategy vary independently of those linked 2012), we use Gower distance, as recommended by Pavo- to resource use and acquisition (Grime et al. 1997; Dı´az ine et al. (2009). et al. 2004), so this assumption may be more or less We compared FRic and FD values relative to random realistic. expectation under a matrix-swap null model (Manly & Sanderson 2002) using the standardized effect size (SES) of Gotelli & McCabe (2002). These indices are termed SES- Priority effects model FRic and SESFD henceforth. FDis and Rao were compared This model randomizes the order of species’ colonization, with values obtained when abundances were randomized whereas species’ traits determine colonization order in the across species but within plots (after Mason et al. 2008a). basic model. In this model, colonization order affects the These indices, termed SESFDis and SESRao, are pure mea- degree of competition experienced by each species, with sures of functional divergence, since the functional rich- early-colonizing species, on average, experiencing less nesses of the observed and randomized communities are intense competition. Species fitness is calculated as in the always identical. We also compared FDiv to this null basic model, but competitive effects on fitness are based on model. We used 104 randomizations in all null-model this randomized colonization order. The desired number of analyses, to ensure accurate estimates of SES values. We species is then selected based on species fitness (so when did not randomize abundances between communities generating communities of ten species, the ten fittest spe- because it is unclear how SES values derived in this way cies are selected). Since colonization order may have a relate to the three primary components of functional strong influence on the level of competition a species expe- diversity. Observed FEve was not compared with either riences, the priority effects model incorporates stochasticity null model since it is independent of species richness, and in the relationship between traits and both species’ occur- it is unclear how SES values generated when randomizing rences and abundances. This stochasticity is highest in the species’ abundances within communities should be least stressed sites where the influence of competition on interpreted.
Journal of Vegetation Science Doi: 10.1111/jvs.12013 © 2013 International Association for Vegetation Science 799 Functional diversity and stress N.W.H. Mason et al.
Under the mass effects model, functional richness indi- influence of niche complementarity on fitness (as indi- ces (FD, FRic, SESFD and SESFRic) should have no cated by the coefficient d). We tested whether the power of power to detect changes in assembly process, since spe- indices was affected by variation in species richness along cies’ occurrences are unrelated to their traits. Similarly, the gradient, by simulating communities of varying species indices that are partially influenced by functional rich- richness. To examine power with increasing species rich- ness (FDis and Rao) might also have low power in the ness, communities containing five species were simulated mass effects model. However, indices of functional diver- for the two most stressed sites (i.e. sites with the lowest d gence (FDiv, SESFDis and SESRao) should have good values). Communities with ten species were simulated for power since they are only influenced by how abundance the two next most stressed sites, and so on until communi- is distributed in occupied trait space. Under the priority ties with 25 species were simulated for the two least effects model, the effect of traits on species’ occurrences stressed sites (i.e. the two sites with the highest d values). is also disrupted, but much less severely than is the mass This provides an indication of power when species richness effects model, so functional richness indices should have increases with the influence of niche complementarity (d). good power to detect changes in assembly processes. Pri- We repeated this process in reverse to estimate power ority effects also disrupt the relationship between traits when species richness declined with d. We also examined and relative abundances, which might influence the power when species richness varied randomly with d. power of functional divergence indices, and indices that Where significant negative correlations were obtained, are partially influenced by functional divergence (FDis these were recorded as ‘negative power’, so that power and Rao). was bound between 1and1.
Simulation framework Results The same simulation framework was applied to the mass For the priority effects models, FRic, FDis, Rao and SES- effects and priority effects models. In each case, it began FRic had power of >0.95 for all levels of species richness with the generation of trait values for the species pool. (Fig. 1a). Thus, these indices had at least a 95% chance Then communities were assembled for each of the ten sites of detecting an increase in niche complementarity with occurring along the hypothetical stress gradient. Func- decreasing stress. FD and SESFD had power of >0.9 for tional diversity values were calculated for each of the sites, the two lowest levels of local species richness (i.e. five along with Pearson correlation and Spearman rank corre- and ten species), but not the three highest (i.e. 15, 20 lation coefficients between each functional diversity index and 25 species). All other indices had low power in and the influence of niche complementarity on fitness (as comparison. indicated by the coefficient d). Significant positive correla- For the mass effects model, SESFDis and SESRao had tions (i.e. two-tailed P < 0.05) between functional diver- power of 0.8 at all levels of local species richness sity indices and d indicate that they are able to detect the (Fig. 1b). The power of FDis and Rao increased with spe- increasing influence of niche complementarity on commu- cies richness, being ~0.8 for the two highest levels of local nity assembly with declining stress. Both these correlation species richness (i.e. 20 and 25 species). All other indices coefficients gave similar power for all indices, so we only had very low power. present the results for Pearson r here. We also examined Variation in species richness along the stress gradient log-linear, quadratic and exponential relationships did not markedly affect the power of FRic, FDis, Rao, SES- between functional diversity indices and d. In no instance FRic, SESRao and SESFDis in the priority effects model did these relationships yield greater power (according to (Fig. 2a). In contrast, power for FD was much higher when Akaike’s information criterion; Burnham & Anderson species richness increased and much lower when it 2002) than Pearson r. decreased with declining stress. Power for SESFD showed We generated trait values for 100 hypothetical regional the opposite pattern, being much higher when species species pools (i.e. 100 separate simulations of trait values richness decreased and much lower when it increased with for the 130 species in our phylogenetic tree), and simulated declining stress. communities at five levels of local species richness: 5, 10, For the mass effects model, the power of SESFD, SES- 15, 20 and 25. For example, with a local species richness FRic, SESRao and SESFDis was unaffected by variation in level of 5, each of the ten simulated local communities species richness along the stress gradient (Fig. 2b). How- (one community for each point along the stress gradient), ever, variation in species richness strongly affected the contains five species. For each species richness level, we power of FD, FRic, FDis and Rao. In the case of FD and recorded the proportion of significant correlations (termed FRic, power was negative when species richness declined power henceforth) obtained between each index and the with declining stress.
Journal of Vegetation Science 800 Doi: 10.1111/jvs.12013 © 2013 International Association for Vegetation Science N.W.H. Mason et al. Functional diversity and stress
5 10 15 20 25 Increasing Random Decreasing (a) 1 (a) 1 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 FDis Rao FDiv SESFD SESFRic SESFDis SESRao SESFDiv FD FRic FEve
Power 0 –0.2 FD FRic FEve FDis Rao FDiv SESFD SESFRic SESFDis SESRao SESFDiv Power –0.2 –0.4 –0.4 –0.6 –0.6 –0.8 –0.8 –1 –1 1 5 10 15 20 25 (b) Increasing Random Decreasing 0.8 (b) 1 0.6 0.8 0.4 0.6 0.2 0.4 0 0.2 FD FRic FEve FDis Rao FDiv SESFD SESFRic SESFDis SESRao SESFDiv
Power 0
–0.2 FD FRic FEve FDis Rao FDiv SESFD SESFRic SESFDis SESRao SESFDiv –0.4 Power –0.2 –0.6 –0.4 –0.8 –0.6 –1 –0.8 –1 Fig. 1. Power of functional diversity indices to detect the increasing influence of niche complementarity with declining stress when: (a)priority Fig. 2. Power of functional diversity indices to detect the increasing effects disrupt the relationship between species traits and abundances, influence of niche complementarity with declining stress with species and (b) mass effects disrupt the relationship between species traits and richness increasing, decreasing or varying randomly as stress declines. In occurrences. Numbers associated with different types of bar shading (a) priority effects disrupt the relationship between species’ traits and indicate species richness of local communities. FD is the sum dendrogram abundances, while in (b) mass effects disrupt the relationship between branch length, FRic is convex hull volume, FEve is functional evenness, species’ traits and occurrences. FD is the sum dendrogram branch length, FDis is functional dispersion, Rao is quadratic entropy and FDiv is FRic is convex hull volume, FEve is functional evenness, FDis is functional functional divergence. SESFD and SESFRic are, respectively, observed FD dispersion, Rao is quadratic entropy and FDiv is functional divergence. and FRic expressed relative to a matrix-swap null model. SESRao, SESFDis SESFD and SESFRic are, respectively, observed FD and FRic expressed and SESFDiv are, respectively, observed Rao, FDis and FDiv expressed relative to a matrix-swap null model. SESRao, SESFDis and SESFDiv are, relative to a null model randomizing abundances across species, but respectively, observed Rao, FDis and FDiv expressed relative to a null within communities. model randomizing abundances across species, but within communities.
context-dependent. Thus, any study examining changes in Discussion assembly processes along gradients should employ several, We are aware of only one simulation study examining the complementary functional diversity indices. They demon- power of a broad range of functional diversity indices to strate that different types of stochasticity have contrasting detect community assembly processes (Mouchet et al. effects on the power of functional diversity to detect 2010). Their study did not consider how null models and changes in assembly processes, with our two assembly variation in species richness affected power. Nor did it models giving highly divergent results. They show that sto- determine which indices maintained power in the face of chastic processes alter the dependence of indices on species trait-independent, stochastic processes. Similarly, we are richness. They also reveal that we need to compare only aware of a single study using simulated data to test if observed functional diversity values to null models (using null models altered the conclusions drawn from functional standardized effect size, SES) to ensure that we draw the diversity patterns (de Bello 2012). However, that study right conclusions about assembly processes. considered a much narrower range of contexts than we Below we discuss what our results for functional diver- have. sity patterns tell us about shifts in assembly processes in Our results extend these existing studies in several key our two models. We specifically outline when and how ways. They show that the power of functional diversity null models improve our ability to detect shifts in assembly indices to detect changes in assembly processes is highly processes along gradients. We finish by suggesting a set of
Journal of Vegetation Science Doi: 10.1111/jvs.12013 © 2013 International Association for Vegetation Science 801 Functional diversity and stress N.W.H. Mason et al. functional diversity indices that should be appropriate for a decreases in the latter. However, when expressing wide range of ecological contexts. In doing this we provide observed values for these indices relative to those expected some guidance about what presence or lack of significant from a matrix-swap null model (i.e. SESFD and SESFRic), variation in these indices along stress gradients may reveal we would conclude, correctly, that the influence of niche about community assembly processes. complementarity on occurrences does not change with stress, irrespective of how species richness varies along the gradient. The power of FDis and Rao was also greatly Null models, stochasticity and power to detect shifting affected by variation in species richness, being much assembly processes higher when species richness increased with declining In the mass effects model, SESRao and SESFDis were the stress, which would affect conclusions about assembly pro- only measures to give reasonable power ( 0.8) for all lev- cesses. This is due to their sensitivity to functional richness, els of species richness. SESRao and SESFDis values were which is itself influenced by species richness, but not partially contingent on the random selection of species assembly processes in this model. Power of SESRao and from the species pool. This explains why we obtained only SESFDis was unaffected by variation in species richness moderate power. This result for SESRao and SESFDis along the gradient. These results show that using null reveals the increasing influence of niche complementarity models to generate SES values helped remove the influ- on species’ abundances with declining stress. That SESRao ence of species richness on functional diversity. This is and SESFDis had greater power than observed Rao and important to prevent drawing spurious conclusions about FDis shows that use of an appropriate null model to obtain assembly processes when species richness varies greatly standardized effect size (SES) improved our ability to between communities. detect changes in assembly processes along the stress gradi- In the priority effects model, variation in species rich- ent. FD, FRic, SESFD and SESFRic had power of ~0 for all ness along the stress gradient did not affect the power of levels of species richness. This is to be expected since trait- FRic, FDis and Rao. Thus, there was no evidence that vari- based assembly processes do not influence species’ occur- ation in species richness would lead to spurious conclu- rence probabilities under this model, so that functional sions under this model. Consequently, there was no need richness varies at random along the stress gradient. to use null models to remove the influence of species In the priority effects model, FRic and SESFRic reliably richness. detected an increase in functional richness with declining stress for all levels of species richness. FDis and Rao had Recommended indices and interpretation of significant higher power than SESFDis and SESRao, especially at low- correlations est levels of species richness. FDis and Rao are sensitive to changes in functional richness, but SESFDis and SESRao Our results permit us to identify a set of indices that give are not. This suggests that the power of FDis and Rao was reasonable power to detect changes in assembly processes largely due to increases in functional richness with declin- across the range of contexts examined. They also allow us ing stress. These results reveal the increasing influence of to suggest interpretations for presence or lack of significant niche complementarity on species’ occurrences (via correlations between these indices and stress gradients. increases in functional richness) with declining stress. The The recommended indices and interpretations of correla- comparatively low power of SESFDis and SESRao also tions between them and stress gradients are summarized indicates that priority effects disrupted the influence of in Table 1. trait-based assembly processes on species’ abundances. Our results indicate that when priority effects influence These results also show that the null models we used to species composition, but mass effects are weak, SESFRic produce SES values provided no increase in power under will have good power to detect increasing niche comple- the priority effects model. mentarity in the priority effects model. SESFRic was also robust against variation in species richness in both our assembly models, and thus satisfies all our criteria. When Null models, stochasticity and dependence of power on using multiple traits, SESFD might also be an appropriate species richness index. Its sensitivity to variation in species richness may In the mass effects model, FD and FRic led to completely have been partly due to saturation of observed FD values different conclusions according to whether species richness at high species richness, which itself could arise from our increased or decreased with declining stress. Using these using only a single trait. Here we have presented a basic indices we would conclude, incorrectly, that the influence matrix-swap null model where swapping of species’ occur- of niche complementarity on species’ occurrences rences between any pair of sites is permitted. Instances increases with declining stress in the former case and where sites are separated by large geographical distances
Journal of Vegetation Science 802 Doi: 10.1111/jvs.12013 © 2013 International Association for Vegetation Science N.W.H. Mason et al. Functional diversity and stress
Table 1. Recommended indices and interpretations of significant correlations between indices and stress gradients.
SESFRic SESRao/SESFDis Interpretation
+ NS Increasing influence of NC on occurrences, but not abundances, with declining stress NS + Increasing influence of NC on abundances, but not occurrences, with declining stress ++ Increasing influence of NC on both occurrences and abundances, with declining stress NS NS No change in influence of NC on either occurrences or abundances, with declining stress SESFRic is the standardized effect size for the FRic of Villeger et al. (2008) obtained using a matrix-swap null model to randomize occurrences. SESRaoand SESFDis are standardized effect size for (Rao) quadratic entropy and functional dispersion (FDis) obtained using a null model randomizing abundances across species but within communities. NC refers to niche complementarity. + indicates a significant positive correlation between functional diversity and declining stress. SESRao/SESFDis indicates that these indices are interchangeable due to their similar behaviour. may require application of spatial constraints (e.g. Rox- We do not recommend directly using observed Rao and burgh & Matsuki 1999) so that the null model incorporates FDis values, for three reasons. First, variation in species dispersal limitation effects (e.g. as proposed by Mason et al. richness strongly affected the power of both indices in the (2007)). However, in testing for changes in assembly pro- mass effects model. Second, it is difficult to interpret corre- cesses along ecological gradients we advise against using lations between these indices and stress gradients since environmentally constrained matrix-swap null models they measure both functional richness and functional (e.g. Peres-Neto et al. 2001). This is because environmen- divergence. This means we cannot use them to disentangle tal constraints on species’ occurrences are relevant in test- the effects of niche complementarity on occurrences from ing for shifts in assembly processes along gradients, and its effects on abundances. Finally, they provide no benefit thus should not be included in null models randomizing for detecting changes in assembly processes if appropriate species’ occurrences. In making these suggestions, we are indices of functional richness and functional divergence none-the-less aware that null models randomizing species’ are available. We have identified SESFRic as an appropri- occurrences have for decades been the subject of intense ate index of functional richness and SESRao or SESFDis as debate (e.g. Connor & Simberloff 1979; Diamond & Gilpin appropriate measures of functional divergence. However, 1982; Harvey et al. 1983; Gotelli 2000; Manly & Sanderson it is possible that Rao and FDis values, calculated using 2002; Hardy 2008). In the end, the appropriate null model occurrences rather than proportional abundances and always depends on the question and the context in which compared to a matrix-swap null model, could also reveal it is posed. the effects of assembly processes on species’ occurrences. SESRao or SESFDis should both have good power to In no instance did FEve provide good power to detect detect changes in assembly processes when mass effects changes in assembly processes. Previous studies have have a strong influence in species’ occurrences. Mathe- found little evidence for change in functional evenness in matically, these indices are very similar and consequently plant communities along ecological gradients (e.g. Mason can be used interchangeably. They were the only indices et al. 2012). It is possible that limited evidence for func- giving reasonable power to detect the increased influence tional evenness as an indicator of local assembly processes of niche complementarity on abundances in the mass could be due to limitations of FEeve (see Pavoine & Bonsall effects model. Either of these indices is required to avoid 2010; Appendix A). It is also possible that variation in func- spurious acceptance of the null hypothesis of no change in tional evenness is simply not associated with changes in trait-based assembly processes when mass effects strongly assembly processes. We require indices of functional even- influence species’ occurrences. While SESRao and SESFDis ness that more closely approximate the concept (as pre- had low power in the priority effects model, they were sented by Mason et al. (2005)) before we can verify this. robust against variation in species richness along the stress We recognize that the modelling approach we have gradient. Thus, variation in species richness will not cause employed is highly simplified, and may not fully reflect these indices to yield spurious conclusions when stochastic processes that occur in real plant communities. We also processes disrupt the relationship between species’ traits recognize that many processes, other than mass and prior- and abundances. The other index of functional divergence ity effects, might prevent relationships between functional we used, FDiv, had low power across all of the contexts we diversity from emerging. Our intention was to build a examined. Pavoine & Bonsall (2010) showed several model where changes in trait-based assembly processes are instances where FDiv values did not vary between com- clearly defined, and to use this model in selecting indices munities that did actually have different functional diver- for testing the hypothesis that these processes change gence (see their Appendix A), which may explain the low along stress gradients. Our study has successfully done this. power we observed for this index. However, it would indeed be interesting to test whether
Journal of Vegetation Science Doi: 10.1111/jvs.12013 © 2013 International Association for Vegetation Science 803 Functional diversity and stress N.W.H. Mason et al. different types of assembly models give similar results. Our References work could most fruitfully be extended using two quite dif- ferent approaches. First, individual-based models would Aikio, S. 2004. Competitive asymmetry, foraging area size and provide useful insights since competitive interactions occur coexistence of annuals. Oikos 104: 51–58. between individuals (e.g. Huston & Smith 1987). This Albert, C.H., de Bello, F., Boulangeat, I., Pellet, G., Lavorel, S. & Thu- might provide a more realistic examination of how niche iller, W. 2012. On the importance of intraspecific variability for complementarity and environmental filtering influence the quantification of functional diversity. Oikos 121: 116–126. the distribution of species and abundance of local commu- de Bello, F. 2012. The quest for trait convergence and divergence nities in trait space. Second, models that simulate meta- in community assembly: are null-models the magic wand? community processes would provide a more realistic Global Ecology and Biogeography 21: 312–317. examination of trait-independent stochastic processes (e.g. de Bello, F., Lavergne, S., Meynard, C., Lepsˇ, J. & Thuiller, W. Mu¨ nkemu¨ ller et al. 2012). Metacommunity models incor- 2010. The spatial partitioning of diversity: showing Theseus porate both priority and mass effects. They thus provide a a way out of the labyrinth. Journal of Vegetation Science 21: 992–1000. suitable avenue for exploring the power of functional de Bello, F., Lavorel, S., Albert, C.H., Thuiller, W., Grigulis, K., diversity indices to detect changes in trait-based assembly Dolezal, J., Janecˇek, Sˇ.&Lepsˇ, J. 2011. Quantifying the rele- processes in the face of stochasticity. vance of intraspecific trait variability for functional diversity. Another potential extension of our modelling frame- Methods in Ecology and Evolution 2: 163–174. work would be to consider whether functional diversity de Bello, F., Carmona, C.P., Mason, N.W.H., Sebastia`,M.-T.& indices might be able to reveal how interactions between Lepsˇ, J. 2012. Trait dissimilarity for functional diversity: trait disturbance and stress influence community assembly. means or trait overlap? Journal of Vegetation Science.In There is emerging evidence that below-ground resource Review. availability strongly moderates the impact of disturbance Berntson, G.M. & Wayne, P.M. 2000. Characterizing the size on taxonomic diversity in plant communities (e.g. Haddad dependence of resource acquisition within crowded plant et al. 2008). However, it remains unclear what assembly populations. Ecology 81: 1072–1085. processes are responsible for this. Simple assembly models Blomberg, S.P., Garland, T. & Ives, A.R. 2003. Testing for phylo- might be helpful in predicting expected patterns of func- genetic signal in comparative data: behavioural traits are tional diversity given hypothesized changes in assembly more labile. Evolution 57: 717–745. processes for different combinations of stress intensity and Burnham, K.P. & Anderson, R.S. 2002. Model selection and multi- disturbance frequency. model inference: a practical information-theoretic approach.2nd edn. Springer, New York, NY, USA. Cahill, J.F. Jr & Casper, B.B. 2000. Investigating the relationship Conclusions between neighbor root biomass and belowground competi- Indices of functional richness and functional divergence tion: field evidence for symmetric competition belowground. – had the highest power to detect changes in assembly pro- Oikos 90: 311 320. cesses along our hypothetical stress gradient. However, Chesson, P. 2000. Mechanisms of maintenance of species diver- – the power of these indices was highly context-dependent. sity. Annual Review of Ecology and Systematics 31: 343 358+C1 + 359–366. Use of appropriate null models increased power and Connor, E.F. & Simberloff, D. 1979. The assembly of species com- removed the influence of species richness on index val- munities: chance or competition? Ecology 60: 1132–1140. ues when mass effects predominated. We recommend Coomes, D.A. & Grubb, P.J. 2000. Impacts of root competition in convex hull volume expressed relative to a matrix-swap forests and woodlands: a theoretical framework and review null model (e.g. SESFRic) to detect the influence of trait- of experiments. Ecological Monographs 70: 171–207. based assembly processes on species’ occurrences. Both Cornwell, W.K. & Ackerly, D.D. 2009. Community assembly and Rao quadratic entropy and functional dispersion shifts in plant trait distributions across an environmental gra- expressed relative to a null model randomizing abun- dient in coastal California. Ecological Monographs 79: 109–126. dances across species but within communities (e.g. SES- Cornwell, W.K., Schwilk, D.W. & Ackerly, D.D. 2006. A trait- Rao or SESFDis) reliably detect the influence of assembly based test for habitat filtering: convex hull volume. Ecology processes on species’ abundances. We must always exer- 87: 1465–1471. cise caution in deriving general predictions from a single Cottenie, K. 2005. Integrating environmental and spatial pro- simulation study. However, we can at least be certain cesses in ecological community dynamics. Ecology Letters 8: that use of these complementary indices together will 1175–1182. provide a reliable test for changes in assembly processes Diamond, J.M. & Gilpin, M.E. 1982. Examination of the “null” along stress gradients for the range of contexts we have model of Connor and Simberloff for species co-occurrences studied. on islands. Oecologia 52: 64–74.
Journal of Vegetation Science 804 Doi: 10.1111/jvs.12013 © 2013 International Association for Vegetation Science N.W.H. Mason et al. Functional diversity and stress
Dı´az, S., Hodgson, J.G., Thompson, K., Cabido, M., Cornelissen, Holdaway, R.J., Richardson, S.J., Dickie, I.A., Peltzer, D.A. & Co- J.H.C., Jalili, A., Montserrat-Marti, G., Grime, J.P., Zarrinka- omes, D.A. 2011. Species- and community-level patterns in mar, F., Asri, Y., Band, S.R., Basconcelo, S., Castro-Diez, P., fine root traits along a 120 000-year soil chronosequence in Funes, G., Hamzehee, B., Khoshnevi, M., Perez-Harguinde- temperate rain forest. Journal of Ecology 99: 954–963. guy, N., Perez-Rontome, M.C., Shirvany, F.A., Vendramini, Huston, M. 1979. A general hypothesis of species diversity. The F., Yazdani, S., Abbas-Azimi, R., Bogaard, A., Boustani, S., American Naturalist 113: 81–101. Charles, M., Dehghan, M., de Torres-Espuny, L., Falczuk, V., Huston, M.A. & DeAngelis, D.L. 1987. Size bimodality in mono- Guerrero-Campo, J., Hynd, A., Jones, G., Kowsary, E., Kaz- specific populations: a critical review of potential mecha- emi-Saeed, F., Maestro-Martinez, M., Romo-Diez, A., Shaw, nisms. The American Naturalist 129: 678–707. S., Siavash, B., Villar-Salvador, P. & Zak, M.R. 2004. The Huston, M. & Smith, T. 1987. Plant succession: life history and plant traits that drive ecosystems: evidence from three conti- competition. The American Naturalist 130: 168–198. nents. Journal of Vegetation Science 15: 295–304. Kohyama, T. & Takada, T. 2009. The stratification theory for Ejrnæs, R., Bruun, H.H. & Graae, B.J. 2006. Community assem- plant coexistence promoted by one-sided competition. Jour- bly in experimental grasslands: suitable environment or nal of Ecology 97: 463–471. timely arrival? Ecology 87: 1225–1233. Kraft, N.J.B., Cornwell, W.K., Webb, C.O. & Ackerly, D.D. 2007. Fukami, T., Bezemer, T.M., Mortimer, S.R. & van der Putten, Trait evolution, community assembly, and the phylogenetic W.H. 2005. Species divergence and trait convergence in structure of ecological communities. The American Naturalist experimental plant community assembly. Ecology Letters 8: 170: 271–283. 1283–1290. Kunin, W.E. 1998. Biodiversity at the edge: a test, of the impor- Gotelli, N.J. 2000. Null model analysis of species co-occurrence tance of spatial “mass effects” in the Rothamsted Park Grass patterns. Ecology 81: 2606–2621. experiments. Proceedings of the National Academy of Sciences of Gotelli, N.J. & McCabe, D.J. 2002. Species co-occurrence: a the United States of America 95: 207–212. meta-analysis of J. M. Diamond’s assembly rules model. Laliberte, E. & Legendre, P. 2010. A distance-based framework Ecology 83: 2091–2096. for measuring functional diversity from multiple traits. Ecol- Grafen, A. 1989. The phylogenetic regression. Philosophical Trans- ogy 91: 299–305. actions of the Royal Society of London Series B – Biological Sciences Lambers, H., Raven, J.A., Shaver, G.R. & Smith, S.E. 2008. Plant 326: 119–157. nutrient-acquisition strategies change with soil age. Trends in Grime, J.P. 1973a. Competition and diversity in herbaceous veg- Ecology & Evolution 23: 95–103. etation – reply. Nature 244: 311. Leibold, M.A., Holyoak, M., Mouquet, N., Amarasekare, P., Grime, J.P. 1973b. Competitive exclusion in herbaceous vegeta- Chase, J.M., Hoopes, M.F., Holt, R.D., Shurin, J.B., Law, R., tion. Nature 242: 344–347. Tilman, D., Loreau, M. & Gonzalez, A. 2004. The metacom- Grime, J.P. 2001. Plant strategies, vegetation processes, and ecosystem munity concept: a framework for multi-scale community properties. 2nd edn. Wiley, Chichester, UK. ecology. Ecology Letters 7: 601–613. Grime, J.P., Thompson, K., Hunt, R., Hodgson, J.G., Cornelis- Manly, B. & Sanderson, J.G. 2002. A note on null models: justi- sen, J.H.C., Rorison, I.H., Hendry, G.A.F., Ashenden, T.W., fying the methodology. Ecology 83: 580–582. Askew, A.P., Band, S.R., Booth, R.E., Bossard, C.C., Camp- Mason, N.W.H. & Pavoine, S. 2013. Does trait conservatism guar- bell, B.D., Cooper, J.E.L., Davison, A.W., Gupta, P.L., Hall, antee that indicators of phylogenetic community structure W., Hand, D.W., Hannah, M.A., Hillier, S.H., Hodkinson, will reveal niche-based assembly processes along stress gradi- D.J., Jalili, A., Liu, Z., Mackey, J.M.L., Matthews, N., ents? Journal of Vegetation Science 24: 820–833. Mowforth, M.A., Neal, A.M., Reader, R.J., Reiling, K., Mason, N.W.H., MacGillivray, K., Steel, J.B. & Wilson, J.B. RossFraser, W., Spencer, R.E., Sutton, F., Tasker, D.E., 2003. An index of functional diversity. JournalofVegetation Thorpe, P.C. & Whitehouse, J. 1997. Integrated screening Science 14: 571–578. validates primary axes of specialisation in plants. Oikos 79: Mason, N.W.H., Mouillot, D., Lee, W.G. & Wilson, J.B. 2005. 259–281. Functional richness, functional evenness and functional Haddad, N.M., Holyoak, M., Mata, T.M., Davies, K.F., Mel- divergence: the primary components of functional diversity. bourne, B.A. & Preston, K. 2008. Species’ traits predict the Oikos 111: 112–118. effects of disturbance and productivity on diversity. Ecology Mason, N.W.H., Lanoiselee, C., Mouillot, D., Irz, P. & Argillier, Letters 11: 348–356. C. 2007. Functional characters combined with null models Hardy, O.J. 2008. Testing the spatial phylogenetic structure of reveal inconsistency in mechanisms of species turnover in local communities: statistical performances of different null lacustrine fish communities. Oecologia 153: 441–452. models and test statistics on a locally neutral community. Mason, N.W.H., Lanoiselee, C., Mouillot, D., Wilson, J.B. & Journal of Ecology 96: 914–926. Argillier, C. 2008a. Does niche overlap control relative abun- Harvey, P.H., Colwell, R.K., Silvertown, J.W. & May, R.M. 1983. dance in French lacustrine fish communities? A new method Null models in ecology. Annual Review of Ecology and Systemat- incorporating functional traits. Journal of Animal Ecology 77: ics 14: 189–211. 661–669.
Journal of Vegetation Science Doi: 10.1111/jvs.12013 © 2013 International Association for Vegetation Science 805 Functional diversity and stress N.W.H. Mason et al.
Mason, N.W.H., Irz, P., Lanoiselee, C., Mouillot, D. & Argillier, evolutionary processes? Journal of Vegetation Science 24: 781– C. 2008b. Evidence that niche specialization explains species 793. –energy relationships in lake fish communities. Journal Peltzer, D.A., Wardle, D.A., Allison, V.J., Baisden, W.T., Bardg- of Animal Ecology 77: 285–296. ett, R.D., Chadwick, O.A., Condron, L.M., Parfitt, R.L., Por- Mason, N.W.H., Carswell, F.E., Richardson, S.J. & Burrows, L.E. der, S., Richardson, S.J., Turner, B.L., Vitousek, P.M., 2011a. Leaf palatability and decomposability increase during Walker, J. & Walker, L.R. 2010. Understanding ecosystem a 200-year-old post-cultural woody succession in New Zea- retrogression. Ecological Monographs 80: 509–529. land. Journal of Vegetation Science 22: 6–17. Peres-Neto, P.R., Olden, J.D. & Jackson, D.A. 2001. Environ- Mason, N.W.H., de Bello, F., Dolezal, J. & Leps, J. 2011b. Niche mentally constrained null models: site suitability as occu- overlap reveals the effects of competition, disturbance and pancy criterion. Oikos 93: 110–120. contrasting assembly processes in experimental grassland Petchey, O.L. & Gaston, K.J. 2002. Functional diversity (FD), communities. Journal of Ecology 99: 788–796. species richness and community composition. Ecology Letters Mason, N.W.H., Richardson, S.J., Peltzer, D.A., Wardle, D.A., de 5: 402–411. Bello, F. & Allen, R.B. 2012. Changes in co-existence mecha- Petchey, O.L., Hector, A. & Gaston, K.J. 2004. How do different nisms along a long-term soil chronosequence revealed by measures of functional diversity perform? Ecology 85: 847–857. functional trait diversity. Journal of Ecology 100: 678–689. Rajaniemi, T.K. 2003. Explaining productivity–diversity rela- Mouchet, M.A., Villeger, S., Mason, N.W.H. & Mouillot, D. tionships in plants. Oikos 101: 449–457. 2010. Functional diversity measures: an overview of their Rao, C.R. 1982. Diversity and dissimilarity coefficients: a unified redundancy and their ability to discriminate community approach. Theoretical Population Biology 21: 24–43. assembly rules. Functional Ecology 24: 867–876. Richardson, S.J., Peltzer, D.A., Allen, R.B., McGlone, M.S. & Par- Mouillot, D., Mason, N.W.H. & Wilson, J.B. 2007. Is the abun- fitt, R.L. 2004. Rapid development of phosphorus limitation dance of species within local communities determined by in temperate rainforest along the Franz Josef soil chronose- their functional characters? Oecologia 152: 729–737. quence. Oecologia 139: 267–276. Mouillot, D., Villeger, S., Scherer-Lorenzen, M. & Mason, Richardson, S.J., Peltzer, D.A., Allen, R.B. & McGlone, M.S. N.W.H. 2011. Functional structure of biological communities 2005. Resorption proficiency along a chronosequence: predicts ecosystem multifunctionality. PLoS One 6: e17476. responses among communities and within species. Ecology Mouquet, N., Leadley, P., Meriguet, J. & Loreau, M. 2004. 86: 20–25. Immigration and local competition in herbaceous plant Richardson, S.J., Williams, P.A., Mason, N.W.H., Buxton, R.P., communities: a three-year seed-sowing experiment. Oikos Courtney, S.P., Rance, B.D., Clarkson, B.R., Hoare, R.J.B., 104: 77–90. St. John, M.G. & Wiser, S.K. 2012. Rare species drive local Mu¨ nkemu¨ ller, T., de Bello, F., Meynard, C.N., Gravel, D., Laver- trait diversity in two geographically disjunct examples of a gne, S., Mouillot, D., Mouquet, N. & Thuiller, W. 2012. From naturally rare alpine ecosystem in New Zealand. Journal of diversity indices to community assembly processes: a test Vegetation Science 23: 626–639. with simulated data. Ecography 35: 468–480. Roxburgh,S.H.&Matsuki,M.1999.Thestatisticalvalidationofnull Pakeman, R.J. 2011. Functional diversity indices reveal the models used in spatial association analyses. Oikos 85: 68–78. impacts of land use intensification on plant community Schleuter, D., Daufresne, M., Massol, F. & Argillier, C. 2010. assembly. Journal of Ecology 99: 1143–1151. A user’s guide to functional diversity indices. Ecological Mono- Paradis, E., Claude, J. & Strimmer, K. 2004. APE: analyses of graphs 80: 469–484. phylogenetics and evolution in R language. Bioinformatics 20: Schwinning, S. & Weiner, J. 1998. Mechanisms determining the 289–290. degree of size asymmetry in competition among plants. Oeco- Pavoine, S. & Bonsall, M.B. 2010. Measuring biodiversity to logia 113: 447–455. explain community assembly: a unified approach. Biological Spasojevic, M.J. & Suding, K.N. 2012. Inferring community Reviews 4: 792–812. assembly mechanisms from functional diversity patterns: the Pavoine, S. & Doledec, S. 2005. The apportionment of quadratic importance of multiple assembly processes. Journal of Ecology entropy: a useful alternative for partitioning diversity in 100: 652–661. ecological data. Environmental and Ecological Statistics 12: 125– Villeger, S., Mason, N.W.H. & Mouillot, D. 2008. New multidi- 138. mensional functional diversity indices for a multifaceted Pavoine, S., Vallet, J., Dufour, A.B., Gachet, S. & Daniel, H. framework in functional ecology. Ecology 89: 2290–2301. 2009. On the challenge of treating various types of variables: Walker, L.R. & Chapin, F.S. 1986. Physiological controls over application for improving the measurement of functional seedling growth in primary succession on an Alaskan flood- diversity. Oikos 118: 391–402. plain. Ecology 67: 1508–1523. Pavoine, S., Gasc, A., Bonsall, M.B. & Mason, N.W.H. 2012. Cor- Wardle, D.A. & Zackrisson, O. 2005. Effects of species and func- relation levels between phylogenetic and functional diver- tional group loss on island ecosystem properties. Nature 435: sity: mathematical artefacts or true ecological and 806–810.
Journal of Vegetation Science 806 Doi: 10.1111/jvs.12013 © 2013 International Association for Vegetation Science Journal of Vegetation Science 24 (2013) 807–819 SPECIAL FEATURE: FUNCTIONAL DIVERSITY Which trait dissimilarity for functional diversity: trait means or trait overlap? Francesco de Bello, Carlos P. Carmona, Norman W. H. Mason, Maria-Teresa Sebastia` & Jan Lepsˇ
Keywords Abstract Biodiversity; Community assembly; Environmental filtering; Functional traits; Question: Many functional diversity indices require the calculation of func- Intraspecific trait variability; Niche tional trait dissimilarities between species. However, very little is known about complementarity; Rao how the dissimilarity measure used might affect conclusions about ecological processes drawn from functional diversity. Received 30 March 2012 Accepted 25 September 2012 Methods: We simulated real applications of functional diversity, to illustrate Co-ordinating Editor: Martin Zobel the key properties of the two most common families of dissimilarity measures: (1) ‘Gower’ distance, using only ‘mean trait’ value per species and then stan- dardizing each trait, e.g. relative to its range; (2) ‘trait overlap’ between species, de Bello, F. (corresponding author, [email protected]): Institute of Botany, which takes into account within-species trait variability. We then examine how Academy of Sciences of the Czech Republic, these approaches could affect conclusions about ecological processes commonly Dukelska´ 135, CZ-379 82, Trˇebonˇ ,Czech assessed with functional diversity. We also propose a new R function (‘trova’, Republic and i.e. TRait OVerlAp) which performs computations to estimate species trait Department of Botany, Faculty of Sciences, dissimilarity with different types of data. University of South Bohemia, Na Zlate Stoce 1, CZ-370 05, Cˇ eske´ Budeˇ jovice, Czech Republic Results: The trait overlap approach generally produces a less context-depen- Carmona, C.P. ([email protected]): dent measure of functional dissimilarity. For example, the results are less depen- Terrestrial Ecology Group, Department of dent on the transformation of trait data (often required in empirical datasets) Ecology, Autonomous University of Madrid, and on the particular pool of species considered (i.e. trait range, regularity and 28049, Madrid, Spain presence of outliers). The results therefore could be more easily compared across Mason, N.W. H. ([email protected]): Landcare studies and biomes. Further, trait overlap more reliably reproduces patterns Research NZ Ltd, Private Bag 3127, Hamilton, expected when niche differentiation structures communities. The Gower New Zealand approach, on the contrary, more reliably detects environmental filtering effects. Sebastia`, M.-T. ([email protected]): Forestry and Technology Centre of Catalonia, Conclusion: The two approaches imply different conceptions of how species E-28240, Solsona, Spain dissimilarities relate to niche differentiation. Trait overlap is suitable for testing Lepsˇ,J.([email protected]): Department of the effect of species interactions on functional diversity within local communi- Botany, Faculty of Sciences, University of ties, especially when relatively small differences in species traits are linked to dif- South Bohemia, Na Zlate´ Stoce 1, CZ-370 05, ferent resource acquisition. Gower is better suited to detecting changes in ˇ Ceske´ Budeˇ jovice, Czech Republic and functional diversity along environmental gradients, as greater differences in trait Institute of Entomology, Biology Centre, values reflect increased niche differentiation. Combining trait overlap and Gow- Academy of Sciences of the Czech Republic, CZ-370 05, Cˇ eske´ Budeˇ jovice, Czech Republic er approaches may provide a novel way to assess the joint effects of environmen- tal filtering and niche complementarity on community assembly. We suggest that attention should be given not only to the index of functional diversity considered but also whether the dissimilarity used is appropriate for the study context.
phylogenetic diversity, many indices have been proposed Introduction to estimate functional diversity, each showing different Functional diversity, the extent of trait differences between properties (Mouchet et al. 2010; Schleuter et al. 2010; species, is a key component of biodiversity (Petchey & Pavoine & Bonsall 2011). A common requirement Gaston 2002; Mason et al. 2005). As for taxonomical and for many functional diversity indices is estimating ‘trait
Journal of Vegetation Science Doi: 10.1111/jvs.12008 © 2012 International Association for Vegetation Science 807 Gower vs overlap for functional diversity F. de Bello et al. differences’ between species (Fig. 1). This is generally the popular packages to compute functional diversity. resolved by computing a pairwise species trait dissimilarity Gower distance is quite useful for combining different matrix (Pavoine et al. 2009; Laliberte & Legendre 2010), types of traits (e.g. quantitative, qualitative, semi-quantita- i.e. a matrix which contains the trait dissimilarity for each tive), it can deal with missing trait values, and the impor- pair of species in a dataset. This matrix can be based on sin- tance of individual traits can be weighted differently gle traits or on the combination of multiple traits (Botta- according to their importance (Pavoine et al. 2009; Lali- Dukat 2005; Lepsˇ et al. 2006; Mason et al. 2011). While berte & Legendre 2010). several studies have explored the behaviour of the existing As mentioned, for a given trait the Gower approach con- functional diversity indices (Villeger et al. 2008; Mouchet siders only a single mean trait value per species. However, et al. 2010; Schleuter et al. 2010), quite surprisingly very it is increasingly recognized that species present a consider- little is known about how the methods used to compute able amount of intraspecific trait variability, both within the trait dissimilarity matrix affect functional diversity val- and across communities (de Bello et al. 2010b; Hulshof & ues. In particular, it remains unclear how different meth- Swenson 2010; Violle et al. 2012). Using only a single ods for calculating dissimilarity might influence mean trait value for species, as with the Gower approach, conclusions about ecological processes drawn from func- obviously ignores this intraspecific trait variability (de tional diversity (see Mason & de Bello this issue, for the set Bello et al. 2010a; Lepsˇ et al. 2011; Albert et al. 2012). of ecological processes most commonly assessed with func- Another potential problem of the Gower approach is that tional diversity). different traits vary on different orders of magnitude. For To our knowledge, there are two main families of example the range of variation in height and seed mass approaches for estimating this trait dissimilarity matrix. described in Cornelissen et al. (2003) is 10À1 to 102 mand For illustration, let us consider here the case of a single 10À3 to 107 mg respectively. Transformations are, there- quantitative trait (we discuss below also cases combining fore, often required to account for lack of normality of trait different types of traits). The first family of approaches was values (Westoby 1998). It is unclear what is the effect of described by Botta-Dukat (2005) and Pavoine et al. (2009) trait range variations and potential transformations on and it is based on the popular Gower distance (Gower describing trait dissimilarity patterns. 1971). This approach, which we will call ‘Gower’ for sim- The second family of approaches to compute trait dis- plicity, considers only a single mean trait value per species similarity, which we call ‘Overlap’, has been proposed by to estimate trait dissimilarity. For quantitative traits, dis- different authors, although with different algorithms (Mac similarity is simply computed as the difference in mean Arthur & Levins 1967; Mouillot et al. 2005b; Lepsˇ et al. trait values between species. To allow comparisons across 2006; Mason et al. 2008, 2011; Geange et al. 2011). For a different traits, this difference is standardized between 0 given quantitative trait, the approach estimates the overlap and 1 for each trait. This standardization can be obtained in trait distribution between species, i.e. considering intra- in different ways, but mostly the trait difference between specific trait values (Fig. 1). The dissimilarity is commonly each species is divided by the spread of trait values existing computed as one minus the overlap in trait distribution in the dataset. The original Gower distance (Gower 1971) (given that the area of the density curve of trait distribu- divides, for example, the trait differences by the trait range tion is equal to one). The approach was designed to esti- (Botta-Dukat 2005). This approach is now used in most of mate overlap in resource use and acquisition between
Fig. 1. The trait dissimilarity between species, normally expressed by a triangular or symmetrical matrix, is essential to compute most functional diversity indices. Depending on the data available two families of approaches exist to compute it. The first basically expresses differences between species’ mean trait values and standardizes trait values based on trait range (or similar). The second expresses trait overlap between the curves of trait distribution for each pair of species (**Here estimated with mean and standard deviation of trait values, but other calculations are possible, see text).
Journal of Vegetation Science 808 Doi: 10.1111/jvs.12008 © 2012 International Association for Vegetation Science F. de Bello et al. Gower vs overlap for functional diversity species, thus providing a means of testing for the effects of illustrate the general behaviour of these methods for sev- interspecific competition on community structure (Mac eral key applications of functional diversity. We claim that Arthur & Levins 1967; Mason et al. 2011). There are gen- no approach is better a priori but that users should be erally two ways to compute trait overlap. The first assumes aware of their potential advantages and pitfalls. Finally, we trait values to have a normal distribution around the trait considered only one trait but we discuss the implication of mean (Mac Arthur & Levins 1967; Lepsˇ et al. 2006). For our tests for combining multiple traits together. this approach is it necessary to estimate, for each species, the mean and standard deviation for each trait to build Test 1: Do Gower and overlap produce analogous normal trait distribution curves (Fig. 1). The second dissimilarity values? approach does not assume that trait density curves follow a Approach normal distribution. Rather, it uses kernel density estima- tors to build trait density curve, which requires no assump- The first test is a simple illustration of the expected distri- tions about curve shape (Mouillot et al. 2005b; Geange bution of dissimilarity values produced by the Gower and et al. 2011; Mason et al. 2011). The second is certainly Overlap approaches. This shows what kind of values are more realistic but probably requires more field measure- generally expected with both methods. We simulated ments. Our experience suggests these two methods give quantitative trait values for 20 species (Fig. 2, left panel). similar results, particularly relative to the Gower approach, We first defined the mean trait value for each species, with and therefore we do not compare them here. mean values increasing across species. For illustration pur- This study aims to comparing the two existing families poses we only show the case of mean values increasing on of approaches (i.e. ‘Gower’ vs ‘Overlap’) to compute trait a logarithmic scales, but similar patterns were observed dissimilarity. We develop several tests, mimicking com- when considering log and non-log scales (Test 2). Then we mon applications of functional diversity, to assess and dis- randomly created 20 individuals per species. For each spe- cuss the main properties and applications of these two cies we randomly selected 20 trait values from a normal approaches. For the different tests described below we fol- distribution, with the standard deviation (SD) proportional low a common approach using simulated trait values for a to trait mean, so that all the species have similar coefficient set of hypothetical species. In the following sections, we of variation (CV, i.e. SD divided by trait mean). The first outline each test (summarized in separate figures; i.e. increase of SD values with increased mean is a common Figs 2–7) and the implications of the results obtained with pattern in many traits (de Bello et al. 2010b). The CV was each test. Then we discuss more generally the applications fixed at 26%, a value within the range of values most often of these approaches and the biological implications of our found for many traits (Cornelissen et al. 2003). For results findings. We selected a number of tests that, we believe, with other values of CV see Appendix S1. After producing
Fig. 2. Left panel: data simulated with 20 species having log-increasing mean trait values and coefficient of variation around the mean of ca. 26%. Central and right panels: typical histograms of trait dissimilarity between species obtained using Gower distance vs trait overlap (central and right panel respectively). Most values are close to zero with trait mean and close to one with trait overlap. For two species i and j (black circles in the left panel) the corresponding dissimilarity is shown for both methods (central and right panel respectively).
Journal of Vegetation Science Doi: 10.1111/jvs.12008 © 2012 International Association for Vegetation Science 809 Gower vs overlap for functional diversity F. de Bello et al. trait values for the 20 individuals per species, and log- Results transforming these values, we computed a new mean and SD of trait values per each species. Given that the individ- For the Overlap approach, transformation of trait data had ual values were generated randomly, the final values of a minimal effect on dissimilarity values (Fig. 3). By con- CV were not exactly 26% but very close to that quantity. trast, for the Gower approach, dissimilarity values were It should be noted also that the trait values within species, greatly affected by transformation of the trait data. We which were normally distributed before log-transforma- expect that they will vary more when the original range of tion, are slightly right-skewed (i.e. there are more small trait values is greater, because the effect of log-transforma- values and few large ones compared to the mean, Fig. 1; tion on raw trait data is greater when the trait range con- however, this skewness is negligible with a CV of 26%). sidered is greater. Comparing Gower vs Overlap values Then we computed trait dissimilarity between species (Fig. 3 lower panels, with and without log-transformation) either using the Gower approach or the Overlap approach shows in which conditions the two approaches are compa- (based on mean and SD). rable. Only after log-transformation and at low levels of dissimilarity (<0.2 in this specific case for the Gower approach) there is a correlation between the distances cal- Results culated with the Gower and Overlap methods. With log- The trait dissimilarity values obtained with the Gower dis- transformation, there seems to be a threshold above which tance were much lower than those obtained with the values with the Overlap approach tend to 1 and the corre- Overlap approach (i.e. the distribution was skewed lation is disrupted. Without transformation of trait data the towards smaller values; Fig. 2). The patterns observed in two approaches provide rather incomparable dissimilarity this example were also observed in all other examples in values. this paper and, most importantly, even across most of the real datasets for which we have made this comparison (not Test 3: How do the range of trait values in a pool of shown for simplicity). With the Gower approach there are species influence dissimilarity? many more values close to zero and only one single pair of Approach species has dissimilarity equal to 1 (i.e. the combination of species with highest and lowest trait values in the dataset). Very often researchers compare functional diversity values With Overlap, many species have trait dissimilarity equal, across different vegetation types or before/after producing or approaching, to one. Since overlap tends to 0 when spe- some experimental modification (Freschet et al. 2011; cies have sufficiently different trait means and small SD, Mason & de Bello 2013). This implies comparing func- dissimilarity will be close to one. An increase in SD values tional diversity across pools of species with, for example, will result in higher overlap and, therefore, generally different trait ranges. We expected that the results, and lower dissimilarity (Appendix S1). consequent biological interpretations, of these compari- sons could indeed depend on the way species dissimilarities Test 2: How much do trait transformations affect are estimated. To show this we simulated four scenarios, dissimilarity? each with a pool of 65 species but with different trait Approach ranges (‘XL’, ‘L’, ‘M’, ‘S’). The trait mean across all 65 spe- cies was the same across all four scenarios. Within each One common problem when using traits to compute func- scenario, species trait mean values were randomly gener- tional diversity is transformation of trait values. Transfor- ated with a normal distribution around this fixed value mation alters the relative distances between species in trait (i.e. the same in all the four scenarios). The difference space, i.e. by magnifying distances between some species across the four scenarios was that the normal distribution pairs and ‘shrinking’ distances between others and their of species trait values around this fixed mean was set from effect will depend on the original spread of trait values. It is larger to smaller (i.e. (‘XL’, ‘L’, ‘M’, ‘S’ scenarios). We unclear to which extent this will affect the results. There- assumed here that the SD of the trait values within species fore, we followed the same approach as for the Test 1, but was constant for all species and scenarios (as in the case of expanding the test to 65 species. The trait values were either log-transformed data in test 1). We did this because we log-transformed (as for Test 1) or not before the calculation wanted to ensure that species with a given difference in of trait mean and SD. It should be noted that transforming trait means would have the same trait dissimilarity, irre- trait values increases here also the evenness of species distri- spective of their trait mean, to allow comparisons across bution along the trait gradient. Using raw or log-trans- different trait ranges. We simulated community samples formed data, therefore, also mimics the case of considering (20 communities per scenario), by randomly selecting 12– a more, or less, even trait distribution in a dataset. 15 species from the whole pool of 65 species (Fig. 4). The
Journal of Vegetation Science 810 Doi: 10.1111/jvs.12008 © 2012 International Association for Vegetation Science F. de Bello et al. Gower vs overlap for functional diversity
Fig. 3. The dissimilarity with the Gower distance (left-above panel) is much more variable, compared to trait overlap (right-above panel) with and without transformation of the data (here log-transformation). Comparing trait dissimilarities based on trait overlap and Gower distance (lower panels) show some correlation between the methods only at lower values of dissimilarity and only with log-transformed data. assumption of a reduction in species richness of around i.e. species relative abundance being equal to 1/species 80% from the species pool to the community level was richness). As for many existing indices of functional diver- based on field observations (Pa¨rtel et al. 1996). Following sity, the trait dissimilarity is an essential parameter for the this first step, we then added a new species to each of the calculation of the Rao index (Pavoine & Bonsall 2011). For four scenarios. This species was intended to be an outlier this and the following tests, we also performed calculations increasing the existing trait range, and was assigned a using another functional diversity index, the one by Pet- mean trait value equal to 1.5 times the maximum value in chey & Gaston (2002). This index is defined as the total the trait range for each scenario (Fig. 4). We then recalcu- branch length in a trait dendrogram connecting all species, lated trait dissimilarity between species (either with Gower and was chosen because it should provide very different or with Overlap). results from the Rao index. We then computed the Rao index of functional diversity (Rao 1982; Botta-Duka´t 2005; Lepsˇ et al. 2006) for each Results and implications sample, using the matrices of dissimilarity between species computed using either the Gower approach or the Overlap This test shows two important patterns, both observed approach. We used the Rao index expressed in terms of with the Rao index (Fig. 4) and with the index of Petchey equivalent numbers (de Bello et al. 2010a) but we also dis- & Gaston (2002); Appendix S2). First, Overlap seems better cuss results in the context of other indices (see Appendix able to detect differences between XL, L, M and S scenar- S2 and Discussion). The Rao index is a key index of func- ios, with functional diversity decreasing from XL to S. This tional diversity, which reflects the property of various simi- confirms that Gower distance comparisons across data sets lar functional diversity indices (Pavoine & Bonsall 2011). should be done carefully as it depends on the pool of spe- It expresses the sum of trait dissimilarities between each cies used for its standardization. For example, consider pair of species in a sample weighted by species relative comparing functional diversity of grasslands and forests abundance (in all our examples with functional diversity, with height as a trait. The range of trait values will be we assume equal abundance of species in a community, clearly lower in grasslands and, intuitively, the functional
Journal of Vegetation Science Doi: 10.1111/jvs.12008 © 2012 International Association for Vegetation Science 811 Gower vs overlap for functional diversity F. de Bello et al.
Fig. 4. Functional diversity computed with Gower (left panel) or trait overlap (right panel), depending on the range of trait values considered (from largest ‘XL’, to smallest ‘S’) with and without the inclusion of a new species outside this range. The scheme above the results shows the simulation approach, with 65 species equally spaced on a trait range. A total of 20 plots per range were built by adding 12–15 species randomly to each plot. The outlier species was set as having 1.5 times the trait value of the species with highest trait value in a given range. diversity should be lower. However if the trait standardiza- tive result arises because range-standardised distances tion is done for each vegetation type separately the values between pre-existing species decrease when the observed of functional diversity obtained with the Gower distance range of trait values increases. Consequently, outliers could be artificially similar across vegetation types (Fig. 4). strongly affect the mean dissimilarity values obtained with With the Overlap approach this problem does not seem to the Gower approach. It also implies that studies considering exist or, more generally, the range of trait values consid- repeated measures of species composition should carefully ered should influence much less the results of functional consider how to compute the dissimilarity with Gower diversity. As for Test 2, this test shows the context-depen- across repeated measures, if they do not want to artificially dence of dissimilarities values obtained using the Gower affect the variation in functional diversity. Most ecologists, approach, because the values depend on the standardiza- including our selves (Hejda & de Bello 2013), would avoid tion of trait differences. As we show with Test 2, standard- these problems by considering both invaded and non izations are not necessary with Overlap. invaded conditions together when estimating the observed Second, with the Gower approach the addition of an out- range of trait values. lier (or any new species in the dataset having sufficiently different traits from existing species) can produce some Test 4: How much do dissimilarity estimations counterintuitive results of functional diversity. Consider influence the tests on species niche differentiation the case of repeatedly measured of plots invaded by an alien and coexistence? species having extreme trait values relative to existing spe- Approach cies. Assuming, for simplicity, that the existing species remained, this should intuitively increase functional diver- One increasingly common application of functional diver- sity in the plots. The Overlap approach shows this pattern, sity is the study of niche differentiation within communi- while the opposite is found with Gower. This counterintui- ties (i.e. differences in alpha niches; Silvertown et al.
Journal of Vegetation Science 812 Doi: 10.1111/jvs.12008 © 2012 International Association for Vegetation Science F. de Bello et al. Gower vs overlap for functional diversity
2006). It is expected that higher functional diversity should tional diversity and the Petchey & Gaston (2002) index, indicate communities with higher divergence between either using Gower or Overlap to determine species dissim- species in terms of traits, consistent with the hypothesis of ilarity. Finally, we also assessed the relationship between limiting similarity, i.e. in order to coexist species with dif- species richness and functional diversity, both with Rao ferent traits occupy different niches within a community and Petchey and Gaston indices. In this case, out of the pool (Mac Arthur & Levins 1967). Niche differentiation could of 65 species we randomly selected 1000 communities with arise, for a given trait, if species are evenly distributed in richness varying from 3 to 40 species. trait space (i.e. functional evenness Mason et al. 2005 or functional regularity Mouillot et al. 2005b). Results and implications Ideally, if we need an index of functional diversity to provide a general indication of niche differentiation Using the Rao index with the Gower approach, the com- between species, it should take higher values in communi- munities in the random assembly scenario wrongly ties with an even trait distribution (all else being equal). showed the same functional diversity as the community While similar tests have been applied to many indices of having even-spacing of trait values. By contrast, with the functional diversity (Mouchet et al. 2010), we further Overlap approach, functional diversity was higher when tested if this expectation was met using either Gower or trait values were evenly distributed. The reason for these Overlap. Again, as for test 4, we considered for this test the patterns is that, as we discussed above (Test 1), the highest Rao index although we discuss further the implications for values of dissimilarities with the Gower index are only considering other functional diversity indices (Appendix obtained between the species with most different trait val- S2 and below). We considered the pool of 65 species simu- ues. The same problem, on the other hand, does not apply lated for Test 2. Then we assembled communities either to the Petchey and Gaston index (Appendix S2). However, random or with even-spacing of trait values. With the ran- it is well known that this index is very strongly influenced dom scenario, species were randomly selected across the 65 by the number of species (which was constrained here). species. With the even-spacing scenario we fixed a minimal Therefore we claim that, with the Gower approach, vari- distance between species with closest trait values (Fig. 5). ous indices of functional diversity similar to Rao (Pavoine This minimal distance was set, with the Gower approach, & Bonsall 2011) will show the highest values only when as to be between 3/65 and 5/65 (in order to produce 13–21 species having maximal and minimal trait values in the species). We then computed both the Rao index of func- species pool are included the community (i.e. functional
Fig. 5. Functional diversity (with the Rao index) patterns in communities assembled randomly or with even spacing in trait values (20 communities per scenario, each containing 13–21 species out of the 65 in the species pool). Random communities were assembled with species sorted randomly from the species pool. Even spaced communities were assembled by excluding all species too close on a trait gradient. Functional diversity, expressed here as the Rao index, should be higher with even spacing scenario reflecting higher niche differentiation.
Journal of Vegetation Science Doi: 10.1111/jvs.12008 © 2012 International Association for Vegetation Science 813 Gower vs overlap for functional diversity F. de Bello et al. diversity will depend, at least partially, on the range of trait tering along ecological gradients (Freschet et al. 2011; values). This example also serves to highlight the problem Mason et al. 2013). Environmental filtering implies that of interpreting a general index of functional diversity like variation across sites in their environmental conditions Rao. We currently lack pure indices of functional evenness causes different trait values to be selected for (i.e. the traits that meet all necessary criteria (Mason et al. 2005; Mouil- conferring greatest fitness vary according to local environ- lot et al. 2005a). For example, the most commonly used mental conditions; Go¨tzenberger et al. 2012). This gives index of functional evenness (Villeger et al. 2008) is rise to the often-observed pattern where species occurring affected by the trait range considered (lower trait range at a given site are more similar in their traits than species having higher chances to detect even-spacing). Other indi- from different sites (Dı´az et al. 1998; Go¨tzenberger et al. ces could be considered, after validation (Kraft & Ackerly 2012). In addition to this, it has long been theorised that 2010; Thuiller et al. 2010b). Consequently, it remains dif- environmental filtering influences species occurrences and ficult to isolate functional evenness from other functional abundances more strongly in stressed environments (Wei- diversity components, but the Rao index using trait Over- her & Keddy 1995; Mason et al. 2008; Pakeman 2011). lap could provide an interesting measure of functional This should lead to lower functional diversity at stressed vs evenness. benign environments (Mason et al. 2008; Carmona et al. Another pattern that we detected is a changing relation- 2012; Munkemuller et al. 2012). ship between species richness and functional diversity. We tested the behaviour of Rao functional diversity Using both the Rao index (Fig. 6) and the Petchey & computed with Gower and Overlap to detect these Gaston index (Appendix S2), we detected more linear pat- expected patterns. We defined a pool of 300 species having terns between species richness and functional diversity mean trait values regularly spaced between 1 and 300. We with overlap. Most of the dissimilarity values are higher then simulated 300 plots having a different mean trait (i.e. many values are close to 1) with overlap and species value (i.e. the mean of species trait values) across an envi- will be considered functionally different. This means that ronmental gradient. The mean trait value of the plots was with overlap there are greater chances to get a positive set to decrease with the environmental gradient. To simu- correlation between species richness and Rao compared to late environmental filtering, within each plot only a the Gower approach. reduced range of trait values, compared to the whole pool of 300 species, was allowed around the plot mean. We then Test 5: How much do dissimilarity estimations considered two scenarios where the range of trait values influence the tests on environmental filtering? around the plot mean was set either (1) to decrease with Approach the gradient (and therefore with the site trait mean) or (2) increase (Fig. 6). This simulates a case where the impor- Another increasingly common application of functional tance of environmental filtering either increases or diversity is to quantify the influence of environmental fil- decreases along the environmental gradient. Thus, the
Fig. 6. Relationship between species richness and functional diversity (with the Rao index) in 1000 communities randomly assembly out of the 65 in the species pool used in previous tests. See Appendix S2 for results on other functional diversity indices.
Journal of Vegetation Science 814 Doi: 10.1111/jvs.12008 © 2012 International Association for Vegetation Science F. de Bello et al. Gower vs overlap for functional diversity range of trait values allowed within a plot was lower at one These results suggest that even a simple study assessing extreme of the trait gradient (8% of the trait range of whole changes in functional diversity across an environmental pool of species) compared to the other (40%). Our idea was gradient should carefully consider various details of trait to simulate a large environment gradient, say an altitudinal dissimilarity computation before deriving conclusions gradient, in which environmental filtering was increasing, about environmental filtering. Users could also consider, or decreasing, towards one extreme of the gradient. For before calculations, whether the trait ranges in communi- each plot, a fixed number of species (15) were randomly ties increase with community trait means and whether selected from within the range of trait values allowed. data transformation is really required. We then computed trait dissimilarity (either with Gower or Overlap) and functional diversity values (again with the Discussion Rao index). For the Gower approach we used either raw or log-transformed trait data (this creates more ‘compacted’ In this study we show that the method used to compute trait values with increasing trait mean). For the Overlap species trait dissimilarity can have profound consequences approach we either considered the SD of trait values for for the detection of expected ecological patterns assessed each species to increase with trait mean (i.e. SD = 30% of using functional diversity. We claim that the two trait mean) or was considered as fixed (i.e. for all species it approaches considered (Gower vs Overlap), when applied was fixed to 10, so that species having a trait difference of carefully, are both viable but one should be aware of the 30 would have a dissimilarity around 0.85). assumptions, potential applications and pitfalls (Table 1). We also claim that, for a given trait, the Overlap approach Results and implications is (1) generally less context dependent for distance-based measures of functional diversity and (2) takes into account Both the Gower and Overlap approaches generally pro- intraspecific trait variability of species, which is most often duced the expected pattern of decreasing functional diver- neglected in functional ecology (Albert et al. 2012; Violle sity toward one end of the environmental gradient. et al. 2012). However in some specific conditions these patterns were We also claim that the two approaches (Gower vs Over- not properly revealed and users should be aware of possi- lap) imply rather different conceptions of the linkage ble existing limitations. The problem, in this case, seems between species dissimilarity and niche differentiation slightly more serious with the Overlap approach and when between species and are therefore suited to different appli- both (1) the SD of species trait values increases with species cations. The Overlap approach was created mainly to trait means and (2) when this occurs in combination with understand differences in terms of niche within a commu- community samples where the range of trait values nity, i.e. what kind of resources are used by coexisting spe- decreases with plot trait mean (for example when vegeta- cies and if species compete or not for the same resource tion with lower species has also a lower range of height (Mac Arthur & Levins 1967; Mouillot et al. 2005b; Lepsˇ values, which is a rather likely scenario; de Bello et al. et al. 2006). This implies that traits, and trait differences, 2012). In this case, species with greater trait values and are used to infer the ‘alpha’ niche of species (Silvertown greater SD will occur in communities with higher trait et al. 2006) and it suggests that trait overlap could help range. However, because of increased SD for a given dis- focussing particularly on niche differentiation within com- tance between species means, overlap will be greater than munities. As such Overlap could help depicting all biotic at the other extreme of the gradient (where species have mechanisms driving the coexistence of species, such as the lower SD in plots within a lower plot trait range). In this effect of competition is increasing, or even decreasing, the case the Gower approach, which is really proportional to functional differentiation between species (Mayfield & differences in trait means, should better discriminate Levine 2010; Mason et al. 2011). changes in FD across a gradient. The main risk with the It should be noted, that the Overlap approach generally Gower approach occurs in the same scenario (i.e. commu- assumes that even small differences between species may nities with higher trait mean having higher trait range), suggest that they occupy a different niche. This might be and when already normally distributed trait values are reasonable in some occasions but not in others. It might be unnecessarily log-transformed. In this case, again, func- mostly appropriate when small differences in trait values tional diversity does not vary along the environmental gra- imply differences in the mode of resource acquisition, or dient as expected. This is because when higher trait ranges the types of resources used. This applies, for instance, food correspond with higher species mean trait values log-trans- type for animals in the original concept of niche overlap formation shrinks distances between species in communi- (May & Mac Arthur 1972). Indeed our examples suggest ties with high range values more than in communities that species could be functionally more different than with low range values. assumed using the Gower approach. However, these
Journal of Vegetation Science Doi: 10.1111/jvs.12008 © 2012 International Association for Vegetation Science 815 Gower vs overlap for functional diversity F. de Bello et al.
Fig. 7. Simulations to show the application of Gower vs Overlap approaches in studying environmental ‘filtering’ effects on functional diversity (see Test 5). Out of a pool of 300 species we assembled 300 plots having each 15 species across an environmental gradient, e.g. an altitudinal gradient. At low altitudes we simulated plots as having higher trait values and either higher trait range (scenario 1) or lower (scenario 2). For the Gower approach the trait data,which were normally distributed, were either log-transformed or not. For the Overlap approach we simulated cases where the standard deviation would increase with trait mean or would be similar for all species. assumptions might not be reasonable when competition and C could be, in both cases, close to 1. In this case the between species increases progressively with differences in Gower approach could be preferred, as greater differences trait values, as for example in the case of size asymmetrical in trait means reflect greater differences in competitive competition. Consider three herbaceous species, species A, ability. More generally, comparing results from the two B and C with heights of 10, 50 100 cm respectively, and a approaches may provide a novel way to differentiate fixed SD for height of 5 cm. In this case let us assume that between different types of competitive impacts on species the tallest species (C) is a superior competitor for light to distribution in trait space. species (B) and B is a superior competitor to the shortest It could be argued that Gower approach might be prefer- species (A). With the Overlap approach discussed here, able when dealing with niche differences across environ- however, the dissimilarity between species A and B and A mental conditions. In this case niche differentiation is
Journal of Vegetation Science 816 Doi: 10.1111/jvs.12008 © 2012 International Association for Vegetation Science F. de Bello et al. Gower vs overlap for functional diversity
Table 1. Summary of the tests conducted and the main conclusions obtained.
Tests Gower Overlap
1. Dissimilarity Lower values, many values approaching zero (distribution Higher values, many values approaching one (distribution values skewed towards lower values) skewed towards one) 2. Trait values Dissimilarity largely dependent on the transformation of Dissimilarity largely independent on the transformation of transformations data (stronger effect when the differences between data species has more order of magnitudes) 3. Range of trait Dissimilarity largely dependent on the pool of species Dissimilarity is largely independent on the pool of species values considered and on the inclusion of outliers, or new considered species in the pool 4. Testing niche differences Even trait spacing is not always easily detected Even trait spacing is more easily detected within communities Species and functional diversity are less correlated Species and functional diversity are more correlated 5. Testing environmental The response of functional diversity to environmental The response of functional diversity to environmental filtering gradients is detected but care is needed gradients is detected but care is needed Conclusion The results are context dependent The results are less context dependent The pool of species considered in the calculation should be The approach can be used to compare across different carefully defined pool of species more easily Preferable for testing environmental filtering on large Preferable for testing niche differentiation and niche environmental gradients overlap within communities
expressed as the beta niche of species, i.e. differentiating dissimilarity matrices; see Botta-Duka´t 2005; Lepsˇ et al. species with a different optima between communities across 2006 and Pavoine et al. 2009). The new R function pro- environmental gradients (Weiher & Keddy 1995; Silver- vided in the supplementary material include all these town et al. 2006). Although even the Overlap approach options, for both the Gower and Overlap approach (for the produced similarly trustable results, in this case it might be Overlap approach both normal and kernel trait distribu- easier to use only trait means because the data are avail- tions are allowed). It should be noted that combining trait able in many existing databases (Kleyer et al. 2008; dissimilarities from multiple traits with the Gower and Klimesˇova´ & de Bello 2009). The approach probably Overlap have, however, a different logic. With Gower, the requires fulfilling fewer requirements in the type of data relative weight of individual traits will be affected by available (see Test 5) and less data collection effort. We spread and distribution of trait value within each trait suggest, anyway, that the species trait mean values used (de Bello et al. 2010b). For example, the relative weight of for calculations should be measured, for each species, in height and specific leaf area (SLA) on functional diversity different environmental condition (Lepsˇ et al. 2011). This values in meadows will depend on whether some forests requires specific field sampling campaigns or simulations, species are in the data set (or if the whole range of trait as discussed by Albert et al. (2012), in order to account for values existing is used for the standardization). If forest changing trait mean values of species along the considered species are included, the effect of height will be down- gradient. When using the Gower approach, however, we weighted in the meadows – because by including forests, recommend special care if using, or not, log-transforma- the range of heights will increase by more than an order of tions. We suggest particularly to log-transform trait values magnitude, whereas no such change will happen for SLA. if SD of species is expected to increase along with trait Therefore, in meadows, species differences in height will mean (as in height). Following of this recommendation have practically no influence on functional diversity calcu- ensures that FD values calculated with the Gower lated using two traits (i.e. height and SLA) if tree species approach behave similarly to those that would have been are used for Gower standardization (Appendix S3). Thus, it calculated using the Overlap approach. might appear that one community is more functionally In this study we show some key properties of the Gower diverse in terms of one trait, while these patterns depend and Overlap approaches based on single traits. Similar only on the mathematical properties of the approach recommendations should are valid when combining (de Bello et al. 2010b). Moreover, the influence of each multiple traits. Combining different traits together, both trait on functional diversity calculated with multiple traits with Gower and Overlap, in fact, implies combining trait will depend on which species are used for standardization. dissimilarities matrices for single traits. This can be done by This is not an issue when distances are calculated using the averaging the trait dissimilarity across all single traits, or Overlap approach. calculating Euclidean distance (i.e. summing the squares Finally, in this paper we dealt mostly with cases of com- of distances and then doing a square root of the result or all puting functional diversity within communities (alpha
Journal of Vegetation Science Doi: 10.1111/jvs.12008 © 2012 International Association for Vegetation Science 817 Gower vs overlap for functional diversity F. de Bello et al. diversity). It is well known that applying functional diver- Freschet, G.T., Dias, A.T.C., Ackerly, D.D., Aerts, R., van Bode- sity indices for measuring functional diversity among com- gom, P.M., Cornwell, W.K., Dong, M., Kurokawa, H., Liu, munities (i.e. beta diversity) further requires that a set G.F., Onipchenko, V.G., Ordonez, J.C., Peltzer, D.A., Rich- properties of trait dissimilarity are fulfilled (Ricotta 2005). ardson, S.J., Shidakov, II, Soudzilovskaia, N.A., Tao, J.P. & Here, we show that alpha functional diversity, which is Cornelissen, J.H.C. 2011. Global to community scale differ- essential for understanding patterns in niche differentia- ences in the prevalence of convergent over divergent leaf tion within and across communities, can be computed trait distributions in plant assemblages. Global Ecology and – with two main families of approaches (Gower and Over- Biogeography 20: 755 765. lap) and that in some cases one approach could be prove Geange, S.W., Piedger, S., Burns, K.C. & Shima, J.S. 2011. A uni- fied analysis of niche overlap incorporating data of different more useful than another (Table 1). types. Methods in Ecology and Evolution 2: 175–184. Go¨tzenberger, L., de Bello, F., Brathen, K.A., Davison, J., Dubuis, Acknowledgements A., Guisan, A., Leps, J., Lindborg, R., Moora, M., Pa¨rtel, M., Pellissier, L., Pottier, J., Vittoz, P., Zobel, K. & Zobel, M. 2012. This research was supported by the Grant Agency of the Ecological assembly rules in plant communities-approaches, Czech Republic (GACR P505/12/1296 and P505/12/1390). patterns and prospects. Biological Reviews 87: 111–127. CPC was supported by a FPI scholarship (BES-2008- Gower, J.C. 1971. General coefficient of similarity and some of 009821). We thank the colleagues in Tomas Herben’s lab its properties. Biometrics 27: 857–871. for the useful discussions on species niche differentiation. Hejda, M. & de Bello, F. 2013. Impact of plant invasion and com- munity invasibility on functional diversity in Central Eur- – References ope. Journal of Vegetation Science 24: 890 897. Hulshof, C.M. & Swenson, N.G. 2010. Variation in leaf func- Albert, C.H., de Bello, F., Boulangeat, I., Pellet, G., Lavorel, S. & tional trait values within and across individuals and species: Thuiller, W. 2012. On the importance of intraspecific vari- an example from a Costa Rican dry forest. Functional Ecology ability for the quantification of functional diversity. Oikos 24: 217–223. 121: 116–126. Kleyer, M., Bekker, R.M., Knevel, I.C., Bakker, J.P., Thompson, Botta-Dukat, Z. 2005. Rao’s quadratic entropy as a measure of K., Sonnenschein, M., Poschlod, P., van Groenendael, J.M., functional diversity based on multiple traits. Journal of Vegeta- Klimesˇ, L., Klimesˇova´, J., Klotz, S., Rusch, G.M., Hermy, M., tion Science 16: 533–540. Adriaens, D., Boedeltje, G., Bossuyt, B., Dannemann, A., Carmona, C.P., Azca´rate, F.M., de Bello, F., Ollero, H.S., Lepsˇ,J. Endels, P., Go¨tzenberger, L., Hodgson, J.G., Jackel, A.K., & Peco, B. 2012. Taxonomical and functional diversity turn- Kuhn, I., Kunzmann, D., Ozinga, W.A., Romermann, C., over in Mediterranean grasslands: interactions between Stadler, M., Schlegelmilch, J., Steendam, H.J., Tackenberg, grazing, habitat type and rainfall. Journal of Applied Ecology, O., Wilmann, B., Cornelissen, J.H.C., Eriksson, O., Garnier, 49: 1084–1093. E. & Peco, B. 2008. The LEDA Traitbase: a database of life- Cornelissen, J.H.C., Cerabolini, B., Castro-Diez, P., Villar-Salva- history traits of the Northwest European flora. Journal dor, P., Montserrat-Marti, G., Puyravaud, J.P., Maestro, M., of Ecology 96: 1266–1274. Werger, M.J.A. & Aerts, R. 2003. Functional traits of woody Klimesˇova´, J. & de Bello, F. 2009. CLO-PLA: the database of clo- plants: correspondence of species rankings between field nal and bud bank traits of Central European flora. Journal adults and laboratory-grown seedlings? Journal of Vegetation of Vegetation Science 20: 511–516. Science 14: 311–322. Kraft, N.J.B. & Ackerly, D.D. 2010. Functional trait and phyloge- de Bello, F., Lavorel, S., Lavergne, S., Albert, C., Boulangeat, I., netic tests of community assembly across spatial scales in an Mazel, F. & Thuiller, W. 2012. Hierarchical effects of envi- Amazonian forest. Ecological Monographs 80: 401–422. ronmental filters on the functional structure of plant com- Laliberte, E. & Legendre, P. 2010. A distance-based framework munities: a case study in the French Alps. Ecography 36: 393– for measuring functional diversity from multiple traits. Eco- 402. logy 91: 299–305. de Bello, F., Lavergne, S., Meynard, C.N., Lepsˇ,J.&Thuiller,W. Lepsˇ, J., de Bello, F., Lavorel, S. & Berman, S. 2006. Quantifying 2010a. The partitioning of diversity: showing theseus a way and interpreting functional diversity of natural communities: out of the labyrinth. Journal of Vegetation Science 21: 992–1000. practical considerations matter. Preslia 78: 481–501. de Bello, F., Lavorel, S., Albert, C.H., Thuiller, W., Grigulis, K., Lepsˇ, J., de Bello, F., Smilauer, P. & Dolezal, J. 2011. Community Dolezal, J., Janecek, S. & Leps, J. 2010b. Quantifying the rel- trait response to environment: disentangling species turn- evance of intraspecific trait variability for functional diver- over vs intraspecific trait variability effects. Ecography 34: sity. Methods in Ecology and Evolution 2: 163–174. 856–863. Dı´az, S., Cabido, M. & Casanoves, F. 1998. Plant functional traits Mac Arthur, R.H. & Levins, R. 1967. Limiting similarity conver- and environmental filters at a regional scale. Journal of Vege- gence and divergence of coexisting species. American Natural- tation Science 9: 113–122. ist 101: 377–385.
Journal of Vegetation Science 818 Doi: 10.1111/jvs.12008 © 2012 International Association for Vegetation Science F. de Bello et al. Gower vs overlap for functional diversity
Mason, N.W.H., Mouillot, D., Lee, W.G. & Wilson, J.B. 2005. application for improving the measurement of functional Functional richness, functional evenness and functional diversity. Oikos 118: 391–402. divergence: the primary components of functional diversity. Petchey, O.L. & Gaston, K.J. 2002. Functional diversity (FD), Oikos 111: 112–118. species richness and community composition. Ecology Letters Mason, N.W.H., Lanoiselee, C., Mouillot, D., Wilson, J.B. & Ar- 5: 402–411. gillier, C. 2008. Does niche overlap control relative abun- Rao, C.R. 1982. Diversity and dissimilarity coefficients – aunified dance in French lacustrine fish communities? A new approach. Theoretical Population Biology 21: 24–43. method incorporating functional traits. Journal of Animal Ricotta, C. 2005. A note on functional diversity measures. Basic Ecology 77: 661–669. and Applied Ecology 6: 479–486. Mason, N.W.H., de Bello, F., Dolezal, J. & Leps, J. 2011. Niche Schleuter, D., Daufresne, M., Massol, F. & Argillier, C. 2010. overlap reveals the effects of competition, disturbance and A user’s guide to functional diversity indices. Ecological Mono- contrasting assembly processes in experimental grassland graphs 80: 469–484. communities. Journal of Ecology 99: 788–796. Silvertown, J., Dodd, M., Gowing, D., Lawson, C. & McConway, Mason, N.W.H. & de Bello, F. 2013. Functional diversity and its K. 2006. Phylogeny and the hierarchical organization of applications. Journal of Vegetation Science 24: 777–780. plant diversity. Ecology 87: S39–S49. May, R.M. & Mac Arthur, R.H. 1972. Niche overlap as a Thuiller, W., Gallien, L., Boulangeat, I., de Bello, F., Munkemul- function of environmental variability. Proceedings of the ler, T., Roquet, C. & Lavergne, S. 2010b. Resolving Darwin’s National Academy of Sciences of the United States of America 69: naturalization conundrum: a quest for evidence. Diversity 1109–1113. and Distributions 16: 461–475. Mayfield, M.M. & Levine, J.M. 2010. Opposing effects of Villeger, S., Mason, N.W.H. & Mouillot, D. 2008. New multidi- competitive exclusion on the phylogenetic structure of com- mensional functional diversity indices for a multifaceted munities. Ecology Letters 13: 1085–1093. framework in functional ecology. Ecology 89: 2290–2301. Mouchet, M.A., Villeger, S., Mason, N.W.H. & Mouillot, D. Violle, C., Enquist, B.J., McGill, B.J., Jiang, L., Albert, C.H., Huls- 2010. Functional diversity measures: an overview of their hof, C.M., Jung, V. & Messier, J. 2012. The return of the vari- redundancy and their ability to discriminate community ance: intraspecific variability in community ecology. Trends assembly rules. Functional Ecology 24: 867–876. in Ecology & Evolution 27: 244–252. Mouillot, D., Mason, W.H.N., Dumay, O. & Wilson, J.B. 2005a. Weiher, E. & Keddy, P.A. 1995. Assembly rules, null models, Functional regularity: a neglected aspect of functional diver- and trait dispersion – new questions from old patterns. Oikos sity. Oecologia 142: 353–359. 74: 159–164. Mouillot, D., Stubbs, W., Faure, M., Dumay, O., Tomasini, J.A., Westoby, M. 1998. A leaf-height-seed (LHS) plant ecology strat- Wilson, J.B. & Chi, T.D. 2005b. Niche overlap estimates egy scheme. Plant and Soil 199: 213–227. based on quantitative functional traits: a new family of non- parametric indices. Oecologia 145: 345–353. Supporting Information Munkemuller, T., de Bello, F., Meynard, C.N., Gravel, D., Laver- gne, S., Mouillot, D., Mouquet, N. & Thuiller, W. 2012. From Additional supporting information may be found in the diversity indices to community assembly processes: a test online version of this article: with simulated data. Ecography 35: 468–480. Pakeman, R.J. 2011. Functional diversity indices reveal the Appendix S1. Further results of the Test 1. impacts of land use intensification on plant community Appendix S2. Test 3 and 4 using the index of func- – assembly. Journal of Ecology 99: 1143 1151. tional diversity of Petchey & Gaston (2002). Pa¨rtel, M., Zobel, M., Zobel, K. & van der Maarel, E. 1996. Appendix S3. Effect of combining two traits for com- The species pool and its relation to species richness: puting functional diversity with the Gower approach. evidence from Estonian plant communities. Oikos 75: 111– 117. Please note: Wiley-Blackwell are not responsible for Pavoine, S. & Bonsall, M.B. 2011. Measuring biodiversity to the content or functionality of any supporting materials explain community assembly: a unified approach. Biological Reviews 86: 792–812. supplied by the authors. Any queries (other than missing Pavoine, S., Vallet, J., Dufour, A.B., Gachet, S. & Daniel, H. material) should be directed to the corresponding author 2009. On the challenge of treating various types of variables: for the article.
Journal of Vegetation Science Doi: 10.1111/jvs.12008 © 2012 International Association for Vegetation Science 819 Journal of Vegetation Science 24 (2013) 820–833 SPECIAL FEATURE: FUNCTIONAL DIVERSITY Does trait conservatism guarantee that indicators of phylogenetic community structure will reveal niche-based assembly processes along stress gradients? Norman W.H. Mason & S. Pavoine
Keywords Abstract Biodiversity; Diversity indices; Environmental filtering; Niche complementarity; Null model; Question: Is the strength of phylogenetic trait conservatism in the species pool Phylogenetic signal; Simpson index; Species reflected in the ability of a-andb-scale phylogenetic diversity indicators to richness; Surrogates; Trait evolution detect niche-based assembly processes along stress gradients?
Nomenclature Methods: We used a simple community assembly model to explore the power NZ Plant Names Database (http:// of indicators of phylogenetic community structure to detect assembly processes. nzflora.landcareresearch.co.nz/default.aspx) The model assumes that with declining stress the influence of niche complemen- tarity on species fitness increases, while that of environmental filtering decreases. Received 29 March 2012 – Accepted 11 November 2012 We separately incorporated two trait-independent stochastic processes mass Co-ordinating Editor: Valerio Pillar and priority effects – in simulating species occurrences and abundances along a hypothetical stress gradient. We measured power for a-andb-scale indices of phylogenetic diversity (PD) as the proportion of simulations yielding a significant Mason, N.W.H. correlation between these indices and the community position along the stress ([email protected]): Landcare gradient. For a-scale indices, we ran simulations where species richness was con- Research, Private Bag 3127, Hamilton, 3240, New Zealand stant along the gradient, or either increased, decreased or varied randomly with Pavoine, S. ([email protected]): Museum declining stress. We used four models of trait evolution, differing in the degree of national d’Histoire naturelle, Departement phylogenetic trait conservatism, to examine how variation in the influence of d’Ecologie et Gestion de la Biodiversite, UMR phylogeny on trait variation impacted the power of PD indices. 7204 CNRS-UPMC, 61 rue Buffon, 75005, Paris, France and Results: None of the PD indices gave high power to detect assembly processes Department of Zoology, Mathematical Ecology for either the mass effects or priority effects model, even when trait variation Research Group, University of Oxford, South was strongly conserved. Those measures of a PD that gave moderate power in Parks Road, Oxford, OX1 3PS, UK the priority-effects model with strong trait conservatism were greatly affected by variation in species richness. Consequently, none of the indices examined met the necessary selection criteria. Additional analyses revealed that trait conserva- tism at the species pool level was poorly maintained at the meta-community scale, with phylogenetically-determined species composition having only mod- erate power to detect functionally-determined species turnover. This lack of a strong phylogenetic signal in trait variation at the meta-community scale is the most likely explanation for the poor ability of PD indices to detect assembly pro- cesses in our study. Conclusion: Phylogenetic diversity indices seem to have limited ability to detect trait-based assembly processes along ecological gradients, even when trait conservatism at the species pool level is strong. Consequently, PD is unlikely to provide a means of avoiding the tough decisions around which traits are most relevant for assembly processes. Rather, our findings emphasize the need to col- lect relevant functional trait data to understand the mechanisms controlling community assembly.
Journal of Vegetation Science 820 Doi: 10.1111/jvs.12033 © 2012 International Association for Vegetation Science N.W.H. Mason & S. Pavoine Phylogenetic diversity and assembly processes
functional evenness and functional divergence. Each com- Introduction ponent provides independent information on the distribu- Phylogeny is increasingly proposed as a means of detecting tion of species and abundance in functional trait space, and niche-based processes of community assembly, especially a separate index is required to quantify each component for species pools where trait variation is phylogenetically (Mouchet et al. 2010). If we imagine phylogenies as conserved (Kraft et al. 2007; Kembel 2009; Swenson & describing the distribution of species relative to each other Enquist 2009). A key advantage of phylogeny over trait- in ‘phylogenetic space’, it is obvious that the concepts based approaches is that it does not require a decision on embodied by the three FD components can be used to which aspects of species function are important for com- define PD as: munity assembly. With trait-based approaches there is 1. Phylogenetic richness: the volume of phylogenetic always a risk that the functional traits measured will fail to space occupied by the species within a community. capture important aspects of function (Violle et al. 2007). 2. Phylogenetic evenness: evenness in the distribution of Many authors have recognized that a potential weakness abundance in phylogenetic space. of phylogenetic approaches to understanding community 3. Phylogenetic divergence: the degree to which the abun- assembly is that strong phylogenetic conservatism in trait dance of a community is distributed toward the extremities variation is far from universal (Blomberg et al. 2003; Cav- of occupied phylogenetic space (Mouchet et al. 2010). ender-Bares et al. 2009; Losos 2011; Pavoine et al. 2013). Indeed, confirmation of trait conservatism is viewed as In this way, we can apply existing theoretical advances necessary justification for studies using phylogeny to in the development of FD indices to obtain robust mea- understand niche-based processes in ecological communi- sures of a-scale PD. This could potentially greatly aid the ties (e.g. Webb et al. 2002; Srivastava et al. 2012). use of phylogeny to study niche-based assembly processes However, there are comparatively few studies testing along environmental gradients, since changes in a-scale whether trait conservatism at the species pool level is suffi- PD along a gradient may be indicative of changes in assem- cient to guarantee that indicators of phylogenetic commu- bly processes (see section ‘phylogenetic diversity, assembly nity structure (i.e. phylogenetic composition and diversity) processes and stress gradients’ below). can detect assembly processes at the community or meta- community scale (but see Kraft et al. 2007; Kembel 2009). Testing the relationship between stress and In particular, it remains unclear whether indicators of phylogenetically determined b-diversity phylogenetic community structure are robust against the stochastic processes, such as dispersal (Kembel 2009), that For decades plant ecologists have demonstrated that turn- influence the assembly of real communities. This study over in functional composition (i.e. mean trait values at uses simple community assembly models, which incorpo- community level) follows a predictable pattern along stress rate stochasticity, to test whether a phylogenetic approach gradients (e.g. Grime 1974). Assuming that the traits influ- can reliably detect changes in assembly processes along a encing species turnover along stress gradients are phyloge- hypothetical stress gradient. netically conserved, we should also be able to retrieve a phylogenetic signal along stress gradients. However, mea- suring changes in phylogenetic composition is more com- Application of functional indices to measure a-scale plicated than for functional composition, because a phylogenetic diversity species’ position in phylogenetic space is defined by its Pavoine & Bonsall (2011) demonstrated that many indices relatedness to other species in the species pool. Recent of functional diversity (FD) can also be used as indices of methodological developments have provided a means of phylogenetic diversity (PD), if we replace functional testing for a phylogenetic signal in species turnover along dissimilarities between species with phylogenetic dissimi- environmental gradients (Pillar & Duarte 2010), using larities between species, functional dendrograms with phy- fuzzy set theory to estimate phylogenetically determined logenetic trees, or functional trait space with phylogenetic species composition. In essence, these methods provide a space. Phylogenies are a means of describing the distribu- test of whether differences in phylogenetic composition tion of species relative to each other in phylogenetic space. between communities are significantly correlated with dif- This is directly analogous to using traits to describe the ferences in environmental variables. A similar approach distribution of species in functional space, so that there is can reveal significant turnover in the composition of func- obvious potential for using existing FD indices as measures tional traits along gradients, so we can test whether results of PD, and vice versa. for phylogenetic composition match those for functional Mason et al. (2005) identified three primary composition. Within the framework presented in Pillar & components of functional diversity – functional richness, Duarte (2010) we can also test for phylogenetic signal at
Journal of Vegetation Science Doi: 10.1111/jvs.12033 © 2012 International Association for Vegetation Science 821 Phylogenetic diversity and assembly processes N.W.H. Mason & S. Pavoine the species pool level (i.e. correlation between pair-wise phylogenetic richness and divergence (Mouchet et al. functional and phylogenetic dissimilarities between 2010), including Rao quadratic entropy (Rao 1982; Pavo- species) and at the meta-community level (i.e. correlation ine et al. 2013) and multivariate dispersion (Laliberte & between pair-wise community dissimilarities in function- Legendre 2010). They should increase when niche com- ally and phylogenetically determined species composi- plementarity enhances either, or both, species’ occurrence tion). In this way, the framework of Pillar & Duarte (2010) probabilities and abundances. provides a means of testing whether phylogenetic signal in An important property of both phylogenetic and the species pool is sufficient to produce a phylogenetic functional diversity indices for understanding ecological signal in functional b-diversity and in species’ turnover processes is that they be independent of existing diversity along environmental gradients. measures (Pavoine et al. 2013). Observed values of func- tional or phylogenetic richness can increase in the absence of any change in assembly processes, simply due to corre- Phylogenentic diversity, assembly processes and stress sponding increases in species richness (Mouchet et al. gradients 2010; Pavoine et al. 2013). Comparing observed values of Stress gradients often represent a shift from size-symmetric phylogenetic richness to those expected from matrix-swap below-ground competition for soil nutrients and water to null models that randomize species’ occurrences (Manly & size-asymmetric above-ground competition for light with Sanderson 2002) can remove any trivial effects of species declining stress (Schwinning & Weiner 1998; Berntson & richness (Mason et al. 2011a; Richardson et al. 2012). Wayne 2000; Cahill & Casper 2000; Coomes & Grubb A null model that randomizes abundances across species 2000). Where size-symmetric below-ground competition but within communities (Mason et al. 2008b) can convert dominates, fitness is enhanced by traits that maximize indices that measure both phylogenetic richness and diver- acquisition and retention of limiting below-ground gence into pure measures of phylogenetic divergence resources (e.g. Richardson et al. 2005; Lambers et al. (Mason et al. 2012b). 2008; Holdaway et al. 2011). Niche differentiation is required for species to co-exist when competition is size- Aims and objectives asymmetric (Aikio 2004; Kohyama & Takada 2009). Thus, We use a simple assembly model to explore the power of niche differences between co-occurring species will indicators of phylogenetic community structure to detect increasingly enhance local fitness as light competition assembly processes along stress gradients under a wide becomes more intense (Mason et al. 2011b). This is con- range of ecological contexts. This model is described in sistent with an increase in the importance of niche detail by Mason et al. (2012a, 2013). It varies the relative complementarity and a decrease in the importance of envi- influence of environmental filtering and niche comple- ronmental filtering (Mason et al. 2008b) with declining mentarity on species fitness along a hypothetical stress gra- stress (see also Mason et al. 2012a,b; for more detailed dient, with niche complementarity becoming more arguments on this topic), and should cause multi-species influential as stress declines. It incorporates two forms of communities, where light competition is intense, to have stochasticity – mass and priority effects – to test whether higher FD (Mouchet et al. 2010; Mason et al. 2011b). indicators of phylogenetic community structure are robust Of the three FD components, functional richness and against trait-independent processes. It also tests whether functional divergence (or indices that combine them) have variation in species richness affects the power of a-scale PD most often been linked to community assembly processes indices to detect changes in assembly processes along the (Mouchet et al. 2010; Spasojevic & Suding 2011; Mason stress gradient. Our ultimate aim is to assess whether mea- et al. 2012b) or ecosystem functioning (Petchey et al. sures of a-andb-scale indicators of phylogenetic diversity 2004; Mouillot et al. 2011) in plant communities and sim- provide a reliable means of testing for niche-based assem- ulation studies. However, one study has found increased bly processes along stress gradients. In doing this, we also functional evenness with declining stress in fish communi- aim to select a set of a-scale PD indices that will provide a ties (Mason et al. 2008a). Assuming that PD is a surrogate reliable test of the hypothesis that niche complementarity for FD, we can use existing work on FD to predict how PD will increase with declining abiotic stress. components should change in response to changes in assembly processes (e.g. Kraft et al. 2007). Phylogenetic Our selection of PD indices used the following criteria: richness should increase when niche complementarity 1. Each index must have reasonable power to detect the enhances species’ occurrence probabilities (Mason et al. increasing influence of niche complementarity with 2012b). Phylogenetic divergence should increase when declining stress (henceforth, power, for brevity), when niche complementarity enhances species’ relative abun- either mass effects or priority effects influence community dances (Mason et al. 2012b). Several indices measure both assembly.
Journal of Vegetation Science 822 Doi: 10.1111/jvs.12033 © 2012 International Association for Vegetation Science N.W.H. Mason & S. Pavoine Phylogenetic diversity and assembly processes
2. Power must be high across a range of species richness Zealand Plant Phylogeny Database (http://plantphy loge- levels. ny.landcareresearch.co.nz/WebForms/Home.aspx) to 3. Variation in species richness along the stress gradient manually graft species that were not available in Phylo- must not affect the power of any index. matic. Branch lengths were computed following Grafen 4. Collectively, the indices must provide reasonable power (1989) using ‘compute.brlen’ in R package ‘ape’ (Paradis across all the contexts examined. et al. 2004; R Foundation for Statistical Computing, Vienna, AT, US). Phylogenetic distances were defined as the sum of branch lengths in the smallest path that Methods connects two species on the phylogenetic tree (patristic Our assembly model and the rationale it is based on distances). Values for a single trait were simulated for the are described in detail in Mason et al. (2013), see also Data 130 indigenous New Zealand species. To do this, we S1. Consequently, we provide here only a brief description applied the Brownian motion (BM) and accelerating (AC) of the model, and its application in testing the power of PD model of trait evolution (Blomberg et al. 2003) to the phy- indices to detect changes in assembly processes. logenetic tree. With the height of the phylogenetic tree standardized to equal one and the variance–covariance matrix of species’ trait values equal to r²C(h,g), where h is Basic assembly model the value at the root node and g is the rate of trait change, The model context is one where the relative influence of we used the following parameter values for the AC model: niche complementarity and environmental filtering r²=1, h = 0, g = exp(À2), g = exp(À10), g = exp(À20). changes along a stress gradient, with the influence of envi- These models are termed AC 2, AC 10 and AC 20, respec- ronmental filtering being greatest at the most stressed sites. tively. Phylogenetic signal increases with g;wheng tends We model the effect of environmental filtering as a decline towards 0, the model tends towards BM (see Pavoine et al. in fitness with increasing difference between a species’ trait 2010 for details). value and the locally optimal trait value. We model the effect of niche complementarity on fitness as the competi- Mass effects model tive effect on potential colonizers of species that are already ‘present’ in the community. Fitness is estimated by com- This is a simple extension of the basic model where, for bining niche complementarity and environmental filtering each site, the desired number of species is randomly effects. Fitness is scaled to abundance so that the most selected from the species pool. The basic model then calcu- abundant species has 103 times the abundance of the least lates the fitness and abundance of the selected species. abundant species. This degree of variation in abundances is Under this model, species’ occurrences are completely typical in plant communities. unrelated to their traits, but abundance is a deterministic We simulated species’ abundances for ten points function of the traits of the species present. This is an (termed ‘sites’ henceforth) along the stress gradient. We extreme form of mass effects. This model tests whether PD chose to use a small number of sites since many studies of indices can detect changes in trait-based assembly vegetation change along stress gradients examine only a processes in communities where these processes have few independent sites (e.g. Mason et al. 2012b), due to dif- weak effects on species’ occurrences, but a strong effect on ficulties in finding suitable sites. At the most stressed site, species’ abundances. 90% of overall fitness is determined by environmental fil- tering and 10% is determined by niche complementarity. Priority effects model Moving from the most stressed to least stressed end of the gradient, we decreased the influence of environmental This model randomizes the order of species’ colonization, filtering and increased the influence of niche complemen- whereas species’ traits determine colonization order in the tarity by intervals of 10%, so that at the least stressed site basic model. In this model, colonization order affects the 10% of overall fitness was due to environmental filtering degree of competition experienced by each species, with and 90% due to niche complementarity. early-colonizing species, on average, experiencing less intense competition. Species fitness is calculated as in the basic model, but with competitive effects on fitness based Generating species pools on this randomized colonization order. The desired num- We used Phylomatic (http://phylodiversity.net/phylomat- ber of species is then selected based on species fitness (so ic/) to construct a hypothesized phylogenetic tree for 130 when generating communities of ten species, the ten fittest indigenous New Zealand herbaceous and woody angio- species are selected). Since colonization order may have a sperm, gymnosperm and fern species, and used the New strong influence on the level of competition a species
Journal of Vegetation Science Doi: 10.1111/jvs.12033 © 2012 International Association for Vegetation Science 823 Phylogenetic diversity and assembly processes N.W.H. Mason & S. Pavoine experiences, the priority effects model incorporates because it is unclear how SES values derived this way stochasticity in the relationship between traits and both relate to the three primary components of PD. Observed species’ occurrences and abundances. This model tests FEve was not compared to either null model since it whether PD indices can detect changes in trait-based appears to be independent of species richness (Mouchet assembly process in communities where priority effects et al. 2010), and it is unclear how SES values generated disrupt the effect of these processes on both occurrences when randomizing species’ abundances within communi- and abundances. ties should be interpreted.
Using FD indices to measure PD Testing for phylogenetic and functional signals in species turnover along stress gradients We selected indices that correspond to the FD components of Mason et al. (2005) to measure the PD components we We implemented the methods of Pillar & Duarte (2010) to identified in the Introduction. We examined one index of test whether phylogeny and function significantly influ- phylogenetic richness – phylogenetic branch length enced species turnover along the stress gradient. Specifi- (termed tree size, TS, henceforth) of Faith (1992). To cally, we used the correlation coefficient roPE (correlation ensure values for each simulated community for this mea- of differences between communities in phylogenetically sure was bound between 0 and 1, they were expressed as a determined species composition with distance along the proportion of the values for the entire species pool. We stress gradient) to test whether differences between com- decided not to use convex hull volume (the FRic of Villeger munities in phylogenetically determined species composi- et al. 2008) because this index requires reduction of tion were significantly correlated with distances between dimensionality, so that the phylogenetic space has fewer communities along the stress gradient. We used roTE dimensions than the species richness of the most species- (correlation of differences between communities in func- poor communities. This risks severe information loss when tionally determined species composition with distance using phylogenetic data. We used FEve as a measure of along the stress gradient) to test whether differences phylogenetic evenness (Villeger et al. 2008). We examined between communities in functionally determined species two indices that are a combination of both phylogenetic composition were significantly correlated with distances richness and phylogenetic divergence – FDis (Laliberte & between communities along the stress gradient. We used Legendre 2010) and Rao quadratic entropy (Pavoine & the coefficient roBF (correlation between matrices of inter- Doledec 2005; de Bello et al. 2010). TS was calculated specific functional and phylogenetic dissimilarities) to test directly from the phylogenetic tree. All other indices were for significant correlation in functional and phylogenetic calculated following principal coordinates analysis (PCoA) distances between pairs of species within the species pool. to convert phylogenetic distances between species to loca- Finally, we used roPT (correlation of differences between tions in Euclidean phylogenetic space (Diniz-Filho et al. communities in functionally and phylogenetically deter- 1998). See Pavoine & Bonsall (2011) for details on how mined species composition) to test whether differences each index was calculated. between communities in phylogenetically determined We compared TS values relative to random expectation species composition were significantly correlated with dif- under a matrix-swap null model that randomizes species ferences in functionally determined species composition. occurrences between sites while keeping the species rich- These correlation tests were achieved using the R imple- ness of sites and frequency of occurrence of species the mentation of Pillar & Duarte’s (2010) methods provided by same as in the observed data (Manly & Sanderson 2002) the SYNCSA package, using function syncsa (Debastiani & using the standardized effect size (SES) of Gotelli & McCa- Pillar 2012). be (2002). This index is termed SESTS henceforth. FDis and Rao were compared with values obtained when abun- The simulation framework dances were randomized across species but within plots (after Mason et al. 2008b). These indices, termed SESFDis The same simulation framework was applied to both the and SESRao, are pure measures of phylogenetic diver- mass effects and priority effects model, and for the Brown- gence, since the phylogenetic richnesses of the observed ian motion and accelerating evolution models studied. In and randomized communities are always identical. We each case, it began with the generation of trait values for used 104 randomizations in all null model analyses, to the species pool. Then communities were assembled for ensure accurate estimates of SES values. We did not use each of the ten sites occurring along the hypothetical stress FDiv (Villeger et al. 2008) as a measure of phylogenetic gradient. Alpha PD values were calculated for each of the divergence since it requires calculation of the convex hull. sites, and the Pearson r correlation coefficient between We did not randomize abundances between communities each PD index and the influence of niche complementarity
Journal of Vegetation Science 824 Doi: 10.1111/jvs.12033 © 2012 International Association for Vegetation Science N.W.H. Mason & S. Pavoine Phylogenetic diversity and assembly processes
on fitness (d) was calculated. Significant positive correla- 1 5 BM tions between a PD indices and d indicate that they are able 0.8 10 0.6 15 to detect the increasing influence of niche complementar- 20 ity on community assembly, with declining stress. We also 0.4 25 0.2 examined log-linear, quadratic and exponential relation- 0 ships between PD indices and d. In no instance did these –0.2 relationships yield greater power than Pearson r. For each –0.4 SESTS SESRao simulation we also recorded the significance (at P < 0.05) –0.6 TS FEve FDis Rao SESFDis of the correlation coefficients proposed in Pillar & Duarte –0.8 –1 (2010). We generated trait values for 100 hypothetical regional 1 5 AC 2 species pools, and simulated communities at five levels of 0.8 10 0.6 15 species richness: 5, 10, 15, 20 and 25. For each species rich- 20 0.4 25 ness level, we recorded the proportion of significant corre- 0.2 lations obtained for each index (termed ‘power’ 0 henceforth). We tested whether the power of a PD indices –0.2 was affected by variation in species richness along the gra- –0.4 TS FEve FDis Rao SESTS SESFDis –0.6 SESRao dient, by simulating communities of varying species rich- –0.8 ness. To examine power with increasing species richness, –1 communities containing five species were simulated for AC 10 the two most stressed sites. Communities with ten species 1 5 were simulated for the two next most stressed sites, and so 0.8 10 0.6 15 on until communities with 25 species were simulated for 20 0.4 the two least stressed sites. This provides an indication of 25 0.2 power when species richness increases with the influence Power (proportion of significant results) 0 of niche complementarity (d). We repeated this process in –0.2 reverse to estimate power when species richness declined –0.4 with d. We also examined power when species richness –0.6 TS FEve FDis Rao SESTS SESFDis SESRao varied randomly with d. Where significant negative corre- –0.8 –1 lations were obtained, these were recorded as ‘negative AC 20 power’, so that ‘power’ was bound between À1and1. 1 5 0.8 10 Results 0.6 15 0.4 20 Alpha-scale phylogenetic diversity 0.2 25 0 In the mass effects model, none of the indices provided –0.2 high power to detect changes in assembly processes for any –0.4 of the evolution models studied (Fig. 1). SESRao and –0.6 TS FEve FDis Rao SESTS SESFDis SESRao –0.8 SESFDis performed best, but still only had power between –1 0.2 and 0.4 in the BM and AC 2 models of evolution. Power for all indices was negligible in the AC 10 and AC 20 Fig. 1. Power of phylogenetic diversity indices to detect the increasing models. influence of niche complementarity with declining stress when mass effects disrupt the relationship between species’ traits and occurrences. In the priority effects model, TS and SESTS had the Results are shown for four models of evolution: Brownian motion (BM), highest power (Fig. 2). However, with power between 0.6 and accelerating evolution with g = exp(À2), exp(À10) or exp(À20) (AC 2, and 0.8 in the BM model and between 0.5 and 0.7 in the AC 10 and AC 20, respectively). Numbers associated with different types AC 2 model, these indices do not provide a high degree of of bar shading indicate species richness of local communities. TS = the reliability in revealing changes in assembly processes. For sum dendrogram branch length, FEve = phylogenetic evenness, the AC 10 and AC 20 evolution models, the power of all FDis = phylogenetic dispersion and Rao = quadratic entropy. SESTS is indices was very low. observed TS expressed relative to a matrix swap null model. SESRao and SESFDis are, respectively, observed Rao and FDis expressed relative to a In the mass effects model, variation in species richness null model randomizing abundances across species, but within strongly influenced the power of TS, FDis and Rao (Fig. 3). communities. For these indices, power was higher when species richness increased with declining stress and lower when species
Journal of Vegetation Science Doi: 10.1111/jvs.12033 © 2012 International Association for Vegetation Science 825 Phylogenetic diversity and assembly processes N.W.H. Mason & S. Pavoine
1 BM 5 20 richness decreased with declining stress. This effect was 10 25 0.8 obvious for all of the evolution models studied. 15 0.6 In the priority effects model, variation in species rich- 0.4 ness strongly influenced the power of TS, SESTS, FDis and 0.2 Rao (Fig. 4). For all of these indices except SESTS, power 0 was higher when species richness increased with declining –0.2 stress and lower when species richness decreased with –0.4 declining stress. The opposite effect was observed for
–0.6 TS FEve FDis Rao SESTS SESFDis SESRao –0.8 SESTS. The influence of species richness on power was evi- –1 dent for each of the evolution models studied.
AC 2 20 1 5 Phylogenetic and functional signals in species turnover 10 25 0.8 15 along stress gradients 0.6 0.4 For the mass effects model, correlations of differences 0.2 between functionally and phylogenetically determined 0 species composition and position along the stress gradient –0.2 (roTE and roPE, respectively) were generally quite low –0.4 (power <0.4; Fig. 5). This is perhaps not so surprising since
–0.6 TS FEve FDis Rao SESTS SESFDis SESRao the mass effects model, by randomly selecting the species –0.8 present in local communities, disrupts the relationship –1 between species’ occurrences and their traits. This stochas- AC 10 ticity in species occurrence may have disrupted the rela- 1 5 20 tionship between stress and both the functional and 0.8 10 25 phylogenetic composition of communities, leading to low 0.6 15 power for both roTE and roPE. 0.4 A more surprising result is that even for the Brownian 0.2 motion model (where trait variation is strongly Power (proportion of significant results) 0 conserved), functionally and phylogenetically determined –0.2 –0.4 species composition were significantly correlated for fewer than 50% of simulations (i.e. power <0.5 for roPT; Fig. 5).
–0.6 TS FEve FDis Rao SESTS SESFDis SESRao –0.8 This is despite almost 100% of simulations yielding a –1 significant correlation for functional and phylogenetic dis- similarities between species (i.e. power ca. 1 for roBF). AC 20 1 5 20 This result suggests that the species pool-level phylogenetic 0.8 10 25 signal was poorly maintained at the meta-community 15 0.6 level, even when traits were strongly phylogenetically 0.4 conserved. Even for the evolution model with the weakest 0.2 phylogenetic signal (AC20) a non-negligible proportion of 0 runs (ca. 0.4) yielded a significant correlation for func- –0.2 tional and phylogenetic dissimilarities between species. –0.4 However, for this model a very small proportion of simula- < –0.6 TS FEve FDis Rao SESTS SESFDis SESRao tions ( 0.1) yielded a significant result for roPT. This –0.8 further emphasizes that phylogenetic signals in trait varia- –1 tion at the species pool level were not maintained at the meta-community level. Fig. 2. Power of phylogenetic diversity (PD) indices to detect the In the priority effects model, power for roTE was very increasing influence of niche complementarity with declining stress when high, with almost 100% of simulations yielding a signifi- priority effects disrupt the relationship between species’ traits and cant result (Fig. 6). This indicates that the functionally abundances. Numbers associated with different types of bar shading determined species composition of communities was indicate species richness of local communities. Results are shown for four models of evolution: Brownian motion (BM), and accelerating evolution almost always significantly correlated with their position with g = exp(À2), exp(À10) or exp(À20) (AC 2, AC 10 and AC 20, along the stress gradient. This result was consistent across respectively). Explanations for PD index codes are given in Fig. 1. all models of trait evolution (i.e. was independent of the
Journal of Vegetation Science 826 Doi: 10.1111/jvs.12033 © 2012 International Association for Vegetation Science N.W.H. Mason & S. Pavoine Phylogenetic diversity and assembly processes
BM BM 1 1 Increasing Increasing 0.8 0.8 Random 0.6 Random 0.6 0.4 Decreasing 0.4 Decreasing 0.2 0.2 0 0 –0.2 –0.2 –0.4 –0.4 –0.6 Rao FEve FDis SESTS SESFDis SESRao –0.6 TS TS Rao SESFDis SESRao FEve FDis SESTS –0.8 –0.8 –1 –1
1 1 AC 2 AC 2 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 –0.2 –0.2 –0.4 –0.4 –0.6 –0.6 FEve Rao SESFDis FDis SESTS SESRao TS TS FDis SESTS SESFDis FEve Rao SESRao –0.8 –0.8 –1 –1
1 1 AC 10 AC 10 0.8 0.8 0.6 0.6 0.4 0.4 0.2
0.2 Power (proportion of significant results) Power (proportion of significant results) 0 0 –0.2 –0.2 –0.4 –0.4 –0.6 TS FEve FDis SESTS SESFDis SESRao –0.6 Rao TS FEve FDis Rao SESTS SESFDis SESRao –0.8 –0.8 –1 –1 1 1 AC 20 AC 20 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 –0.2 –0.2 –0.4 –0.4 –0.6 TS FDis Rao –0.6 FEve SESTS SESFDis SESRao
TS FEve FDis Rao SESTS SESFDis SESRao –0.8 –0.8 –1 –1 Fig. 4. Power of phylogenetic diversity (PD) indices with species richness Fig. 3. Power of phylogenetic diversity (PD) indices with species richness increasing, decreasing or varying randomly as stress declines when increasing, decreasing or varying randomly as stress declines when mass priority effects disrupt the relationship between species’ traits and effects disrupt the relationship between species’ traits and occurrences. abundances. Results are shown for four models of evolution: Brownian Results are shown for four models of evolution: Brownian motion (BM), motion (BM), and accelerating evolution with g = exp(À2), exp(À10) or = À À À and accelerating evolution with g exp( 2), exp( 10) or exp( 20) (AC 2, exp(À20) (AC 2, AC 10 and AC 20, respectively). Explanations for PD index AC 10 and AC 20, respectively). Explanations for PD index codes are given codes are given in Fig. 1. in Fig. 1.
Journal of Vegetation Science Doi: 10.1111/jvs.12033 © 2012 International Association for Vegetation Science 827 Phylogenetic diversity and assembly processes N.W.H. Mason & S. Pavoine level of trait conservatism). By contrast, power for roPE composition of communities was consistently significantly was moderate to low (<0.5), even for the Brownian correlated with their position on the stress gradient, phylo- motion model. This indicates that while the functional genetic composition was not. Results for roPT and roBF were the same as for the mass effects model, indicating 1 that, irrespective of the assembly model used, phylogenetic BM signals in trait variation at the species pool level were 0.8 poorly maintained at the meta-community level. 5 10 0.6 15 Discussion 20 0.4 25 Our results suggest that indicators of phylogenetic commu- nity structure do not provide a reliable means of detecting 0.2 changes in niche-based assembly processes or niche-based species turnover along gradients. Measures of a-level phy- 0 logenetic diversity (PD) did not reliably detect an increase roTE roPE roPT roBF in niche complementarity with declining stress. Measures 1 of b PD did not reliably detect functionally determined AC 2 changes in species composition along the stress gradient. 0.8 Our results also suggest that the weak phylogenetic signal 5 at the meta-community level may be responsible for the 10 low power of a-scale FD indicators to detect assembly 0.6 15 processes. 20 0.4 25 Consequently, non-significant results for correlations between stress gradients and a PD are not sufficient to reject the hypothesis that niche-based assembly processes
cant results) 0.2 fi change along stress gradients. Nor is a non-significant cor- b 0 relation between PD and stress (i.e. roPE) sufficient to roTE roPE roPT roBF reject the hypothesis that functionally determined species on of signi
Ɵ composition changes along stress gradients. 1 AC 10 The only other simulation study testing the power of PD indices to detect assembly processes that we are aware of 0.8 5 (Kraft et al. 2007) obtained equivocal results. Therefore, it 10 may be premature to declare that PD indices are of no use Power (propor 0.6 15 for revealing assembly processes, especially since our study 20 was not designed to provide a definitive general statement 0.4 25 on this question. Below we provide a synthesis of our results, with special reference to our selection criteria for a 0.2 PD indices. We then explore general explanations for the
0 roTE roPE roPT roBF Fig 5. Proportion of runs giving significant results for the four matrix correlation coefficients studied when mass effects disrupt the relationship 1 AC 20 between species’ traits and occurrences. Numbers associated with different types of bar shading indicate species richness of local 0.8 communities. The correlations studied were: roTE = correlation of 5 differences between communities in functionally determined species 10 composition with distance along the stress gradient; roPE = correlation of 0.6 15 differences between communities in phylogenetically determined species 20 composition with distance along the stress gradient; roPT = correlation of 0.4 25 differences between communities in functionally and phylogenetically determined species composition; roBF = correlation of functional and 0.2 phylogenetic dissimilarity between species. Results are shown for four models of evolution: Brownian motion (BM), and accelerating evolution 0 with g = exp(À2), exp(À10) or exp(À20) (AC 2, AC 10 and AC 20, roTE roPE roPT roBF respectively).
Journal of Vegetation Science 828 Doi: 10.1111/jvs.12033 © 2012 International Association for Vegetation Science N.W.H. Mason & S. Pavoine Phylogenetic diversity and assembly processes
1 low power of the PD indices examined and consider BM whether the lack of power may be due to limitations in the 0.8 modelling framework used. We finish by discussing sce- narios where phylogenetic data may have more potential 5 0.6 10 to reveal assembly processes, and by providing suggestions 15 for future modelling studies to investigate these scenarios. 0.4 20 25 0.2 Phylogenetic diversity indices do not satisfy criteria for detecting changes in assembly processes 0 roTE roPE roPT roBF None of the a PD indices examined satisfied our selection
1 criteria. They generally had very low power, and those that AC 2 had reasonable power in some contexts were strongly influ- 0.8 enced by variation in species richness. Consequently, none of these indices provides a reliable test of the hypothesis that 5 0.6 the influence of niche complementarity on community 10 15 assembly will increase with declining stress. Based on these 0.4 20 findings, we must advise against the use of PD indices to 25 reveal changes in niche-based assembly processes along 0.2 stress gradients. For the priority effects model, observed cant results)
fi phylogenetic dendrogram branch length (TS) and TS com- 0 roTE roPE roPT roBF pared with a matrix swap null model (SESTS) both had some power to detect changes in assembly processes along
on of signi 1 Ɵ AC 10 the hypothetical stress gradients. This was true for all of the species richness levels studied in the two evolution models 0.8 where the phylogenetic influence on trait variation was
5 strongest (BM and AC 2). However, variation in species 0.6 Power (propor 10 richness strongly influenced the power of both TS and 15 SESTS in both the mass effects and priority effects model. 0.4 20 Consequently, using these indices to infer changes in 25 0.2 assembly processes along a gradient could lead to spurious conclusions if species richness also varies along the gradient. 0 This susceptibility to variation in species richness means roTE roPE roPT roBF that these PD indices are not suitable for revealing trait- based assembly processes, since a variety of trait-indepen- 1 AC 20 dent assembly processes might influence species richness. 0.8 None of the other indices provided much power in either the mass effects or priority effects model. Thus, none 5 0.6 of the indices examined satisfied all of our selection 10 15 criteria, so we cannot suggest their use to detect shifts in 0.4 20 assembly processes along stress gradients. By extension, 25 we also failed to find a set of indices that collectively pro- 0.2 vide power to detect shifts in assembly processes across all of the contexts studied. Based on these findings, it seems 0 roTE roPE roPT roBF that a PD indices are unlikely to provide a useful means for detecting changes in community assembly processes along Fig 6. Proportion of runs giving significant results for the four matrix stress gradients in ecological timescales. correlation coefficients studied when priority effects disrupt the relationship between species’ traits and occurrences. Numbers associated with different types of bar shading indicate species richness of local General problems – trait conservatism at species pool communities. Explanations of correlation coefficient codes are given in level not maintained during community assembly Fig. 5. Results are shown for four models of evolution: Brownian motion (BM), and accelerating evolution with g = exp(À2), exp(À10) or exp(À20) Pavoine et al. (2013) demonstrate that a PD is a poor (AC 2, AC 10 and AC 20, respectively). surrogate for a FD, since correlations between FD and PD
Journal of Vegetation Science Doi: 10.1111/jvs.12033 © 2012 International Association for Vegetation Science 829 Phylogenetic diversity and assembly processes N.W.H. Mason & S. Pavoine are generally not very strong, even when phylogeny munity assembly (e.g. see results for plant phylogenies in strongly determines inter-specific variation in trait values. Blomberg et al. 2003; appendix 6), although some studies This may limit the ability of PD indices to detect assembly have shown that strong trait conservatism can occur in processes that act via species’ traits. Using the same plants (Feild & Arens 2007; Donoghue 2008). modelling framework as we have used here, Mason et al. One potential limitation of our modelling approach is (2013) showed that a FD indices had very high power to that we consider only a single functional trait. Some sim- detect changes in trait-based assembly processes along the ulation studies (Kraft et al. 2007; Kembel 2009) have hypothetical stress gradient. Consequently, it seems that found that, in some instances, the power of PD indices the low power of a PD indices was largely due to their increased as the number of traits on which niche comple- inability to accurately capture variation in a FD. Our mentarity or environmental filtering acted increased. results, considered in light of Pavoine et al. (2013) and Thus, our use of a single trait may have limited the abil- Mason et al. (2013), suggest that species pool-level trait ity of PD to capture FD, since our functional trait space conservatism may generally be poorly maintained at the a has much lower dimensionality than the phylogenetic (i.e. local community) scale, although further work is space. However, Pavoine et al. (2013) used a large num- required to provide a definitive answer. If this is true, it is ber of traits and still found correlations between FD and possible that a-scale indicators of phylogenetic community PD to be relatively weak. Consequently, it seems unlikely structure may generally be poorly suited to detecting that including more trait dimensions in our modelling niche-based assembly processes in local communities. framework would have improved the power of PD indi- Our work extends the work of Pavoine et al. (2013) and ces markedly. Mason et al. (2013) to suggest that b-scale (i.e. meta-com- The shape of the phylogenetic tree we used, being quite munity) measures of phylogenetic community structure asymmetric, may also have limited the power of PD indices are poorly able to detect niche-based species turnover to detect assembly processes. However, Kraft et al. (2007) along stress gradients. We have shown that the low power found that tree topology had only a subtle effect on the of b PD indices is due to weak phylogenetic signal in trait power of PD indices. Consequently, it seems unlikely that variation at the meta-community level. If phylogenetic sig- the shape of our phylogenetic tree can provide the main nals at the meta-community level (b scale) were generally explanation for the low power of PD indices. More much weaker than those at the species-pool level (c scale), recently, evidence has emerged that poorly resolved phy- this would raise serious doubts about the overall ability of logenies with many soft polytomies (e.g. nodes joining phylogenetic approaches to detect niche-based assembly many species within a genus, with equal, arbitrary branch processes. However, it seems that studies examining the length) can artificially inflate trait conservatism at species ability of phylogenetic community structure to detect pool level (as measured using Blomberg et al.’s K statistic; assembly processes (including Kraft et al. 2007; Kembel Davies et al. 2011). Our tree does include several soft poly- 2009) have ignored this point. Consequently, it is impossi- tomies, so it is possible that we have over-estimated species ble to say whether this result represents a general pattern pool-level trait conservatism (though it is unclear whether or is due to some artefact in our modelling frame work. the measure of conservatism we used, roBF, is subject to this effect). This could partially explain the discrepancy we observed between species pool- and meta-community- Potential problems with the modelling framework level trait conservatism. It is unclear whether soft polyto- Other simulation studies (Kraft et al. 2007; Kembel 2009) mies may have also influenced variation in PD indices. We have found that, in some instances, PD may have good are unaware of any studies examining this question, and power to detect assembly processes. This raises the possibil- the influence of tree shape on measures of phylogenetic ity that the low power of PD indices that we have demon- community structure was beyond the scope of the present strated may be due to problems with our modelling study. framework. However, Kraft et al. (2007) did not include Another potential limitation to our treatment of the any trait-independent stochasticity in their trait-based phylogeny is that we assigned branch lengths after select- assembly models. Consequently, their simulations proba- ing our subset of species from the complete phylogeny of bly over-emphasize the role of traits (and hence phylog- vascular plants. This may have artificially constrained the eny) in assembly relative to real plant communities. phylogenetic space spanned by our species pool. An alter- Further, in both Kraft et al. (2007) and Kembel et al. native approach would be to ensure all our species were (2009), PD indices generally only had good power when inserted into a full phylogeny (e.g. like that provided by trait variation was extremely phylogenetically conserved. Angiosperm Phylogeny Group 2009) and then obtain our The level of phylogenetic conservatism that they used is phylogenetic distances directly from the full phylogeny. probably unrealistic for the plant traits that control com- However, since we used several strongly conserved models
Journal of Vegetation Science 830 Doi: 10.1111/jvs.12033 © 2012 International Association for Vegetation Science N.W.H. Mason & S. Pavoine Phylogenetic diversity and assembly processes of trait evolution, this should also have constrained varia- a poor shortcut for capturing functionally relevant trait tion in functional trait values. variation, there is no way of avoiding the hard work Our findings do not provide definitive evidence that required to demonstrate the significance of plant traits for indicators of phylogenetic community structure will gen- individual-level functions. However, there are many excit- erally have limited power to detect assembly processes. ing opportunities for phylogeny to advance community However, they do build on the work of previous authors ecological studies. In particular, there is great potential for demonstrating that caution must be exercised when inter- phylogeny to help us understand how the history of trait preting observed patterns of phylogenetic composition and evolution contributes to observed correlations between diversity. Indeed, there is growing evidence that rather stress gradients and measures of functional community than using phylogeny to derive surrogate measures of structure. functionally determined community structure, a more fruitful line of research might be to consider how phylog- eny affects observed patterns in functional diversity and References composition (e.g. Safi et al. 2011; Srivastava et al. 2012). With respect to examining patterns along environmental Aikio, S. 2004. Competitive asymmetry, foraging area size and – gradients, it may be interesting to consider whether trait coexistence of annuals. Oikos 104: 51 58. Angiosperm Phylogeny Group 2009. An update of the Angio- evolution plays a major role in correlations between envi- sperm Phylogeny Group classification for the orders and ronment and functional diversity and composition. For families of flowering plants: APG III. Botanical Journal of the example, one could simulate traits for the species pool Linnean Society 161: 105–121. using models of varying trait conservatism and record the Berntson, G.M. & Wayne, P.M. 2000. Characterizing the size proportion of the simulated species pools yielding correla- dependence of resource acquisition within crowded plant tions as or more extreme than those observed. It would be populations. Ecology 81: 1072–1085. particularly interesting to compare observed correlations Blomberg, S.P., Garland, T. & Ives, A.R. 2003. Testing for phylo- with those for simulated species pools with similar levels of genetic signal in comparitive data: behavioural traits are trait conservatism (as measured by Blomberg et al.’s K or more labile. Evolution 57: 717–745. Pillar & Duarte’s roBF). If the observed correlations are Cahill, J.F. Jr & Casper, B.B. 2000. Investigating the relationship within the 95% confidence interval of the simulated between neighbor root biomass and belowground competi- species pools, then one might conclude that the role of trait tion: field evidence for symmetric competition belowground. evolution in generating the observed correlations cannot Oikos 90: 311–320. be ignored. This is just one of many exciting opportunities Cavender-Bares, J., Kozak, K.H., Fine, P.V.A. & Kembel, S.W. for using phylogeny to better understand how historical 2009. The merging of community ecology and phylogenetic processes influence observed patterns in plant communi- biology. Ecology Letters 12: 693–715. ties (Cavender-Bares et al. 2009). Coomes, D.A. & Grubb, P.J. 2000. Impacts of root competition in forests and woodlands: a theoretical framework and review of experiments. Ecological Monographs 70: 171–207. Conclusions Davies, T.J., Kraft, N.J.B., Salamin, N. & Wolkovich, E.M. 2011. Our findings raise doubts about the ability of indicators of Incompletely resolved phylogenetic trees inflate estimates of – phylogenetic community structure to reveal niche-based phylogenetic conservatism. Ecology 93: 242 247. assembly processes along stress gradients. Consequently, de Bello, F., Lavergne, S., Meynard, C., Lepsˇ, J. & Thuiller, W. 2010. The spatial partitioning of diversity: showing Theseus they are unlikely to provide a means of avoiding the tough a way out of the labyrinth. Journal of Vegetation Science 21: decisions around which traits are most relevant for assem- 992–1000. bly processes. Rather, our findings emphasize the need to Debastiani, V.J. & Pillar, V.D. 2012. SYNCSA – R tool for analysis collect relevant functional trait data to understand the of metacommunities based on functional traits and phylog- mechanisms controlling community assembly. Identifying eny of the community components. Bioinformatics 28: 2067– relevant traits requires an understanding of their signifi- 2068. cance for functions occurring at the individual scale, such Diniz-Filho, J.A., De Sant’ana, C.E.R. & Bini, L.M. 1998. An as growth rate and stress tolerance. We now thoroughly eigenvector method for estimating phylogenetic inertia. understand the functional significance of some plants Evolution 52: 1247–1262. traits, such as the influence of foliar traits on photosyn- Donoghue, M.J. 2008. A phylogenetic perspective on the distri- thetic rates and leaf longevity (Reich et al. 1999; Wright bution of plant diversity. Proceedings of the National Academy et al. 2004). However, the functional significance of of Sciences of the United States of America 105: 11549–11555. variation in other traits, such as those relating to root mor- Faith, D.P. 1992. Conservation evaluation and phylogenetic phology, is much less clear. Since phylogeny appears to be diversity. Biological Conservation 61: 1–10.
Journal of Vegetation Science Doi: 10.1111/jvs.12033 © 2012 International Association for Vegetation Science 831 Phylogenetic diversity and assembly processes N.W.H. Mason & S. Pavoine
Feild, T.S. & Arens, N.C. 2007. The ecophysiology of early angio- contrasting assembly processes in experimental grassland sperms. Plant, Cell and Environment 30: 291–309. communities. Journal of Ecology 99: 788–796. Gotelli, N.J. & McCabe, D.J. 2002. Species co-occurrence: a Mason, N.W.H., de Bello, F., Dray, S., Mouillot, D. & Pavoine, S. meta-analysis of J. M. Diamond’s assembly rules model. 2013. A guide for using functional diversity indices to reveal Ecology 83: 2091–2096. changes in assembly processes along ecological gradients Grafen, A. 1989. The phylogenetic regression. Philosophical Trans- Journal of Vegetation Science 24: 794–806. actions of the Royal Society of London. Series B – Biological Sciences Mason, N.W.H., Richardson, S.J., Peltzer, D.A., Wardle, D.A., 326: 119–157. de Bello, F. & Allen, R.B. 2012b. Changes in co-existence Grime, J.P. 1974. Vegetation classification by reference to strate- mechanisms along a long-term soil chronosequence gies. Nature 250: 26–31. revealed by functional trait diversity. Journal of Ecology 100: Holdaway, R.J., Richardson, S.J., Dickie, I.A., Peltzer, D.A. & 678–689. Coomes, D.A. 2011. Species- and community-level patterns Mouchet, M.A., Villeger, S., Mason, N.W.H. & Mouillot, D. in fine root traits along a 120 000-year soil chronosequence 2010. Functional diversity measures: an overview of their in temperate rain forest. Journal of Ecology 99: 954–963. redundancy and their ability to discriminate community Kembel, S.W. 2009. Disentangling niche and neutral influences assembly rules. Functional Ecology 24: 867–876. on community assembly: assessing the performance of Mouillot, D., Villeger, S., Scherer-Lorenzen, M. & Mason, community phylogenetic structure tests. Ecology Letters 12: N.W.H. 2011. Functional structure of biological communities 949–960. predicts ecosystem multifunctionality. PLoS ONE 6: e17476. Kohyama, T. & Takada, T. 2009. The stratification theory for Paradis, E., Claude, J. & Strimmer, K. 2004. APE: analyses of plant coexistence promoted by one-sided competition. phylogenetics and evolution in R language. Bioinformatics 20: Journal of Ecology 97: 463–471. 289–290. Kraft, N.J.B., Cornwell, W.K., Webb, C.O. & Ackerly, D.D. 2007. Pavoine, S. & Bonsall, M.B. 2011. Measuring biodiversity to Trait evolution, community assembly, and the phylogenetic explain community assembly: a unified approach. Biological structure of ecological communities. The American Naturalist Reviews 4: 792–812. 170: 271–283. Pavoine, S. & Doledec, S. 2005. The apportionment of quadratic Laliberte, E. & Legendre, P. 2010. A distance-based framework entropy: a useful alternative for partitioning diversity in for measuring functional diversity from multiple traits. ecological data. Environmental and Ecological Statistics 12: 125– Ecology 91: 299–305. 138. Lambers, H., Raven, J.A., Shaver, G.R. & Smith, S.E. 2008. Plant Pavoine, S., Baguette, M. & Bonsall, M.B. 2010. Decomposition nutrient-acquisition strategies change with soil age. Trends in of trait diversity among the nodes of a phylogenetic tree. Ecology & Evolution 23: 95–103. Ecological Monographs 80: 485–507. Losos, J.B. 2011. Seeing the forest for the trees: the limitations of Pavoine, S., Gasc, A., Bonsall, M.B. & Mason, N.W.H. 2013. Cor- phylogenies in comparative biology. The American Naturalist relations between phylogenetic and functional diversity: 177: 709–727. mathematical artefacts or true ecological and evolutionary Manly, B. & Sanderson, J.G. 2002. A note on null models: justi- processes? Journal of Vegetation Science 24: 781–793. fying the methodology. Ecology 83: 580–582. Petchey, O.L., Hector, A. & Gaston, K.J. 2004. How do Mason, N.W.H., Mouillot, D., Lee, W.G. & Wilson, J.B. 2005. different measures of functional diversity perform? Ecology 85: Functional richness, functional evenness and functional 847–857. divergence: the primary components of functional diversity. Pillar, V.D. & Duarte, L.D.S. 2010. A framework for metacom- Oikos 111: 112–118. munity analysis of phylogenetic structure. Ecology Letters 13: Mason, N.W.H., Lanoiselee, C., Mouillot, D., Wilson, J.B. & 587–596. Argillier, C. 2008a. Does niche overlap control relative abun- Rao, C.R. 1982. Diversity and dissimilarity coefficients: a unified dance in French lacustrine fish communities? A new method approach. Theoretical Population Biology 21: 24–43. incorporating functional traits. Journal of Animal Ecology 77: Reich, P.B., Ellsworth, D.S., Walters, M.B., Vose, J.M., Gresham, 661–669. C., Volin, J.C. & Bowman, W.D. 1999. Generality of leaf trait Mason, N.W.H., Irz, P., Lanoiselee, C., Mouillot, D. & Argillier, C. relationships: a test across six biomes. Ecology 80: 1955–1969. 2008b. Evidence that niche specialization explains species– Richardson, S.J., Peltzer, D.A., Allen, R.B. & McGlone, M.S. energy relationships in lake fish communities. Journal 2005. Resorption proficiency along a chronosequence: of Animal Ecology 77: 285–296. responses among communities and within species. Ecology Mason, N.W.H., Carswell, F.E., Richardson, S.J. & Burrows, L.E. 86: 20–25. 2011a. Leaf palatability and decomposability increase during Richardson, S.J., Williams, P.A., Mason, N.W.H., Buxton, R.P., a 200-year-old post-cultural woody succession in New Courtney, S.P., Rance, B.D., Clarkson, B.R., Hoare, R.J.B., Zealand. JournalofVegetationScience22: 6–17. St. John, M.G. & Wiser, S.K. 2012. Rare species drive local Mason, N.W.H., de Bello, F., Dolezal, J. & Leps, J. 2011b. Niche trait diversity in two geographically disjunct examples of a overlap reveals the effects of competition, disturbance and
Journal of Vegetation Science 832 Doi: 10.1111/jvs.12033 © 2012 International Association for Vegetation Science N.W.H. Mason & S. Pavoine Phylogenetic diversity and assembly processes
naturally rare alpine ecosystem in New Zealand. Journal Violle, C., Navas, M.L., Vile, D., Kazakou, E., Fortunel, C., of Vegetation Science 23: 626–639. Hummel, I. & Garnier, E. 2007. Let the concept of trait be Safi, K., Cianciaruso, M.V., Loyola, R.D., Brito, D., Armour- functional!. Oikos 116: 882–892. Marshall, K. & Diniz-Filho, J.A.F. 2011. Understanding Webb, C.O., Ackerly, D.D., McPeek, M.A. & Donoghue, M.J. global patterns of mammalian functional and phylogenetic 2002. Phylogenies and community ecology. Annual Review diversity. Philosophical Transactions of the Royal Society of of Ecology and Systematics 33: 475–505. London. Series B – Biological Sciences 366: 2536–2544. Wright, I.J., Reich, P.B., Westoby, M., Ackerly, D.D., Baruch, Z., Schwinning, S. & Weiner, J. 1998. Mechanisms determining the Bongers, F., Cavender-Bares, J., Chapin, T., Cornelissen, degree of size asymmetry in competition among plants. J.H.C., Diemer, M., Flexas, J., Garnier, E., Groom, P.K., Gu- Oecologia 113: 447–455. lias, J., Hikosaka, K., Lamont, B.B., Lee, T., Lee, W., Lusk, C., Spasojevic, M.J. & Suding, K.N. 2011. Inferring community Midgley, J.J., Navas, M.L., Niinemets, U., Oleksyn, J., Osada, assembly mechanisms from functional diversity patterns: the N., Poorter, H., Poot, P., Prior, L., Pyankov, V.I., Roumet, C., importance of multiple assembly processes. Journal of Ecology Thomas, S.C., Tjoelker, M.G., Veneklaas, E.J. & Villar, R. 100: 652–661. 2004. The worldwide leaf economics spectrum. Nature 428: Srivastava, D.S., Cadotte, M.W., MacDonald, A.A.M., Marushia, 821–827. R.G. & Mirotchnick, N. 2012. Phylogenetic diversity and the functioning of ecosystems. Ecology Letters 15: 637–648. Supporting Information Swenson, N.G. & Enquist, B.J. 2009. Opposing assembly mecha- nisms in a Neotropical dry forest: implications for phylogenetic Additional supporting information may be found in the and functional community ecology. Ecology 90: 2161–2170. online version of this article: Villeger, S., Mason, N.W.H. & Mouillot, D. 2008. New multidi- mensional functional diversity indices for a multifaceted Data S1. Details of assembly model. framework in functional ecology. Ecology 89: 2290–2301.
Journal of Vegetation Science Doi: 10.1111/jvs.12033 © 2012 International Association for Vegetation Science 833 Journal of Vegetation Science 24 (2013) 834–842 SPECIAL FEATURE: FUNCTIONAL DIVERSITY Contrasting effects of productivity and disturbance on plant functional diversity at local and metacommunity scales Etienne Laliberte, David A. Norton & David Scott
Keywords Abstract Assembly rules; Disturbance; Environmental filtering; Functional diversity; Land-use change; Questions: Is trait convergence more intense when soil resource availability Limiting similarity; Null models and disturbance constrain productivity and limit above-ground competition? Do the effects of productivity and disturbance on functional diversity differ between Abbreviations the local and metacommunity scales? ANPP = above-ground net primary productivity; FDis = functional dispersion; Location: Semi-arid grasslands in New Zealand (43°59′ S, 170°27′ E). FRic = functional richness; NPP = net primary productivity Methods: We measured trait convergence and divergence in grasslands along gradients of primary productivity and disturbance at local (i.e. 1 m 9 1m)and Nomenclature metacommunity (8 m 9 50 m) scales, using long-term (27-yr) manipulations Moore & Edgar 1970; Healy & Edgar 1980; of soil resource availability and grazing intensity. We compared trait dispersion Allan 1982; Webb et al. 1988; Edgar & Connor metrics to those expected under different null models. 2000 Results: At the metacommunity scale, we found stronger trait convergence Received 2 November 2011 with increasing productivity and grazing intensity, where all short, slow- Accepted 14 December 2012 growing species were excluded from the potential species pool. However, Co-ordinating Editor: Norman Mason once this broad-scale filter on species pool was taken into account, we found that at the local scale, abundance-weighted functional dispersion of co-occur- Laliberte, E. (corresponding author, etienne. ring species was stronger than expected under our null model, thereby suggest- [email protected]): School of Plant Biology ing limiting similarity. Moreover, trait divergence became stronger at higher (M090), The University of Western Australia, 35 productivity and lower grazing intensity, where size-asymmetric competition Stirling Highway, Crawley, WA, 6009, Australia for light is likely to have been more intense. Norton, D.A. ([email protected]): School of Conclusions: At the metacommunity scale, environmental filtering led to Forestry, University of Canterbury, Private Bag species with particular traits being excluded from the species pool. In contrast, at 4800, Christchurch, 8013, New Zealand the local community scale where individuals interact, there was evidence of lim- Scott, D. ([email protected]): iting similarity. Our results suggest that environmental filtering and limiting 4 Murray Place, Lake Tekapo, New Zealand similarity are not mutually exclusive and jointly determine community structure, but can operate at different spatial scales.
(2) ‘limiting similarity’, whereby species that co-occur at a Introduction local scale differ significantly from each other in particular The assembly of plant communities is thought to be largely traits related to resource acquisition and/or use (Weiher & driven by two distinct processes: (1) ‘environmental filter- Keddy 1995; Grime 2006). The importance of these two ing’, whereby the pool of species that can potentially colo- distinct forces during community assembly is commonly nize a site (i.e. the potential species pool) is filtered in such inferred from patterns of trait dispersion (Gotzenberger€ a way that the plant species that can actually grow at that et al. 2012). Trait convergence is typically associated with site (i.e. the ‘realized’ species pool) are more functionally environmental filtering (Weiher & Keddy 1995; D ıaz et al. similar to each other than expected by chance, and 1998; Lebrija-Trejos et al. 2010), because species with
Journal of Vegetation Science 834 Doi: 10.1111/jvs.12044 © 2012 International Association for Vegetation Science E. Laliberteetal. Resource availability, grazing, trait dispersion traits that are poorly suited to local environmental condi- metacommunity-scale environmental filtering on poten- tions cannot establish or occur at very low abundance. On tial species pools is taken into account (de Bello et al. the other hand, trait divergence is thought to result from 2012). Testing these hypotheses requires an experimental local competition during community assembly (Weiher & design where both productivity and disturbance intensity Keddy 1995; Stubbs & Wilson 2004; Mason et al. 2011) are manipulated independently. In particular, it has been since classical competition theory states that there is a limit argued that trait-based community assembly is best studied to how similar co-occurring species can be (MacArthur & by constructing synthetic communities from a common Levins 1967). species pool, experimentally altering environmental condi- Environmental filtering and associated trait conver- tions and/or disturbance regimes, and following long-term gence is generally expected to become more intense where changes in species composition and trait distributions strong environmental constraints to plant establishment or (Grime 2002). growth are present (e.g. low productivity and/or high dis- In this study, we explore trait-based plant community turbance; Weiher & Keddy 1995). On the other hand, assembly in a 27-yr experiment conducted in New Zealand Grime (2006) predicted that higher productivity should grasslands, where a common pool of 25 plant species was lead to trait convergence due to competitive displacement sown into the resident vegetation within a 3-ha area in of poorer competitors, and that this should be most appar- 1982, after which soil resource availability and grazing ent for ‘core’ vegetative traits that strongly relate to growth intensity were experimentally manipulated (Scott 1999). rate and competitive ability, such as specific leaf area, leaf We evaluate plant trait dispersion patterns at both the dry matter content, leaf nutrient concentration or plant metacommunity scale (8 m 9 50 m) as well as the local height (Grime 2006). In contrast, higher disturbance (1 m 9 1 m) scale in order to distinguish environmental intensity should promote trait divergence by suppressing filtering effects on the potential species pool from local bio- competitive dominance (Grime 2006). Such divergence tic processes (de Bello et al. 2012). This unique long-term should be stronger for traits related to regeneration and experiment allowed us to explore in a controlled setting dispersal because disturbance increases the likelihood that how productivity and disturbance together influence pat- species with different regeneration niches could co-occur terns of trait convergence and divergence in plant commu- (Grubb 1977). An additional complexity is that some stud- nities. Specifically, we ask the following questions: ies have found evidence for convergence in vegetative 1 Is trait convergence at the metacommunity scale more traits and divergence in regenerative traits (Swenson & En- intense under strong abiotic regimes (e.g. low productivity quist 2009), whereas others found convergence for several and/or high disturbance)? traits under both higher productivity and disturbance 2 Is trait divergence at the local community scale more intensity (Pakeman et al. 2011), the latter of which does intense under environmentally benign conditions (e.g. not support predictions from Grime (2006). In addition, high productivity and/or low disturbance), where asym- Mason et al. (2012) found convergence of leaf traits with metric competition for light may become more important? decreasing soil fertility (i.e. total phosphorus concentra- tion), while Spasojevic & Suding (2012) observed greater Methods plant functional diversity at both low and high soil Study area and site resource availability, although for different functional traits. Clearly, whether trait convergence and divergence The study was conducted on the AgResearch trial site at vary with productivity and disturbance in the way pre- Mount John, west of Lake Tekapo in the Mackenzie Basin dicted by Grime (2006) still remains unclear. of New Zealand’s South Island (43°59′S, 170°27′E, 820 m One possibility is that the effects of productivity on trait a.s.l.). Details about the climate, soils and settlement convergence or divergence depend on disturbance inten- history of the study area and study site are available sity, and vice versa, as predicted by some models of species elsewhere (Scott 1999). The dominant vegetation type co-existence (Huston 1979, 1994). For example, at low prior to human settlement in the area is likely to have been productivity, high disturbance intensity may lead to trait short-tussock grassland with a variable woody component convergence if only specific plant functional types can per- (McGlone 2001). Prior to the start of the experiment, vege- sist under such extreme conditions. Conversely, at high tation was a depleted native fescue tussock (F. novae-zelan- productivity, greater disturbance intensity may prevent diae) grassland dominated by the exotic mouse-ear trait convergence by suppressing competitive dominance hawkweed (Hieracium pilosella) and around 30–40 other and allowing the co-existence of functionally dissimilar plant species. This vegetation type is representative of large species. In addition, trait convergence and divergence areas of New Zealand rangelands (Wardle 1991). Detailed depend on how species pools are defined, such that local vegetation analyses are available elsewhere (Scott 2001, trait divergence may only become apparent once any 2007; Laliberte et al. 2010).
Journal of Vegetation Science Doi: 10.1111/jvs.12044 © 2012 International Association for Vegetation Science 835 Resource availability, grazing, trait dispersion E. Laliberteetal.
Experimental design adjusted based on residual bulk of the moderate grazing treatment (i.e. height of 1–2 cm). The grazing intensity The experiment is described in detail by Scott (1999) and treatment is relative (i.e. within each whole plot), and not is only summarized here. In 1982, 25 agricultural grass absolute, as the annual sheep grazing days achieved and legume pasture species were sown using a rotary hoe depended on the forage growth of the different fertilizer drill within a 3-ha area. The trial followed a split-plot treatments. This is a key component of the experimental design consisting of two spatial replications (blocks; ca. design that increases realism and relevance, since herbi- 1.5 ha each), each split into five whole plots (50 m vore density and production depends on primary produc- 9 50 m) receiving one of the following five nominal fer- tion in ecosystems (McNaughton et al. 1989). Grazing tilizer treatments: 0, 50, 100, 250 and 500 kg ha 1 yr 1 of occurs in the period November–May. sulphur-fortified superphosphate (i.e. a P/S fertilizer). Pre- vious studies from this study site showed that pasture Vegetation sampling growth responded to the addition of P and S, and that irri- gation of fertilized plots strongly increased production Sampling of all vascular plant species present within each (Scott 2000). A recent study (Laliberte et al. 2012b) 8m9 50-m plot was undertaken in November 2007 and showed that unfertilized plots were dominated by plants 2008 (Laliberte et al. 2010). Twenty 1 m 9 1-m quadrats with a conservative nutrient-use strategy, whereas plants were randomly positioned along two longitudinal tran- with leaf traits associated with rapid growth were more sects (ten quadrats per transect) in each plot. Transects abundant in fertilized plots. The fertilizer was applied were 3 m apart from each other and 2 m from the closest annually in early spring. The whole plots receiving fence. Cover (i.e. vertical projection of canopy, including 500 kg ha 1 yr 1 were also irrigated fortnightly from living and non-living components) of all vascular plant November to May of each year to represent the highest- species present in each 1 m 9 1-m quadrat was recorded intensity agriculture typical of the region. Fertilizer was using a seven point semi-quantitative scale (1, 0.1%; 2, applied each year for the first 19 yr of the experiment and 0.1–0.9%; 3, 1–5%; 4, 5–25%; 5, 25–50%; 6, 51–75%; 7, not applied since. 76–100%). Percentage cover per species per quadrat was Each whole plot was further split into six 8 m 9 50-m taken as the median of the percentage cover of the class. subplots (thereafter simply referred to as ‘plots’) corre- Cover values for 2007 and 2008 were averaged for each sponding to a two-way factorial design involving sheep species/quadrat combination. Although cover values are grazing intensity (lax, moderate and hard) and stocking not as reliable as biomass data for estimating species rela- type (mob vs. sustained). We consider this ‘plot’ (i.e. tive abundances, a previous study found that measures of 8m9 50 m) scale to represent a metacommunity since it functional diversity derived from cover values were simi- is larger than the scale at which individual plants interact lar to those derived from biomass data (Lavorel et al. and includes a set of similar local communities that are 2008). linked through dispersal (Leibold et al. 2004). In mob-grazing plots, a larger number of sheep (with Plant functional traits actual numbers depending on available feed-on-offer of the moderate plot) were introduced to plots for 3–4days, A previous principal components analysis (PCA) measured while sustained grazing plots received a lower number of from 14 plant functional traits measured on all plant sheep for a longer period (i.e. several weeks). However, species from our experiment showed that most of the trait only plots corresponding to the mob stocking type were variation (74%) between species was represented by the used in the present study as this made it possible to quan- first two PCA axes (Laliberte et al. 2012b). Leaf nitrogen tify the proportion of above-ground net primary produc- (N) concentration had the highest loading on the first PCA tion (ANPP) consumed (or destroyed) by sheep (Laliberte axis (Laliberte et al. 2012b), which is associated with the & Tylianakis 2012). Lax, moderate and hard grazing inten- fundamental leaf economics spectrum (Wright et al. sity levels corresponded to a ratio of 1:2:4 sheep grazing 2004). On the other hand, plant height had the strongest days, respectively, in the years 2–4 of the experiment, and loading on the second PCA axis, which we interpreted as a a ratio of 2:3:4 in subsequent years. Plots were always gradient of competitive ability for light resources (Laliberte grazed in groups of three, corresponding to the three mob- et al. 2012b). In this study, we restricted our analyses to stocked grazing intensity levels per whole plot. For each these two ecologically relevant traits, which together rep- grazing event (i.e. when vegetation had reached ca. 30 cm resented the most important axes of functional variation in height), sheep numbers were adjusted based on avail- among our species. Details on trait measurements are able feed-on-offer of the moderate grazing treatment. The available in previous studies (Laliberte & Tylianakis 2012; duration of grazing was the same for all three plots but was Laliberte et al. 2012b).
Journal of Vegetation Science 836 Doi: 10.1111/jvs.12044 © 2012 International Association for Vegetation Science E. Laliberteetal. Resource availability, grazing, trait dispersion
Productivity and grazing intensity indicate trait divergence, while negative values suggest trait convergence or environmental filtering (Ingram & We measured total (i.e. above- and below-ground) net Shurin 2009). Finally, we used linear mixed models to primary productivity (NPP) over an 18-mo period for each test for the interactive effects of productivity and grazing plot. Grazing intensity was measured as the proportion of intensity on functional richness and FRicdev,withran- ANPP consumed (or destroyed) by sheep over the 18-mo dom intercepts per block and whole plot in order to take sampling period (Laliberte & Tylianakis 2012; Laliberte into account the split-plot nature of the experiment. In et al. 2012a). Details of these measurements are available these linear mixed models, NPP and grazing intensity elsewhere (Laliberte & Tylianakis 2012; Laliberteetal. were first log-transformed and centred. Log transforma- 2012a). tion was used to linearize relationships, whereas centring was used so that unstandardized coefficients for a predic- Statistical analyses tor represented its average effect on the response variable when the conditioning variable (i.e. the other predictor Following Mason et al. (2005), we measured different fac- involved in the interaction) is held at its mean value ets of functional diversity to focus on different community (Aiken & West 1991). assembly processes at different scales. First, we focused on functional richness to detect environmental filtering at the Local-scale interactions and functional dispersion metacommunity scale. Second, we used a functional diver- gence measure to detect the influence of biotic interactions To explore whether fine-scale interactions between plants on trait dispersion patterns at local scales. (e.g. resource competition) could lead to trait divergence within local communities, we used the vegetation cover data at the 1 m 9 1-m scale, the scale at which most indi- Community-scale environmental filtering and functional richness vidual plants in the quadrat are expected to interact below- We measured functional richness (FRic; Villeger et al. and/or above-ground. We measured abundance-weighted 2008; Laliberte & Legendre 2010) for each 8 m 9 50-m functional dispersion for all 1 m 9 1-m quadrats and plot in order to quantify the strength of trait-based envi- averaged it over the 20 quadrats at the metacommunity ronmental filtering at the metacommunity scale (Cornwell (8 m 9 50-m plot) level. Functional dispersion is the et al. 2006), and to determine whether the strength or weighted average distance of individual species to the cen- direction of environmental filtering could depend on pro- troid of all species within a community, where weights are ductivity and/or grazing intensity. Leaf [N] and plant species relative abundances, and is the multivariate ana- height were first log-transformed to reduce the importance logue of the weighted mean absolute deviation (Laliberte of species with large trait values. & Legendre 2010). FDis is related to Rao’s quadratic We used a null model approach to test whether the entropy (Botta-Dukat 2005; Laliberte & Legendre 2010), observed functional richness in each plot was lower or which is the multivariate analogue of the weighted vari- higher than expected if metacommunity assembly was ance, but FDis is by construction less sensitive than Rao’s not influenced by species traits (i.e. leaf [N] and plant Q to species with extreme trait values. height). To do so, we created 999 random species compo- To test for trait divergence, we used a different null sition matrices where species richness per plot (i.e. row) model than that described above for functional richness. was preserved, and where column marginal frequencies First, and most importantly, we restricted the randomization were used as probabilities. The randomization was done between 1 m 9 1-m quadrats within each 8 m 9 50-m using the ‘commsimulator’ function (‘r1′ algorithm) in plot, in order to detect local-scale assembly processes the vegan R package (R Foundation for Statistical Com- independent of any filtering effect on species pools (Mason puting, Vienna, AT). We then computed standardized et al. 2011; de Bello et al. 2012). To do so, we kept species deviations from the null expectation (FRicdev)foreach abundance matrices for each plot (i.e. 20 quadrats as rows) plot: constant but randomized rows within the species 9 trait matrix (Stubbs & Wilson 2004), using 999 randomizations. l ¼ FRicobs null We then computed standardized deviations from the null FRicdev r null expectation (FDisdev) using the same approach described
above for FRicdev, and also used linear mixed models to test where FRicobs is the observed functional richness for a for the interactive effects of productivity and grazing inten- 9 l metacommunity (i.e. 8 m 50 m plot), null is the mean sity on FDisdev. Again, variables were log-transformed to of the null distribution of functional richness values, and remove the influence of large values and to linearize rnull is its standard deviation. Positive FRicdev values relationships, and predictors were centred.
Journal of Vegetation Science Doi: 10.1111/jvs.12044 © 2012 International Association for Vegetation Science 837 Resource availability, grazing, trait dispersion E. Laliberteetal.
Results Table 1. Results of linear mixed-effect models for (a) functional richness (FRic) and (b) FRicdev at the metacommunity (i.e. 8 m 9 50-m plot) scale. Community-scale environmental filtering and functional b = unstandardized regression coefficient; r = standard error; richness df = degrees of freedom; Prod = productivity.
At the metacommunity (i.e. 8 m 9 50-m plot) scale, there brdf t P was a highly significant (P 0.001) decrease in func- (a) tional richness with increasing productivity (Fig. 1a, Intercept 5.3374 0.3018 17 17.6824 <0.0001 Table 1a). Patterns of functional richness were not simply Productivity 5.3648 1.3544 17 3.9611 0.0010 Grazing intensity 1.5353 2.0144 17 0.7622 0.4564 driven by changes in species richness, since FRicdev across 9 all plots was significantly lower than zero (lower tail t-test, Prod Grazing 1.9752 7.3483 17 0.2688 0.7913 (b) P 0.0001; Fig. 1b). Importantly, FRic decreased with dev Intercept 0.1129 0.0274 17 4.1153 0.0007 = greater productivity (P 0.003; Table 1b), and the rela- Productivity 0.4800 0.1382 17 3.4728 0.0029 tionship between FRicdev and productivity depended on Grazing intensity 0.2812 0.2128 17 1.3215 0.2039 grazing intensity (productivity 9 grazing intensity interac- Prod 9 Grazing 2.0925 0.7095 17 2.9491 0.0090 tion, P = 0.009; Table 1b), such that reductions in FRicdev with higher productivity became more important under higher grazing intensity (Fig. 1b). The reduction in func- Discussion tional richness under greater productivity was clearly due to the loss of species with low height and/or lower leaf [N] Our study found evidence for both environmental filtering at high productivity, which interestingly were all native (trait convergence) and limiting similarity (trait diver- species (Fig. 2). gence), but at different spatial scales. At the metacommu- nity scale, plant species that encompassed almost the entire spectrum of functional variation from the potential Local-scale interactions and functional dispersion species pool were present at low productivity. Conversely, At the local (i.e. 1 m 9 1-m quadrat) scale, there were no with higher productivity, functional richness decreased significant (P > 0.29; Table 2a) changes in abundance- and was much smaller than expected under a null model, weighted functional dispersion with productivity or graz- indicating that metacommunity membership was ing intensity. However, by comparing the observed func- restricted to a non-random subset of ecological strategies tional dispersion to the within-metacommunity null relative to those found across the entire species pool. Still, expectation, we found that FDisdev was significantly higher once this metacommunity-scale environmental filtering than zero (t-test, P 0.0001; Fig. 3) across all plots, sug- was taken into account, we found evidence for limiting gesting that trait divergence was predominant at the local similarity at the local scale (where individual plants are scale. Importantly, FDisdev became significantly (P expected to interact), in that abundance-weighted func- 0.02; Table 2b) more positive with greater productivity tional dispersion was higher than expected under random (Fig. 3a) and lower grazing intensity (Fig. 3b), suggesting community assembly. Importantly, at the local scale we stronger local trait divergence where competition for light found stronger-than-expected trait divergence at higher is thought to be most intense. productivity and at lower grazing intensity, conditions
(a) (b)
Fig. 1. Changes in (a) functional richness (FRic) and (b) standardized deviations from the null model (FRicdev) with net primary productivity (NPP). Fert = fertilizer level. Graz = grazing intensity.
Journal of Vegetation Science 838 Doi: 10.1111/jvs.12044 © 2012 International Association for Vegetation Science E. Laliberteetal. Resource availability, grazing, trait dispersion
Fig. 2. Functional richness under increasing productivity (columns, from left to right) and grazing intensity (rows, from top to bottom). Each panel represents an individual 8 m 9 50-m plot (i.e. ‘metacommunity’). In each panel, all species present in each plot are plotted in a two-dimensional niche space defined by leaf [N] and plant height. Axes are on logarithmic scales. Black circles indicate native species, while grey triangles indicate exoticspecies.
Table 2. Results of linear mixed-effect models for (a) functional dispersion competition on patterns of trait convergence (at the meta- 9 (FDis) and (b) FDisdev at the local community (i.e. 1 m 1-m quadrat) community scale) and divergence (at the local scale) b = r = scale. unstandardized regression coefficient; standard error; become stronger at higher productivity, presumably due to = = df degrees of freedom; Prod productivity. shifts from size-symmetric below-ground competition to brdf t P asymmetric above-ground competition (Aerts 1999). (a) The strong trait convergence at higher productivity at Intercept 0.1813 0.0130 17 13.9872 <0.0001 the metacommunity scale was due to the disappearance of Productivity 0.0550 0.0500 17 1.1009 0.2863 (1) all short, slow-growing (i.e. low leaf [N]) native species; Grazing intensity 0.0408 0.0769 17 0.5314 0.6020 (2) Hieracium pilosella, the short exotic species with moder- Prod 9 Grazing 0.1145 0.2563 17 0.4470 0.6605 ately high leaf [N] that dominated all plots prior to the start (b) of the experiment (Scott 2007; Laliberte et al. 2012b); and Intercept 0.3930 0.0699 17 5.6185 <0.0001 (3) Festuca novae-zelandiae and Poa colensoi, two relatively Productivity 0.8754 0.3287 17 2.6629 0.0164 Grazing intensity 1.3156 0.4963 17 2.6505 0.0168 tall native tussock grass species with traits associated with Prod 9 Grazing 1.6251 1.7458 17 0.9309 0.3649 slow growth rates (Laliberte et al. 2012b). In trait-based community assembly analyses, a restricted range (or restricted multivariate volume) in functional attributes has which are expected to lead to stronger above-ground com- generally been attributed to abiotic environmental filter- petition between plants (Huston 1994; Grime 2002; Mason ing, whereby unusually stressful abiotic conditions only et al. 2011). Overall, our results show that the effects of allow species with evolved tolerances to these conditions
Journal of Vegetation Science Doi: 10.1111/jvs.12044 © 2012 International Association for Vegetation Science 839 Resource availability, grazing, trait dispersion E. Laliberteetal.
(a) (b)
Fig. 3. Changes in abundance-weighted functional dispersion (FDis) and standardized deviations from the null model (FDisdev)with(a) net primary productivity (NPP) and (b) grazing intensity. Fert = fertilizer level. to persist in the community (Van der Valk 1981; Weiher & faster growth rates (Craine & Lee 2003). This may reflect Keddy 1995; Weiher et al. 1998; Cornwell et al. 2006; the particular evolutionary history of New Zealand grass- Kraft et al. 2008; Cornwell & Ackerly 2009; Ingram & Shu- lands, which prior to human settlement were confined to rin 2009). However, in our study, we suggest that such sites with marginal environmental conditions (McGlone trait convergence resulted from strong competition for 2001). light and/or soil resources from faster-growing species at In contrast to trait convergence found at the metacom- high productivity, which over the >27 yr duration of the munity scale under greater productivity, we observed trait experiment led to the competitive exclusion of plant spe- divergence at the local scale. This local trait divergence cies with inherently slower growth, particularly those of became stronger at higher biomass and lower grazing short stature. Stronger trait convergence with higher pro- intensity, perhaps due to stronger competition for light ductivity is consistent with predictions from Grime (2006), (Mason et al. 2012). Trait divergence in plant communities showing that environmental filtering is not always due to has generally been interpreted as evidence for a dominant changes in abiotic conditions but can also result from role of competitive interactions in structuring local assem- inter-specific interactions. However, contrary to our blages (Weiher & Keddy 1995; Weiher et al. 1998; Stubbs hypothesis, higher grazing intensity strengthened the & Wilson 2004), whereby competition for resources entails environmental filtering at higher productivity, which limiting similarity with respect to strategies of resource could be due to negative effects of herbivores, such as acquisition and use (MacArthur & Levins 1967). Our find- trampling (Olff & Ritchie 1998). ing that trait divergence was higher at higher productivity As productivity and grazing intensity increased, we is consistent with the view that competition intensity found that environmental filtering was strongest for native increases with productivity (Grime 2002), even though we species. Native species in this study were functionally dis- found evidence for limiting similarity across all plots, tinct from exotic species, with exotic species exhibiting a regardless of productivity. In addition, our study showed rapid growth trait syndrome, while the reverse was true stronger evidence for limiting similarity under lower graz- for natives. These results are consistent with those from a ing intensity, where competition for light is also expected recent global meta-analysis comparing leaf traits among to be higher because grazing can suppress competitive co-occurring exotic and native species, where exotic dominance (Grime 1973; Laliberte et al. 2012a). Our species were positioned further along the acquisition–con- results are consistent with those of Mason et al. (2011), servation axis towards a faster growth strategy (Leishman who found lower niche overlap in more productive grass- et al. 2007). The difference in leaf traits between native lands and in the absence of mowing. and exotic species observed in our study may be partly Our results suggest that environmental filtering and explained by the fact that all sown species at the start of limiting similarity are both important in determining com- the experiment were exotic species that had been selected munity structure, but operate differently. Environmental based on their potential suitability as pasture species, of filtering shapes species pools by excluding species that which high intrinsic growth rate is a key characteristic. have poor local fitness. In our study, the stronger filtering Other studies that compared co-occurring herbaceous at high productivity and grazing intensity likely resulted native and exotic species in New Zealand grasslands also from competitive exclusion and negative effects associated found that exotic species had faster intrinsic growth rates with grazing over the >27-yr duration of the experiment. compared with native ones (Scott 1970; King & Wilson On the other hand, limiting similarity acts at a local scale 2006) or possessed leaf and root attributes associated with and allows species to co-exist by limiting niche overlap
Journal of Vegetation Science 840 Doi: 10.1111/jvs.12044 © 2012 International Association for Vegetation Science E. Laliberteetal. Resource availability, grazing, trait dispersion
(Mason et al. 2011), thus reducing inter-specific competi- Cornwell, W.K., Schwilk, D.W. & Ackerly, D.D. 2006. A trait- tion relative to intra-specific competition (Chesson 2000). based test for habitat filtering: convex hull volume. Ecology An important finding of our study was that both the 87: 1465–1471. strength of environmental filtering and limiting similarity Craine, J.M. & Lee, W.G. 2003. Covariation in leaf and root traits were not mutually exclusive, but depended on spatial scale for native and non-native grasses along an altitudinal gradi- – as well as productivity and disturbance. Further work ent in New Zealand. Oecologia 134: 471 478. along productivity and disturbance gradients in other Dıaz, S., Cabido, M. & Casanoves, F. 1998. Plant functional traits non-grassland systems is required to evaluate the generality and environmental filters at a regional scale. Journal of Vege- – of this finding. tation Science 9: 113 122. Edgar, E. & Connor, H.E. 2000. Flora of New Zealand, Vol. V. Maa- naki Whenua Press, Lincoln, NZ. € Acknowledgements Gotzenberger, L., de Bello, F., Brathen, K.A., Davison, J., Dubuis, A., Guisan, A., Leps, J., Lindborg, R., Moora, M., Partel,€ M., We wish to thank P. Fortier for help with fieldwork, Pellissier, L., Pottier, J., Vittoz, P., Zobel, K. & Zobel, M. 2012. L. Kirk, A. Leckie, N. Pink and J. M. Tylianakis for Ecological assembly rules in plant communities—approaches, logistical and academic support, and A. Simpson for patterns and prospects. Biological Reviews 87: 111–127. the use of stock. We also thank R. K. Didham, N. Grime, J.P. 1973. Competitive exclusion in herbaceous vegeta- Gross,H.Lambers,W.GLee,N.W.H.Mason,M.M. tion. Nature 242: 344–347. Mayfield, B. Shipley, E. Weiher, J. A. Wells and anon- Grime, J.P. 2002. Plant strategies, vegetation processes, and ecosystem ymous reviewers for providing insightful comments on properties, 2nd ed. John Wiley & Sons, Chichester, UK. previous versions of the manuscript. Field research was Grime, J.P. 2006. Trait convergence and trait divergence in her- funded by the Miss E. L. Hellaby Indigenous Grassland baceous plant communities: mechanisms and consequences. – Research Trust and the School of Forestry, University Journal of Vegetation Science 17: 255 260. of Canterbury. E. L. was supported by the University Grubb, P.J. 1977. The maintenance of species-richness in plant of Western Australia, the University of Canterbury, the communities: the importance of the regeneration niche. Bio- 52: 107–145. Fonds queb ecois de recherche sur la nature et les tech- logical Reviews Healy, A.J. & Edgar, E. 1980. Flora of New Zealand, Vol. III: adven- nologies (FQRNT), Education New Zealand, and the tive cyperaceous, petalous & spathaceous monocotyledons.Botany Australian Research Council (ARC). Division, Department of Scientific and Industrial Research, Lincoln, NZ. References Huston, M.A. 1979. A general hypothesis of species diversity. American Naturalist 113: 81–101. Aerts, R. 1999. Interspecific competition in natural plant com- Huston, M.A. 1994. Biological diversity. Cambridge University munities: mechanisms, trade-offs and plant–soil feedbacks. Press, Cambridge, UK. Journal of Experimental Botany 50: 29–37. Ingram, T. & Shurin, J.B. 2009. Trait-based assembly and phylo- Aiken, L.S. & West, S.G. 1991. Multiple regression: testing and inter- genetic structure in northeast Pacific rockfish assemblages. preting interactions. Sage, Newbury Park, CA, US. Ecology 90: 2444–2453. Allan, H.H. 1982. Flora of New Zealand, Vol. I: indigenous trache- King, W.M. & Wilson, J.B. 2006. Differentiation between native ophyta - psilopsida, lycopsida, filicopsida, gymnospermae, dicotyle- and exotic plant species from a dry grassland: fundamental dons. Botany Division, Department of Scientific and responses to resource availability, and growth rates. Austral Industrial Research, Wellington, NZ. Ecology 31: 996–1004. de Bello, F., Price, J.N., Munkem€ uller,€ T., Liira, J., Zobel, M., Kraft, N.J.B., Valencia, R. & Ackerly, D.D. 2008. Functional traits Thuiller, W., Gerhold, P., Gotzenberger,€ L., Lavergne, S., and niche-based tree community assembly in an Amazonian Lepsˇ, J., Zobel, K. & Partel,€ M. 2012. Functional species pool forest. Science 322: 580–582. framework to test for biotic effects on community assembly. Laliberte, E. & Legendre, P. 2010. A distance-based framework Ecology 93: 2263–2273. for measuring functional diversity from multiple traits. Ecol- Botta-Dukat, Z. 2005. Rao’s quadratic entropy as a measure of ogy 91: 299–305. functional diversity based on multiple traits. Journal of Vegeta- Laliberte, E. & Tylianakis, J.M. 2012. Cascading effects of long- tion Science 16: 533–540. term land-use changes on plant traits and ecosystem func- Chesson, P. 2000. Mechanisms of maintenance of species diver- tioning. Ecology 93: 145–155. sity. Annual Review of Ecology and Systematics 31: 343–366. Laliberte, E., Norton, D.A., Tylianakis, J.M. & Scott, D. 2010. Cornwell, W.K. & Ackerly, D.D. 2009. Community assembly Comparison of two sampling methods for quantifying and shifts in plant trait distributions across an environmental changes in vegetation community structure under range- gradient in coastal California. Ecological Monographs 79: 109– land development. Rangeland Ecology & Management 63: 537– 126. 545.
Journal of Vegetation Science Doi: 10.1111/jvs.12044 © 2012 International Association for Vegetation Science 841 Resource availability, grazing, trait dispersion E. Laliberteetal.
Laliberte, E., Lambers, H., Norton, D., Tylianakis, J. & Huston, Scott, D. 1970. Relative growth rates under controlled tempera- M. 2012a. A long-term experimental test of the dynamic tures of some New Zealand indigenous and introduced equilibrium model of species diversity. Oecologia 171: 439– grasses. New Zealand Journal of Botany 8: 369–379. 448. Scott, D. 1999. Sustainability of New Zealand high-country pas- Laliberte, E., Shipley, B., Norton, D.A. & Scott, D. 2012b. Which tures under contrasting development inputs. 1. Site, and plant traits determine abundance under long-term shifts in shoot nutrients. New Zealand Journal of Agricultural Research soil resource availability and grazing intensity? Journal of 42: 365–383. Ecology 100: 662–677. Scott, D. 2000. Sustainability of New Zealand high-country pas- Lavorel, S., Grigulis, K., McIntyre, S., Williams, N.S.G., Garden, tures under contrasting development inputs. 4. Organic mat- D., Dorrough, J., Berman, S., Quetier, F., Thebault, A. & Bo- ter components. New Zealand Journal of Agricultural Research nis, A. 2008. Assessing functional diversity in the field – 43: 387–414. methodology matters!. Functional Ecology 22: 134–147. Scott, D. 2001. Sustainability of New Zealand high-country pas- Lebrija-Trejos, E., Perez-Garc ıa, E.A., Meave, J.A., Bongers, F. & tures under contrasting development inputs. 7. Environ- Poorter, L. 2010. Functional traits and environmental filter- mental gradients, plant species selection, and diversity. New ing drive community assembly in a species-rich tropical sys- Zealand Journal of Agricultural Research 44: 59–90. tem. Ecology 91: 386–398. Scott, D. 2007. Sustainability of high-country pastures under Leibold, M.A., Holyoak, M., Mouquet, N., Amarasekare, P., contrasting development inputs. 9. Vegetation dynamics. Chase, J.M., Hoopes, M.F., Holt, R.D., Shurin, J.B., Law, R., New Zealand Journal of Agricultural Research 50: 396–406. Tilman, D., Loreau, M. & Gonzalez, A. 2004. The metacom- Spasojevic, M.J. & Suding, K.N. 2012. Inferring community munity concept: a framework for multi-scale community assembly mechanisms from functional diversity patterns: the ecology. Ecology Letters 7: 601–613. importance of multiple assembly processes. Journal of Ecology Leishman, M.R., Haslehurst, T., Ares, A. & Baruch, Z. 2007. Leaf 100: 652–661. trait relationships of native and invasive plants: community- Stubbs, W.J. & Wilson, J.B. 2004. Evidence for limiting similarity and global-scale comparisons. New Phytologist 176: 635–643. in a sand dune community. Journal of Ecology 92: 557–567. MacArthur, R.H. & Levins, R. 1967. The limiting similarity, con- Swenson, N.G. & Enquist, B.J. 2009. Opposing assembly mecha- vergence, and divergence of coexisting species. American Nat- nisms in a Neotropical dry forest: implications for phyloge- uralist 101: 377–385. netic and functional community ecology. Ecology 90: 2161– Mason, N.W.H., Mouillot, D., Lee, W.G. & Wilson, J.B. 2005. 2170. Functional richness, functional evenness and functional Van der Valk, A.G. 1981. Succession in wetlands: a Gleasonian divergence: the primary components of functional diversity. approach. Ecology 62: 688–696. Oikos 111: 112–118. Villeger, S., Mason, N.W.H. & Mouillot, D. 2008. New multi- Mason, N.W.H., de Bello, F., Dole zal, J. & Lepsˇ,J.2011.Niche dimensional functional diversity indices for a multifac- overlap reveals the effects of competition, disturbance and eted framework in functional ecology. Ecology 89: contrasting assembly processes in experimental grassland 2290–2301. communities. Journal of Ecology 99: 788–796. Wardle, P. 1991. Vegetation of New Zealand. Cambridge University Mason, N.W.H., Richardson, S.J., Peltzer, D.A., de Bello, F., Press, Cambridge, UK. Wardle, D.A. & Allen, R.B. 2012. Changes in coexistence Webb, C.J., Sykes, W.R. & Garnock-Jones, P.J. 1988. Flora of mechanisms along a long-term soil chronosequence New Zealand, Vol. IV: naturalised pteridophytes, gymnosperms, revealed by functional trait diversity. Journal of Ecology dicotyledons. Botany Division, Department of Scientific and 100: 678–689. Industrial Research, Wellington, NZ. McGlone, M.S. 2001. The origin of native grasslands of south- Weiher, E. & Keddy, P.A. 1995. Assembly rules, null models, eastern South Island in relation to pre-human woody eco- and trait dispersion: new questions from old patterns. Oikos systems. New Zealand Journal of Ecology 25: 1–15. 74: 159–164. McNaughton, S.J., Oesterheld, M., Frank, D.A. & Williams, K.J. Weiher, E., Clarke, G.D.P. & Keddy, P.A. 1998. Community 1989. Ecosystem-level patterns of primary productivity and assembly rules, morphological dispersion, and the coexis- herbivory in terrestrial habitats. Nature 341: 142–144. tence of plant species. Oikos 81: 309–322. Moore, L.B. & Edgar, E. 1970. Flora of New Zealand, Vol. II: indige- Wright, I.J., Reich, P.B., Westoby, M., Ackerly, D.D., Baruch, Z., nous tracheophyta - monocotyledons except graminae.Botany Bongers, F., Cavender-Bares, J., Chapin, T., Cornelissen, Division, Department of Scientific and Industrial Research, J.H.C., Diemer, M., Flexas, J., Garnier, E., Groom, P.K., Gu- Wellington, NZ. lias, J., Hikosaka, K., Lamont, B.B., Lee, T., Lee, W., Lusk, C., Olff, H. & Ritchie, M.E. 1998. Effects of herbivores on grassland Midgley, J.J., Navas, M.L., Niinemets, U., Oleksyn, J., Osada, plant diversity. Trends in Ecology & Evolution 13: 261–265. N., Poorter, H., Poot, P., Prior, L., Pyankov, V.I., Roumet, C., Pakeman, R., Lennon, J. & Brooker, R. 2011. Trait assembly in Thomas, S., Tjoelker, M.G., Veneklaas, E.J. & Villar, R. 2004. plant assemblages and its modulation by productivity and The worldwide leaf economics spectrum. Nature 428: disturbance. Oecologia 167: 209–218. 821–827.
Journal of Vegetation Science 842 Doi: 10.1111/jvs.12044 © 2012 International Association for Vegetation Science Journal of Vegetation Science 24 (2013) 843–852 SPECIAL FEATURE: FUNCTIONAL DIVERSITY Functional and phylogenetic community assembly linked to changes in species diversity in a long-term resource manipulation experiment Pille Gerhold, Jodi N. Price, Kersti Pussa,€ Rein Kalamees, Kaia Aher, Ants Kaasik & Meelis Partel€
Keywords Abstract Fertilizer addition; Functional diversity; Habitat filtering; Limiting similarity; Phylogenetic Question: There are contrasting opinions about how communities assemble diversity; Productivity; Resources; Sucrose along a productivity gradient, particularly in relation to competitive interactions. addition; Trait convergence; Trait divergence; One view is that functionally similar, and closely related species, are less likely Weaker competitor exclusion to co-exist (limiting similarity). Alternatively, competitive exclusion may act on dissimilar species because species bearing traits associated with low competitive Abbreviations C = control treatment; F = fertilizer addition ability are excluded (weaker competitor exclusion). We ask if patterns of treatment; FD = functional diversity; HF functional and phylogenetic diversity are related to changes in species diversity = habitat filtering; LS = limiting similarity; PD in response to fertility manipulations. = phylogenetic diversity; S = sucrose addition treatment; WCE = weaker competitor exclusion Location: Species-rich mesophytic grassland in Estonia. Methods: The grassland has been manipulated from 2002 to 2011 to increase Nomenclature (fertilizer addition) and decrease productivity (sucrose addition) in small-scale Ku¨ hn et al. 2004 (50 cm 9 50 cm) plots. We linked annual increases and decreases in species Received 5 April 2012 diversity to changes in functional and phylogenetic diversity. We used abun- Accepted 9 January 2013 dance-weighted mean pair-wise functional or phylogenetic distance of all possi- Co-ordinating Editor: Francesco de Bello ble species pairs. Results: We found convergence in four traits (plant height, leaf distribution, lat- Gerhold, P. (corresponding author, pille. eral spread, type of reproduction) and a decrease in mean functional and phylo- [email protected]), Price, J.N. genetic diversity, in support of weaker competitor exclusion or habitat filtering. € ([email protected]), Pussa, K. There was less support for limiting similarity, with divergence found for two ([email protected]), Kalamees, R. traits associated with decreasing species diversity (leaf distribution in the sucrose ([email protected]), Aher, K. ([email protected]), Kaasik, A. treatment and lateral spread in the fertilizer treatment). € ([email protected]) & Partel, M. Conclusions: Our results support the view that competition can lead to the ([email protected]): Department of Botany, exclusion of weaker competitors, rather than increasing functional and phyloge- Institute of Ecology and Earth Sciences, University of Tartu, Lai 40, 51005, Tartu, Estonia netic diversity, as expected from the principle of limiting similarity. However, Gerhold, P. : Department of Experimental multiple assembly processes, which are generally seen as mutually exclusive, Plant Ecology, Institute for Water and Wetland are operating simultaneously, albeit on different traits and at different stages of Research, Radboud University Nijmegen, PO community assembly. Box 9010, 6500, GL Nijmegen, The Netherlands Price, J.N.: School of Plant Biology, The University of Western Australia, WA 6009, Australia
ecological research, and remain largely unresolved (e.g. Introduction Keddy 1992; Weiher & Keddy 1995; Chesson 2000; Ack- The mechanisms driving non-random assembly patterns erly 2003; Gotzenberger€ et al. 2012; Price et al. 2012). in plant communities have long been of interest in Community assembly is often thought to result from
Journal of Vegetation Science Doi: 10.1111/jvs.12052 © 2013 International Association for Vegetation Science 843 Functional diversity in species loss and gain P. Gerhold et al. abiotic and biotic filters, with habitat filtering (abiotic) (2012), using a new methodology which included the increasing similarity among species, and biotic filters dark diversity component of the species pool (Partel€ et al. (mostly competition) preventing co-existing species from 2011), found evidence for both limiting similarity and being too similar (limiting similarity; MacArthur & Levins weaker competitor exclusion, depending on the traits 1967). Based on these traditional co-existence theories it and the community type considered. is often assumed that functional diversity should be low Conflicting results about the increase or decrease in in regions of strong abiotic stress (abiotic filters causing functional and phylogenetic diversity are also likely due to convergence), and high in regions where competitive intrinsic differences among community types (i.e. grass- interactions are stronger (biotic filters causing divergence; lands vs forests), and along environmental gradients (de Weiher & Keddy 1995). The idea that competition should Bello et al. 2012). The importance of competition differs leave an ‘imprint’ on patterns in communities can be along productivity gradients, but few studies have exam- traced back to Darwin, who noted that species of the ined functional and phylogenetic diversity along environ- same genus tend to be more similar, and hence should mental gradients (but see Pakeman et al. 2011; Mason compete more intensely, than more distantly related spe- et al. 2012; Spasojevic & Suding 2012), and as such, these cies (competition relatedness hypothesis; Darwin 1859). hypotheses remain largely untested. Weiher et al. (1998) Both limiting similarity and the competition relatedness found increased functional diversity with higher levels of hypothesis are based on the same mechanism and, to soil phosphate, in support of limiting similarity. Similarly, date, both have limited empirical support, as evidence for Mason et al. (2012) found divergence at high fertility and habitat filtering or random patterns has often been found convergence at low fertility in rain forest communities (see recent meta-analyses of Cahill et al. 2008; Gotzen-€ along a long-term soil chronosequence. Spasojevic & Sud- berger et al. 2012; Price & Partel€ 2013). Phylogenetic ing (2012) found support for multiple assembly processes, relatedness can be used as a surrogate for functional simi- with abiotic filtering occurring at the low end of a resource larity, although this approach has been criticized (e.g. gradient, producing convergence of some traits, but they from this issue Bernard-Verdier et al. 2013; Pavoine et al. also found increased functional diversity in other traits 2013), it can be useful if trait data are not available and if related to below-ground resource competition. At the the observed traits are phylogenetically conserved (Prin- resource-rich end of the gradient, they found increased zing et al. 2001; Kraft et al. 2007). functional diversity for height and leaf area, in support of The principle of limiting similarity is widely accepted limiting similarity or facilitation. Pakeman et al. (2011) in community assembly theory, but other mechanisms examined functional diversity along productivity and dis- can influence assembly in competitive communities (Hil- turbance gradients, and found for most traits that func- leRisLambers et al. 2012). For example, co-existence tional diversity declined with increased productivity and among species may be due to ‘weaker competitor exclu- disturbance, inconsistent with limiting similarity. The sion’, whereby functionally similar species co-exist in a authors suggest that this could be due to either habitat fil- sort of balanced competition (increased similarity) and tering or weaker competitor exclusion. weaker competitors are excluded (Chesson 2000; May- Experimental approaches can provide greater insight field & Levine 2010; Kunstler et al. 2012). One of the into mechanisms driving community assembly. It is well reasons this mechanism has been overlooked is partly established that increasing productivity (e.g. via fertiliza- methodological, as increased similarity among co-existing tion) in temperate grasslands decreases species richness, by species is usually attributed to habitat filtering, and many increasing biomass and light competition (e.g. Lep s 1999; existing methods fail to separate these processes (de Bello Liira & Zobel 2000; Silvertown et al. 2006; Price & Morgan et al. 2012). These hypotheses are commonly tested 2007; Gross et al. 2009; Hautier et al. 2009). Fewer using null models to compare the trait dissimilarity and/ studies, however, have examined the effect of decreasing or phylogenetic relatedness among co-occurring species productivity on species richness, but opposite effects are to a model expected if communities were assembled at expected to occur, with less competition and decreased random (Weiher & Keddy 1995; Gotelli 2000; de Bello biomass resulting in an increase in species richness 2012; Gotzenberger€ et al. 2012). Null model tests gener- (Blumenthal et al. 2003; Perry et al. 2010). While resource ally assign one of three mechanisms to the observed pat- manipulation experiments are commonly performed, less terns: either patterns do not differ from random is known about the relationship between productivity and (attributed to stochastic or dispersal-based assembly), functional and phylogenetic diversity. Recently, Cadotte & traits/relatedness are more similar/closely related than Strauss (2011) examined phylogenetic patterns of natural expected (attributed to habitat filters), or species are more colonization and extinction in experimentally assembled dissimilar/distantly related than expected (attributed to communities over 2 yr. They found that successful biotic filters; Weiher & Keddy 1995). de Bello et al. colonists tended to be either closely or distantly related to
Journal of Vegetation Science 844 Doi: 10.1111/jvs.12052 © 2013 International Association for Vegetation Science P. Gerhold et al. Functional diversity in species loss and gain community residents, whereas extinctions did not exhibit any strong relatedness patterns. – + We used a long-term resource manipulation experi- (a) LS (c) LS ment to determine patterns of trait diversity and phyloge- Persistent netic relatedness in relation to changes in species species diversity (annual turnover). In a previous study at this Lost experimental site, Liira et al. (2012) found that fertilizer 0 species addition increased biomass and decreased species rich- + – Change in FD Change (b) WCE/HF (d) WCE/HF Gained ness. After 10 yr of manipulations, sucrose addition species decreased biomass, but no changes in species richness were detected. The authors suggest this is likely due to dispersal limitation, as species adapted to less fertile con- ditions are not available in the vicinity (Partel€ et al. 0 2000; Partel€ & Zobel 2007). The ‘missing’ species can also Change in species diversity experience a delay in colonization, and establish only Fig. 1. Hypotheses on the mechanisms of decreased (a, b) and increased afteranevenlongertime(>10 yr; Liira et al. 2012). We (c, d) species diversity in regard to change in functional diversity (FD, expand on this study by examining community assembly calculated as mean pair-wise distance between all species) in a patterns in relation to traits and phylogeny. Liira et al. community. Positive (+)andnegative(À) relationships are indicated, both (2012) examined absolute changes in species richness, for a decrease and increase of diversity. (a) Limiting similarity (LS). Negative relationship. Functionally similar species out-compete each and we examine the relationship between annual other, leading to trait divergence. (b) Weaker competitor exclusion or changes in species diversity (increases and decreases), habitat filtering (WCE/HF). Positive relationship. Species with similar trait functional diversity (hereafter FD) and phylogenetic values advantageous in competition will be favoured, leading to trait diversity (hereafter PD). The reduction found in species convergence. Alternatively, the habitat favours functionally similar species richness in fertilized plots in the previous paper was a with specific adaptations to the new conditions. (c) Limiting similarity (LS). mean trend, but all plots were characterized by inter- Positive relationship. Functionally dissimilar species are favoured due to annual species turnover. In stable communities, species limiting similarity, leading to trait divergence. (d) Weaker competitor exclusion or habitat filtering (WCE/HF). Negative relationship. Species with turnover might be due to random community processes similar trait values advantageous in competition will be favoured, leading or climatic variability across years, but in our case, turn- to trait convergence. Alternatively, habitat favours functionally similar over should be at least partly a deterministic response to species. the experimental treatments. We examined functional and phylogenetic diversity in that a decrease in species diversity is associated with a relation to changes in species diversity. We used multiple decrease in FD and PD. trait comparisons along an experimentally manipulated fertility gradient to distinguish between opposing forces Second, despite the fact that there was no mean increase governing community assembly. First, we hypothesize that in species richness in sucrose addition treatments com- decreased species diversity via increased productivity can pared to control plots (but higher richness and evenness result from different types of competition, which result in than in the fertilized treatment; Liira et al. 2012), there contrasting patterns of FD and PD. may be changes in FD and PD, prior to an increase in spe- cies diversity in the future. An increase in species diversity Limiting similarity (Fig. 1a). Competitive exclusion can be accompanied by an increase or decrease in FD and occurs between functionally similar and closely related PD, depending on the dominant assembly process. species, which leads to increased FD and PD (i.e. trait divergence and a phylogenetically over-dispersed com- Limiting similarity (Fig. 1c). Species increasing in abun- munity). In this case, we expect that a decrease in spe- dance are dissimilar and distantly related to resident spe- cies diversity is associated with an increase in FD and PD. cies due to increased competition between functionally Weaker competitor exclusion or habitat filtering similar and closely related species. Hence, we expect an (Fig. 1b). Closely related species with similar trait values increase in species diversity is associated with an advantageous in competition will be favoured, which increase in FD and PD. leads to decreased FD and PD (i.e. trait convergence and Weaker competitor exclusion or habitat filtering a phylogenetically clustered community). Alternatively, (Fig. 1d). Closely related species with similar trait values altered habitat conditions via changed productivity advantageous in competition will increase in abun- favour functionally similar species with specific adapta- dance, leading to trait convergence. Alternatively, tions to the habitat conditions. In this case, we expect altered environmental conditions favour functionally
Journal of Vegetation Science Doi: 10.1111/jvs.12052 © 2013 International Association for Vegetation Science 845 Functional diversity in species loss and gain P. Gerhold et al.
similar species with specific adaptations to the habitat mean FD and PD in the experimental plots in our data set conditions. This leads to decreased FD and PD (i.e. trait (r = 0.113, P = 0.06; Appendix S1) demonstrating that PD convergence and a phylogenetically clustered commu- and FD are not varying in parallel. nity). In this case, we expect an increase in species diver- The phylogeny of species was based on the phylogeny sity is associated with a decrease in FD and PD. for higher plants of Central Europe (Durka & Michalski 2012). We used the phylogenetic tree pruned down to the species list across the plots (n = 58) as the reference list to Methods measure PD in single plots. The study area is part of a long-term resource manipula- We calculated both FD and PD using the species′ cover tion experiment examining the effect of fertilizer and in the plots and the dissimilarity measure – mean pair- sucrose addition on a mesophytic grassland community in wise distance of all possible species pairs (MPD; Pavoine & southeast Estonia. The grassland has been experimentally Bonsall 2011). For FD, MPD was computed from the Gow- manipulated from 2002 to 2011 to increase (fertilizer er distance (Gower 1971; modified by Podani 1999) addition) and decrease (sucrose addition; see Liira et al. between each pair of taxa in a sample in the means of trait 2012 for full details) productivity. The experimental design states, and weighted by species relative abundances. For consisted of three treatments: fertilizer (24 replicates), PD, MPD was computed from the distance matrix of the sucrose (24 replicates) and control (seven replicates), pruned phylogenetic tree as the average branch length which were applied on randomly located permanent plots. between each pair of taxa in the means of phylogenetic The location of the plots was determined using the distance, and weighted by species relative abundances. homogeneity principle to ensure that plots were of similar We used the program R and a modified version of the relative elevation and hence situated in similar moisture function ′mpd′ in the package picante (Kembel et al. conditions. Treatments were fully randomized. In all plots 2010; R Foundation for Statistical Computing, Vienna, (50 cm 9 50 cm), the presence of all vascular species was AT). The modification was necessary because the original ′ recorded, and the percentage cover estimated visually in mpd′ function in the picante package (v. 1.5-2) with the July (mid-summer) from 2002 to 2011. option ′abundance.weighted = TRUE′ included in the We linked annual changes in species diversity to MPD calculation of the diagonal in the dissimilarity matrix changes in functional diversity (FD), phylogenetic diver- between species (which is always 0). This creates a spuri- sity (PD) and mean trait values. We calculated species ous positive correlation with number of species (with a diversity using Simpson’s reciprocal index, 1/D (D = Σ [ni/ larger dissimilarity table, the relative importance of the 2 N] ), where ni is the cover of each species and N is the total diagonal is smaller). Our modified function (Appendix S2) cover in a plot. The diversity values were log-transformed did not use the diagonal with the weighted MPD, and is to obtain a normal distribution. We selected two groups of not intrinsically related to the number of species in the functional traits: (1) vegetative traits related to local com- dissimilarity matrix. munity processes (D ıaz et al. 1998; plant height, leaf distri- We analysed the relationship between annual changes bution, leaf area, lateral spread); (2) regenerative traits in species diversity, and changes in mean FD, PD and FD of related to colonization ability (D ıaz et al. 1998; type of single traits using a linear mixed model (generalized linear reproduction, start of flowering). The vegetative and mixed models using Markov chain Monte Carlo tech- regenerative traits are also associated with competitive niques: MCMC), taking into account the effects of treat- response (i.e. the ability of a species to tolerate competi- ment, year and experimental plot as random factors. All tion; sensu Goldberg & Landa 1991). We obtained plant analyses were performed using R. Models were fitted using trait values from the databases CLO-PLA (Klime sova&de