Received: 20 September 2017 | Revised: 24 May 2018 | Accepted: 29 May 2018 DOI: 10.1111/geb.12783 RESE ARCH PAPER Traditional plant functional groups explain variation in economic but not size‐related traits across the tundra biome H. J. D. Thomas1 | I. H. Myers‐Smith1 | A. D. Bjorkman1,2,3 | S. C. Elmendorf4 | D. Blok5 | J. H. C. Cornelissen6 | B. C. Forbes7 | R. D. Hollister8 | S. Normand2 | J. S. Prevéy9 | C. Rixen9 | G. Schaepman‐Strub10 | M. Wilmking11 | S. Wipf9 | W. K. Cornwell12 | J. Kattge13,14 | S. J. Goetz15 | K. C. Guay16 | J. M. Alatalo17 | A. Anadon‐Rosell11,18,19 | S. Angers‐Blondin1 | L. T. Berner15 | R. G. Björk20,21 | A. Buchwal22,23 | A. Buras24 | M. Carbognani25 | K. Christie26 | L. Siegwart Collier27 | E. J. Cooper28 | A. Eskelinen14,29,30 | E. R. Frei31 | O. Grau32 | P. Grogan33 | M. Hallinger34 | M. M. P. D. Heijmans35 | L. Hermanutz27 | J. M. G. Hudson36 | K. Hülber37 | M. Iturrate‐Garcia10 | C. M. Iversen38 | F. Jaroszynska39 | J. F. Johnstone40 | E. Kaarlejärvi41,42,43 | A. Kulonen9,39 | L. J. Lamarque44 | E. Lévesque44 | C. J. Little10,45 | A. Michelsen46,47 | A. Milbau48 | J. Nabe‐Nielsen2 | S. S. Nielsen2 | J. M. Ninot18,19 | S. F. Oberbauer49 | J. Olofsson41 | V. G. Onipchenko50 | A. Petraglia25 | S. B. Rumpf37 | P. R. Semenchuk28,37 | N. A. Soudzilovskaia51 | M. J. Spasojevic52 | J. D. M. Speed53 | K. D. Tape54 | M. te Beest41,55 | M. Tomaselli25 | A. Trant27,56 | U. A. Treier2 | S. Venn57,58 | T. Vowles20 | S. Weijers59 | T. Zamin33 | O. K. Atkin57 | M. Bahn60 | B. Blonder61,62 | G. Campetella63 | B. E. L. Cerabolini64 | F. S. Chapin III65 | M. Dainese66 | F. T. de Vries67 | S. Díaz68 | W. Green69 | R. B. Jackson70 | P. Manning3 | Ü. Niinemets71 | W. A. Ozinga35 | J. Peñuelas32,72 | P. B. Reich73,74 | B. Schamp75 | S. Sheremetev76 | P. M. van Bodegom51 1 School of Geosciences, University of Edinburgh, Edinburgh, United Kingdom 2 Ecoinformatics and Biodiversity, Department of Bioscience, Aarhus University, Aarhus, Denmark 3 Senckenberg Gesellschaft für Naturforschung, Biodiversity and Climate Research Centre (SBiK‐F), Frankfurt, Germany 4 Institute of Arctic and Alpine Research, University of Colorado, Boulder, Colorado 5 Department of Physical Geography and Ecosystem Science, Lund University, Lund, Sweden 6 Department of Ecological Science, Vrije Universiteit, Amsterdam, The Netherlands 7 Arctic Centre, University of Lapland, Rovaniemi, Finland 8 Biology Department, Grand Valley State University, Allendale, Michigan 9 WSL Institute for Snow and Avalanche Research SLF, Davos, Switzerland [Correction added on 10 December 2018, after first online publication: There were errors throughout this article previously and this article has been corrected in this current version.] This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2018 The Authors Global Ecology and Biogeography Published by John Wiley & Sons Ltd Global Ecol Biogeogr. 2018;1–18.  wileyonlinelibrary.com/journal/geb | 1 2 | THOMAS et al. 10 Department of Evolutionary Biology and Environmental Studies, University of Zurich, Zurich, Switzerland 11 Institute for Botany and Landscape Ecology, Greifswald University, Greifswald, Germany 12 School of Biological Earth and Environmental Sciences, University of New South Wales, Sydney, New South Wales, Australia 13 Max Planck Institute for Biogeochemistry, Jena, Germany 14 German Centre for Integrative Biodiversity Research (iDiv), Halle‐Jena‐Leipzig, Germany 15 School of Informatics, Computing, and Cyber Systems, Northern Arizona University, Flagstaff, Arizona 16 Bigelow Laboratory for Ocean Sciences, Boothbay, Maine 17 Department of Biological and Environmental Sciences, Qatar University, Doha, Qatar 18 Department of Evolutionary Biology, Ecology and Environmental Sciences, University of Barcelona, Barcelona, Spain 19 Biodiversity Research Institute, University of Barcelona, Barcelona, Spain 20 Department of Earth Sciences, University of Gothenburg, Gothenburg, Sweden 21 Gothenburg Global Biodiversity Centre, Gothenburg, Sweden 22 Institute of Geoecology and Geoinformation, Adam Mickiewicz University, Poznan, Poland 23 Department of Biological Sciences, University of Alaska Anchorage, Anchorage, Alaska 24 Forest Ecology and Forest Management, Wageningen University and Research, Wageningen, Netherlands 25 Department of Chemistry, Life Sciences and Environmental Sustainability, University of Parma, Parma, Italy 26 The Alaska Department of Fish and Game, Juneau, Alaska 27 Department of Biology, Memorial University, St John’s, Newfoundland and Labrador, Canada 28 Department of Arctic and Marine Biology, UiT‐The Arctic University of Norway, Tromsø, Norway 29 Department of Physiological Diversity, Helmholtz Centre for Environmental Research – UFZ, Leipzig, Germany 30 Department of Ecology and Genetics, University of Oulu, Oulu, Finland 31 Department of Geography, University of British Columbia, Vancouver, British Columbia, Canada 32 Global Ecology Unit, CREAF‐CSIC‐UAB‐UB, Bellaterra, Spain 33 Department of Biology, Queen’s University, Kingston, Ontario, Canada 34 Biology Department, Swedish Agricultural University (SLU), Uppsala, Sweden 35 Plant Ecology and Nature Conservation Group, Wageningen University & Research, Wageningen, The Netherlands 36 British Columbia Public Service, British Columbia, Canada 37 Department of Botany and Biodiversity Research, University of Vienna, Vienna, Austria 38 Climate Change Science Institute and Environmental Sciences Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 39 Department of Biology, University of Bergen, Bergen, Norway 40 Department of Biology, University of Saskatchewan, Saskatoon, Canada 41 Department of Ecology and Environmental Sciences, Umeå University, Umeå, Sweden 42 Department of Biology, Vrije Universiteit Brussel (VUB), Brussels, Belgium 43 Faculty of Biological and Environmental Sciences, University of Helsinki, Helsinki, Finland 44 Département des Sciences de l’Environnement and Centres d’études nordiques, Université du Québec à Trois‐Rivières, Trois‐Rivières, Quebec, Canada 45 Eawag Swiss Federal Institute of Aquatic Science & Technology, Dubendorf, Switzerland 46 Department of Biology, University of Copenhagen, Copenhagen, Denmark 47 Center for Permafrost (CENPERM), University of Copenhagen, Copenhagen, Denmark 48 Research Institute for Nature and Forest (INBO), Brussels, Belgium 49 Department of Biological Sciences, Florida International University, Miami, Florida 50 Department of Geobotany, Lomonosov Moscow State University, Moscow, Russia 51 Environmental Biology, Department Institute of Environmental Sciences, CML, Leiden University, Leiden, The Netherlands 52 Department of Biology, University of California Riverside, Riverside, California 53 NTNU University Museum, Norwegian University of Science and Technology, Trondheim, Norway 54 Water and Environmental Research Center, University of Alaska, Fairbanks, Alaska 55 Environmental Sciences, Copernicus Institute of Sustainable Development, Utrecht University, Utrecht, The Netherlands 56 School of Environment, Resources and Sustainability, University of Waterloo, Waterloo, Ontario, Canada 57 Research School of Biology, Australian National University, Acton, ACT, Australia 58 Centre for Integrative Ecology, School of Life and Environmental Sciences, Deakin University, Burwood, Victoria, Australia 59 Department of Geography, University of Bonn, Bonn, Germany 60 Department of Ecology, University of Innsbruck, Innsbruck, Austria 61 Environmental Change Institute, School of Geography and the Environment, University of Oxford, Oxford, United Kingdom THOMAS et al. | 3 62 Rocky Mountain Biological Laboratory, Crested Butte, Colorado 63 School of Biosciences & Veterinary Medicine ‐ Plant Diversity and Ecosystems Management Unit, University of Camerino, Camerino, Italy 64 DiSTA, University of Insubria, Varese, Italy 65 Institute of Arctic Biology, University of Alaska, Fairbanks, Alaska 66 Department of Animal Ecology and Tropical Biology, University of Würzburg, Würzburg, Germany 67 School of Earth and Environmental Sciences, The University of Manchester, Manchester, United Kingdom 68 Instituto Multidisciplinario de Biología Vegetal (IMBIV), CONICET and FCEFyN, Universidad Nacional de Córdoba, Córdoba, Argentina 69 Department of Organismic and Evolutionary Biology, Harvard University, Cambridge, Massachusetts 70 Department of Earth System Science, Stanford University, Stanford, California 71 Institute of Agricultural and Environmental Sciences, Estonian University of Life Sciences, Tartu, Estonia 72 CREAF, Cerdanyola del Vallès, Spain 73 Department of Forest Resources, University of Minnesota, St. Paul, Minneapolis, Minnesota 74 Hawkesbury Institute for the Environment, Western Sydney University, Penrith, NSW, Australia 75 Department of Biology, Algoma University, Sault Ste. Marie, Ontario, Canada 76 Komarov Botanical Institute, St Petersburg, Russia Correspondence H. J. D. Thomas, School of Geosciences, Abstract University of Edinburgh, Crew Building, Aim: Plant functional groups are widely used in community ecology and earth system Edinburgh EH9 3FF, United Kingdom. Email: haydn.thomas@ed.ac.uk modelling to describe trait variation within and across plant communities. However, this approach rests on the assumption that functional groups explain a large propor‐ Funding information Natural Environment Research Council, tion of trait variation among species. We test whether four commonly used plant Grant/Award Number: NE/M016323/1 functional groups represent variation in six ecologically important plant traits. and NE/L002558/1; Academy of Finland, Grant/Award Number: 256991; ArcticNet; Location: Tundra biome. The Arctic Research Centre; Biotechnology Time period: Data collected between 1964 and 2016. and Biological Sciences Research Council; Carlsberg Foundation, Grant/Award Major taxa studied: 295 tundra vascular plant species. Number: 2013‐01‐0825; Danish Council Methods: We compiled a database of six plant traits (plant height, leaf area, specific leaf for Independent Research, Grant/Award Number: DFF 4181‐00565; European area, leaf dry matter content, leaf nitrogen, seed mass) for tundra species. We exam‐ Research Council, Grant/Award Number: ined the variation in species‐level trait expression explained by four traditional func‐ ERC‐SyG‐2013‐610028 IMBALANCE‐P; Synthesis Centre of the German Centre tional groups (evergreen shrubs, deciduous shrubs, graminoids, forbs), and whether for Integrative Biodiversity Research (iDiv) variation explained was dependent upon the traits included in analysis. We further Halle‐Jena‐Leipzig, Grant/Award Number: DFG FZT 118; JPI Climate, Grant/Award compared the explanatory power and species composition of functional groups to al‐ Number: 291581; Marie Skłodowska ternative classifications generated using post hoc clustering of species‐level traits. Curie Actions, Grant/Award Number: INCA 600398; Montagna di Torricchio Results: Traditional functional groups explained significant differences in trait expres‐ Nature Reserve; National Aeronautics sion, particularly amongst traits associated with resource economics, which were con‐ and Space Administration; US National Science Foundation, Grant/Award sistent across sites and at the biome scale. However, functional groups explained 19% of Number: DEB‐1637686, DEB‐1234162 overall trait variation and poorly represented differences in traits associated with plant and DEB‐1242531; Natural Sciences and Engineering Research Council of Canada; size. Post hoc classification of species did not correspond well with traditional functional Organismo Autónomo Parques Nacionales; groups, and explained twice as much variation in species‐level trait expression. Polar Continental Shelf Program; Royal Canadian Mounted Police; Russian Science Main conclusions: Traditional functional groups only coarsely represent variation in Foundation, Grant/Award Number: well‐measured traits within tundra plant communities, and better explain resource 14‐50‐000290; Swedish Research Council, Grant/Award Number: 2015‐00465 economic traits than size‐related traits. We recommend caution when using func‐ 015‐00498; Swiss National Science tional group approaches to predict tundra vegetation change, or ecosystem func‐ Foundation; University of Zurich; U.S. Department of Energy tions relating to plant size, such as albedo or carbon storage. We argue that alternative classifications or direct use of specific plant traits could provide new insights for ecological prediction and modelling. KEYWORDS cluster analysis, community composition, ecosystem function, plant functional groups, plant functional types, plant traits, tundra biome, vegetation change 4 | THOMAS et al. 1 | I NTRO D U C TI O N are most commonly categorized into four functional groups: ev‐ ergreen shrubs, deciduous shrubs, graminoids and forbs. This Many ecosystems around the world are responding rapidly to grouping structure is rooted in Chapin, Bret‐Harte, Hobbie, and global change drivers, including warming (IPCC, 2013), chang‐ Zhong’s (1996) demonstration that clustering of 37 species based ing precipitation patterns (Weltzin et al., 2003), increased nu‐ on 21 plant traits aligned with growth form‐based groupings. The trient availability (Galloway et al., 2008), elevated atmospheric use of functional groups is thus inherently a trait‐based approach, CO2 (Cramer et al., 2001) and altered herbivory regimes (Díaz et based on the hypothesis that plant species within functional al., 2007). Perhaps nowhere will ecosystem response to climate groups possess similar traits and act in ecologically similar ways change be greater than in the tundra, which is warming at twice (Lavorel & Garnier, 2002; McGill et al., 2006). This hypothesis the global average rate (IPCC, 2013; Serreze & Barry, 2011) and has so far only been tested at the site scale (Chapin et al., 1996) undergoing rapid vegetation change (Elmendorf, Henry, Hollister, or for individual traits (Dorrepaal, Cornelissen, Aerts, Wallén, & Björk, Boulanger‐Lapointe, et al., 2012; Myers‐Smith et al., 2011). Logtestijn, 2005; Körner, Leuzinger, Riedl, Siegwolf, & Streule, Predicting how plant communities will respond to environmen‐ 2016), yet continues to underpin a wide range of studies examin‐ tal change, and the resulting impact on ecosystem structure ing tundra plant community responses to environmental change and function, has been described as the “holy grail” of ecology (Figure 1). (Lavorel & Garnier, 2002). However, the responses of different There is evidence that functional groups display distinct dif‐ species and environments are often highly complex, representing ferences in their response to environmental change in the tundra. a major challenge for the prediction of community response to Experimental warming and fertilization are associated with increases environment change (Díaz et al., 2016; McGill, Enquist, Weiher, & in cover and biomass of deciduous shrubs and graminoids, often at Westoby, 2006). the expense of other functional groups (Dormann & Woodin, 2002; One approach to reducing complexity in ecological commu‐ Elmendorf, Henry, Hollister, Björk, Bjorkman, et al., 2012). In turn, nities is to classify species with similar characteristics into plant the relative abundance of different functional groups influences functional groups or plant functional types (Harrison et al., 2010). multiple ecosystem properties, including biomass accumulation, light Species are commonly grouped based on a priori classification by interception, soil moisture and soil nutrients (McLaren & Turkington, growth form (e.g., forb, shrub), life history (e.g., evergreen, decid‐ 2010, 2011). Functional groups also integrate multiple plant traits uous) or other morphological characteristics (Wright et al., 2006; and may therefore better explain ecosystem function and commu‐ Wullschleger et al., 2014). In the tundra, vascular plant species nity change compared to single trait‐based approaches (Laughlin & F I G U R E 1 Studies employing an “evergreen shrub ‐ deciduous shrub ‐ graminoid – forb” functional group classification (or close variant) to examine the response of tundra communities to environmental change over the past two decades. Studies were identified based on a literature search on Web of Science using the search terms “tundra" and “plant functional group” or “plant functional type”. For a list of studies see Appendix A. Studies are grouped by whether they found clear differences in functional group response (Yes: clear differences were found between some (but not necessarily all) functional groups; Not clear: differences between groups were inconsistent amongst sites or over time; No: No significant differences in functional group response). Studies vary in duration from 2–30 years and incorporate a range of bioclimatic contexts and experimental types. For full meta‐analyses of functional group response see Dormann and Woodin (2002) and Dorrepaal (2007) THOMAS et al. | 5 Messier, 2015; Soudzilovskaia et al., 2013). By extension, plant func‐ alpine tundra species, and whether explanatory power is sensi‐ tional groups may integrate information from traits that are difficult tive to: (a) differences in species composition among sites or (b) to collect, including root structure or mycorrhizal association, that the use of different plant traits in analyses. We examine six traits, may be critical to explaining vegetation change (Cornelissen, Aerts, plant height (PH), seed mass (SM), leaf area (LA), specific leaf area Cerabolini, Werger, & Heijden, 2001; Soudzilovskaia et al., 2015). (SLA), leaf dry matter content (LDMC) and leaf nitrogen (LN), that Despite their prevalence in ecological analysis, functional groups are the most commonly collected plant traits in the tundra biome have often displayed low explanatory power and inconsistent re‐ (Bjorkman et al., 2018a) and considered to be cornerstones of sponses across experiments (Bret‐Harte et al., 2008; Dorrepaal, plant ecological strategy (Díaz et al., 2016). We hypothesize that 2007). In a meta‐analysis of 36 environmental manipulation experi‐ plant functional groups will exhibit distinct trait distributions, and ments in the tundra, Dormann and Woodin (2002) found that plant that traits associated with plant economics (SLA, LDMC, LN) will functional groups did not predict community response, except in the be better explained by traditional functional groups than traits case of fertilization and warming treatments. Even amongst these associated with plant size (PH, SM, LA), reflecting consistent func‐ treatment types, differences in functional group response have not tional group responses in resource addition experiments (fertiliza‐ always been clear in the literature (Figure 1). Functional groups have tion and warming), but not in other experimental types (Dormann also shown highly conflicting responses across studies; for example, & Woodin, 2002). evergreen shrubs have shown positive, neutral and negative re‐ sponses to warming (Elmendorf, Henry, Hollister, Björk, Boulanger‐ 1.1.2 | Does functional group composition align Lapointe, et al., 2012; Hollister, Webber, & Tweedie, 2005; Zamin, with post hoc trait‐based clustering of species? Bret‐Harte, & Grogan, 2014). Finally, functional groups have shown inconsistent responses among and within experiments, in differ‐ We compare the species composition and explanatory power of tra‐ ent years (Cornelissen & Makoto, 2014), time‐scales (Saccone & ditional functional groups with two statistically derived, trait‐based Virtanen, 2016), environmental conditions (Dorrepaal, 2007) and clustering approaches, which represent optimal grouping of spe‐ spatial scales (Mörsdorf et al., 2015). cies within multivariate trait‐space. Given that traditional functional Low explanatory power may arise from high trait variation within groups were formulated using trait‐based clustering, albeit with a functional groups, such that group differences are not significant, smaller species pool, we hypothesize that post hoc classification will particularly among small species pools (Cornelissen et al., 2004). produce similar species groupings to traditional functional groups. For example, Körner et al. (2016) found that tissue carbon and ni‐ This approach directly addresses calls to compare traditional func‐ trogen did not vary by functional group in European alpine plants, tional groups with other trait‐based classifications (Boulangeat et al., whilst Iversen et al. (2017) reported greater variation in fine‐root 2012; Dorrepaal, 2007; Hudson et al., 2011), and provides the first carbon‐to‐nitrogen ratios within groups than among groups in bi‐ trait‐based assessment of traditional functional groups at the tundra omes spanning the globe. Many studies have instead found that biome scale. tundra species respond highly individualistically to change (Hollister et al., 2005; Hudson, Henry, & Cornwell, 2011; Lavorel & Garnier, 2 | M ATE R I A L S A N D M E TH O DS 2002), and that functional group responses instead reflect strong species‐specific responses, often of dominant species (Bret‐Harte 2.1 | Tundra biome definition et al., 2008; Little, Jagerbrand, Molau, & Alatalo, 2015; Shaver et al., 2001). An alternative hypothesis is, therefore, that traditional In line with previous biome‐scale assessments of tundra vegetation functional groups do not represent key dimensions of trait variation community change, we considered the tundra biome as the vegetated among species, and thus may obscure certain aspects of ecosystem regions above tree line, both at high latitude and high altitude (Bliss, function and change. Given that much of our current understanding Heal, & Moore, 1981; Elmendorf, Henry, Hollister, Björk, Boulanger‐ of tundra vegetation change is based on functional group responses Lapointe, et al., 2012). Tundra plant communities include many widely (Elmendorf, Henry, Hollister, Björk, Boulanger‐Lapointe, et al., 2012; distributed common species, and functional groups are considered to McLaren & Turkington, 2010; Myers‐Smith et al., 2011), testing this be consistent across the large geographical gradients and variety of hypothesis is critical to understanding the mechanisms and future environments within the tundra (Henry & Molau, 1997). patterns of tundra vegetation change. 2.2 | Dataset 1.1 | Research questions We established a database of tundra plant traits by combining 18,613 plant trait records from the TRY database (Kattge et al., 2011; 1.1.1 | How well do functional groups represent Appendix B) with 37,435 records from Tundra Trait Team (TTT) con‐ species trait variation? tributors (Bjorkman et al., 2018a), forming the largest database of tun‐ In this study, we test whether traditional functional groups ex‐ dra plant traits compiled to date. We considered all species present at plain differences in six plant functional traits among Arctic and International Tundra Experiment (ITEX) and associated plots as tundra 6 | THOMAS et al. species (Bjorkman et al., 2018b; Henry & Molau, 1997; Elmendorf, separating graminoids into grasses, sedges and rushes and a (b) Henry, Hollister, Björk, Boulanger‐Lapointe, et al., 2012). We included seven‐group classification further separating evergreen and de‐ all available trait records for tundra species, but excluded records from ciduous shrubs into dwarf and tall shrubs. To examine the distri‐ manipulated locations such as experiments or botanical gardens. Of bution of individual traits within and among functional groups, we the 449 species in the ITEX dataset, 386 (86%) had trait data available. plotted the distribution of species‐level mean traits for each of Species lacking trait data were generally rare or uncommon species the six plant traits studied and tested the significance of distri‐ unique to single sites, and on average represented <3% of total plant butions using pairwise Wilcoxon signed‐rank tests. To visualize cover across all sites. multivariate trait distributions and examine the weighting of dif‐ We combined taxonomic synonyms following The Plant List ferent traits, we performed principal components analysis (PCA) (www.theplantlist.org) to ensure consistent taxonomy across all on multivariate trait distributions using the “prcomp” function in studies. As sampling problems inevitably arise from compiling trait the R “stats” package, and plotted the first two component axes. data from a large number of disparate studies (Jetz et al., 2016), we We conducted PERMANOVA analysis to test the significance of removed duplicate entries, obviously erroneous values (e.g., values and variance explained by functional groups to estimate how well <0), and observations more than four standard deviations from each traditional functional groups represent trait characteristics. We species mean (see Bjorkman et al., 2018a for more information). used Euclidian distance with 999 permutations for the combina‐ For seed mass, which is prone to measurement error due to the tion of all six traits using the “adonis” function in the R package small masses involved and large variation within individuals (Pérez‐ “vegan” (Oksanen et al., 2013). Harguindeguy et al., 2013), we manually checked values more than We performed all analyses at the biome scale using all trait data, three standard deviations from each species’ mean and removed val‐ encompassing 1,333 unique georeferenced locations and non‐georef‐ ues that had clear measurement or transcription error. erenced trait data for tundra species. To examine if functional group significance was affected by species composition, we also conducted analyses at three unique geographical locations: Abisko (northern 2.3 | Trait selection Sweden, 68°N, 18°E, 98 species available) representing European sub‐ We selected six plant traits for analyses: plant height (maximum arctic tundra, Davos (the Swiss Alps, 47°N, 10°E, 67 species available) measured height), seed mass (dry mass), leaf area per leaf (fresh leaf representing European alpine tundra, and Qikiqtaruk‐Herschel Island area), specific leaf area (ratio of fresh leaf area to dry leaf mass), leaf (northern Canada, 69°N, −139°E, 16 species available) representing dry matter content (ratio of leaf dry mass to fresh leaf mass) and leaf North American arctic tundra. We chose these sites to represent vari‐ nitrogen (nitrogen per unit leaf dry mass). A total of 295 species had ation in geography and species richness across the tundra. We also data available for all six traits. A review of the ecological associations repeated all analyses using a subset of only georeferenced trait data of each trait can be found in Díaz et al. (2016). We additionally tested collected north of 60°N to examine if findings were influenced by en‐ two traits with low data availability, stem density (ratio of stem dry vironmental variation across collection locations. mass to fresh stem volume) and leaf life span. These traits align with To examine if the variation explained by functional groups key characteristics of functional groups, but are rarely measured was dependent on the traits included in analysis, we repeated for tundra species (Supporting Information Table S1). We log‐trans‐ PERMANOVA analysis for every possible multivariate combination formed trait values to account for log‐normal distributions, stand‐ of traits. This enabled us to test whether particular trait combina‐ ardized between 0 and 1 using variance scaling, and aggregated tions were well differentiated by functional groups. We also differ‐ traits at the species level to allow multivariate comparison among entiated between size‐related and economic traits, reflecting the species and different units of measurement. Within‐species varia‐ two major dimensions of trait variation amongst global plant species tion cannot be captured using this approach, but is assumed not to (Díaz et al., 2016). As some traits were available for more species contribute to a large proportion of trait variation at the biome scale than others, resulting in unequal sample sizes among different trait (Siefert et al., 2015). However, we also re‐ran analysis using the 25th combinations, we randomly selected 295 species (the minimum and 75th percentile of species‐level trait data, representing the low‐ number of species for which all six traits were available) for each est and highest quarter of trait values for each species, respectively, trait combination and calculated the mean variance explained over to test whether results were altered by within‐species variation in 999 replications for each combination. the dataset as a whole. 2.5 | Comparison with post hoc classifications 2.4 | Trait variation explained by functional group We compared the species composition and explanatory power of We assigned species to four functional groups—evergreen shrubs, functional groups to post hoc species classifications created using deciduous shrubs, graminoids and forbs—based on previous clas‐ statistical clustering of species‐level plant traits. We grouped spe‐ sification of ITEX species (Elmendorf, Henry, Hollister, Björk, cies using two contrasting clustering approaches, k‐means clus‐ Boulanger‐Lapointe, et al., 2012). We also examined two more de‐ tering (k‐means) and hierarchical agglomerative clustering (HCA). tailed functional group classifications: (a) a six‐group classification K‐means clustering employs a top‐down approach, assigning THOMAS et al. | 7 species to groups based on multivariate distance from group means Functional groups explained 18.5% of multivariate trait expres‐ (Ding & He, 2004). Hierarchical agglomerative clustering employs sion among species across all six traits (four‐cluster PERMANOVA, a bottom‐up approach, iteratively combining groups with similar R 2 = 0.185, p < 0.001), and were significant both for the tundra traits (Lukasová, 1979). We performed clustering using the R pack‐ biome and at the site level. The direction of trait weightings indi‐ age “vegan” and selected a four‐cluster solution for both methods cated that economic traits (SLA, LDMC, LN; greater association to correspond with the number of functional groups. When test‐ with PCA axis 1) and size‐related traits (PH, SM, LA; greater as‐ ing alternative six‐ and seven‐functional group classifications we sociation with PCA axis 2) comprised distinct axes of trait varia‐ selected six‐cluster and seven‐cluster solutions, respectively. For tion, with functional groups primarily differentiated along the first HCA clustering, we used Euclidian distance and Ward’s criterion PCA axis. The relative position of functional groups was consistent to measure linkage. We compared differences in species compo‐ among sites, regardless of species composition or geographical lo‐ sition between post hoc trait‐based classifications and traditional cation (Figure 3). functional groups by calculating the maximum possible number of The explanatory power of functional groups was strongly de‐ consistently categorized species amongst grouping methods. We pendent on the traits included in the analysis. Trait combinations also estimated the relative abundance of consistently grouped spe‐ including only economic traits (SLA, LN, LDMC) were better ex‐ cies within the ITEX database (Elmendorf, Henry, Hollister, Björk, plained by functional groups than size‐related traits (PH, SM, LA), Boulanger‐Lapointe, et al., 2012, (Polar Data Catalogue; CCIN regardless of the number of traits included in analysis (Figure 4a). 10786)) using the most recent year for all plots and aggregating at This was largely driven by LDMC, as combinations containing this the site level. trait were best explained by functional groups (Figure 4b). In con‐ Finally, we repeated PERMANOVA analysis for post hoc trait‐ trast, trait combinations containing PH or SM were comparatively based classifications and examined the variance explained by groups poorly explained by functional groups (Figure 4c). Inclusion of leaf for all traits, for only size‐related and for only economic traits. This life span and stem density traits reduced data availability by over enabled us to: (a) test the variation remaining unexplained when 80% (Supporting Information Table S1) but improved the explana‐ using post hoc classification of species, and thus (b) test the explan‐ tory power of groups from 19% to 55% and 41%, respectively. This atory power of traditional functional groups compared to optimal improvement was driven by economic differences, and primarily dif‐ four‐group clustering of species, acknowledging that it is unlikely ferentiated shrubs (wood density) or evergreen shrubs (leaf life span) that all trait variation will be explained, and (c) examine whether post from other groups (Supporting Information Figure S4). hoc trait‐based classifications could differentiate between axes of trait variation. 3.2 | Comparison of post hoc trait‐based All analyses were conducted in R version 3.3.2 (R Core Team, classifications with functional groups 2017). Trait data have been submitted to the TRY database (https:// www.try-db.org) and are publicly available at https://github.com/ Post hoc trait‐based classification of species did not correspond TundraTraitTeam/TraitHub. Code is available at https://github.com/ well with traditional functional group composition. The four groups hjdthomas/Tundra_functional_groups identified by post hoc classification were consistently located within trait‐space across clustering methods, and were differentiated by the two axes of trait variation, although more strongly by size‐re‐ lated traits (Figure 5). Post hoc classifications thus represented: 3 | R E S U LT S (a) tall species with large leaves and seeds (high PH, SM and LA), (b) mid‐sized species with economically acquisitive strategies (low 3.1 | Trait variation explained by traditional LDMC, high SLA and LN), (c) small species with economically acquisi‐ functional groups tive strategies, and (d) small species with economically conservative We found large overlap between the trait distributions of functional strategies. groups for the majority of traits examined, such that trait distribu‐ Forty‐two per cent of species were consistently classified be‐ tions were often not significantly different among functional groups tween traditional functional groups and k‐mean clustering, and (Figure 2, Supporting Information Figure S1). The significance of 43% between traditional functional groups and HCA clustering functional group distributions was strongly trait dependent, for (Figure 5f, Table 1). In contrast, 74% of species were consistently example with significant differences among all groups for specific classified between post hoc clustering methods. Evergreen shrubs, leaf area, but no significant differences between any groups for approximately half of graminoids and one third of forbs were largely seed mass. Among functional groups, evergreen shrubs exhibited assigned to consistent groups across the three clustering methods the most distinct differences in trait expression compared to other (Figure 5f). Deciduous shrubs showed very low correspondence be‐ tundra plants, primarily driven by economic traits (Figures 2 and 3). tween functional groups and post hoc classifications due to large In contrast, deciduous shrubs and graminoids exhibited largely over‐ trait overlap with both graminoids and forbs, but showed high cor‐ lapping trait distributions for many individual traits and in multivari‐ respondence between clustering methods (Table 1, Supporting ate trait‐space. Information Table S2). 8 | THOMAS et al. 1.25 0.8 Functional Group 1.00 0.6 Deciduous Shrub Density 0.75 Density 0.4 Evergreen Shrub 0.50 0.2 Graminoid 0.25 0.0 Forb 0.00 1 100 10000 0.01 0.10 1.00 Plant Height (m) Leaf Area (mm2) 0.8 4 0.6 3 Density Density 0.4 2 0.2 1 0 0.0 1 10 100 1e 03 1e 01 1e+01 1e+03 Seed Mass (mg) Specific Leaf Area (mm2/mg) 3 4 3 Density Density 2 2 1 1 0 0 0.1 1.0 10 100 Leaf Dry Matter Content (g/g) Leaf N (mg/g) F I G U R E 2 Smoothed distribution of species‐level traits represented by the four traditional tundra plant functional groups. Distributions are based on species‐level mean traits for the 295 tundra species for which data are available for all six plant traits of interest. Trait values are presented on the x axis in untransformed units on a log scale. Significance of distributions is indicated by symbols (pairwise Wilcoxon rank sum test; * = p < 0.05; ** = p < 0.01, *** = p < 0.001). Pairs of traits that are significantly different from each other, but not different from other functional groups, are indicated by black bars connecting the centre of those two distributions. Abundant species were more likely to be consistently classified Post hoc classifications explained 45% (k‐means, R 2 = 0.448, across grouping methods (Supporting Information Figure S2a), and the p < 0.001) and 37% (HCA, R 2 = 0.366, p < 0.001) of trait variation relative abundance of consistently classified species within tundra plant amongst tundra species, compared to 19% for traditional func‐ communities (51%) was greater than would be expected if all species tional groups (Figure 5d–f). Despite derivation using all six plant had equal abundance (35%). Although abundant species had more avail‐ traits, post hoc classifications explained greater variation in size‐ able trait observations, and thus may have more representative species‐ related traits than traditional functional groups for both cluster‐ mean traits, the number of trait observations did not significantly affect ing methods (functional groups: R 2 = 0.080, p < 0.001; k‐means: whether a species was consistently classified (Supporting Information R 2 = 0.474, p < 0.001; HCA: R 2 = 0.406, p < 0.001), whilst k‐means Figure S2b). Species that were consistently categorized across grouping sampling also slightly better explained variation in economic traits methods occupied a distinct region of trait‐space (p < 0.001) and were (functional groups: R 2 = 0.339, p < 0.001, k‐means: R 2 = 0.343, mostly large (taller, larger leaves or larger seeds) with extreme economic p < 0.001; HCA: R 2 = 0.266, p < 0.001, Figure 5d–f). Our results traits (i.e., highly conservative or highly acquisitive species, Supporting demonstrate that unexplained trait variation does not solely arise Information Figure S2d). Inconsistently classified species had traits due to aggregation of species into a small number of groups, and closer to the centre of the overall distribution of tundra species within that functional groups have less than half the explanatory power functional trait space, suggesting that the traits of these species may be of optimal species classification for the six most commonly col‐ poorly represented by traditional functional groups. lected tundra plant traits. THOMAS et al. | 9 F I G U R E 3 Distribution of tundra species in trait space. Inset plots indicate principal components analysis (PCA) multivariate distribution of six plant traits for three tundra sites, (a) Qikiqtaruk, (b) Abisko (c) Davos, and for (d) the whole tundra biome. Trait space was defined based on plant height (PH), seed mass (SM), leaf area (LA), specific leaf area (SLA), leaf dry matter content (LDMC) and leaf nitrogen content (LN). Individual species are represented by points and functional groups by point colour (blue = evergreen shrub, green = deciduous shrub, yellow = graminoid, purple = forb). Ellipses represent 95% confidence interval of functional group distributions. Arrows indicate direction and weighting of each trait. Georeferenced trait collection locations are indicated on the map by grey circles and modelled site locations by red circles 4 | D I S CU S S I O N plant functional groups represent 19% of variation in the six most com‐ monly measured plant traits amongst tundra species. Furthermore, 4.1 | Trait variation is poorly explained by traditional the species composition of functional groups did not align well with functional groups post hoc trait‐based classification of species. Together, our findings To be meaningful for ecological analyses, plant functional groups indicate that traditional functional groups poorly represent species‐ should accurately and consistently represent differences in species level variation in the six plant traits considered by this study, and characteristics that underpin their environmental preferences and highlight potential limitations of functional group approaches to pre‐ responses (Chapin et al., 1996). In this study, we find that traditional dicting community responses to environmental change in the tundra. 10 | THOMAS et al. (a) Our findings for the six most commonly measured traits in part 30 contradict Chapin et al.’s (1996) finding that growth‐form based Variance Explained (%) functional groups can be reproduced from trait information. This 20 Trait Combination discrepancy could arise from the greater number of species and in‐ Only Economic Traits dividual trait records represented in our study, which may increase Only Morphological Traits Economic & Morphological Traits variability within functional groups and species, or the greater num‐ 10 ber of traits included in Chapin et al. (1996). Trait variation may also be better represented by alternative classifications such as those 2 3 4 5 6 distinguishing between tall and dwarf shrubs, or between grasses and sedges. Although alternative six‐group and seven‐group clas‐ (b) sification schemes did slightly increase the explanatory power of 30 functional groups (from 18.5% to 21.4% and 24.9%, respectively, Variance Explained (%) Supporting Information Figure S3), the overall variance explained 20 remained low and substantially less than post hoc classifications 20 Trait Combination Includes LDMC (53.6% and 56.8%, respectively). Excludes LDMC Low explanatory power of functional groups could also arise 10 10 from the choice of traits included in analysis. The traits investi‐ gated in this study are considered critical determinants of ecolog‐ 2 3 4 5 6 ical processes (Díaz et al., 2016; Pérez‐Harguindeguy et al., 2013), and represent both available tundra trait data and the focus of (c) trait‐based research in tundra ecosystems (Bjorkman et al., 2018a). 30 Nevertheless, we found that the explanatory power of functional Variance Explained (%) groups was highly trait‐specific (Figure 4), and thus functional groups may represent differences amongst plant traits not inves‐ 20 20 Trait Combination tigated here that are nonetheless critical to ecosystem function in Includes Height / Seed Mass Excludes Height / Seed Mass the tundra (Figure 6). For example, inclusion of stem density in‐ 10 10 creased the explanatory power of traditional functional groups to over 50% (Supporting Information Figure S4), but reduced species representation by 80% (n = 53) and did not improve representation 2 3 4 5 6 Number of traits of size‐related traits. F I G U R E 4 Trait variation explained by functional groups for all possible trait combinations. Functional groups best explained 4.2 | Functional groups align with economic traits combinations of (a) only economic traits, or (b) those containing leaf dry matter content (LDMC), and worst explained combinations Among tundra species, traditional functional groups better repre‐ of only morphological traits or (c) those containing plant height sented variation in economic traits (SLA, LDMC, LN) than size‐re‐ or seed mass. Points indicate the mean variance explained lated traits (PH, SM, LA). Indeed, functional groups explained (PERMANOVA R2) by functional groups and coloured to visualize roughly equal variation in economic traits to post hoc clustering the importance of different trait combinations (33.5% compared to 34.3% for k‐means clustering). As such, ecosys‐ tem functions related to resource economics such as photosynthetic rate or nutrient cycling may be well represented using functional Our findings support a previous trait‐based criticism of tradi‐ group approaches (Lavorel & Garnier, 2002). This difference may tional functional groups in European alpine species (Körner et al., also explain why studies focusing on community responses to re‐ 2016), and may explain low explanatory power and contradictory source addition (Dormann & Woodin, 2002; Elmendorf, Henry, responses of functional groups in previous tundra studies (Dormann Hollister, Björk, Bjorkman, et al., 2012; Zamin et al., 2014) or litter & Woodin, 2002; Dorrepaal, 2007; Figure 1). Although it is possible quality (Carbognani, Petraglia, & Tomaselli, 2014; Cornelissen et al., that the tundra is unusual in the global context due to small plant 2007; Dorrepaal et al., 2005) find the clearest differences between growth‐forms and harsh environmental conditions, our study is in functional groups. line with findings that functional groups poorly describe trait vari‐ Low representation of size‐related traits may arise due to con‐ ation in tropical forests (Wright et al., 2013), temperate grasslands vergence of growth forms in the tundra; all functional groups con‐ (Forrestel et al., 2017; Fry, Power, & Manning, 2014; Wright et al., tain both comparatively large (e.g., the tall deciduous shrub Salix 2006), and among certain traits at the global scale (Iversen et al., glauca or forb Chamaenerion angustifolium) and comparatively small 2017; Kattge et al., 2011; Reichstein, Bahn, Mahecha, Kattge, & (eg, the dwarf deciduous shrub Salix polaris or forb Saxifraga bryoi‐ Baldocchi, 2014; Wright et al., 2005). des) species. As a result, functional groups may poorly represent THOMAS et al. | 11 (a) (d) (g) (b) (e) (c) (f) F I G U R E 5 Comparison of group structure, trait variation explained, and group composition between traditional functional groups and post hoc classifications. (a–c) principal components analysis (PCA) visualization of species clusters as defined by (a) traditional functional groups, (b) k‐means clustering, and (c) hierarchical‐agglomerative clustering (HCA). Species are indicated by points and group distribution by ellipses. Colours indicate groups (dark blue = evergreen shrub, green = deciduous shrub, yellow = graminoid, purple = forb). Post hoc classifications are matched with functional groups based on maximum species correspondence between grouping methods, such that each post hoc classification corresponds with a traditional functional group. Post hoc groups approximately represent (i) tall species with large leaves and seeds (purple), (ii) mid‐sized species with economically acquisitive strategies (yellow), (iii) small species with economically acquisitive strategies (green) and (iv) small species with economically conservative strategies (blue). (d–f) Trait variation explained by (d) traditional functional groups, (e) k‐means, and (f) hierarchical agglomerative clustering (HCA) for multivariate combinations of all six plant traits (white), size‐related traits only (red) and economic traits only (light blue). (g) Comparison of group composition across clustering methods. The stacked bars represent individual species and are ordered by traditional functional group (species order remains consistent across columns). The colour of each stacked bar represents the group to which species were assigned by each classification method (classification can change across columns). For example, a species categorized as a graminoid by traditional functional groups can be categorized in the group most corresponding to forbs by post hoc classifications ecosystem functions or properties relating to size‐related traits, morphological characteristics, such as distinguishing between tall such as albedo, carbon storage, seed dispersal or competitive ability and dwarf shrubs (Elmendorf, Henry, Hollister, Björk, Boulanger‐ (Lavorel & Garnier, 2002; Loranty, Goetz, & Beck, 2011; Westoby, Lapointe, et al., 2012; Vowles et al., 2017), may better charac‐ Falster, Moles, Vesk, & Wright, 2002). Such properties are impli‐ terize differences in trait expression, although we found limited cated as key drivers of community‐level vegetation change in the evidence for this (Supporting Information Figure S3). As such, post tundra (Kaarlejärvi, Eskelinen, & Olofsson, 2017; Mekonnen et al., hoc classification of species or direct use of trait data may identify 2018). Functional group classifications that explicitly recognize differences amongst size‐related traits, and associated drivers of 12 | THOMAS et al. TA B L E 1 Top: Similarity in species composition between traditional functional groups and post hoc trait‐based classifications (k‐means = k‐means clustering; HCA = hierarchical agglomerative clustering), calculated as the proportion of consistently classified species out of all species. Bottom: Relative abundance of consistently classified species within tundra (International Tundra Experiment, ITEX) vegetation communities, calculated as the proportion of the summed abundance of consistently classified species out of the summed abundance of all species for which trait data are available across all ITEX plots Functional groups versus Functional groups versus k‐means versus Functional group k‐means (%) HCA (%) HCA (%) All methods (%) Similarity between group species composition All groups 42 43 74 35 Evergreen shrubs 89 94 94 89 Deciduous shrubs 0 13 87 0 Graminoids 52 51 78 42 Forbs 37 37 69 30 Relative abundance of consistent species All groups 56 59 87 51 Evergreen shrubs 99 100 99 99 Deciduous shrubs 0 21 79 0 Graminoids 74 65 84 62 Forbs 24 32 82 22 community change and ecosystem function, that are obscured by differences amongst individual species is low. Species that were variation within traditional functional groups (Matesanz, Escudero, consistently categorized (Supporting Information Table S3) pos‐ & Fernando, 2009). sessed similar traits including a larger structure (tall with large leaves and seeds) and either highly conservative or acquisitive resource economic traits. However, some species that were inconsistently 4.3 | Trait‐based approaches as an alternative to classified, notably deciduous shrubs such as Betula nana and gram‐ functional groups inoids such as Agrostis spp., have demonstrated the greatest vegeta‐ Our findings contribute to growing support for the use of trait‐ tion responses at many tundra sites (Bret‐Harte et al., 2001; Venn, based approaches as an alternative to functional groups within Pickering, & Green, 2014), suggesting that traditional functional ecological research and earth system modelling. Trait‐based ap‐ groups may obscure some important trait characteristics associated proaches include post hoc grouping of species according to with vegetation change (Saccone et al., 2017). common traits (Suding et al., 2008), common responses to environ‐ mental conditions (Cornwell & Ackerly, 2010) or common effects 4.4 | Underpinning assumptions on ecosystem processes (Cornwell et al., 2008; Laughlin, 2011), as well as direct use of trait data in analysis (McGill et al., 2006). In The findings of this study are based on several key assumptions. this study, post hoc classifications explained more than twice as First, we assume that the species for which trait data are available much trait variation as functional groups, and were distinguished are representative of all tundra species. Species lacking trait data along two global axes of trait variation (Díaz et al., 2016), repre‐ are often rare (low abundance) or endemic (occur at few sites). senting large versus small species, and economically “fast” versus The data gap for these missing species could represent unusual “slow” species (Díaz et al., 2016; Reich, 2014). Post hoc classifica‐ trait combinations not easily captured by trait‐based classification tions thus better captured the multidimensionality of trait varia‐ (Sandel et al., 2015). We also do not examine mosses and lichens, tion compared to traditional groupings (Maire, Grenouillet, Brosse, which play an important role in ecosystem function in the tundra & Villéger, 2015), and produced relatively robust species groupings (Turetsky, Mack, Hollingsworth, & Harden, 2010). Nevertheless, across the two clustering methods. the species included in this study reflect the majority of tundra Post hoc approaches have nevertheless been criticized on the plant biomass and include the species known to be most rapidly basis of inconsistencies across methodologies and ecological com‐ responding to climate change (Elmendorf, Henry, Hollister, Björk, munities (Dyer, Goldberg, Turkington, & Sayre, 2001; Fry et al., 2014), Boulanger‐Lapointe, et al., 2012). and could be biased towards representing rarer species with more Second, we assume that plant traits are meaningful predictors of extreme traits. In this study, functional groups better represented species’ responses to environmental dynamics or effects on ecosystem differences amongst more abundant species (Table 1), and thus may function. In this study, we do not examine whether traits or alternative capture community‐level characteristics even if representation of trait‐based classifications better predict community dynamics than THOMAS et al. | 13 Traditional Both Alternative trait-based Functional Groups approaches classifications (this study) Woodiness Economic traits (shrub versus non-shrub) Leaf dry matter content Size-related traits Specific leaf area Plant height Leaf habit Leaf nitrogen Leaf area (shrub species) (although see Körner et al., Seed mass 2016) Phylogeny Multiple trait axes (with exceptions e.g. Size-economic trait Data gaps Vaccinium sp.) combinations Wood traits Stem density (limited data) Chemical traits Micronutrients, leaf lignin Physiological traits Leaf photosynthetic rates Leaf life span Below ground traits Root traits Mycorrhizal association Phenological traits F I G U R E 6 Functional groups and post hoc trait‐based classifications capture different characteristics of tundra plant communities. Solid circles enclose characteristics represented by functional groups, post hoc classifications, and by both approaches, according to the findings of this study. The dotted circle encloses the data gaps for traits that are not well represented in tundra trait databases or trait‐based analysis yet are suggested to be important in the literature (Bardgett, Mommer, & Vries, 2014; Chave et al., 2009; Cleland et al., 2012; Eckstein et al., 1999) functional groups. Traditional functional groups may better predict invalidate species‐level clustering (Shipley et al., 2016; Violle et al., certain ecological dynamics than trait‐based approaches as they inte‐ 2012). The species considered in this study have large geographical grate multiple measured and unmeasured traits across plant organs, ranges, encompassing both Arctic and alpine tundra, and nontundra ecological strategy, and life cycle (Grime et al., 1997). Nevertheless, locations. However, our findings are robust when using individual there is widespread evidence to support trait‐based approaches to trait‐data (Supporting Information Figure S1), across site‐specific modelling ecosystem dynamics (Suding et al., 2008; Violle & Jiang, species assemblages (Figure 3), for the 25th and 75th percentile of 2009; Cornwell & Ackerly, 2010; Soudzilovskaia et al., 2013, but see species‐level trait data (Supporting Information Figure S5), and for Clark, 2016). Single traits, such as plant height, have also predicted veg‐ only trait collection locations north of 60°N (Supporting Information etation responses to change that are obscured within traditional func‐ Figures S6–S9). Furthermore, most studies have found within‐species tional groups (Elmendorf, Henry, Hollister, Björk, Boulanger‐Lapointe, variation to be small compared to among‐species variation (Anderegg et al., 2012). Continuing to assess the extent to which trait‐based ap‐ et al., 2018; Kattge et al., 2011; Siefert et al., 2015), including in the proaches can meaningfully describe and predict ecosystem processes tundra biome (Thomas et al., in prep, manuscript available upon re‐ therefore remains an essential research focus (McGill et al., 2006). quest). Nevertheless, within‐species trait variation may be an import‐ Differentiating community responses or ecosystem processes using ant driver of community change, particularly at small spatial scales, post hoc trait‐based classifications would provide a direct test of this and may explain highly individualistic species responses to change question, and could offer valuable insight into the relative importance (Hollister et al., 2005). Thus, we advocate that studies should recog‐ of different traits for prediction and modelling. nize and account for the extent of trait variation within communities. Third, we assume that the majority of trait variation occurs among Finally, attempts to classify species into functional groups may be species. Should large trait variation occur within species this could impossible if trait expression or species response is dependent upon 14 | THOMAS et al. environmental and ecological context (Dorrepaal, 2007; Laughlin & use of underlying trait data, could provide new insights for predict‐ Messier, 2015). Group classifications and even growth strategies ing vegetation change and ecosystem processes in response to global may change depending on resource availability (Bret‐Harte et al., drivers of environmental change. 2001), such that division into discrete classifications may obscure the variability inherent to natural environments (Westoby & Wright, AC K N OW L E D G M E N T S 2006). Although differences between functional groups were sta‐ tistically significant in this study, the majority of trait variation was The project was funded by the UK Natural Environment Research not explained by classifications, whether using traditional functional Council [ShrubTundra Project NE/M016323/1 (IMS, AB, HT, SAB, groups (81% of variance unexplained) or post hoc classification (55% DG) & PhD Studentship NE/L002558/1 (HT)], the Synthesis Centre of of variance unexplained). We, therefore, join those who advocate the German Centre for Integrative Biodiversity Research (iDiv) Halle‐ that ecological analyses should continue to move towards incor‐ Jena‐Leipzig (DFG FZT 118; sTundra working group [postdoctoral fel‐ porating explicitly trait‐based approaches, focusing on traits them‐ lowship to AB]). The study has been supported by the TRY initiative selves as the fundamental units of analysis (Laughlin, 2014; McGill on plant traits (https://www.try-db.org). The TRY initiative and data‐ et al., 2006; Violle & Jiang, 2009; Weiher et al., 2011; Westoby & base is hosted at the Max Planck Institute for Biogeochemistry, Jena, Wright, 2006). Germany. TRY is currently supported by DIVERSITAS/Future Earth and the German Centre for Integrative Biodiversity Research (iDiv) Halle‐Jena‐Leipzig. Authors were supported by the Swedish Research 4.5 | Future priorities Council (2015‐00465) (DB) and (2015‐00498) (EK), Marie Skłodowska Our findings suggest that new trait data collection campaigns should Curie Actions (INCA 600398) (DB), the National Science Foundation focus on traits that distinguish among ecological strategies and re‐ (USA; RH), the Carlsberg Foundation (2013‐01‐0825) (SN), the Danish sponses to changing growing conditions. Whilst existing trait records Council for Independent Research ‐ Natural Sciences (DFF 4181‐00565) have been informed by standardized protocols and contemporary re‐ (SN), European Research Council Synergy grant ERC‐SyG‐2013‐610028 search priorities (Cornelissen et al., 2003; Pérez‐Harguindeguy et al., IMBALANCE‐P (JP), University of Zurich Research Priority Program 2013), these have tended to focus on easily measurable leaf traits. on Global Change and Biodiversity (GSS, MIG), the Office of Biological Future trait collection campaigns should therefore focus on ecologi‐ and Environmental Research in the U.S. Department of Energy’s Office cally important traits for which we have few records, including chemi‐ of Science (Next‐Generation Ecosystem Experiments in the Arctic cal and physiological traits (Eckstein, Karlsson, & Weih, 1999), and ‐ NGEE Arctic) (CMI), NASA Arctic Boreal Vulnerability Experiment ‐ whole‐plant measurements, incorporating stem (Chave et al., 2009) ABoVE (LB, SG), The Swiss National Science Foundation (EF, AK, SV), and belowground (Iversen et al., 2015) characteristics. Finally, pheno‐ NSERC Canada (EL, JJ, AP, BSPG, TZ), ArcticNet (EL, AP, GH), The US logical traits such as leaf out or flowering time are rarely integrated into National Science Foundation Niwot Ridge LTER (DEB‐1637686) (MJ), wider trait‐based approaches, yet may be critical to predicting ecologi‐ Long‐Term Ecological Research (DEB‐1234162) (PR) and Long‐Term cal responses, particularly in a warming tundra (Cleland et al., 2012). Research in Environmental Biology (DEB‐1242531) (PR), Organismo Autónomo Parques Nacionales (JMN), the Arctic Research Centre, Denmark (JNN), RSF (#14‐50‐000290) (VO), the Polar Continental 5 | CO N C LU S I O N Shelf Program (AP, EL, GH), the Royal Canadian Mounted Police (GH), the Montagna di Torricchio Nature Reserve (Italy) (GC) the Academy In this study, we demonstrate that traditional plant functional groups of Finland Decisions no. 256991 (BF), JPI Climate no. 291581 (BF), poorly represent differences in the six most commonly measured plant and the BBSRC David Phillips Fellowship (BB/L02456X/1) (FTdV). traits among tundra vascular plant species. Although functional groups Additional data and contributions were provided by L. Andreu‐Hayles, were statistically distinct and consistent among sites, they explained P. Beck, A. Blach Overgaard, B. Bond‐Lamberty, J. Craine, J. Dickie, only 19% of overall trait variation and primarily differentiated between S. Dullinger, B. Eberling, B. Enquist, J. Fang, K. Fleischer, H. Ford, resource economic traits rather than size‐related traits. Post hoc trait‐ G. Freschet, E. Garnier, D. Georges, R. Halfdan Jørgensen, K. Harper, based classification of species did not align with functional group S. Harrison, M. Harze, G. Henry, S. Jansen, J. Hille Ris Lambers, R. Klady, classification, but produced robust alternative groupings that aligned M. Kleyer, S. Kuleza, T. Lantz, A. Lavalle, F. Louault, B. Medlyn, R. Milla, with two global axes of trait variation. Together, our findings indicate J. Ordonez, C. Pladevall, H. Poorter, P. Poschlod, C. Price, N. Rueger, B. that traditional functional groups may not characterize trait variation Sandel, F. Schweingruber, B. Shipley, A. Siefert, L. Street, K. Suding, J. within tundra vegetation communities, particularly among size‐related Tremblay, M. Tremblay, M. Vellend, E. Weiher, C. Wirth, P. Wookey and I. traits. We therefore argue that: (a) traditional functional groups should Wright and the Royal Botanic Gardens Kew Seed Information Database be used with caution when testing ecological responses or ecosystem (SID). We thank innumerable field technicians, logistics teams, graduate functions associated with size‐related traits; (b) functional group ap‐ and undergraduate assistants for help with data collection, and parks, proaches require sufficient species and trait measurements to capture wildlife refuges, field stations, and the local and indigenous people for variation within groups, within species and among traits; and (c) the the opportunity to conduct research on their land. Finally, we thank the use of alternative classifications based on trait expression, or direct referees and editors for their constructive comments on the manuscript. THOMAS et al. | 15 AU T H O R C O N T R I B U T I O N S vegetation to global change. Journal of Vegetation Science, 7, 347– 358. https://doi.org/10.2307/3236278 HT and IMS conceived the study. HT performed statistical analysis Chave, J., Coomes, D., Jansen, S., Lewis, S. L., Swenson, N. G., with additional input from IMS and AB. HT wrote the manuscript with & Zanne, A. E. (2009). Towards a worldwide wood eco‐ nomics spectrum. Ecology Letters, 12, 351–366. https://doi. input from IMS and AB with contributions from all authors. AB com‐ org/10.1111/j.1461-0248.2009.01285.x piled the TTT database with assistance from IMS, SE and AB led the Clark, J. S. (2016). Why species tell more about traits than traits about sTundra working group. IMS supervised HT and acquired funding for species: Predictive analysis. Ecology, 97, 1–7. https://doi.org/10.1002/ the project. Authorship order was based on total contribution to the ecy.1453 manuscript for the first four authors (see above), and then (a) input Cleland, E. E., Allen, J. M., Crimmins, T. M., Dunne, J. A., Pau, S., Travers, S. E., … Wolkovich, E. M. (2012). Phenological tracking enables pos‐ from the sTundra working group and contribution to TTT (alphabeti‐ itive species responses to climate change. Ecology, 93, 1765–1771. cal), (b) input from the sTundra working group and contribution to https://doi.org/10.1890/11-1912.1 TRY (alphabetical), (c) input from the sTundra working group only (al‐ Cornelissen, J., Aerts, R., Cerabolini, B., Werger, M., & van der Heijden, phabetical), (d) contribution to the TTT database (alphabetical), and M. (2001). Carbon cycling traits of plant species are linked with my‐ corrhizal strategy. Oecologia, 129, 611–619. https://doi.org/10.1007/ (e) contribution to TRY (alphabetical). s004420100752 Cornelissen, J. H. C., Lavorel, S., Garnier, E., Díaz, S., Buchmann, N., Gurvich, D. E., … Poorter, H. (2003). A handbook of protocols for DATA AC C E S S I B I L I T Y standardised and easy measurement of plant functional traits worldwide. Australian Journal of Botany, 51, 335–380. https://doi. Trait data have been submitted to the TRY database (https:// org/10.1071/BT02124 www.try-db.org) and are publicly available at https://github.com/ Cornelissen, J. H. C., & Makoto, K. (2014). Winter climate change, plant TundraTraitTeam/TraitHub. Composition data are available in the Polar traits and nutrient and carbon cycling in cold biomes. Ecological Data Catalogue (https://www.polardata.ca/ CCN 10786). Code is avail‐ Research, 29, 517–527. https://doi.org/10.1007/s11284-013-1106-1 Cornelissen, J. H. C., Quested, H. M., Gwynn‐Jones, D., Van Logtestijn, able at https://github.com/hjdthomas/Tundra_functional_groups R. S. P., De Beus, M. A. H., Kondratchuk, A., … Aerts, R. (2004). Leaf digestibility and litter decomposability are related in a wide range of subarctic plant species and types. Functional Ecology, 18, 779–786. REFERENCES https://doi.org/10.1111/j.0269-8463.2004.00900.x Anderegg, L. D. L., Berner, L. T., Badgley, G., Sethi, M. L., Law, B. E., & Cornelissen, J. H. C., van Bodegom, P. M., Aerts, R., Callaghan, T. V., van HilleRisLambers, J. (2018). Within‐species patterns challenge our Logtestijn, R. S. P., Alatalo, J., … Zhao, X. (2007). Global negative veg‐ understanding of the leaf economics spectrum. Ecology Letters, 21(5), etation feedback to climate warming responses of leaf litter decom‐ 734–744. https://doi.org/10.1111/ele.12945 position rates in cold biomes. Ecology Letters, 10, 619–627. https:// Bardgett, R. D., Mommer, L., & De Vries, F. T. (2014). Going underground: doi.org/10.1111/j.1461-0248.2007.01051.x Root traits as drivers of ecosystem processes. Trends in Ecology and Cornwell, W. K., & Ackerly, D. D. (2010). A link between plant Evolution, 29, 692–699. https://doi.org/10.1016/j.tree.2014.10.006 traits and abundance: Evidence from coastal California Bjorkman, A., Myers‐Smith, I., Elmendorf, S., Normand, S., Thomas, H., Alatalo, woody plants. Journal of Ecology, 98, 814–821. https://doi. J., … Zamin, T. (2018a). Tundra Trait Team: A database of plant traits span‐ org/10.1111/j.1365-2745.2010.01662.x ning the tundra biome. Global Ecology and Biogeography. 27, 1402–1411. Cornwell, W. K., Cornelissen, J. H. C., Amatangelo, K., Dorrepaal, E., Bjorkman, A. D., Myers‐Smith, I. H., Elmendorf, S. C., Normand, S., Rüger, Eviner, V. T., Godoy, O., … Westoby, M. (2008). Plant species traits N., Beck, P. S., … Georges, D. (2018b). Plant functional trait change are the predominant control on litter decomposition rates within across a warming tundra biome. Nature, 562, 57–62. biomes worldwide. Ecology Letters, 11, 1065–1071. https://doi. Bliss, L. C., Heal, O. W., & Moore, J. J. (1981). Tundra ecosystems: A com‐ org/10.1111/j.1461-0248.2008.01219.x parative analysis. New York, NY: Cambridge University Press. Cramer, W., Bondeau, A., Woodward, F. I., Prentice, I. C., Betts, R. A., Boulangeat, I., Philippe, P., Abdulhak, S., Douzet, R., Garraud, L., Lavergne, Brovkin, V., … Young‐Molling, C. (2001). Global response of terres‐ S., … Thuiller, W. (2012). Improving plant functional groups for dy‐ trial ecosystem structure and function to CO2 and climate change: namic models of biodiversity: At the crossroads between functional Results from six dynamic global vegetation models. Global Change and community ecology. Global Change Biology, 18, 3464–3475. Biology, 7, 357–373. https://doi.org/10.1111/j.1365-2486.2012.02783.x Díaz, S., Kattge, J., Cornelissen, J. H. C., Wright, I. J., Lavorel, S., Dray, S., Bret‐Harte, M. S., Mack, M. C., Goldsmith, G. R., Sloan, D. B., DeMarco, … Gorné, L. D.. (2016). The global spectrum of plant form and func‐ J., Shaver, G. R., … Chapin, F. S. (2008). Plant functional types do tion. Nature, 529, 167–171. https://doi.org/10.1038/nature16489 not predict biomass responses to removal and fertilization in Díaz, S., Lavorel, S., McIntyre, S., Falczuk, V., Casanoves, F., Milchunas, Alaskan tussock tundra. Journal of Ecology, 96, 713–726. https://doi. D. G., … Campbell, B. D. (2007). Plant trait responses to grazing ‐ org/10.1111/j.1365-2745.2008.01378.x A global synthesis. Global Change Biology, 13, 313–341. https://doi. Bret‐Harte, M. S., Shaver, G. R., Zoerner, J. P., Johnstone, J. F., Wagner, org/10.1111/j.1365-2486.2006.01288.x J. L., Chavez, A. S., … Laundre, J. A. (2001). Developmental plasticity Ding, C., & He, X. (2004). K‐means clustering via principal component allows Betula nana to dominate tundra subjected to an altered envi‐ analysis. Proceedings of the Twenty‐First International Conference on ronment. Ecology, 82, 18–32. https://doi.org/10.2307/2680083 Machine Learning, Cl, 29. ACM. Carbognani, M., Petraglia, A., & Tomaselli, M. (2014). Warming ef‐ Dormann, C. F., & Woodin, S. J. (2002). Climate change in the fects and plant trait control on the early‐decomposition in alpine Arctic: Using plant functional types in a meta‐analysis of snowbeds. Plant and Soil, 376, 277–290. https://doi.org/10.1007/ field experiments. Functional Ecology, 16, 4–17. https://doi. s11104-013-1982-8 org/10.1046/j.0269-8463.2001.00596.x Chapin, F. S., Bret‐Harte, M. S., Hobbie, S. E., & Zhong, H. L. (1996). Dorrepaal, E. (2007). Are plant growth‐form‐based classifications use‐ Plant functional types as predictors of transient responses of arctic ful in predicting northern ecosystem carbon cycling feedbacks 16 | THOMAS et al. to climate change? Journal of Ecology, 95, 1167–1180. https://doi. Iversen, C. M., Sloan, V. L., Sullivan, P. F., Euskirchen, E. S., McGuire, A. org/10.1111/j.1365-2745.2007.01294.x D., Norby, R. J., … Wullschleger, S. D. (2015). The unseen iceberg: Dorrepaal, E., Cornelissen, J. H. C., Aerts, R., Wallén, B., & Van Plant roots in arctic tundra. New Phytologist, 205, 34–58. https://doi. Logtestijn, R. S. P. (2005). Are growth forms consistent predictors org/10.1111/nph.13003 of leaf litter quality and decomposability across peatlands along a Jetz, W., Cavender‐Bares, J., Pavlick, R., Schimel, D., Davis, F. W., Asner, latitudinal gradient? Journal of Ecology, 93, 817–828. https://doi. G. P., … Ustin, S. L. (2016). Monitoring plant functional diversity from org/10.1111/j.1365-2745.2005.01024.x space. Nature Plants, 2, 1–5. Dyer, A. R., Goldberg, D. E., Turkington, R., & Sayre, C. (2001). Effects Kaarlejärvi, E., Eskelinen, A., & Olofsson, J. (2017). Herbivores rescue of growing conditions and source habitat on plant traits and func‐ diversity in warming tundra by modulating trait‐dependent species tional group definition. Functional Ecology, 15, 85–95. https://doi. losses and gains. Nature Communications, 8. https://doi.org/10.1038/ org/10.1046/j.1365-2435.2001.00487.x s41467-017-00554-z Eckstein, R. L., Karlsson, P. S., & Weih, M. (1999). Leaf life span and nu‐ Kattge, J., Díaz, S., Lavorel, S., Prentice, I. C., Leadley, P., Bönisch, trient resorption as determinants of plant nutrient conservation in G., … Wirth, C. (2011). TRY – A global database of plant temperate‐arctic regions. New Phytologist, 143, 177–189. https://doi. traits. Global Change Biology, 17, 2905–2935. https://doi. org/10.1046/j.1469-8137.1999.00429.x org/10.1111/j.1365-2486.2011.02451.x Elmendorf, S. C., Henry, G. H. R., Hollister, R. D., Björk, R. G., Bjorkman, Körner, C., Leuzinger, S., Riedl, S., Siegwolf, R. T., & Streule, L. (2016). A. D., Callaghan, T. V., … Wookey, P. A. (2012). Global assessment of Carbon and nitrogen stable isotope signals for an entire alpine flora, experimental climate warming on tundra vegetation: Heterogeneity based on herbarium samples. Alpine Botany, 126, 153–166. https:// over space and time. Ecology Letters, 15, 164–175. doi.org/10.1007/s00035-016-0170-x Elmendorf, S. C., Henry, G. H. R., Hollister, R. D., Björk, R. G., Boulanger‐ Laughlin, D. C. (2011). Nitrification is linked to dominant leaf traits rather Lapointe, N., Cooper, E. J., … Wipf, S. (2012). Plot‐scale evidence than functional diversity. Journal of Ecology, 99, 1091–1099. https:// of tundra vegetation change and links to recent summer warming. doi.org/10.1111/j.1365-2745.2011.01856.x Nature Climate Change, 2, 453–457. Laughlin, D. C. (2014). The intrinsic dimensionality of plant traits and its Forrestel, E. J., Donoghue, M. J., Edwards, E. J., Jetz, W., du Toit, J. C. relevance to community assembly. Journal of Ecology, 102, 186–193. O., & Smith, M. D. (2017). Different clades and traits yield similar https://doi.org/10.1111/1365-2745.12187 grassland functional responses. Proceedings of the National Academy Laughlin, D. C., & Messier, J. (2015). Fitness of multidimensional pheno‐ of Sciences USA, 114, 705–710, 201612909. https://doi.org/10.1073/ types in dynamic adaptive landscapes. Trends in Ecology and Evolution, pnas.1612909114 30, 487–496. https://doi.org/10.1016/j.tree.2015.06.003 Fry, E. L., Power, S. A., & Manning, P. (2014). Trait‐based classification and Lavorel, S., & Garnier, E. (2002). Predicting changes in community com‐ manipulation of plant functional groups for biodiversity‐ecosystem position and ecosystem functioning from plant traits: Revisiting the function experiments. Journal of Vegetation Science, 25, 248–261. Holy Grail. Functional Ecology, 16, 545–556. https://doi.org/10.1111/jvs.12068 Little, C. J., Jagerbrand, A. K., Molau, U., & Alatalo, J. M. (2015). Galloway, J. N., Townsend, A. R., Erisman, J. W., Bekunda, M., Cai, Z., Community and species‐specific responses to simulated global Freney, J. R., … Sutton, M. A. (2008). Transformation of the nitrogen change in two subarctic‐alpine plant communities. Ecosphere, 6, 227. cycle: Recent trends, questions, and potential solutions. Science, 320, https://doi.org/10.1890/ES14-00427.1 889–892. https://doi.org/10.1126/science.1136674 Loranty, M., Goetz, S., & Beck, P. (2011). Tundra vegetation effects on Grime, J. P. P., Thompson, K., Hunt, R., Hodgson, J. G., Cornelissen, J. pan‐Arctic albedo. Environmental Research, 6, 029601. H. C., Rorison, I. H., … Whitehouse, J. (1997). Integrated screening Lukasová, A. (1979). Hierarchical agglomerative clustering pro‐ validates primary axes of specialisation in plants. Oikos, 79, 259–281. cedure. Pattern Recognition, 11, 365–381. https://doi. https://doi.org/10.2307/3546011 org/10.1016/0031-3203(79)90049-9 Harrison, S. P., Prentice, I. C., Barboni, D., Kohfeld, K. E., Ni, J., & Sutra, Maire, E., Grenouillet, G., Brosse, S., & Villéger, S. (2015). How many J. P. (2010). Ecophysiological and bioclimatic foundations for a global dimensions are needed to accurately assess functional diver‐ plant functional classification. Journal of Vegetation Science, 21, 300– sity? A pragmatic approach for assessing the quality of functional 317. https://doi.org/10.1111/j.1654-1103.2009.01144.x spaces. Global Ecology and Biogeography, 24, 728–740. https://doi. Henry, G. H. R., & Molau, U. (1997). Tundra plants and climate change: org/10.1111/geb.12299 The International Tundra Experiment (ITEX). Global Change Biology, 3, Matesanz, S., Brooker, R. W., Valladares, F., & Klotz, S. (2009). Temporal 1–9. https://doi.org/10.1111/j.1365-2486.1997.gcb132.x dynamics of marginal steppic vegetation over a 26‐year period of Hollister, R. D., Webber, P. J., & Tweedie, C. E. (2005). The response of substantial environmental change. Journal of Vegetation Science, Alaskan arctic tundra to experimental warming: Differences between 20(2), 299–310. short‐ and long‐term responses. Global Change Biology, 11, 525–536. McGill, B. J., Enquist, B. J., Weiher, E., & Westoby, M. (2006). Rebuilding https://doi.org/10.1111/j.1365-2486.2005.00926.x community ecology from functional traits. Trends in Ecology and Hudson, J. M. G., Henry, G. H. R., & Cornwell, W. K. (2011). Taller and Evolution, 21, 178–185. https://doi.org/10.1016/j.tree.2006.02.002 larger: Shifts in Arctic tundra leaf traits after 16 years of experi‐ McLaren, J. R., & Turkington, R. (2010). Ecosystem properties deter‐ mental warming. Global Change Biology, 17, 1013–1021. https://doi. mined by plant functional group identity. Journal of Ecology, 98, 459– org/10.1111/j.1365-2486.2010.02294.x 469. https://doi.org/10.1111/j.1365-2745.2009.01630.x IPCC. (2013). Climate change 2013: The physical science basis: McLaren, J. R., & Turkington, R. (2011). Biomass compensation Working Group I contribution to the Fifth assessment report of the and plant responses to 7 years of plant functional group re‐ Intergovernmental Panel on Climate Change. In T. F. Stocker, D. Qin, movals. Journal of Vegetation Science, 22, 503–515. https://doi. G.‐K. Plattner, M. Tignor, S. K. Allen, J. Boschung, A. Nauels, Y. Xia, V. org/10.1111/j.1654-1103.2011.01263.x Bex, & P. M. Midgle (Eds.). Cambridge University Press. Mekonnen, Z. A., Riley, W. J., & Grant, R. F. (2018). Accelerated nutri‐ Iversen, C. M., McCormack, M. L., Powell, A. S., Blackwood, C. B., ent cycling and increased light competition will lead to 21st cen‐ Freschet, G. T., Kattge, J., … Violle, C. (2017). A global fine‐root tury shrub expansion in North American Arctic tundra. Journal of ecology database to address below‐ground challenges in plant Geophysical Research: Biogeosciences, 123(5), 1683–1701. ecology. New Phytologist, 215(1), 15–26. https://doi.org/10.1111/ Mörsdorf, M. A., Ravolainen, V. T., Støvern, L. E., Yoccoz, N. G., Jónsdóttir, nph.14486 I. S., & Bråthen, K. A. (2015). Definition of sampling units begets THOMAS et al. | 17 conclusions in ecology: The case of habitats for plant communities. through the community‐level: A trait‐based response‐and‐effect PeerJ, 3, 1–17. https://doi.org/10.7717/peerj.815 framework for plants. Global Change Biology, 14, 1125–1140. https:// Myers‐Smith, I. H., Forbes, B. C., Wilmking, M., Hallinger, M., Lantz, T., doi.org/10.1111/j.1365-2486.2008.01557.x Blok, D., … Hik, D. S. (2011). Shrub expansion in tundra ecosystems: Turetsky, M. R., Mack, M. C., Hollingsworth, T. N., & Harden, J. W. (2010). Dynamics, impacts and research priorities. Environmental Research The role of mosses in ecosystem succession and function in Alaska’s Letters, 6, 045509. https://doi.org/10.1088/1748-9326/6/4/045509 boreal forest. Canadian Journal of Forest Research, 40, 1237–1264. Oksanen, J., Blanchet, F. G., Kindt, R., Legendre, P., Minchin, P. R., O’hara, Venn, S., Pickering, C., & Green, K. (2014). Spatial and temporal func‐ R. B., … Oksanen, M. J. (2013). Package ‘vegan’. Community ecology tional changes in alpine summit vegetation are driven by increases package, version, 2(9). in shrubs and graminoids. AoB PLANTS, 6, 1–15. https://doi. Pérez‐Harguindeguy, N., Díaz, S., Garnier, E., Lavorel, S., Poorter, H., org/10.1093/aobpla/plu008 Jaureguiberry, P., … Cornelissen, J. H. C. (2013). New handbook for Violle, C., Enquist, B. J., McGill, B. J., Jiang, L., Albert, C. H., Hulshof, C., … standardised measurement of plant functional traits worldwide. Messier, J. (2012). The return of the variance: Intraspecific variability Australian Journal of Botany, 61, 167–234. https://doi.org/10.1071/ in community ecology. Trends in Ecology and Evolution, 27, 244–252. BT12225 https://doi.org/10.1016/j.tree.2011.11.014 R Core Team. (2017). R: A language and environment for statistical comput‐ Violle, C., & Jiang, L. (2009). Towards a trait‐based quantification of spe‐ ing. Vienna, Austria: R Foundation for Statistical Computing. cies niche. Journal of Plant Ecology, 2, 87–93. https://doi.org/10.1093/ Reich, P. B. (2014). The world‐wide “fast‐slow” plant economics spec‐ jpe/rtp007 trum: A traits manifesto. Journal of Ecology, 102, 275–301. https:// Vowles, T., Gunnarsson, B., Molau, U., Hickler, T., Klemedtsson, L., & Björk, doi.org/10.1111/1365-2745.12211 R. G. (2017). Expansion of deciduous tall shrubs but not evergreen Reichstein, M., Bahn, M., Mahecha, M. D., Kattge, J., & Baldocchi, D. dwarf shrubs inhibited by reindeer in Scandes mountain range. Journal D. (2014). Linking plant and ecosystem functional biogeography. of Ecology, 38, 42–49. https://doi.org/10.1111/1365-2745.12753 Proceedings of the National Academy of Sciences USA, 111, 13697– Weiher, E., Freund, D., Bunton, T., Stefanski, A., Lee, T., & Bentivenga, 13702. https://doi.org/10.1073/pnas.1216065111 S. (2011). Advances, challenges and a developing synthesis of eco‐ Saccone, P., & Virtanen, R. (2016). Extrapolating multi‐decadal plant logical community assembly theory. Philosophical Transactions of community changes based on medium‐term experiments can be the Royal Society B: Biological Sciences, 366, 2403–2413. https://doi. risky: Evidence from high‐latitude tundra. Oikos, 125, 76–85. https:// org/10.1098/rstb.2011.0056 doi.org/10.1111/oik.02399 Weltzin, J. F., Loik, M. E., Schwinning, S., Williams, D. G., Fay, P. A., Saccone, P., Hoikka, K., & Virtanen, R. (2017). What if plant functional Haddad, B. M., … Zak, J. C. (2003). Assessing the response of ter‐ types conceal species‐specific responses to environment? Study on restrial ecosystems to potential changes in precipitation. BioScience, arctic shrub communities. Ecology, 98(6), 1600–1612. https://doi. 53, 941. https://doi.org/10.1641/0006-3568(2003)053[0941:ATRO org/10.1002/ecy.1817 TE]2.0.CO;2 Sandel, B., Gutierrez, A., Schrodt, F., Banerjee, A., Reich, P., & Kattge, Westoby, M., Falster, D. S., Moles, A. T., Vesk, P. A., & Wright, I. J. (2002). J. (2015). Estimating the missing species bias in plant trait mea‐ Plant ecological strategies: Some leading dimensions of variation be‐ surements. Journal of Vegetation Science, 25, 828–838. https://doi. tween species. Annual Review of Ecology and Systematics, 33, 125– org/10.1111/jvs.12292 159. https://doi.org/10.1146/annurev.ecolsys.33.010802.150452 Serreze, M. C., & Barry, R. G. (2011). Processes and impacts of Arctic Westoby, M., & Wright, I. J. (2006). Land‐plant ecology on the basis of amplification: A research synthesis. Global and Planetary Change, 77, functional traits. Trends in Ecology and Evolution, 21, 261–268. https:// 85–96. https://doi.org/10.1016/j.gloplacha.2011.03.004 doi.org/10.1016/j.tree.2006.02.004 Shaver, G. R., Syndonia Bret‐Harte, M., Jones, M. H., Johnstone, J., Wright, S. J., Kitajima, K., Kraft, N. J. B., Reich, P. B., Wright, I. J., Bunker, Gough, L., Laundre, J., & Chapin, F. S. (2001). Species composition D. E., … Zanne, A. E. (2013). Functional traits and the growth—mor‐ interacts with fertilizer to control long‐term change in tundra pro‐ tality trade‐off in tropical trees. Ecology, 91, 3664–3674. https://doi. ductivity. Ecology, 82, 3163–3181. https://doi.org/10.1890/0012-96 org/10.1890/09-2335.1 58(2001)082[3163:SCIWFT]2.0.CO;2 Wright, J. P., Naeem, S., Hector, A., Lehman, C., Reich, P. B., Shipley, B., De Bello, F., Cornelissen, J. H. C., Laliberté, E., Laughlin, Schmid, B., & Tilman, D. (2006). Conventional functional clas‐ D. C., & Reich, P. B. (2016). Reinforcing loose foundation stones in sification schemes underestimate the relationship with eco‐ trait‐based plant ecology. Oecologia, 1–9. https://doi.org/10.1007/ system functioning. Ecology Letters, 9, 111–120. https://doi. s00442-016-3549-x org/10.1111/j.1461-0248.2005.00850.x Siefert, A., Violle, C., Chalmandrier, L., Albert, C. H., Taudiere, A., Wright, I. J., Reich, P. B., Cornelissen, J. H. C., Falster, D. S., Garnier, E., Fajardo, A., … Wardle, D. A. (2015). A global meta‐analysis of the Hikosaka, K., … Westoby, M. (2005). Assessing the generality of relative extent of intraspecific trait variation in plant communi‐ global leaf trait relationships. New Phytologist, 166, 485–496. https:// ties. Ecology Letters, 18, 1406–1419. https://doi.org/10.1111/ doi.org/10.1111/j.1469-8137.2005.01349.x ele.12508 Wullschleger, S. D., Epstein, H. E., Box, E. O., Euskirchen, E. S., Goswami, Soudzilovskaia, N., Douma, J., Akhmetzhanova, A., van Bodegom, P., S., Iversen, C. M., … Xu, X. (2014). Plant functional types in Earth sys‐ Cornwell, W., Moens, E., … Cornelissen, J. H. C. (2015). Global pat‐ tem models: Past experiences and future directions for application terns of plant root colonization intensity by mycorrhizal fungi ex‐ of dynamic vegetation models in high‐latitude ecosystems. Annals of plained by climate and soil chemistry. Global Ecology and Biogeography, Botany, 114, 1–16. https://doi.org/10.1093/aob/mcu077 24, 371–382. https://doi.org/10.1111/geb.12272 Zamin, T. J., Bret‐Harte, M. S., & Grogan, P. (2014). Evergreen shrubs Soudzilovskaia, N. A., Elumeeva, T. G., Onipchenko, V. G., Shidakov, dominate responses to experimental summer warming and fertil‐ I. I., Salpagarova, F. S., Khubiev, A. B., … Cornelissen, J. H. C. ization in Canadian mesic low arctic tundra. Journal of Ecology, 102, (2013). Functional traits predict relationship between plant abun‐ 749–766. https://doi.org/10.1111/1365-2745.12237 dance dynamic and long‐term climate warming. Proceedings of the National Academy of Sciences USA, 110, 18180–18184. https://doi. SUPPORTING INFORMATION org/10.1073/pnas.1310700110 Suding, K. N., Lavorel, S., Chapin, F. S., Cornelissen, J. H. C., Díaz, S., Additional Supporting Information may be found online in the Garnier, E., … Navas, M.‐L. (2008). Scaling environmental change supporting information tab for this article. 18 | THOMAS et al. BIOSKETCH How to cite this article: Thomas HJD, Myers‐Smith IH, This work was led by Haydn J. D. Thomas as part of the Bjorkman AD, et al. Traditional plant functional groups sTUNDRA working group. Haydn is a plant ecologist inter‐ explain variation in economic but not size‐related traits ested in how the tundra biome is changing. His work primar‐ across the tundra biome. Global Ecol Biogeogr. 2018;00:1–18. ily focuses on whether, and how, plant traits can be used to https://doi.org/10.1111/geb.12783 predict change to plant communities and ecosystem function in tundra ecosystems. The sTUNDRA working group is an international collaboration exploring tundra vegetation and trait change across biome‐scale climate gradients and over three decades of ecological monitoring led by Isla H. Myers‐ Smith, Anne D. Bjorkman and Sarah C. Elmendorf. Trait data were provided by the “Tundra Trait Team”, compiled by Anne D. Bjorkman. Additional trait data were provided by TRY, a network of vegetation scientists headed by Future Earth and the Max Planck Institute for Biogeochemistry, providing a global archive of curated plant traits.