Plant Ecology & Diversity on the Delineation of Tropical Vegetation

This article was downloaded by: [JAMES COOK UNIVERSITY] On: 27 October 2013, At: 17:27 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Plant Ecology & Diversity Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/tped20 On the delineation of tropical vegetation types with an emphasis on forest/savanna transitions Mireia Torello-Raventos a b , Ted R. Feldpausch b , Elmar Veenendaal c , Franziska Schrodt b , Gustavo Saiz a , Tomas F. Domingues d , Gloria Djagbletey e , Andrew Ford f , Jeanette Kemp g , Beatriz S. Marimon h , Ben Hur Marimon Junior h , Eddie Lenza h , James A. Ratter i , Leandro Maracahipes h , Denise Sasaki j , Bonaventure Sonké k , Louis Zapfack v , Hermann Taedoumg k , Daniel Villarroel l , Michael Schwarz m , Carlos A. Quesada b n , F. Yoko Ishida n , Gabriela B. Nardoto o p , Kofi Affum-Baffoe q , Luzmilla Arroyo l , David M.J.S. Bowman r , Halidou Compaore s , Kalu Davies a , Adama Diallo t , Nikolaos M. Fyllas b , Martin Gilpin b , Fidèle Hien s , Michelle Johnson b , Timothy J. Killeen l u , Daniel Metcalfe f , Heloisa S. Miranda p , Mark Steininger u , John Thomson a , Karle Sykora c , Eric Mougin w , Pierre Hiernaux w , Michael I. Bird a , John Grace d , Simon L. Lewis b x , Oliver L. Phillips b & Jon Lloyd a b a School of Earth and Environmental Science, James Cook University , Cairns , Australia b Earth and Biosphere Institute, School of Geography, University of Leeds , Leeds , UK c Centre for Ecosystem Studies, University of Wageningen , Wageningen , the Netherlands d School of Geosciences, University of Edinburgh , Edinburgh , UK e Forest Research Institute of Ghana , Kumasi , Ghana f CSIRO Ecosystem Sciences Tropical Forest Research Centre , Atherton , Australia g Queensland Herbarium , Townsville , Australia h Universidade do Estado de Mato Grosso , Nova Xavantina , Brazil i Royal Botanic Garden , Edinburgh , UK j Fundação Ecológica Cristalino , Alta Floresta , Brazil k Plant Systematics and Ecology Lab, Department of Biology , University of Yaoundé I , Yaoundé , Cameroon l Museo Noel Kempff Mercado , Santa Cruz , Bolivia m Fieldwork Assistance , Jena , Germany n Instituto Nacional de Pesquisas da Amazônia , Manaus , Brazil o Centro de Energia Nuclear na Agricultura , São Paulo , Brazil p Universidade de Brasilia , District Federal , Brazil q Resource Management Support Centre, Forestry Commission of Ghana , Kumasi , Ghana r School of Plant Science University of Tasmania , Hobart , Australia s Institut de l'Environnement et de Recherches Agricoles , Ouagadougou , Burkina Faso t Centre National des Semences Forestières , Ouagadougou , Burkina Faso u Center for Applied Biodiversity Science, Conservation International , Washington , DC , USA v Deparment of Plant Biology , University of Yaounde I , Yaoundé , Cameroon w Géosciences Environnement Toulouse , Toulouse , France x Department of Geography , University College London , London , UK Accepted author version posted online: 03 Jan 2013.Published online: 20 Mar 2013.

To cite this article: Mireia Torello-Raventos , Ted R. Feldpausch , Elmar Veenendaal , Franziska Schrodt , Gustavo Saiz , Tomas F. Domingues , Gloria Djagbletey , Andrew Ford , Jeanette Kemp , Beatriz S. Marimon , Ben Hur Marimon Junior , Eddie Lenza , James A. Ratter , Leandro Maracahipes , Denise Sasaki , Bonaventure Sonké , Louis Zapfack , Hermann Taedoumg , Daniel Villarroel , Michael Schwarz , Carlos A. Quesada , F. Yoko Ishida , Gabriela B. Nardoto , Kofi Affum-Baffoe , Luzmilla Arroyo , David M.J.S. Bowman , Halidou Compaore , Kalu Davies , Adama Diallo , Nikolaos M. Fyllas , Martin Gilpin , Fidèle Hien , Michelle Johnson , Timothy J. Killeen , Daniel Metcalfe , Heloisa S. Miranda , Mark Steininger , John Thomson , Karle Sykora , Eric Mougin , Pierre Hiernaux , Michael I. Bird , John Grace , Simon L. Lewis , Oliver L. Phillips & Jon Lloyd (2013) On the delineation of tropical vegetation types with an emphasis on forest/savanna transitions, Plant Ecology & Diversity, 6:1, 101-137, DOI: 10.1080/17550874.2012.762812 To link to this article: http://dx.doi.org/10.1080/17550874.2012.762812

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 Plant Ecology & Diversity, 2013 Vol. 6, Issue 1, 101–137, http://dx.doi.org/10.1080/17550874.2012.762812

On the delineation of tropical vegetation types with an emphasis on forest/savanna transitions Mireia Torello-Raventosa,b, Ted R. Feldpauschb , Elmar Veenendaalc , Franziska Schrodtb , Gustavo Saiza , Tomas F. Dominguesd , Gloria Djagbleteye , Andrew Fordf , Jeanette Kempg , Beatriz S. Marimonh , Ben Hur Marimon Juniorh , Eddie Lenzah , James A. Ratteri , Leandro Maracahipesh , Denise Sasakij , Bonaventure Sonkék , Louis Zapfackv , Hermann Taedoumgk , Daniel Villarroell , Michael Schwarzm , Carlos A. Quesadab,n,F.YokoIshidan , Gabriela B. Nardotoo,p, Kofi Affum-Baffoeq , Luzmilla Arroyol , David M.J.S. Bowmanr , Halidou Compaores , Kalu Daviesa , Adama Diallot , Nikolaos M. Fyllasb , Martin Gilpinb , Fidèle Hiens , Michelle Johnsonb , Timothy J. Killeenl,u, Daniel Metcalfef , Heloisa S. Mirandap , Mark Steiningeru , John Thomsona , Karle Sykorac , Eric Mouginw , Pierre Hiernauxw , Michael I. Birda , John Graced , Simon L. Lewisb,x, Oliver L. Phillipsb and Jon Lloyda,b* aSchool of Earth and Environmental Science, James Cook University, Cairns, Australia; bEarth and Biosphere Institute, School of Geography, University of Leeds, Leeds, UK; cCentre for Ecosystem Studies, University of Wageningen, Wageningen, the Netherlands; dSchool of Geosciences, University of Edinburgh, Edinburgh, UK; eForest Research Institute of Ghana, Kumasi, Ghana; fCSIRO Ecosystem Sciences Tropical Forest Research Centre, Atherton, Australia; gQueensland Herbarium, Townsville, Australia; hUniversidade do Estado de Mato Grosso, Nova Xavantina, Brazil; iRoyal Botanic Garden, Edinburgh, UK; jFundação Ecológica Cristalino, Alta Floresta, Brazil; kPlant Systematics and Ecology Lab, Department of Biology, University of Yaoundé I, Yaoundé, Cameroon; lMuseo Noel Kempff Mercado, Santa Cruz, Bolivia; mFieldwork Assistance, Jena, Germany; nInstituto Nacional de Pesquisas da Amazônia, Manaus, Brazil; oCentro de Energia Nuclear na Agricultura, São Paulo, Brazil; pUniversidade de Brasilia, District Federal, Brazil; qResource Management Support Centre, Forestry Commission of Ghana, Kumasi, Ghana; rSchool of Plant Science University of Tasmania, Hobart, Australia; sInstitut de l’Environnement et de Recherches Agricoles, Ouagadougou, Burkina Faso; tCentre National des Semences Forestières, Ouagadougou, Burkina Faso; uCenter for Applied Biodiversity Science, Conservation International, Washington DC, USA; vDeparment of Plant Biology, University of Yaounde I, Yaoundé, Cameroon; wGéosciences Environnement Toulouse, Toulouse, France; xDepartment of Geography, University College London, London, UK (Received 9 November 2011; final version received 26 December 2012)

Background: There is no generally agreed classification scheme for the many different vegetation formation types occurring in the tropics. This hinders cross-continental comparisons and causes confusion as words such as ‘forest’ and ‘savanna’ have different meanings to different people. Tropical vegetation formations are therefore usually imprecisely and/or ambiguously defined in modelling, remote sensing and ecological studies. Aims: To integrate observed variations in tropical vegetation structure and floristic composition into a single classification scheme. Methods: Using structural and floristic measurements made on three continents, discrete tropical vegetation groupings were defined on the basis of overstorey and understorey structure and species compositions by using clustering techniques. Results: Twelve structural groupings were identified based on height and canopy cover of the dominant upper stratum and the extent of lower-strata woody shrub cover and grass cover. Structural classifications did not, however, always agree with those based on floristic composition, especially for plots located in the forest–savanna transition zone. This duality is incorporated into a new tropical vegetation classification scheme. Conclusions: Both floristics and stand structure are important criteria for the meaningful delineation of tropical vegetation formations, especially in the forest/savanna transition zone. A new tropical vegetation classification scheme incorporating this information has been developed. Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 Keywords: canopy cover; cluster analysis forest; savanna; tropics; vegetation categorisation

1. Introduction Beard 1955; Fosberg 1961; Williams 1965; Eiten 1968, Tropical vegetation covers an area of over 5 × 107 km2, Ribeiro and Walter 1998) and even for vitriol (Maxwell ca. 42% of the Earth’s terrestrial vegetated surface 2004). Particular sticking points have been the reaching (Walter 1975). A wide range of different structural and of an agreement on what exactly constitutes a ‘savanna’ physiognomic forms exist across this region, ranging from (Dyksterhuis 1957; Eiten 1986; Ratnam et al. 2011) and evergreen forest, through seasonally dry deciduous for- the appropriate means by which to differentiate ‘forest’ and est, savanna and thorn scrub to grasslands and semi-desert ‘woodland’, or even to use the latter term at all (Lawesson vegetation (Schimper 1903). The appropriate means to 1994). classify and describe these diverse vegetation types have Here, vegetation science differs from soil science where long been a fertile ground for speculation and debate a universally agreed international classification system (Burtt-Davy 1938; Richards et al. 1940; Dansereau 1951; exists (the World Reference Base (WRB) IUSS 2004).

*Corresponding author. Email: [email protected]

© 2013 Botanical Society of Scotland and Taylor & Francis 102 M. Torello-Raventos et al.

It is thus not uncommon for soils classified according to of ‘implied causality’ – a problem already pointed out national schemes also to be given a WRB classification (e.g. by Hult (1881), who argued for significant defects in the McKenzie et al. 2004), thus allowing analyses of patterns ‘deductive school of thought’, where the physical environ- in the causes and consequences of wide-ranging varia- ment rather than the vegetation itself sometimes emerged as tions in soil properties across international boundaries (e.g. the main classification criterion (Nicolson 1996). This crit- Quesada et al. 2010, 2011). icism is certainly relevant to the approaches advocated by The current work was motivated through a require- the prominent British botanists of the early twentieth cen- ment generated by the ‘Tropical Biomes in Transition’ tury (Moss 1910; Tansley 1913) as well as to those of the (TROBIT) project (www.geog.leeds.ac.uk/TROBIT) con- USA (Nichols 1923), the concepts from which still unfor- cerned with identifying the principal determinants of trop- tunately permeate many ideas about how vegetation should ical vegetation structure and, in particular, the relative be classified today. distributions of forest versus savanna on a global scale. Numerous definitions of exactly what constitutes the In the course of sampling the structure of a range of tropical ‘tropics’ exist – including some guided by the nature of the vegetation types (Figure 1), it became clear that a uniform vegetation itself (Oliver 1979). Here, with the sites sampled terminology was required to describe and categorise the as part of this analysis ranging in latitude from 20.5◦ Sto various formations occurring on three continents. There is 15.4◦ N (with the most equatorial being 2.5◦ S) we sim- also a need to describe both in situ and in silico vegeta- ply take an unambiguous astronomical definition of ‘the tion formation types in an unambiguous manner to allow tropics’ being that region of the earth lying between the comparisons of model predictions with observations at a Tropics of Cancer and Capricorn, to describe the geograph- range of scales (regional, continental and global), espe- ical scope of the work undertaken here. Much of the work cially where the aim has been to simulate the structure and presented here also revolves around a comparison of the physiognomic characteristics of plant communities using relationships between stand floristic composition and veg- climatological and/or soils data (Fisher et al. 2010). etation structure within forest/savanna mosaics located in To that end, the principal objective of this paper was to Australia, Brazil, Bolivia, Cameroon and Ghana, and unfor- develop a pan-tropical vegetation classification and nomen- tunately we were not able to sample representative stands of clature scheme derived from analyses of a dataset of tropi- the many other vegetation types occupying this floristically cal vegetation structure and species composition. Data were diverse region. Nevertheless, as is explained in Section 6.5, collected with a uniform methodology for over 60 plots it is anticipated that the scheme developed here may have a across seven tropical countries on three continents, with a wider applicability to all tropical vegetation types and, with series of statistical analyses then used to help define a pan- some refinements, even be applicable outside the tropics. tropical vegetation classification and nomenclature scheme. We anticipate that not all readers will be familiar with With our choice of sampling localities specifically designed the different lowland vegetation types of the tropics and/or to elucidate any cross-continental convergence in vegeta- how these different vegetation formation types have gen- tion structural characteristics, we refer to the primary units erally been classified to date. Thus, prior to presenting we are seeking to define as ‘vegetation formation types’ our own analysis, current issues and controversies in tropi- (Nichols 1923). The concept of vegetation formations goes cal vegetation classification and nomenclature schemes are all the way back to Humboldt (1805) who applied this idea considered in some detail, with major schemes currently to group together assemblages of plants possessing similar employed to describe the main vegetation formation types physiognomic characteristics, such as a meadow or forest. on each continent briefly outlined. Some readers may thus Here we adopt the ‘ideal’ definition of Whittaker (1962) choose to skip the remainder of the Introduction and move viz. “a community-type defined by dominance of a given straight to Section 3. Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 growth form in the uppermost stratum (or the uppermost closed stratum) of the community, or by combinations of dominant growth forms”. 2. Background With one main purpose for the development of a pan-tropical rule-based non-hierarchical vegetation scheme 2.1. The ambiguous nature and definition of many being for the provision of a consistent classification frame- vegetation descriptors work facilitating inter-comparisons of studies interpreting As noted by Eiten (1992), there are many different ways and/or modelling how vegetation changes in response to in which many of the words used to describe vegetation variations in climate soils and/or climate, an a priori deci- can be applied. And even when applied in the same way, sion was made towards the start of this exercise to exclude definitions may vary substantially. For example, the word climate or soil type as part of the vegetation classification ‘forest’ can be applied to describe a woody vegetation for- itself. This is somewhat at odds with past practice, espe- mation type, usually (but not always) meeting a criterion cially amongst the ecologists of the United Kingdom and of minimum height and/or vegetative cover. By way of her (ex-)colonies where both climate and soil type have illustration, in its classification of Australian vegetation, long been considered important factors in any vegetation AUSLIG (1990) considers as ‘forest’ any vegetation for- description (Whittaker 1962). Nevertheless, as is explained mation type for which the dominant woody stratum has further in Section 2.4, that approach introduces the problem both a maximum height (H+) greater than 5 m and a foliar Delineation of tropical vegetation types 103

projected cover (ζ ; see Appendix 1 for a precise defini- this is their use when defining ‘biomes’ (defined here as in tion) of more than 0.3. Although theoretically independent Campbell and Reece 2002) “the world’s major communi- of the area, for some applied purposes a minimum area ties, classified according to the predominant vegetation and can be invoked. For example, FAO (2010) consider that as characterized by adaptations of organisms to that particular well as having to meet the joint criterion of H+ > 5m environment”). In order to avoid confusion with the termi- and a canopy crown projected cover (ς: defined as in nology above and consistent with correct English usage, we Appendix 1) of more than 0.1, a forest must also have suggest a capitalisation of these words when using in this a minimum area of 0.5 ha. As the upper canopy layer is large-scale geographical context (e.g. ‘Forest Biome’) and implicitly considered to be woody, both AUSLIG (1990) suggest this also be done even when colloquially abbre- and FAO (2010) constitute structural-physiognomic def- viated terms are used (e.g. ‘Cerrado’ when describing the initions where both form and structure are considered. ‘Cerrado Region’ of Brazil). More complex structural-physiognomic definitions have been proposed. For example, Eiten (1968) goes to some pains to provide an unambiguous delineation between what 2.2. Leaf habit and leaf size as additional classification constitutes a ‘shrub’ versus a ‘tree’, this enabling him to criteria define as ‘forest’ any vegetation formation type that is One of the most obvious delineators between the various ‘closed’ (ς>0.6) and for which ’trees’ constitute at least vegetation types can be their different foliage characteris- 60% of the total woody cover. Even more complex is the tics. This can be as simple as a differentiation between the structural-physiognomic-demographic definition suggested broadleaf versus needle leaf types (e.g. Küchler 1949) to by de Laubenfels (1975), which consists of a ‘forest’ being the distinction of three or more leaf size classes (Raunkiaer a woody stand with a continuous canopy which is open 1934; Webb 1959). underneath and with juvenile trees randomly distributed or Although in temperate regions leaf habit (deciduous concentrated in gaps. This is as opposed to ‘woodland’, vs. evergreen) and leaf life-span (τ ) generally correlate where the canopy is discontinuous and the juveniles avoid L closely, with evergreen foliage typically having τ > 1 year open areas. L and deciduous species with τ sometimes of only a few Even more confusing are the many definitions of L months (Chabot and Hicks 1982), no such congruence ‘savanna’. These can be vague and simple, as in the between leaf lifetimes and leaf habit exists for tropical for- structural-physiognomic definition of Lehmann et al. est species (Ackerly 1996). This is because many evergreen (2011): “The presence of a dominant C grass layer and a 4 species continually produce leaves, but with τ < 1 year discontinuous tree cover”, with the critical adjectives ‘dom- L (Reich et al. 2004; Williams et al. 2008) with an ever- inant’ and ‘discontinuous’ left undefined. At the other end green tropical tree or shrub canopy potentially maintained of the spectrum are the very precise, such as suggested across a wide range of leaf lifetimes (Kikuzawa 1991). The by Walker and Gillison (1982) for the woodland savan- distinction between ‘evergreen’ and ‘deciduous’ species nas of Australia as “a formation where single-stemmed therefore really exists only at the level of ‘leaf habit’, woody plants over 3 m tall occur in excess of 0.2% and i.e. whether or not the plant is typically leafless for some less than 90% crown cover and where there is a graminoid portion of the year (Richards 1996). component greater than 2% cover”. Both leaf habit and leaf size of the dominant strata A second way in which words such as ‘forest’ and have been used as prefixed qualifiers in vegetation clas- ‘savanna’ are often used is as a descriptor for the vegetation sification schemes. For example, Webb (1959) refers to formation type in which a particular taxon is usually found ‘semi-evergreen mesophyll (vine) forest’, and leaf habit (e.g. Hopkins 1966; Boland et al. 2006), with the com-

Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 has long proved a useful descriptive metric for both for- parison of ‘forest species’ versus ‘savanna species’ being est and savanna vegetation types (Eyre 1963; Whitmore a common investigative tool (e.g. Hoffmann et al. 2005a). 1984; Cochrane et al. 1985; Hoffmann et al. 2005b; Lloyd Such classifications, however, may still lack definition of et al. 2009). We therefore suggest this can be useful if what is actually meant by the words ‘forest’ and ‘savanna’. carefully applied at the stand level (i.e. when considering As well as being used as concrete nouns to define the total leaf area) in a way similar to that proposed by a particular vegetation formation type or as an adjec- Huber (1995) viz. evergreen (> 75% of trees are evergreen); tive describing the structural-physiognomic formation with semi-evergreen (between 50% and 75% of trees are ever- which a particular taxon is usually associated, words such green); semi-deciduous (between 25% and 50% of trees are as ‘savanna’, ‘woodland’ and ‘forest’ (or local variants evergreen); and deciduous (< 25% of trees are evergreen). thereof) are often less precisely employed in a more collo- quial context to describe a particular large-scale vegetation type or region. Examples from the tropics include the use of the word ‘cerrado’ which, as well as describing a partic- 2.3. Floristics as an additional classification criterion ular structural form of savanna (cerrado sensu stricto), is It is not uncommon for species names to be included also often applied (as cerrado sensu lato) to describe the in many vegetation descriptions. Indeed, this is less than entire region over which savannas are the dominant veg- surprising, as most inventories of tropical vegetation are etation type (Eiten 1968; de Laubenfels 1975). Similar to generally undertaken by field botanists more interested in 104 M. Torello-Raventos et al.

floristic composition than the vegetation structure per se. in any meaningful vegetation description (e.g. Williams Examples of such floristic-structural-physiognomic def- 1965; Blasco et al. 2000). These essentially are bioclimatic initions include Beard’s (1953) ‘Eschweilera - Licania descriptors of which Schimper’s ‘rain forest’ and ‘mon- rainforest association’ of Guyana or ‘Hirtella glandu- soonal forest’ are simple examples, as is the use of losa cerradão’ in the transition between Cerrado and descriptions, such as ‘montane forest’ to describe a dis- Amazon Forest in central Brazil (Ratter et al. 1973) tinct vegetation formation type at some high elevation (e.g. and the usual classifications of many African botanists, Grubb 1977) or ‘seasonally dry forest’ (e.g. Pennington for example: ‘Pterocarpus antunesii dry deciduous for- et al. 2004). In such cases, the clear implication is that est’ in Malawi (Hall-Martin 1975), ‘Anogeissus-Strychnos- one (or all) of the floristic, structural and/or foliar habit Combretum open savanna’ in Sudan (Keay 1949) or characteristics of the vegetation under investigation is a ‘Pennisetum/Andropogon long grasslands’ in the Serengeti consequence of climate. Soil characteristics have also been (Anderson and Talbot 1965). The inclusion of floristic used in some vegetation descriptions with causation sim- information in terms of a dominant and/or characteristic ilarly implied; for example, ‘hydromorphic grassland’ in species is not, however, possible for most forest and many southern Africa (Cronje et al. 2008). Similarly, Ratter et al. savanna formations in the tropics, due to their high species (1977) and Furley and Ratter (1988) discussed two forms diversity and the frequent lack of clearly dominant genera. of similarly structured woodland within Brazil (cerradão) As well as a stand’s dominant taxa potentially con- characterised by very different species compositions asso- tributing to its classification as a prefixed refiner, there ciated with fertile versus infertile soils. They thus labelled are also examples of floristic composition influencing the them as ‘mesotrophic cerradão’ and ‘dystrophic cerradão’, structural classification itself. For example, White (1983) respectively. In this case, the soil fertility prefix reflects an concluded that circumstances in the forest/savanna transi- implied causation in species differences. tion zone where the vegetation structure was ‘forest-like’ Although the reasons for a particular structural, floristic but with species normally associated with ‘woodlands’ that and/or physiognomic form existing where it does will often the term ‘forest’ was inappropriate because such stands dif- be correctly identified through the use of such climatic fer from what is usually considered ‘forest’ in all ways apart or edaphic modifiers, this moves the classification beyond from their height and canopy cover. He suggested instead the domain of vegetation description, and more towards the term ‘transitional woodland’. Although such a flexible a functional biogeographic perspective (i.e. to where the approach to the use of words such as woodland in White objective is to account for why a particular vegetation for- (1983) has been severely criticised by Lawesson (1994), we mation type is found where it is). Moreover, the climatic or see nothing wrong with that approach, as long as the mean- edaphic property invoked may not necessarily be the exclu- ing of the different vegetation groupings and the nature sive driver. For example, although the term ‘seasonally dry of their delineations remain clear. Indeed, White’s (1983) forest’ may correctly point out that many of the charac- scheme seems well constructed and Lawesson’s (1994) teristics of such forests are due to a prolonged dry season paper merely serves to demonstrate that it is much easier (stature, deciduousness and a unique species composition), to criticise a classification scheme than to construct one in South America at least, such forests are found growing (Richards et al. 1940). in a mosaic with evergreen savannas (cerrado) and with variations in soil depth and fertility or texture being the most likely reasons for the differences in vegetation for- 2.4. On the inclusion of site characteristics in vegetation mation type (Marimon-Junior and Haridasan 2005; Lloyd classifications et al. 2009). Indeed, as pointed out by Oliveira-Filho and There are essentially two viewpoints as to the extent to Ratter (2002), there are strong structural and floristic sim- Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 which descriptors of the environment in which a vegetation ilarities between what are termed ‘dry-deciduous forests’ formation type is found should be included in a vegetation on one hand and ‘mesotrophic cerradão’ on the other. classification. For example, in one of the first attempts to This provides an example of how, for two similar and generate a universally accepted vegetation scheme, Fosberg often intergrading vegetation types, different names can (1961) argued that although habitat information provided end up implying two very different reasons for why they important information on landscapes or even ‘ecosystems’, are actually found where they are. A failure to appreci- there was no place for this in the description of vegeta- ate such subtleties has the potential to lead to questionable tion structure per se; a position also taken by some other predictions of the nature of changes in vegetation distri- authors (Dansereau 1951; de Laubenfels 1975; Lawesson butions in response to climate change (e.g. Malhi et al. 1994). Nevertheless, Fosberg (1961) then went on to contra- 2009), and we suggest that although climatic or edaphic dict himself by defining numerous vegetation forms which contextualisations are potentially important, they should included some description of the habitat in which the veg- not be included as primary vegetation descriptors for this etation formation type was found, such as ‘desert forest’ or reason. ‘swamp scrub’. We also note that sometimes what seem to be On the other hand, as climate has a clear influence bioclimatic descriptors may actually contain floristic on vegetation structure (Schimper 1903), it has often and/or structural information. For example, in discussing been argued that environment should therefore be included lower montane forests in northern Thailand, Santisuk Delineation of tropical vegetation types 105

(1988) differentiates the bioclimatically defined ‘rain (and perhaps why) a vegetation formation type is located, forest’ from ‘oak forest’ or ‘oak-pine forest’ on the basis there seems little logic in incorporating them (or their of both structure and species composition. This is despite climatic and/or geographical components) in a primary these lower montane ‘rain forests’ having little in common vegetation classification descriptor. Their use also tends to in terms of species composition when compared with their under-emphasise the unique nature of contact areas where lowland ‘rain forest’ counterparts. two or more large morpho-climatic domains (‘extensive ecoregions’) meet. For example, in south-central Brazil the Cerrado and Amazon Forest Biomes interpenetrate each 2.5. Ecoregions, lifezones and biomes other, with ‘zones of tension’ or ‘zones of transition’ (ZOT) As a variant of the bioclimatic approach, Holdridge (1947) being formed. These ecotones occur on all tropical conti- suggested that there were unique relationships between nents and are uniquely spatially heterogeneous landscapes vegetation formation type and climate, with each being where processes such as speciation may occur at rates predictable from the other as long as mean annual precipi- appreciably faster than deeper into the individual biomes tation, temperature and potential evaporation were known. themselves (Smith et al. 1997). Together, these climatic variables were seen to give rise Whilst intrinsically placing less emphasis on region or to a range of ‘humidity provinces’ for any given temper- prevailing climate, biogeographers and evolutionary ecol- ature regime which were then used as prefixed modifiers. ogists often also take a ‘large-scale approach’ to vegeta- For example, ‘rain forest’ in Holdridge’s scheme exists tion classification, linking together phylogenetically related under a relatively moister climatic regime than ‘wet forest’, species into large-scale groupings which may, or may which is in turn under a more favourable regime than not, include climate as part of the biome descriptor (e.g. ‘moist forest’ and then ‘dry forest’ and so on. However, ‘Seasonally Dry Tropical Forests’). It is, however, clear in testing, this became an almost inverse method with the that many different vegetation formation types may occur ‘lifezone’ first defined from climatological measurements within any one ‘biome’ (Section 2.6), as is also evidenced and with the vegetation characteristics within that lifezone by a lack of congruence between point-level comparisons then categorised (Holdridge et al. 1971). Moreover, such of mapped vegetation formation types and herbarium spec- analyses also showed the importance of edaphic condi- imens (Särkinen et al. 2011). tions in modifying the vegetation structure within any one lifezone. Despite these considerations, this scheme, or vari- ants thereof, are still used by some workers as part of their classification criteria (e.g. Blasco et al. 2000), with 2.6. Current vegetation classification schemes one interesting variant being the ‘Sistema de Clasificación As described above, vegetation schemes vary widely in Bioclimática Mundial’ (Rivas-Martínez 1997) where pre- both the nature of their classification criteria and some- cipitation seasonality also forms part of the description times with structural characteristics not defined at all. through an ‘ombrothermal index’. For example, using This is because there is no universally accepted veg- the Rivas-Martínez (1997) classification scheme in their etation classification scheme used by terrestrial ecolo- analysis of the vegetation formation types of Bolivia gists as a whole, or for the tropical regions specifically. and Peru, Josse et al. (2007) refer to various forms of Though not, it would seem, for want of trying. For ‘tropical pluviseasonal humid forest’ as well as having example, as part of the International Biosphere Program ‘tropical pluvial’ and ‘tropical xeric’ vegetation formation of the 1960s a general vegetation classification scheme types. was proposed (Fosberg 1961), and under the auspices of Related to the lifezone concept is that of ‘ecoregions’ UNESCO a structural-physiognomic-habitat scheme has Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 where climate and vegetation characteristics are often also been created (Ellenberg and Mueller-Dombois 1966), mixed. For example, in their conservation assessment of both being reviewed in Mueller-Dombois and Ellenberg the Indo-Pacific Region, Wikramanayake et al. (2002) first (1974). A third attempt was made by Eiten (1968) who took generalised biome definitions to characterise the dis- used an approach first dividing vegetation forms on the tribution of dominant vegetation types and then applied basis of structure into 26 classes and then, on the basis climatic criteria to specific regions, arriving at terms like of growth forms (e.g. broadleaf versus conifer), embedding ‘Sri-Lanka dry-zone dry evergreen forests’. Likewise, Josse within these subdivisions a further differentiation based on et al. (2007) referred to vegetation mapping units such as periodicity of leaf production giving in total 216 structural- ‘Boliviano-Tucumano lower montane subhumid pluvisea- physiognomic units. Probably because these classification sonal forest’; this being an example of a geographical- schemes involved literally hundreds of different vegetation bioclimatic-physiognomic-structural classification. groups, many of which proved difficult to delineate eas- Whilst one can see the immense benefit of such hybrid ily in the field, none is widely used today, especially in the approaches for applied purposes, such as determining con- tropical regions, with workers on individual continents con- servation area priorities and ‘biodiversity hotspots’, as for tinuing to use more or less independent schemes which are the general bioclimatic approach discussed in Section 2.4, considered much more convenient and practical. These pro- although lifezones or ecoregion categorisations can in some vide an essential background to the approach taken later in cases provide useful contextual information about where this paper and are thus here briefly described. 106 M. Torello-Raventos et al.

(a) (b) Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013

(c) (d)

Figure 1. Vegetation formation types of (a) Australia, (b) Africa, (c) South America and (d) South-East Asia. These are expressed as structurograms with the vertical axis representing the upper stratum canopy height and the horizontal axis a range of woody canopy cover metrics (see Appendix I). Also shown for (a)–(c) are the locations of sites sampled as part of this study. Delineation of tropical vegetation types 107

2.6.1. Australia. The Australian National Vegetation diverse; for example, in the Cerrado Region of Brazil Information System (AUSLIG 1990) defines a hierarchi- (Ratter et al. 2003). cal classification system for describing the structural and floristic patterns of groups of plants. Designed primarily for 2.6.2. Africa. Although several national or regional clas- mapping purposes, it is a structural-physiognomic-floristic sifications have long existed (e.g. Gossweiler 1939; classification with forest defined in terms of dominant Harrison and Jackson 1958; Acocks 1988; Low and Rebelo physiognomic form (trees being taken, as in many other 1998), an agreed vegetation classification and mapping schemes, as any woody plant > 5 m in height) with the dif- convention for the African continent was developed under ferentiation of woodland versus forest made on the basis of the auspices of UNESCO (White 1983). At its core this a critical foliar projective cover (ζ ) of 0.30 and with height is a structural differentiation between ‘grassland’, ‘wooded (H) differences not important in differentiating these two grassland’, ‘woodland’ and ‘forest’, these being made pri- forms. Scrubland is defined as woody vegetation < 5m marily on the basis of canopy cover (ς), with the already in height, and within that height domain there is also a mentioned additional proviso that a vegetation form with special category of ‘heath’ (H < 2.0 m; ζ>0.30). The the structure of a forest, but with a species composition nor- general structural component of the scheme is illustrated mally associated with woodlands is termed a ‘transitional in Figure 1(a), which also shows how words like ‘open’ woodland’ (Figure 1(b)). and ‘closed’ and ‘high’ and ‘low’ are used to define the Partially overlapping this classification along a con- domains. As well as including ζ on the lower axis of tinuum of increased arboreal cover and height occurring this structurogram, we have also included the equivalent with an accompanying decrease in herbaceous cover is the values of ς as suggested by Gillison (1994), this being separately considered ‘bushland’ and thicket’, for which the proportion of the ground area covered by the verti- the minimum ς is quoted as 0.4. Thicket canopy cover, cal projection of tree or shrub crowns (Appendix 1). Also although undefined, is obviously greater than ‘bushland’, as shown on the diagram are the equivalent values for the pro- thicket is described as “forming an impenetrable commu- jected canopy area per unit ground area of woody species nity except along tracks made by animals” (White 1983). (canopy area index; CW) based on the equations presented Affiliated with this group are ‘scrub forests’, where an open in the Appendix 1. In this classification, the dominant canopy of trees occurs above the scrub layer. The White (almost always upper) stratum forms the basis of the (1983) classification also makes allowances for ‘scrub categorisation. woodland’, this including areas occupied by dwarf forms of The AUSLIG classification uses the growth form and trees such as Colophospermum mopane (Veenendaal et al. foliage cover classification as summarised in Figure 1(a) as 2008) as well as a transitional vegetation sometimes occur- the base for a more detailed vegetation classification with ring between woodlands and waterlogged depressions and floristic codes, including understorey genera as appropri- with all woody vegetation < 2 m classified as ‘scrubland’, ate. For example, the base structural designation of ‘L1’ the most notable areas of which occur in the extra-tropical means a dominant stratum of low trees (H < 10 m) with Fynbos region of South Africa. All the bushland, thicket a ζ W < 0.10 (i.e. ‘low open woodland’) and an expanded and scrubland forms are considered to have ‘subordinate code such as ‘wL1kZ’ translates as low Acacia trees with ground cover’. a foliage cover less than 0.1 and an understorey of low Although clearly an impressive achievement (appar- shrubs (saltbush). Here the ‘w’ represents Acacia, ‘Z’ indi- ently taking some 15 years to compile) the vegetation cates the presence of low shrubs in the understorey and classification approach of White (1983) has not been with- ‘k’ indicates that these shrubs are members of the family out its critics (e.g. Lawesson 1994), especially as that author Chenopodiaceae. sometimes seems to break his own rules. For example, Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 The AUSLIG system has the advantage of automati- White (1983) refers occasionally to ‘short forests’ which cally covering all possible height/canopy cover combina- can be as low as 8 m in height and which therefore, should, tions, with the area below the solid black line in Figure 1(a) according to his own definitions, be classified as woodland. indicating the canopy/cover height combinations naturally Likewise, reference is made to forests as having ‘open occurring in the Australian tropics (Adam 1992; Fox et al. canopies’, even though the minimum ς for forest is quoted 2001). The coding system also allows for a good degree as 0.7, for which all tree canopies must be close to touch- of structural and floristic information to be conveyed pre- ing (Appendix 1). White (1983) also used climatic terms cisely. The use of foliar projective cover in the Australian occasionally, arguing that these represented a “convenient system has, however, its own unique legacy (Specht 1972), nomenclatural shorthand for important physiognomic and and one therefore wonders whether it is really necessary floristic differences . . . impossible to designate concisely to use it as an alternative to the easier-to-measure canopy in purely physiognomic terms”. This is fine, of course, just cover metrics (Appendix 1) as is generally applied else- as long as one knows how to translate the shorthand. where. The coding system works well for the Australian Using a chorological system based on geographic dis- situation where a few genera tend to dominate the land- tribution patterns, White (1983) then went on to provide an scape, except for the relatively minor tropical forest zone. array of different major vegetation categories, ranging from But it might not work so well in other systems where the climate-regional-physiognomic-chorological-structural the species composition of stands tends to be much more (e.g. ‘Drier peripheral semi-evergreen Guineo-Congolian 108 M. Torello-Raventos et al.

rain forest’) to the simply structural (e.g. ‘wooded between the dry forest and cerrado vegetation types is grassland’). Often subdivisions employing floristic infor- that the latter typically contain many evergreen woody mation were also introduced (e.g. ‘Sudanian Isoberlina species (Eiten 1972; Franco 2002; Hoffmann et al. 2005b). woodland’), and sometimes even edaphic factors are Nevertheless, even though the upper stratum of dry forest invoked (e.g. ‘Sudanian wooded grassland on Pleistocene may be almost entirely deciduous (Lloyd et al. 2009), at clays’). least in some cases there may be a substantial evergreen In summary, White’s Vegetation of Africa does not rep- understorey present (Gentry 1995; Killeen et al. 1998; resent a classification system at all, but rather constitutes a Cochrane and Cochrane 2010). Although savanna areas series of comprehensive vegetation descriptions for which occurring within the main Amazon forest and in north- a consistency of syntax was not the main priority, and for ern South America are floristically distinct from those of which a rigorous and unambiguous categorisation scheme the (southern) Cerrado Region (Eiten 1983, Ratter et al., was also not required. We also note that when the author did 2003, 2006, 2011), the general structure of Figure 1(c) still refer to forest formations whose structural characteristics holds. are at odds with his own classification (e.g. ‘Guineo- We also note that within the main Amazon Basin, Congolian short forest’) it is because, although these forest various forest classifications have been proposed. For formations would normally be referred to as woodlands example, Daly and Mitchell (2000) differentiated at one according to the main scheme (Figure 1(b)), they also con- end of the spectrum ‘dense forest’ (mata pesada) with a 2 −1 sist of species normally found in forests, having a stunted basal area (AB)of> 40 m ha and ‘open forest’ with 2 −1 2 −1 stature attributable, for example, to their growing on shal- 18 m ha < AB < 24 m ha at the other. Others, low and/or rocky soils (Schnell 1952; Richards 1957). This such as Huber (1995) and Cochrane et al. (1985) pri- suggests the need for having a dual classification system marily invoked some measure of stand-level foliar habit, considering not only the structure of the vegetation, but also the latter for example referring to ‘tropical semi-evergreen the phytochoria present. seasonal forest’ as opposed to ‘tropical evergreen forest’. As detailed by Veloso et al. (1991) the Brazilian classifica- 2.6.3. South America. The UNESCO publication Vege- tion includes dense, open or mixed ‘ombrophilous forests’ tation of South America (UNESCO 1981) differs sub- and deciduous or semi-deciduous seasonal forests; the term stantially from its African counterpart (White 1983) in ‘ombrophilous’ implying adapted to, or characteristic of, that vegetation structure and chorology play a background high rainfall (Wiesner 1894; Holm 1896), this then essen- role, with the primary classifications based on climatic tially being a classification related to both structure and associations. climate. We therefore provide in Figure 1(c) our own structur- Although sometimes considered to be part of the ogram of tropical South American vegetation types based Cerrado Biome (Cole 1960), the vegetation of the semi- on our own experiences, as well as descriptions as available arid region of north-eastern Brazil commonly known as in Goodland and Pollard (1973), Eiten (1972, 1983, 1986), caatinga is usually considered as a discrete vegetation Ratter et al. (1973), Daly and Mitchell (2000), Oliveira- grouping distinct from both savanna and forest (Andrade- Filho and Ratter (2002) and Cochrane and Cochrane Lima 1981; Eiten 1983; Daly and Mitchell 2000). This (2010). Firstly, as for Africa, from the bottom left to the vegetation formation type encompasses a range of forms top right of the diagram can be a savanna/forest continuum with the common elements being trees and/or shrubs that (varying shades of green to brown) of increasing domi- are often spiny and mostly deciduous. The most common nance of woody vegetation. This starts with the ‘clean’ form is a closed canopy thornscrub 3–6 m in height, though or open grassland ‘campo limpo’ (grassland), where tree arboreal forms may have canopies as high as 10 m and with Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 cover is virtually absent (not shown), the ‘campo sujo’ for- a canopy crown coverage as low as 10%. Generally, grasses mations; the ‘campo cerrado’ formations; then ‘cerrado’ are subordinate, although Eiten (1983) does note the exis- (sensu stricto) with the tallest and most closed canopy tence of a short-grass savanna form of caatinga dominated stage being ‘cerradão’ (Oliveira-Filho and Ratter 2002). by deciduous shrubs (seridó). It should also be noted that Although cerradão is considered by some to be a climax the use of the word ‘caatinga’ to describe a vegetation for- form of forest within the Cerrado Region (Coutinho 1978), mation type is not consistent across South America. For some evidence points to this, at least in some cases, as also example, in Venezuela it is usually taken to describe a sometimes being a transitional vegetation type reflecting a low-stature forest usually found on poorly drained sandy specific successional stage in the continual advancement of soils and often with sclerophyllous leaves (e.g. Coomes and forest into cerrado, such as has been observed on the south- Grubb 1996). ern edge of Amazonia (Ratter 1992). A transition is then Not included in Figure 1(c) is the Chaco formation, a made to the foliar habit characteristic of semi-deciduous grouping of trees and shrubs found in the transition between and dry deciduous forests, although with some structural the tropical and temperate zones to the south, principally overlap with cerradão, but which represent a fundamentally in Paraguay and Argentina (Bucher 1982). This generally different phytochoria to nearby (non-mesotrophic) savan- xerophyllous vegetation formation type includes a wide nas (Daly and Mitchell 2000; Oliveira-Filho and Ratter range of physiognomic formations similar to that found for 2002; Killeen et al. 2006). Another important difference Caatinga (Daly and Mitchell 2000). Delineation of tropical vegetation types 109

2.6.4. Asia. The classification of forest types across as 2 m, with the vegetation structure ranging from rela- Asia stems mostly from the work of Champion (1936) tively open with isolated and/or clumped groupings of trees and Champion and Seth (1968) in India who, utilis- through to a dense closed canopy. The herbaceous ground ing Schimper’s (1903) bioclimatic approach, first divided layer, when present, is described as much less abundant lowland forests into ‘moist tropical’ and ‘dry tropical’ and than the ‘savanna forests’ described above, with arboreal with a further four montane/alpine classifications. Within grasses such as bamboo sometimes dominant (Stamp and these bioclimatic classes, several subgroups are recognised, Lord 1923; Vidal 1960a,b; Hundley 1961; Williams 1965; some of which are based on leaf habit differences (e.g. ‘dry Champion and Seth 1968). tropical deciduous forest’), some using other physiognomic Based on numerous vegetation descriptions and characteristics (e.g. ‘thorn forest’) and others reflecting spe- photographs (Stamp and Lord 1923; Hundley 1961; cific habitat characteristics, such as ‘swamp forest’ and Bunyavejchewin 1983; Whitmore 1984; Stott 1992; kerangas forest; the latter also termed ‘heath forests’ and MacKinnon et al. 1996; Laumonier 1997; Monk et al. 1997; typically being found on coarsely textured soils derived Koy et al. 2005) structurograms incorporating the above from siliceous parent materials (Whitmore 1984), and with features are shown in Figure 1(d), again with heights and such classifications having been applied, with minor modi- canopy cover estimates necessarily approximate. As for fications, in Burma (Hundley 1961), Cambodia (Tani et al. the other continents, this shows a broad continuum along 2007), Laos (Vidal 1960a,b), Indonesia (MacKinnon et al. the main diagonal stretching from grasslands through to 1996; Whitten et al. 1997; 2001) and Thailand (Santisuk tall evergreen forest. As for scrub in Australia, bushland 1988) as well as in regional analyses (e.g. Whitmore 1984). and/or thicket in Africa and caatinga and/or chaco in South The Champion and Seth (1968) scheme also allows for America, we also find in the bottom right-hand corner a typ- further subdivisions. For example, Santisuk (1988) intro- ically short but often dense vegetation type (usually with duced community floristics into his classification of the few if any grasses), this sort of vegetation variously being vegetation of northern Thailand, referring to three decidu- referred to as thorn shrub, thorn forest or scrub forest. ous forest types: tropical mixed deciduous forest, deciduous dipterocarp forest and pine-deciduous dipterocarp forest. The latter two formations are particularly interesting, as 2.7. Cross-continental comparisons and global datasets they are also sometimes referred to as ‘savanna forest’, this From a comparison of Figures 1(a)–(d), there are, not term being as originally proposed for a park-like (i.e. open surprisingly, certain common features to be seen in the canopy) woody xerophyllous vegetation formation type less structurograms. For example, there is a common increase than 20 m high and with a substantial grass understorey in maximum woody plant height as the various mea- by Schimper (1903). Its origin is apparently in the work sures of canopy cover increase from grasslands, through of Kurz (1875), who used it to describe some low for- savanna formations of increasing woodiness to forests. est formations in Burma (Stott 1992). ‘Savanna forests’ There is also a common vegetation formation type found characterised by ‘flat-topped crowns of spreading Acacia towards the lower right corner of the structurograms, this trees’ are also referred to by Whitten et al. (1997) for West being variously referred to as caatinga in Brazil, scrub Bali, while Hundley (1961) describes ‘savanna forests’ in forest or thorn scrub in Asia, bushland, thicket or scrub Burma dominated by species usually characteristic of ‘dry woodland/shrubland in Africa and low forest or scrub deciduous forests’. All these vegetation types seem of a (either open or closed) in Australia. Whilst such compar- similar structure to the deciduous dipterocarp forests of isons indicate at least broad similarities in structure, they mainland South-East Asia in being of a relatively open also point to the wide differences in terminology generally canopy and characterised by a grassy understorey with fre- employed for the different continents. For example, for an Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 quent fires. On the basis of the fire-tolerance characteristics 18 m-high stand with a canopy area index of 0.70, the most of the species present in such forests, Ratnam et al. (2011) likely terms applied would be ‘woodland’ in Australia and have argued that they should really be classified as savan- Africa, cerradão or ‘forest’ for South America (depend- nas. A similar argument could be made for the fire-tolerant ing on species composition), but for South-East Asia, this Eucalyptus urophylla S.T. Blake formations in Timor that would most likely be referred to as some form of forest. have previously been categorised as evergreen forest (Monk There is also considerable overlap. For example, at a height et al. 1997), as these generally have a distinct grass layer of 12 m and with a canopy area index of 1.0, in Africa (Martin and Cossalter 1976) and with apparently similar (and depending mostly on species composition) a vegeta- vegetation formation types considered to be ‘savanna’ in tion formation type could be any of woodland, transitional New Guinea (Paijmans 1975). The term ‘savanna’ is usu- woodland, scrub forest or forest. For South America, this ally only applied in South-East Asia to grasslands with a could be any of caatinga, cerradão or (semi)-deciduous for- woody component limited to isolated and/or scattered trees est. One other feature that emerges from the comparison of (Monk et al. 1997; Corlett 2009). Figure 1 is that the dry dipterocarp and similar ‘savanna Also occurring throughout mainland Asia are areas of forests’ of South-East Asia are of a structure more usually what is often classified as scrub forest or thorn scrub. considered woodland or savanna formations on the other These vegetation formation types are usually less than continents (cerrado sensu stricto and cerradão in South 6 m tall and with the upper canopy sometimes as low America). 110 M. Torello-Raventos et al.

Given the above complexities and often imprecise char- (3) use this information to develop a new pan-tropical acterisations of the structural and/or floristic characteristics vegetation classification scheme. in the initial vegetation descriptions in the first place, it is therefore not surprising that attempts to date to develop Although an emphasis is placed on savanna/forest global maps of vegetation characteristics (Matthews 1983; ecotones where precise definitions are to be most likely Olson et al. 1983; Wilson and Henderson-Sellers 1985) required, suggestions are also made as to how the classifica- have been shown to have only a low degree of concor- tion scheme may be extended to cover vegetation formation dance in terms of the vegetation predicted at any one types of the tropics and subtropics and even with a possible location (DeFries and Townshend 1994a; Leemans et al. extension to the more temperate zones (Section 6.5). 1997) and with a range of remotely sensed vegetation classifications now advocated for global or continental- scale studies (e.g. DeFries and Townshend 1994b; Hansen et al. 2000; Loveland et al. 2000; Eva et al. 2002), 4. Materials and methods including simulations of surface fluxes of carbon dioxide, With the overarching objective of determining the princi- water and heat (Sellers et al. 1996). Some attempt at pal controls on both where ZOTs occur and how vegetation unification of vegetation classes has been made through types are distributed within individual ZOTs, sampling the International Geosphere Biosphere Program (IGBP) areas in Australia, Africa and South America were selected (Loveland and Belward 1997), but still definitions differ, with a view to sampling at least one ZOT on each of especially for tropical regions. For example, the IGBP the major continents containing appreciable areas of both methodology defines forest as having a height of > 2m savanna and forest (Figure 2). Our sampling design was and a tree canopy cover (ζ T) of more the 0.6, but for to maximise differences in climate and soils, with the their map of South America, Eva et al. (2002) stipulate a intention that our sampling would lead to globally applica- height of > 5 m and ‘open forests’ starting at a canopy ble results. With a logistically feasible area first identified cover of 0.3. In terms of savanna-type systems generally (based on collaborators, politics and vegetation) a general similar approaches are taken, but with very different values. region of interest was then pinpointed, this usually being For example, the University of Maryland scheme (Hansen an area where both savanna and forest were known to et al. 2000) defines ‘wooded grasslands’ as stands of a occur in close proximity, but also in some cases (for exam- height > 5 m and with 0.4 >ζT > 0.1. On the other ple West Africa and Bolivia) to also allow us to generate hand, the IGBP system defines ‘woody savannas’ as hav- a transect to examine precipitation effects on vegetation ing a height > 2 m and with 0.6 >ζT > 0.3; both structure. In both these cases we designed the transect to with an herbaceous or other understorey. Taking this with extend from (semi-) evergreen forest formations to the dri- the ‘forest’ criteria of Eva et al. (2002) above, this then est savanna-type vegetation practicable. For West Africa, means that, assuming the proportion of light intercepted this was an area of ‘Sahelian savanna’ about 100 km south on average by the tree crowns to be 0.6, then the stand of the Sahara Desert in Mali (Hiernaux et al. 2009; Mougin shown in Figure 6(g)) could variously be described as ‘open et al. 2009) with three roughly equidistant sample clusters forest’, ‘woody savanna’ or ‘wooded grassland’ (see also between those plots and the main ZOT cluster in Ghana. For Figure 1(a)). Bolivia, the driest areas were taken to represent the south- ernmost limit of a dry forest/savanna mosaic just north of where the Chaco vegetation formation type begins to 3. Objectives dominate in central Bolivia, extending about 400 km north From Section 2 it is clear that, being more structural into a well-studied ZOT located near the Bolivian–Brazilian Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 than floristic, vegetation classification criteria used by border (Killeen et al. 2001). global remote sensors and/or modellers differ from those Within each region (or transect location) plots were generally employed by vegetation scientists. Yet more selected with an aim to encompass the diversity of vege- importantly, substantial contradictions in definitions exist tation found in the general area, with an additional criterion between various workers even within the one field and being long-term ‘security’ from future intensive anthro- with there being little, if any, consistency in definitions of pogenic modifications as far as practical. Thus many plots vegetation type. The study here therefore attempted to: were located in areas that were subjected to a varying degree of official protection. Many had already been estab- (1) address this problem by using numerical techniques lished by local partners with a view to monitoring long- to assess the differences and similarities between term changes in vegetation structure and function. Some the vegetation formation type observed on three of plots were specifically chosen as apparent ‘outliers’ where the four continents of interest; the vegetation formation type observed was not that nor- (2) assess the relationships between the predominantly mally expected on the basis of climate. Examples of this structural approach taken by those interested in were open woodlands at relatively high rainfall such as large-scale patterns in vegetation structure and in Australia (EKP-01) and Brazil (ALC-01 and ALC-02), function and the actual floristics of the same veg- and vegetation dominated by tropical forest species but in etation formation types; and low-rainfall areas in Australia (FMS-02 and RSC-01). Delineation of tropical vegetation types 111

Figure 2. Location of the sampling sites in Africa, Australia and South America and their index of soil water deficit (-W)asdefinedby Equation 1.

Measurements were made in five field campaigns, each where PA is mean annual precipitation, Qs is mean annual over a period of ca. 2 months from July 2006 to March global solar radiation, ρ is the density of liquid water and λ 2009, with as many plots as possible sampled within the is the latent heat of evaporation for H2O. W typically ranges allocated time and are summarised as follows (see also from −0.5to0myr−1 for moist forests, extending (in our Figure 2): West Africa (Ghana, Burkina Faso and Mali): dataset) to nearly −3.1 m yr−1 for arid (Sahel) savanna 14 plots; August to October 2006 across a rainfall gra- just south of Timbuktu. Within our sampled ZOT it ranges −1 −1 dient with mean annual precipitation (PA) varying from from around −1.0 m yr (South America) to −1.3 m yr 0.35 m yr−1 at Hombori to 1.2 m yr−1 at Asukese in Ghana; (Africa). Bolivia (February to May 2007): 11 plots across a rainfall gradient from 0.82 m yr−1 at Tucavaca to 1.45 m yr−1 at Los Fierros (Noel Kempff Mercado National Park); Cameroon 4.1. Plot selection (10 plots; November to December 2007 with all plots in Of the 63 permanent sample plots established, 55 were of close proximity and with a similar precipitation regime an area of 1 ha, but due to time constraints towards the end −1 −1 1.59 m yr < PA < 1.62 m yr in Mbam Djerem National of two field campaigns, eight plots were sampled for areas Park); Brazil (17 plots; April to June 2008 with a range of of only between 0.4–0.6 ha (Supplementary Information −1 −1 PA from 1.51 m yr at Nova Xavantina to 2.35 m yr A). In terms of natural and human-mediated disturbances, at Alta Floresta); and Australia (11 plots; February to some sites were fire-protected and with domestic animal April 2009 across a rainfall gradient from 0.67 m yr−1 grazing specifically excluded, but some others, especially at Richton Scrub to 3.5 m yr−1 at Cape Tribulation). All in West Africa, farm animals were observed grazing in or sampling campaigns had been timed to coincide with the nearby the sample plots. For all plots, there were no bar- Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 end of the wet season and the timing of expected maxi- riers placed to the grazing of vegetation by the natural mum plant physiological activity and standing herbaceous fauna. Table A1 gives details of plot histories (in terms biomass. As is detailed below, for each plot (typically of of those previously or newly established), plot protection 1 ha size) a series of stratified measurements of vegeta- status and perceived anthropogenic influences (grazing and tion cover was made using both total inventory and transect fire protection or promotion). approaches. Detailed measurements of soil and plant phys- iological characteristics were also made in the same plots 4.2. Plot establishment at the same time (e.g. Domingues et al. 2010; Saiz et al. × 2012). Most plots were 100 m 100 m and established according The locations of the sampled plots are shown in to standard protocols (Phillips et al. 2010), with naviga- × Figure 2, which also shows an index of plant water supply tion within each plot via 25 numbered quadrats of 20 m in relation to evaporative demand, W (Berry and Roderick 20 m, irrespective of vegetation formation type. The plots 2002) estimated as follows: for which only smaller areas could be sampled due to time constraints (Table A1), were BFI-01 to BFI-04 in Ghana (0.5 ha), MDJ-09 and MDJ-10 in Cameroon (0.4 ha), W = PA − Qs/(ρλ)(1)VCR-01 in Brazil (0.6 ha) and OTT-04 in Bolivia (0.5 ha). 112 M. Torello-Raventos et al.

4.3. Woody vegetation structure outlined in Walker and Hopkins (1990). The herbaceous A stratified sampling approach was used, separating the layer was defined to include all monocots and herbaceous woody cover into three layers: dicotyledonous plants occurring along the 10 transects within each plot, with our estimates separated into ϕ3 (herbaceous species with the C photosynthetic pathway) (1) Trees and shrubs with a diameter at breast height 3 and ϕ (predominately grass species with the C photo- (1.3 m), d,of> 10 cm. For this layer, henceforth 4 4 synthetic pathway) usually being the mean cover estimates we use the symbol U (for upper). averaged across 110 × 1.0 m2 quadrats per plot. (2) Trees with d ranging from 2.5–10 cm. These are Total projected cover of both shrubs and tree denoted M (for mid-stratum). seedlings < 1.5 m high was also recorded at this time in (3) All other trees and shrubs. This layer includes each quadrat sampled, these estimates contributing to C . small trees and shrubs taller than 1.5 m but with S d < 25 mm as well as all woody species less than 1.5 m tall (i.e. treelets, small shrubs and seedlings) and is denoted as S (for subordinate). 4.5. Estimation of woody species stand-level crown areas To estimate CU we used site-specific allometric equations For U, measurements of d were made for all stems developed on the basis of data from the 50 or more concur- within each sample plot (irrespective of area) with H rent determinations of d and crown area with a simple linear also being determined for a subset of (usually more than ordinary least squares (OLS) regression found suitable for 2 > 50) trees/shrubs within each plot, stratified according all sites and always giving an r 0.6 (Supplementary to d using an ultrasonic-laser hypsometer (Vertex Laser Information B). For each site, the resulting equation was > VL400 III, Haglöf Sweden) as described in Feldpausch then applied to all trees with d 0.1 m within each plot, et al. (2011). For the same subset used for H determina- the sum of which (after dividing by the plot area) was taken tions, crown areas were also estimated using the average to be CU. crown radii measured from the centre of the bole to the four For CM, we had also measured crown diameters of / > cardinal points at the distance furthest from the trunk. trees shrubs 25 mm along zigzag lines across transects For the characterisation of M and S we employed a tran- as described for the calculation of crown separation ratio sect approach based on Gentry (1988), this being referred (Penridge and Walker 1988), deriving site-specific equa- to here as the ‘Gentry method’, with the transect layout tions relating variations in crown area to d utilising all mea- ≥ ≥ shown in Appendix 2. In brief, we established 10 transects sured trees and shrubs with 25 mm d 0.1 m and then of 50 m length in each 1 ha plot where we measured d for all scaling up to the whole-plot level as for CU above. In some woody species with d > 25 mm and projected crown areas plots, however, this measure could not be executed due to determined from measurements of their average crown radii time pressure. We applied the site-specific equations of CU taken from the centre of the stem to the four cardinal points or, when possible, from a nearby similarly structured plot. at the distance furthest from the stem. For woody species For CS, we directly calculated the plot-level value by with d < 25 mm but H > 1.5 m (contributing to the subor- simple multiplication of the total transect area covered by sampled ± 1 m along the (normally) 10 50 m transects in dinate stratum canopy area index, CS) the canopy projected area was estimated based on visual estimates of the crown relation to the total plot area. diameters along two perpendicular axes. In all cases we identified all individuals to species level where possible (see also Section 4.6) and divided them into 4.6. Species identifications and photosynthetic pathways

Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 life forms: trees and shrubs. Shrubs we took to be woody Woody and herbaceous species were usually identified in species with either a single stem (bole) of length at least the field by local botanists, but where necessary specimens 1.5 m, but with height less than 3 m, or a woody species were collected and verified against herbarium collections. with a stem length prior to branching of less than 1.5 m, Herbaceous species were subdivided according to photo- also being < 5 m height. All woody species of ≥ 5 m height synthetic pathways (C4 versus C3) from reference literature were thus considered trees, allowing for the large woody for both C3 and C4 Poaceae (e.g. Watson and Dallwitz species of drier savannas such as Acacia tortilis subsp. rad- 1992 onwards), Cyperaceae (e.g. Wilson and Bruhl 2007) diana (Savi) Brenen to be considered as such. Species with and dicotyledonous herbs (e.g. Johnson and Hatch 1968), a single stem of 1.5 m or more above the soil surface and although these determinations were not always possible for with a height > 3 m were thus also considered trees. These all sites. criteria are similar to those proposed by Eiten (1968).

4.7. Cluster analysis of sites 4.4. Characterisation of lower layer 4.7.1. Structural characteristics. Taking as input vari- Along each Gentry-type transect the fractional foliage ables the mean and 0.95 quantile of measured canopy cover (ϕ) of the herbaceous ground layer was visually heights for all trees and shrubs with d > 0.1 m (i.e. in 2 recorded every 5 m ina1m plot, using the procedures the U stratum) along with the plot-level estimates of CU, Delineation of tropical vegetation types 113

Table 1. Details of vegetation structure measurements used as input into the clustering analysis. Note that for CM trees and shrub cover were treated as separate input variables for the clustering analysis with CS trees and shrubs further divided according to plant height giving four subordinate subgroups.

Means of Measurement

∗ Upper stratum canopy height, HU (m) From height measurements on usually more than 50 trees within each   > Average upper stratum height, H U (m) plot with d 0.1 m an allometric equation was derived relating H to D and with 0.95 quantile and mean values for each plot then calculated > ∗   for all trees with d 0.1 m to yield HU and H U respectively. 2 −2 Canopy Area Index, CW (m m ) Sum of CU, CM and CS as defined below (including both tree and shrubs, but excluding seedlings) 2 −2 Upper-stratum tree crown area, CU (m m ) Stand-level canopy projected area estimated based on the site-specific allometric equations presented in Supplementary Information B (Section 4.5) for all trees within the plot with d > 0.1 individually measured. 2 −2 Mid-stratum crown area, CM (m m ) Stand-level projected area estimated based on the site-specific allometric equations presented in Supplementary Information B and according to the life form classification (Section 4.3), separated for trees/shrubs for all woody vegetation of 100 mm > d > 25 mm. with measurements made along (usually) ten 50 m-long transects (Section 4.3 and Appendix 2) 2 −2 Subordinate-stratum crown area, CS (m m ) Projected crown area, estimated separately for trees and shrubs with d < 25 mm and H > 1.5 m. This stratum also includes seedling projected crown areas, again separated for trees and shrubs, estimated visually for H < 1.5 m in quadrants along (usually) ten 50 m-long transects (Section 4.3, 4.4 and Appendix 2). Fractional cover of C3 and C4 herbaceous plants From visual estimates of foliar interception made in 1 × 1 m quadrants 2 −2 ϕ3 and ϕ4 (m m ) every five metres along (usually) ten 50 m long transects. For each quadrant, separate estimates were made for C3 versus C4 herbaceous plants (Section 4.4 and Appendix 2).

CM, CS (with the latter divided into subcomponents of previously sampled in Ghana (ASN-02 and ASN-04: see trees and shrubs and with seedling as treated separately), Lewis et al. 2009) as well as several forest sites from ϕ3 and ϕ4 (Table 1) we applied the clustering method Bolivia (CHO-01, CRP-01, CRP-02, HCC-01, HCC-02: of Complete Linkage (McQuitty 1960; Sokal and Sneath see Honorio Coronado et al. 2009). One forest in Ghana 1963; MacNaughton-Smith 1965) using the routine daisy (BFI-03) was excluded from this analysis due to strong available within the package cluster (Maechler et al. 2005) anthropogenic influences on species composition being as deployed in the R statistical environment using as input apparent (e.g. presence of a Mangifera indica L. tree) as a dissimilarity matrix estimated as described by Gower well as two savannas in Brazil (ALC-01 and ALC-02) for (1971). This component of the analysis was limited to which species determinations were not of sufficient cover- those 39 plots for which a full dataset had been collected age or fidelity. The fuzzy clustering approach used allows (including identification of grasses and herbs for alloca- sites to potentially belong to more than one cluster, and tion of photosynthetic pathways in the ground layer and associated with each site is thus a likelihood of membership Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 the separate identification of tree and shrub species in the in each cluster. These probabilities of cluster membership subordinate layer). indicate the strength of the association between the site species composition and that of the various clusters it could potentially be assigned to: a procedure we deemed espe- 4.7.2. Species composition. We employed a ‘fuzzy’ clus- cially suitable for this analysis where stands representing a tering approach (as described by Kaufman and Rousseeuw mixture of forest and savanna species were anticipated. 2008) after appropriate transformations of the data (Legendre and Gallagher 2001) assessing the results for both the ‘Chord’ and ‘Hellinger’ transformed distance 4.7.3. Analysis of the tree (U) biodiversity. For the cal- matrices using the routine fanny available within the culation of tree diversity we consider here only the trees package cluster (Maechler et al. 2005). We then used constituting the U layer in each plot (d > 0.1 m), cal- the R package ade4 to present the ordination of the clusters culating Chao’s estimator corrected for unseen species using a principal coordinate analysis (PCoA) as suggested for the Jaccard abundance-based similarity index (Chao by Borcard et al. (2011). The analysis was conducted et al. 2005) using the programme EstimateS (Colwell at the species level with the three continents examined 2009). Jaccard’s coefficient (J), originally formulated for treated separately and, in addition to the sites included presence–absence data, in its abundance-based form may as part of this study, we also included two forest sites be written as: 114 M. Torello-Raventos et al.

MN J = ,(2)the number of species per site and the number of species (M + N − MN) shared between sites, complete the species information for all the sites. For the calculation of the rarefaction curves in each plot (d > 0.1 m) we used the function specaccum where M is the total relative abundances of individuals with the sites added in random order using the R package belonging to the shared species in one plot and N is the total vegan. relative abundances of individuals belonging to the shared species in the other (Chao et al. 2005). As shown by Chao et al. (2005), the abundance-weighted Jaccard estimator 5. Results is considerably less biased than classic indices, especially when a substantial proportion of species making up the 5.1. Structural differences and a preliminary total community assemblage have not been detected; such classification as might be expected to be the case for the relatively small The clustering procedure gave rise to 11 discrete groups, 1 ha plots sampled here. Additional parameters, such as all taken below a relative height (h) of 0.25 (Figure 3).

Height 0.0 0.1 0.2 0.3 0.4 0.5 0.6

MDJ-04 Θ Long-grass savanna woodland MDJ-06 3 MDJ-02 MDJ-08 BDA-01 Θ Shrub-rich savanna woodland BDA-02 4 FMS-01 LFB-03 Θ Grassland savanna HOM-01 1 HOM-02 IBG-01 Θ Scrub savanna 2 IBG-02 IBG-03 RSC-01 Θ Stunted forest 5 TUC-01 EKP-01 Θ Tall savanna woodland DCR-01 6 DCR-02 KBL-02 Θ Shrub-rich woodland 7 TUC-02 ALC-02 NVX-01 BBI-02 Θ Savanna woodland IBG-04 8 ALC-01 TUC-03 BBI-01 MLE-01

Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 MDJ-05 Θ Shrub-rich forests 9 FMS-02 MDJ-03 TAN-04 Θ Forest /Tall (closed) woodland NVX-02 10 FLO-01 VCR-02 KBL-03 VCR-01 MDJ-01 ALF-01 LFB-01 Θ Tall forest 11 MDJ-07 KBL-01 KCR-01 ALF-02 CTC-01 0.16 0.24 0.29 0.44

Figure 3. Complete linkage clustering dendrogram showing the segregation pathways to the 11 defined woody clusters using as discriminating variables the stand structure measurements as detailed in Table 1. Delineation of tropical vegetation types 115

This grouping was considered to provide our most intuitive only 0.17 (Figure 6(j)). We therefore took a minimum delineation and with the clusters numbered in approximate ϕG of 0.1 as a reasonable limit to define a savanna-type order of increasing dominance of woody vegetation. ecosystem. Especially at the lowest canopy covers, ϕ3 could The clustering results (Figure 3) revealed two clearly be comparable to, or even greater than ϕ4. Photograph distinct structural configurations of the vegetation at h = examples of such sites are shown in Figure 6(a) (BDA- 0.62 suggesting 28 savanna and 17 forest sites. The divi- 03) and Figure 6(b) (HOM-01). One further example of sion of the forest cluster was higher than the savanna cluster this is the high-rainfall site EKP-01 in northern Australia at h = 0.44 where 9 separated from 10 and 11. These (Figure 6(h)) where ϕ3 was 0.19, but with ϕ4 only 0.02: two then separated at h = 0.29. The first savanna divi- this being due to the ground later being dominated by sion within the main cluster started at h = 0.37 height, Xanthorrhoea johnsonii A.T. Lee (grass tree) and the C3 with 1, 2, 3 and 4 separating from 5, 6, 7, 8. sedge Scleria terrestris (L.) Fassett. Separation appeared to be mainly on the basis of different crown covers. Subsequent subdivisions occurred at 0.24 for 5.1.2. Defining the clusters. As a guide to the develop- groups 1, 2, 3 and 4 (followedby0.21for1 and ment of a delineation scheme, differentiating characteristics 2 and 0.16 for 3 and 4) with a second subdivision at of the 11 clusters were examined; with (but not to unnec- 0.32 for groups 5, 6 7 and 8; further divisions being essarily pre-empt our final classification scheme as detailed at 0.23 for 5,0.22for6 and 0.15 for 7 and 8. in Figure 9) the following simple ‘rules’ suggested from the Although some of the derived clusters consisted only labelling of the clusters as follows: of plots located in close proximity within the one region (e.g. HOM-01 and HOM-02 in 1 and IBG-02, IBG-02 and (1) To be termed a savanna, ϕG > 0.1. Otherwise IBG-03 in 2) five out of the 11 clusters encompassed veg- (unless spuriously bare soil) the vegetation forma- etation formation types sampled in more than one region, tion type must be scrub, woodland or forest. and often across different continents (for example, RSC- (2) ϕG > 0.1 does not, however, necessarily preclude 01 from Australia and TUC-01 from Bolivia in 5 and the existence of a grassland, scrub, woodland or BDA-01, BDA-02 (Burkina Faso) grouped together with even a forest. For scrub or woodland the existence FMS-01 (Australia) and LFB-03 (Bolivia) in 4). of a herbaceous fraction > 0.1 is simply indi- cated by the use of the word ‘savanna’ as a suffix 5.1.1. Cluster structural differences. Differentiation or prefix. On the rare occasions this is observed through the clustering algorithm was made on the basis for a forest, the word ‘axylale rich’ is used as a of several factors; for example, whilst of the same mean prefix; the term axylale originating from the clas- canopy height, HU, 1 and 2 differ in their 0.95 quantile sification of Du Rietz (1931) based on the extent height H∗ (Figure 4(b)) as well as in the extent of their of lignification and unambiguously encompassing herbaceous (ground) cover (ϕG). Intermediate tree and all herbaceous plant types (Ingrouille and Eddie shrub layers were also important; for example, 4 and 5 2006). were differentiated mostly on the basis of CM, although differences in herbaceous ground cover extent no doubt 5.1.3. Labelling and characterisation of the clusters. 0: contributed. The extent of the total woody layer at ground Not included in our ordination due to a virtual absence of level (tree and shrubs) was also important in contrasting the trees and shrubs, this vegetation formation type has CW, different groups. For example, although the upper layers (the summated projected canopy areas for all woody vege- were uniformly low for 1 and 2, they differed markedly tation with H > 1.5 m and d > 25 mm) of less than 0.02 and in CS. The magnitude of grass cover also varied widely with a distinct herbaceous ground layer present (ϕG > 0.1). Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 between clusters, primarily being the difference between One example from our dataset is near Dano, Burkina Faso 3 and 8 (discussed below) but otherwise seemed not to (BDA-03; Figure 6(a)), where despite a precipitation (PA) be a key delineating variable. Generally speaking, although of 0.98 m yr−1, only grasses occur, most likely due to the ϕ3 and ϕ4 showed a clear tendency to lower values for the shallow seasonally waterlogged soil. As other examples: for higher cluster numbers which were also of greater height Australia there is the Mitchell (tussock) grassland domi- and, generally speaking, also with higher woody canopy nated by Astrebla spp., associated with alkaline cracking- covers. clay soils (vertisols) and covering an area of nearly 5 × 106 2 Although there was an obvious tendency for both ϕ4 km (Orr 1975) or 6% of the total area of Australia. Of a to decline with increasing CW, with no C4 grasses being similar magnitude, the Llanos savannas of Venezuela and present for CW > 1.5 (Figure 5), some C3 herbaceous Colombia can have extensive grasslands, with absence of species persisted, even at the highest CW (this being MDJ- woody plants, named sabana abierta or sabana lisa, these 01, a forest in the Cameroon ZOT for which ϕG = 0.10 areas also being of restricted drainage (Beard 1953). Within (see also Figure 6(o)). It was also clear that for some veg- South America there are also areas of ‘campo limpo’ within etation formation types that would normally be termed the Brazilian Cerrado region usually associated with shal- ‘savanna’ did not have a value for ϕG much greater than 0.1. low lithosol soil types or a high water table for part of the For example, MLE-01 in Mole, Ghana, would be locally year (Eiten 1972) that also fall within this category. Within described as an ‘open Guinea savanna’ but has a ϕG of Africa, the most notable area dominated by this vegetation

116 M. Torello-Raventos et al.

11 11

10 10

9 9

8 8

7 7

6 6

5 5

4 4

3

3 ); crown of middle

2

2 U

1 1 C

1.5 1.0 0.5 0.0 0.8 0.6 0.4 0.2 0.0

4 4 M M

C ) m (m cover grass

Shrub C Shrub ) m (m

-2 2 -2 2

11 11

10 10

9 9

8 8

7 7

6 6

5 5

4 4

3 3

2 2

1.5 m and for trees and shrub seedlings; and of the 1 1 > ); crown cover of upper stratum ( W H

C

0.8 0.6 0.4 0.2 0.0 1.5 1.0 0.5 0.0

M M 3 3

C Tree ) m (m

C ) m (m cover grass

-2 2 -2 2 25 mm

), cluster 9 ( 10 11 )

11 < 11

10 10 d

9 9

8 8

7 7

6 6

5 5

4 4

3 3

2 2

1 1

0.8 0.6 0.4 0.2 0.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0

U U

S S C ) m (m

C seedling Shrub ) m (m

-2 2 -2 2

11 11

10 10

9 9

8 8

7 7

6 6

5 5

4 4

) shown separately for trees and shrubs 3 3

S ) of the upper stratum; Total stand level crown area ( 2 2

∗ U C 1 1 H

3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 0.8 0.6 0.4 0.2 0.0

W W

) m (m C

S S

C seedling Tree ) m (m

-2 2

-2 2

11 11

10 10

9 9

8 8

7 7

6

Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 6

; 0.95 quantile height ( 5 5

U

4

4 

3

3

H

2

2 

1 1 species as differentiated by the clustering procedure. 4 5 0 25 20 15 10

0.8 0.6 0.4 0.2 0.0

Cluster 1 ( ), cluster 2 3 4 (+), 5 6 7 8

U

S S

C Shrub ) m (m (m) and C

-2 2

3

11 11

10 10

9 9

8 8

7 7

6 6

5 5

4 4

3 3

2 2

1 1 ) shown separately for trees and shrubs; crown of lower stratum ( M 0

50 40 30 20 10

0.8 0.6 0.4 0.2 0.0

C

U

H S S

(m)

) m (m C Tree

-2 2 * Figure 4. Average height of the upper stratum fractional herbaceous cover of both C stratum ( Delineation of tropical vegetation types 117

3.0

2.5

2.0 ) -2 m 2 )

1.5 -2 (m 4 m ϕ 2 (m and W 3

ϕ C

1.0

0.5 0.0 0.2 0.4 0.6 0.8 1.0 1.2

0.0

0 0.035 0.12 0.21 0.321 0.34 0.37 0.45 0.55 0.68 0.74 0.775 0.85 1.07 1.3 1.631 1.75 2.22 2.29 2.38 2.46 2.85 3

FMS-01 ALC-01 MLE-01 MDJ-04

MDJ-02 TUC-02

ALF-01 ALL-02 KCR-01 VCR-01 VCR-02 MDJ-05 MDJ-01

BDA-02 IBG-01 BDA-01

BDA-03

IBG-02

HOM-01 HOM-02

IBG-03

DCR-02

CTC-01 KBL-03

ALC-02

MDJ-06 MDJ-08 BBI-01 EKP-01 LFB-01 IBG-04 KBL-02 NVX-01 TUC-01 KBL-01 DCR-01 MDJ-07 NVX-02 TAN-04 MDJ-03

FLO-01

TUC-03

BBI-02 LFB-03 FMS-02 RSC-01

Cluster 0 ( ), cluster 1 ( ), cluster 2 ( ), cluster 3 ( ), cluster 4 (+), cluster 5 (x), cluster 6 ( ), cluster 7 ( ), cluster 8 ( ), cluster 9 ( ), cluster 10 ( ), cluster 11 ( )

Figure 5. Variations in the fractional herbaceous cover of C3 and C4 species (stacked bars) with sites ranked according to their total woody canopy crown cover, CW, which is also shown. Cluster symbols for CW are as in Figure 4.

type is the Serengeti Plains of Tanzania where grasslands As well as this formation being present in north- extend over an area of ca. 7 × 103 km2. As for our other ern South America, for example the ‘shrub savannas’ examples, this absence of trees is attributable to adverse of Bolivar State, Venezuela (Huber 1995), and in wood- soil physical conditions, this being attributable in this case ier parts of the seasonally flooded Llanos de Orinoco to the relatively young soils there and the presence of rela- of Venezuela/Colombia (Sarmiento 1984), scrub savannas tively shallow calcareous hardpans (Anderson and Talbot also occur in seasonally flooded areas south of the Amazon 1965). Although, as shown later, in some cases mono- forest region in both the Llanos de Moxos of Bolivia (Haase cotyledonous and dicotyledonous herbs may also make a and Beck 1989) and the Pantanal region of Brazil (Prance substantial contribution to the ground layer of this vegeta- and Schaller 1982; Pinder and Rosso 1998). They are also tion formation type, we term this vegetation type, simply found in Australia, for example as tall shrublands with tus- ‘grassland’. sock grasses (AUSLIG 1990), often dominated by Acacia 1:HereCW is > 0.02 but < 0.30 and with ϕG > 0.1. aneura F.Muell. ex Benth. Analogous communities occur Our example (Figure 6(b)) is from the Sahel region of in East Africa, often on seasonally flooded soils (Harrison Africa, being located at Hombori in Mali (HOM-02) with and Jackson 1958; Greenway; 1973) but also in drier areas −1 PA ∼ 0.35 m yr . This plot is typical of the vegetation for- on hardpan soils as exemplified by the ‘savanna thick- mation type occurring along an extended belt between the ets’ of Burtt (1942), these described as having a relatively Sahara desert and the more woody ‘Sudanian savannas’, as dense grass layer co-existing with a relatively sparse shrub well as occurring in southern Africa as, for example ‘turf or stunted tree layer, usually also dominated by Acacia

Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 thornveld’ and ‘sweet bushveld’ (Acocks 1988). There are species. Stamp and Lord (1923) also note in central Burma also many examples in Australia, these being mostly ‘open the presence of an Acacia catechu (L.) Willd., Oliv. thorn woodland with tussock grasses’ and often dominated by scrub (also apparently known as ‘Sha jungle’) of about Eucalyptus (AUSLIG 1990) and with much of the cerrado 2 m height and with ‘grass as the only undergrowth’ which aberto (literally ‘closed open savanna’) formation of the would likely also come under this category.    > < Cerrado region of Brazil also falling into this class. This 3:Here H U is 6 m and 12 m with vegetation formation type differs from 0 by virtue of hav- CW > 0.30 and ϕG > 0.5, this being a typical characteris- ing a canopy area index > 0.10 (i.e. a more than notional tic of the tall grasses which have an average height greater amount of woody vegetation cover of H > 1.5 m being than 1.0 m. The illustrated example in Figure 6(d) is for present), and we term it ‘grassland savanna’. Mbam Djerem National Park in Cameroon (MDJ-04) for −1 2: Vegetation in this category is classified by an aver- which PA ∼ 1.6 m yr . We term this vegetation type >   age height for the woody elements with d 0.1 m, H U, ‘long-grass savanna woodland’. Such tall grass woodlands of less than 6 m. This grouping also has CW > 0.30 and are abundant in Africa, for example Acacia-Commiphora ϕG > 0.1. We show as an example in Figure 6(c) a ‘woodland’ in the Serengeti (Herlocker 1975), although an frequently burned campo sujo near Brasilia in Brazil (IBG- appreciable fraction of this ecosystem may have a canopy −1 02) for which PA ∼ 1.6 m yr . This is termed a ‘scrub cover less than 0.3 (White 1983) and thus be more suit- savanna’. ably termed in our case ‘long grassland savanna’, this 118 M. Torello-Raventos et al.

(a) DAN-03, a grassland (b) HOM-02, a grassland savanna

(c) IBG-02, a shrub savanna (d) MDJ-04, a long grassland savanna Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013

(e) DAN-01, a shrub-rich savanna woodland (f) TUC -01, a stunted shrub-rich forest

Figure 6. Representative examples for each cluster type as defined in Section 5.1.3.

    > < being a variant of 1. Such tall grasses are, however, only 4:Here H U is 6mbut 12 m with a rarely encountered in South America, being limited to cer- CW > 0.3 and ϕG > 0.1. This formation differs from tain areas of the Llanos and Pantanal (Sarmiento 1984; 3 in that it has an appreciable subordinate woody layer Haase and Beck 1989; Pinder and Rosso 1998). Long- with CS > 0.1. Examples of this cluster are grassy grass savanna woodlands dominated by an understorey of open woodlands in eastern Australia, such as Forty Mile −1 Sorghum species are also found in Australia, not only in Scrub National Park (FMS-01) where PA ∼ 0.78 m yr . areas of reasonably high precipitation around Darwin and Examples in South America are LFB-03, a cerrado sensu −1 the Kimberly Region (PA ∼ 1.6 m yr , but also extend- stricto at Noel Kempff National Park, Bolivia (PA ∼ ing into drier areas on more fertile basalt-derived soils 1.2 m yr−1) and for Africa the open ‘Sudan savannas’ as (Mapping Units D4, H4 and F4 in Fox et al. 2001). BBI-01 and, as our illustrated example BDA-01 in Dano, Delineation of tropical vegetation types 119

(g) RSC-01; a stunted axylale-rich forest (h) EKP-01; a tall savanna woodland

(i) TUC-02, , a shrub-rich woodland (j) MLE-01, a savanna woodland Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013

(k) MDJ-03, a shrub-rich forest (l) MDJ-05, a shrub-rich forest

Figure 6. (Continued).

−1 < < BurkinaFasoforwhichPA ∼ 0.98 m yr (Figure 6(e)). class (25 mm d 0.1 m). One example is RSC-01 in −1 Essentially this is a savanna vegetation type where some Australia (PA ∼ 0.67 m yr ), locally referred to as ‘dry trees are reasonably tall, but with the bulk of its woody forest scrub’ (Figure 6(g)). Interestingly, although exist- vegetation in the lower strata. One other example would be ing at a low rainfall, these ‘inland forests’ share many the more xeric forms of dry dipterocarp forest in Thailand canopy species with ‘beach scrub’ formations in higher- (Santisuk 1988). In almost all cases, the subordinate layer precipitation regions close to the Australian coast (Fensham is dominated by shrub species rather than treelets or tree 1995). Our second example of this interesting vegetation seedlings (Figure 4) we term this formation ‘shrub-rich form, TUC-01 (Figure 6(f)), located at the drier end of the −1 savanna woodland’. Chiquitano forest region of Bolivia (PA ∼ 0.89 m yr )also    > < 5:Here H U is 6 m and 12 m with has strong floristic associations with taller forest further CW > 0.7 and with over half of CW in the middle size north (Killeen et al. 2006). We thus refer to this primary 120 M. Torello-Raventos et al.

(m) NXV-02, a tall closed woodland

(n) TAN-01, a forest

(o) MDJ-01, a tall axylale-rich forest

Figure 6. (Continued).

formation group as ‘stunted forest’ with both RSC-01 and woodland formations south of the Central African forest TUC-01 being more precisely defined as ‘stunted axylale- (Wild and Fernandes, 1967). We denote this common veg- rich forests’ due to ϕG > 0.1. It should also be noted etation formation type as ‘tall savanna woodland’.    > < that both these forests also have an appreciable shrub and 7:Here H U 6 m and with CW 0.70, but with shrub-seedling layer in addition to the herbaceous ground almost all of this in the subordinate layer (CS > 0.6). cover present. Other examples would include some of the As an example of this type of vegetation in this cluster we taller caatinga formations on deep and moist soils in Brazil, only have one plot TUC-02, being an ‘Abajoy savanna’ in Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 referred to as ‘arboreal caatinga’ by Daly and Mitchell Bolivia (PA ∼ 1.1 m). This is ‘shrub-rich woodland’. (2000), some of the African stunted forests mentioned in This formation is dominated by shrubs and palms Section 2.6.2 and most likely also some of the formations (Figure 6(i)), with only a few trees with d > 0.1 m and described in Burma by Stamp and Lord (1923), in par- with those tree species present in the discontinuous upper ticular a Tectona hamiltonii-Terminalia oliveri formation stratum, usually occurring in nearby forests. It is there- usually less than 10 m high being described as “a splen- fore possible that this unusual site was actually a savanna did example of a stunted and xerophyllous forest produced derived from forest through human intervention (Veldman by a physiologically dry soil”. and Putz 2011).    > > ϕ     6:Here H U is 12 m with CW 0.30 and G 8: Here, as for 3, H U is between 6 m and 12 m, usually > 0.1. This vegetation formation type is very com- with CW > 0.30 and grasses present. But in this case with mon in Australia (where they are usually classified as ‘open ϕG < 0.5. In our dataset there are many examples from forests’; Figure 1(a)) with the shown example here at PA ∼ South America (both Amazon savannas and those from the 2.9 m yr−1 being EKP-01 (Figure 6(h)) and with many of Cerrado region), and also from West Africa. The exam- the taller ‘dry dipterocarp’ forests of southeast Asia seem- ple shown in Figure 6(j) (MLE-01) is a Guinea savanna ingly also falling within the category (Bunyavejchewin in Mole, Ghana (PA ∼ 1.2 m). Across our dataset, PA for 1983), as would some of the taller but more open miombo this vegetation type varies from 0.69 m yr−1 for BBI-01 in Delineation of tropical vegetation types 121

Burkina Faso to around 2.0 m yr−1 for the Amazon savanna 5.2. Taxonomic clustering and stand structure ALC-01. This vegetation formation type is simply termed Although some sites could not be included in the clus- ‘savanna woodland’. tering analysis based on structure due to insufficient data This is likely to be a very common vegetation formation being collected (usually due to time constraints and with type, including much of the woodland region that stretches missing data at the shrub or ground-layer level), many over 1000 km from southern Mali in West Africa to Uganda of these had complete floristic inventories of trees under- in the east as well as the drier miombo and mopane taken (d > 0.1 m) and so could be included in the floristic formations of southern Africa (White 1983), also includ- clustering and, on the basis of recorded observations and ing the more open ‘dry dipterocarp forests’ of southeast photographic records, unambiguously placed in one of the Asia (Blasco 1983; Bunyavejchewin 1983; Stott 1992) and 12 categories above. Results of the fuzzy clustering of much of the Australian savanna region (Fox et al. 2001). species compositions for this expanded dataset (also includ- As cerradão, it was probably also much more abundant ing species composition data from additional forests from in the Cerrado region of Brazil than is now observed, as Africa and South America) are shown separately for each a consequence of human disturbances (Ratter et al. 2006). continent in Figure 7.    9:Here H U can be either greater or less than 12 m but with ϕG < 0.1 and with a significant subordinate layer 5.2.1. Africa. Six clusters were identified, with a sil- (CS > 0.6) in all cases. Our shown example (Figure 6(k)) is houette plot illustrating cluster membership shown in −1 MDJ-03, a dry forest in Cameroon (PA ∼ 1.6 m yr ) with Figure 7(d). Here, for ease of discussion we have also this formation also being found in Australia (FMS-02) at a given each colour-coded cluster a simple geographic- −1 much lower rainfall (PA ∼ 0.7 m yr ). As for 4,themost physiognomic label consistent with local usage (e.g. notable feature here is the relatively dominant subordi- ‘Sudan-Guinea savanna’), with the length of the silhouette nate layer, mostly consisting of shrubs, and these constitute (S) of each plot within a given cluster giving an indication ‘shrub-rich forests’, with the smaller statured FMS-02 a of the certainty of membership in that cluster and for which   < ‘stunted shrub-rich forest’ because of H U 12 m and negative values (i.e. as for MDJ-05) indicate that the plot to be consistent with our descriptions of 5. has probably been misclassified – or at least misclassified Also of interest in this grouping, and thus shown in in the cluster to which it has actually been assigned. Here Figure 6 additional to the nearby MDJ-03, is MDJ-05, the Ghanaian and Cameroonian forests group into different which might be described as a ‘successional forest’ devel- clusters with three savanna groupings for western Africa oping in an area that previously consisted of savanna sites. vegetation (Figure 6(l)). In the associated PCoA ordination (Figure 7(a)) the    > 10: Within this cluster H U is usually 12 m with probability profile of cluster membership was again rep- CW > 0.7, with the herbaceous ground layer virtually resented, this time being proportional to the size of each absent (ϕG< 0.1) and also with a considerable subordinate ‘wing’. This showed most sites to be located reasonably layer (CS). As is explained below, this vegetation forma- close to their other cluster members and with the six clus- tion type which we observed within ZOT in Africa, South ters generally well separated and with one segment usually America and Australia includes what might generally be dominant. However, three sites were less well defined in termed as ‘transitional forest’ or, without habitat or geo- terms of species composition. The Ghana plot BFI-04 was graphical implications, simply ‘forest’. But as we show a forest in a zone of transition located some distance away later, this vegetation formation type may also include stands from the other forests in the dataset, with its poor clas- where the predominant floristic mix is of species usu- sification due to a different forest species mix than the ally associated with savanna woodlands or even grassland other forests. Also of a mixed composition was BDA-02, Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 savannas. In which case, as we argue below, then the term that appeared to have affiliations with both the ‘Guinea ‘closed woodland’ or ‘tall closed woodland’ may be more savanna’ and ‘Sudan-Guinea savanna’ floristic associa- appropriate. Due to its dichotomous nature, we therefore tions. However, MDJ-05, our ‘successional forest’ (shown show here two examples: NXV-02 a cerradão located near in Figure 6(l)) lying equidistant between the ‘Cameroon Nova Xavantina, Brazil (Figure 6(m)), for which PA ∼ forest’ and ‘Cameroon savanna floristic associations’ and − 1.50 m yr 1, and a nearby transitional forest, this being the ‘Guinea savanna floristic association’, was clearly mis- TAN-01 located some 100 km north (Figure 6(n)) and with classified, as also illustrated by its negative silhouette length −1 a slightly higher precipitation (PA ∼ 1.65 m yr ). (Figure 7(d)).     > 11:Asfor 10, H U 12 m and with the herbaceous Figure 8(a) shows these taxonomically defined clusters ground layer virtually absent (ϕG< 0.1) and the subordi- overlain on the structural cluster grouping of the previous nate much less evident than 10, especially in terms of section. Here, within each ovoid (coloured as in Figure 7(a) ∗ shrubs. The 0.95 percentile tree height, HU, is also greater except for MDJ-05 which is of an unclassifiable status) the than for the 10 group, with the threshold for the difference name of each cluster member overlays its structural clas- being 36 m. This category may be considered typical forests sification. The total number of species within each cluster of the moist tropical regions and referred to here as ‘tall as well as the number of common species shared between forests’. The example shown is MDJ-01, an axylale-rich two or more plots as well as the number of species found forest occurring in Cameroon (Figure 6(o)). in all plots within that cluster is also shown. The various 122 M. Torello-Raventos et al. Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013

Figure 7. C-Means fuzzy clustering of plot floristic composition by continent. Right-side panels: clusters identified by colour in a prin- cipal coordinate ordination associated with star plots showing probability of group membership: Africa, South America and Australia. Left-side panels: plots of sillouete width (for which the greater the value, the stronger the probability of correct cluster identification) with cluster labels (as used in the text) also given: Africa, South America and Australia. ‘J’ gives the numbers of clusters, ‘nj’ the number of observations and ‘aveiCj’ is the average of the dissimilarities between clusters.

lines connecting the different clusters indicate the numbers Here we see that the ‘Sahel floristic association’, the of shared species, where in all cases we have also made ‘Cameroon savanna floristic association’ and the ‘Ghana allowances for unseen species (Chao et al. 2005) in the forest floristic association’ have all their members falling numbers presented (Section 4.9). within the same structural classification. On the other hand, Delineation of tropical vegetation types 123

(a) Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013

(b)

Figure 8. Overlap of structural and floristic clustering results for each continent. Encircled integers detail the number of species shared by the floristic clusters as connected by arrows. Within each floristic cluster is given the total number of species contained within that cluster and the number of species that are common to two or more sites within that cluster and the number of species that occur at all sites within that cluster: (a) Africa; (b) South America; (c) Australia. 124 M. Torello-Raventos et al.

(c)

Figure 8. (Continued).

of the five members of the ‘Cameroon forest floristic group- our measures of the Chao Jaccard abundance-based similar- ing’, three are structurally ‘tall’ but with MDJ-03 a shrub- ity index, J (lower diagonal of Supplementary Information rich forest and MDJ-10 in the ambiguous 10: a struc- Table D1). For example, for plots within the ‘West Africa tural grouping also includes the floristically very different forest floristic association’ the strongest association with BFI-01; the latter a clear member of the ‘Sudan-Guinea any plot contained within a savanna floristic association is savanna floristic grouping’ (Figure 7(d); Supplementary J = 0.02 (BFI-01 and ASN-04) and with values generally Information Table C1). This grouping, as for the ‘Guinea <0.001. On the other hand, comparing plots of this same woodland floristic association’, is for all other cases asso- ‘West Africa Forest floristic association’ with those within ciated with the woodland vegetation formation types; viz. the ‘Central Africa forest floristic association’ we find J to shrub-rich savanna woodland, savanna woodland or tall be as much as 0.23. woodland (Figure 7(d)). Nevertheless, some 15 of the 455 species we recorded In terms of shared species, the floristically ambigu- in Africa were found in both forest and savanna floristic ous status of MDJ-05 is illustrated by sharing 14 of its association plot members. An examination of the actual 28 species with plots aligned with the ‘Central Africa abundances of these species occurring in both groups forest floristic association’ and 11 with the ‘Central Africa (Supplementary Information Table E1) suggests that it is savanna floristic association’. just as likely that a species usually occurring in one or more Despite their geographic distance, the two African of the forest-aligned plots occasionally occurs in one of the forest clusters show much more floristic similarity with savanna-aligned plots as the other way around. each other than they do with their nearby savanna floris- tic groups. Specifically, of the 127 species found in the 5.2.2. South America. Fuzzy clustering results and the ‘Central Africa forest floristic association’, 50 species are associated PCoA of the results are shown in Figure 7(b) also shared with the ‘West African forest floristic associa- and Figure 7(e), respectively. Here, seven clusters were Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 tion’ located some 1500 km away but only 10 species over- identified for which we have again given simple designa- lap with the plots aligned with the nearby ‘Central Africa tions for the floristic associations. Firstly we note that the savanna floristic association’. Likewise, of the 199 species silhouette analysis suggested no likely misclassification of in the West African forest floristic association only between any plot, although OTT-02 was only a marginal member of six and 26 species occur in nearby plots aligned with the the ‘Bolivia core savanna floristic association’. This proba- ‘Guinea savanna floristic association’ and ‘Sudan-Guinea bly because it also has some species associated usually not savanna floristic association’, respectively. This can also only with all the other savanna groupings, but also with the be seen by an examination of species sharing on an indi- Bolivian semi-deciduous forest groupings (Supplementary vidual plot basis (Supplementary Information Table D1). Information Tables C2 and D2), as is also the case for For example, the tall forest plot MDJ-01 shares a mini- the only ‘ZOT savanna association member’ from Bolivia, mum of 10 and a maximum of 20 of its 41 species with LFB-03. The ordination also suggests that it is hard to other plots with the same ‘Central African forest floristic classify correctly the cerradão NXV-02, which is clearly association’, and between five and 14 species for plots in the placed away from its own ‘ZOT savanna floristic associa- ‘West African forest floristic association’. But for the plots tion’ in Figure 7(e), this being due to affinities with both of the ‘Central Africa savanna floristic association’ (all less the ‘Brazil ZOT forest floristic association’ and ‘Bolivia than 100 km away), the number of shared species varies semi-deciduous forest floristic association’ (Supplementary from zero to four. These differences can also be seen in Information Table C2). Indeed, of the 27 forest and savanna Delineation of tropical vegetation types 125

plots sampled in South America, only two did not have at As for Africa, it seems that this forest/savanna floristic least one species in common with NXV-02, these being overlap is as much due to species primarily associated with TUC-01 (the southernmost member of the ‘Bolivia semi- a forest floristic association being occasionally found in a deciduous forest floristic association’) and ALF-02, one of plot more floristically aligned with a savanna as the other the members of the northernmost ‘Core Southern Amazon way around (Supplementary Information Table E2). It also forest floristic association’ (Supplementary Information seems that there may be some species in South America Table D2). that can be described as ‘forest-savanna generalists’. One Figure 8(b) shows the floristic associations with each such species is Bowdichia virgilioides Kunth which was plot also overlain upon its structural classification. This observed in five of the seven floristic associations, with shows that although the ‘Core Southern Amazon forest another candidate being Machaerium acutifolium Vogel floristic association’ and ‘Bolivia ZOT forest floristic asso- (Supplementary Information Table E2). ciation’ affiliated plots all aligned with the same structural grouping (tall forest, 11), the ‘Brazil ZOT forest floristic 5.2.3. Australia. The fuzzy clustering analysis for association’ had only one of its affiliated plots (VCR-01) in Australia (Figure 7(c)) gave rise to four floristic association this structural grouping, with the other three all situated in groupings, but with the high-rainfall savanna woodland 10. The ‘Bolivia semi-deciduous forest floristic associa- EKP-01 (Figure 6(g)) having little in common floristically tion’ includes plots that are even more diverse structurally, with the other sites and almost certainly misclassified as with one axylale-rich stunted forest (TUC-01; 11), three part of the ‘Broadleaf savanna floristic association’ as can tall forests (ACU-01, CRP-01, CRP-02) and one member also be seen from the PCoA plot of Figure 7(f). Treating of the forest/tall woodland structural group (OTT-01; 10). this site as a single member group then, Figure 8(c) shows Also in this structural group is one ‘ZOT savanna floristic the overlay of the structural and floristic classifications association’ plot member NXV-02 (whose mix of forest and which reveals a much better congruence between the savanna species has already been noted), though with most two classifications than for Africa and South America. of the other members of this floristic association being in Moreover, with the exception of nine stray Eucalyptus the savanna woodland structural classification. crebra F. Muell. trees in the stunted shrub-rich forest Overall, the extent of species sharing between the for- FMS-02 (located only 1 km away from the tall savanna est association versus savanna association affiliated plots woodland FMS-01 which was dominated by this species) was greater for South America than Africa, with 19 of there was no exchange of species between the different the 109 species in the ‘Brazil ZOT forest floristic asso- floristic associations. ciation’ affiliated plots also found in the ‘Brazil ZOT savanna floristic association’ plots, as were 14 of the 129 species in the ‘Bolivia ZOT forest floristic association’ 5.3. Proposed classification scheme for tropical and 12 of the 143 species of the ‘Bolivia semi-deciduous vegetation forest floristic association’, of which 15 were also shared Through the clustering procedure (Figures 3 and 4), our with the ‘Bolivia savanna floristic association’. On the analysis of structural differences showed that not only was other hand, although sharing 30 and 14 species with the total woody plant canopy cover and upper stratum canopy ‘Brazilian ZOT forest floristic association’ and ‘Bolivian height important in differentiating between the various ZOT forest floristic association’ plots respectively, the groupings, but that the relative abundance of trees and ‘Core Southern Amazon forest floristic association’ shared shrubs in the understorey, the extent of herbaceous cover only two of its 143 species with the ‘Bolivian ZOT savanna and 0.95 quantile height were also important factors in floristic association’ and only eight with the ‘Bolivian accounting for the differences between clusters. In some Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 semi-deciduous forest floristic association’. cases differences between the structural groupings could be This sharing of species across the forest/savanna floris- attributed to more than one factor. For example, although tic associations was not solely attributable to a high number not readily differentiable in terms of their mean canopy of forest species in NXV-02 (Supplementary Information height, 10 and 11 differed in the 0.95 quantile height, ∗ Table D2). For example, the ‘Brazilian ZOT savanna floris- HU as well as in the nature of their woody understorey ele- tic association’ aligned SMT-02 shares over 10% of its ments (shrub versus tree seedlings). Although this provides species with the ‘Brazilian ZOT forest floristic associa- for some flexibility in designating criteria for the differen- tion’ aligned VCR-01 (J = 0.20). Nevertheless, of the tiation of the various structural groupings, it also means ‘Brazilian ZOT savanna floristic association’ aligned plots, that careful consideration needs to be taken as to which it was always NXV-02 which had the strongest species over- characteristics to invoke with both pragmatic and scientific lap with the ‘Brazilian ZOT forest floristic association’ considerations of importance (i.e. ease of measurement ver- aligned plots: the highest J for such a comparison being sus the likely relative importance of the various differences 0.43 for NXV-02←→ TAN-01, which is actually greater in a functional sense). than many of the linkages within the various forest or From Figures 7 and 8 it is also clear that, although savanna associations, and with 15 of the 48 species in in many cases there was no major overlap of the TAN-01 also found amongst the 55 species of NXV-02. floristic versus structural groupings, both forest and 126 M. Torello-Raventos et al.

savanna/woodland floristic associated plots were observed (1968) in his vegetation classification as well as in many within the 10 structural grouping, this being the case earlier schemes (e.g. Du Rietz 1931). for both Africa (MDJ-10 and BFI-01) and South America Although the current study was very much focussed (NXV-02, FLO-01, TAN-01, VCR-02 and OTT-01), but not on forest and savanna ecosystems, the right-hand side of for Australia. Following the suggestion of White (1983) in Figure 9 extends the classification table to more open- the classification of Figure 9, we have therefore adopted canopied systems but for where a grass layer is absent and the convention that, neither structure nor phylogeny being where, for consistency, we use semantically similar vege- alone capable of sufficiently describing differences in such tation definitions. For the ‘grassless’ vegetation types we situations, a hybrid nomenclature scheme be adopted. Here have also used the same boundaries for the cover classes as the term ‘woodland’ is applied in all cases where the stand suggested by our cluster analysis of the forests and savan- species makeup is predominantly of taxon usually associ- nas investigated as part of the current study (CW = 0.05, ated with non-forest formations (e.g. the ‘tall woodlands’ 0.30 and 0.70). We here also remind the reader that CW as NXV-02 and BFI-01). By corollary, the term ‘forest’ is only referred to throughout this paper refers to the total canopy applied where upper stratum floristic composition is clearly area index. This is not the same thing as ‘canopy crown more similar to stands usually of a taller forest stature (e.g. cover’ or just ‘woody canopy cover’ (ς) as often used the ‘shrub-rich forest’ MDJ-03 or the ‘stunted axylale-rich by both vegetation scientists and remote-sensing scientists. forest’ TUC-01). The relationship between the two is given in Appendix 1. Though critical for the accurate description of the dif- This shows that the above boundaries in CW extent relate to ferent structural groups examined, the words ‘axylale rich’ values of ς = 0.05, ς = 0.25 and ς = 0.50, respectively. and ‘shrub-rich’ play a role as prefixed modifiers rather Compared with many other previous schemes (e.g. than as principal descriptors, with ‘axylale’ as defined by Fosberg 1961; Ellenberg and Mueller-Dombois 1966; Eiten Du Rietz (1931) meaning any plant whose stem does not 1968) the classification of Figure 9 is relatively simple, produce woody, persistent tissue and generally dies back at involving only 20 different groupings for which the base the end of each growing season; this definition thus includ- measurements required are only the mean and 0.95 quan- ing most members of the Poaceae (grass family) as well tile height of the upper woody stratum (d > 0.1 m), the as herbaceous dicots. We also use as a prefixed modifier total canopy cover of the woody component of the vege- the word ‘long’ to describe grasslands and savannas whose tation (excluding seedlings) and the extent of herbaceous maximum height (at the time of maximum biomass towards ground cover which, for many purposes can all probably be the end of the wet season, as was usually the date of sam- estimated with sufficient accuracy by simple visual assess- pling for this study) was greater than 1.0 m, this segregation ment by experienced botanists. Likewise, the prefix terms of the two grassland types also having being made by Eiten require only the simple measurements of maximum grass Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013

Figure 9. Proposed structural-physiognomic scheme for the classification of tropical vegetation on the basis of the canopy cover of tree, shrubs and grasses and the mean and upper canopy tree height (d > 0.1 m). Delineation of tropical vegetation types 127

height (where appropriate) and shrub canopy cover. The dif- drier tropical forests (Gentry 1995), but with the taller  ∗ ferentiation of ‘woodland’ versus ‘forest’ at high values of 11 grouping (of a greater HU and generally associated CW does, however, also require a knowledge of the species with slightly moister climates) having an understorey con- composition and whether it is forest- or savanna-type for- sisting mostly of tree seedlings, this reflecting a ground mations where these species are usually found. layer dominated by regenerating seedlings of middle- and upper-canopy species. This suggests that, together with the tendency for the herbaceous layer to not only become less 6. Discussion abundant but also to be increasingly dominated by species Although by no means sampling the full possible spec- of the C3 pathway, there is also an increase in specialised trum of vegetation structures, the approach taken here has understorey shrub species at or around CW = 0.7 for both been to sample a wide range of different forms of tropi- forest and savanna. This varying understorey shrub layer cal vegetation across a range of continents and, through a is reflected through our numerical clustering procedure, numerical analysis, to determine if discrete grouping could and suggests a need to differentiate forest and savanna be observed and to determine, in particular, if a struc- formations with a significant shrub understorey and those tural categorisation could be obtained that was applicable without. As pointed out by Gentry (1995), there is an unfor- at a pan-tropical scale. To a large degree, this approach tunate tendency for scientists working in tropical forests in proved successful, with our final categorisation (Figure 9) particular to concentrate on the larger upper canopy species giving rise to a grouping based on both woody cover and and ignore the species of the lower layers. Nevertheless, for plant height, similar to that already employed in Australia the drier forests (e.g. TUC-01 in Figure 6(f)) this under- (Section 2.6.1) and, for woody cover at least, also simi- storey is likely to be important in terms of productivity as it lar to currently employed empirical global classifications is often predominantly evergreen in contrast to a generally (Section 2.6.4). Unlike the latter group of classifications deciduous upper stratum (Killeen et al. 1998). This reten- which are mostly based on remote-sensing products, our tion of some active leaf area during the period for which analysis does, however, suggest that even for open canopies the upper canopy is leafless would potentially allow bet- with a grass layer present, differences in woody plant ter access to water from both early and late wet season upper-stratum height are important in differentiating the rains – under higher light conditions – as well as utilisa- vegetation formation. For example, at a canopy crown tion of water from any significant episodic rainfall events cover of 0.5, the upper stratum 0.95 quantile height can during the drier months. range from just over 6 m (ALC-02: a savanna woodland near Santarém, Brazil) to 28 m (KBL-02: a tall savanna woodland in Northern Queensland). A similar conclusion 6.2. The role of grasses in the vegetation classification was also reached by Eiten (1968) who, in a well-argued Although most workers agree that the term ‘savanna’ and clear structural classification scheme, also consid- should only be applied to vegetation formation types char- ered canopy height to be as important a consideration as acterised by at least some herbaceous cover, there have been tree/shrub cover in categorisation of non-forest woody veg- many variations upon this theme. Here, we have used our etation formation types. Interestingly, Eiten (1968) also own data to help define it, suggesting a minimum fractional reached the same conclusion as that reached from this herbaceous cover of 10% as required for any woody vege- study, that it was important to differentiate stands with a tation form to be termed ‘savanna’. Interestingly, we found herbaceous cover more than about 1.0 m tall from those again a close coincidental concordance of our classification of a smaller stature. Although the former are expected to scheme with that of Eiten (1968), as he also placed a cri- often be found on swampy areas on all continents, they also terion of a minimum herbaceous cover of 10% in order for Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 seem to occur naturally on well-drained soils in areas of rel- the term ‘savanna’ to be applied. His scheme, does, how- atively high precipitation, especially in Africa, where their ever, differ from ours in that the term ‘savanna’ was only to standing peak biomass may be as much as double that of be applied for a woody crown cover of less than 10%. As the shorter grassland forms (E. M. Veenendaal, unpublished presented here, we propose the use of the word ‘savanna’ data 2006–2008). It thus seems apposite to retain the dis- as a precise descriptor that when used (always as a suffix or tinction between short and long grasses when considering prefix) has the meaning that the vegetation formation type the structure of both savanna and grassland communities as in question has both trees/shrubs present and a herbaceous suggested by the clustering analysis of vegetation structure cover > 10%; the presence of such a grass layer, for exam- in Section 5.1. ple, differentiating a ‘savanna woodland’ from a (virtually grassless) woodland. Some workers (e.g. Lehmann et al. 2011) have sug- 6.1. The importance of the shrub understorey gested that the presence of the C4 pathway in the Although differentiated in Figure 9 in terms of their herbaceous layer also be the requirement for a vegeta- 0.95 quantile height, 10 and 11 were also differenti- tion formation type to be defined a ‘savanna’. We see ated on the basis of their understorey elements, these being no reason for this, especially as some stands we sam- mainly shrubs and shrub seedlings for 10, consistent with pled in Africa were actually dominated by C3 herbaceous the taxonomically distinct understorey stratum of many species, most likely due to intense grazing. Moreover, a 128 M. Torello-Raventos et al.

marked presence of C3 grasses in tropical savannas can Within individual stands, we generally observed only a occur even in the absence of anthropogenic influences, as very limited overlap between forest and savanna-affiliated shown for example for the sites close to Brasilia (IBG- species, even within ZOT. For example, from Table G2 it 01/02/03/04) and elsewhere in the Cerrado region where seems reasonable to conclude – even without a formal sta- the C3 species Echinolaena inflexa (Poir.) Chase is often the tistical analysis – that < 5% of the 555 species we identified most abundant grass present (Lloyd et al. 2008). in South America were ambiguous in terms of being obvi- ously neither ‘forest’ or ‘savanna’ affiliated. Nevertheless, it is worth noting that, as for all such analyses, the results of the inferred overlap must depend on the number and nature 6.3. Structure versus floristics of the sites sampled. For example, numbers would have One of the most important results from the current work is been higher had some of the deciduous or semi-deciduous the demonstration that there is not necessarily a congruence forests of the Atlantic Coast been included. This is because between structural and floristic groupings for vegetation their community compositions show a far greater overlap formation types within the forest/savanna transition zone. with that of the Cerrado (sensu lato) than is the case for In particular, tall and dense stands can occur which are the forests of the Amazon Basin (Méio et al. 2003; Ratter composed principally of species usually associated with the et al. 2006). One exception was NXV-02, a tall closed more open savanna formations and not with the nearby for- woodland which presented an unusually equal mix of both est with which they have the strongest structural affinity forest and savanna species. This ‘unusual’ species mix may (Figure 8). This is as noted before by White (1983) for be related to its close proximity to an unusually southerly what he termed ‘transitional woodlands’, and we believe intrusion of Amazon forest along the Rio das Mortes and this is an important distinction. It is accounted for in Araguaia flood plain predominantly in an area dominated our classification scheme through the term ‘tall’ and/or by savanna vegetation. Indeed it may structurally and floris- ‘closed woodlands’ for these formations, and especially tically reflect a distinct successional stage (see Ratter 1992), given some large and systematic structural and physiolog- in which the presence of some species is only temporary. ical differences between species usually associated with For example, the high abundance of pioneer tree Tachigali forest versus savanna (Hoffmann et al. 2004, 2005a; Lloyd paniculata L.G. Silva & H.C. Lima in NVX-02 is con- et al. 2009; Ratnam et al. 2011) should be a useful dis- sidered to lead to high rates of gap formation, leading to tinction when modelling biome shifts at the forest/savanna increased probabilities for new species establishment, many ecotone. As well as our own examples given in the text, of them from the nearby forest (Franczak et al. 2011). one further example of where the word ‘forest’ may not Although it will be often clear as to whether a stand be appropriate for such a vegetation formation type is for is composed primarily of forest or savanna species, there ‘closed miombo woodlands’ of northern Zambia. These are situations where it may not be so obvious. For exam- can be over 20 m high and with a minimal grass cover, ple, the genus Eucalyptus occupies vast areas of Australia, but still with much closer floristic association with more dominating many forests of the temperate zone, and there open miombo savanna formations further south than with is thus often confusion as to whether to refer to the denser the nearby forest formations (Wild and Fernandes 1967). stands in northern Australia (which almost always have an In contrast, we have observed both within our own appreciable grass understorey) as forest or savanna, as is dataset and in the literature cases where a vegetation for- for example exemplified by the occasional use of the term mation type, more savanna and/or woodland in a structural ‘open-forest savanna’ (e.g. Hutley et al. 2000). There is, sense, actually consisted mostly of species more usually however, an almost total absence of eucalypt species in associated with denser and/or taller forest formations. what are normally considered (for want of a better word) the Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 Thus, following Stamp and Lord (1923) we have proposed ‘rain forests’ of the Australian moist tropics (Figure 8(c), the use of the term ‘stunted forest’ to describe the vege- Adam 1992), and thus we would suggest that eucalypt for- tation in such situations. Often such forests will also have mations with a grassy understorey should be considered as some sort of grass layer, and hence we have introduced the savanna woodland (or tall savanna woodland), even though term ‘axylale rich’ as a modifier to be included (only for many are classified as ‘forests’ through the AUSLIG (1990) forests) when the herbaceous cover is > 10%, as for exam- system (Figure 1(a)). ple for RSC-01 (Figure 6(g)). As shown in Figure 5 and A second area of ambiguity is for South-East Asia as has already been noted by Killeen and Hinz (1992), the (Section 2.6.4), where it has been common to refer to grasses in such axylale-rich forests (as well as for closed ‘dry dipterocarp forest’ or ‘savanna forest’ for over a cen- woodlands such as NXV-02) are most likely of the C3 tury (Kurz 1875). These ‘forests’ are clearly of a structure photosynthetic pathway, which may be attributable to the much more usually associated with savanna formations relative advantage of C4 photosynthesis existing only at (Figure 1), including a clearly discernible grass layer (Stott high light levels (Grace et al. 1998; Sage and Kubien 2003). 1992). Nevertheless, it is understandable that they have This relative disadvantage of the C4 photosynthetic mode traditionally been considered as forests, as the dominant in the shade is perhaps also relevant to the increased domi- species of these formations are of the same genera as the nance of shrubs in the ground layer of the more open forests Dipterocarpaceae that dominate the tall forests of much of of the ZOT, as discussed in Section 6.1. the moister parts of South-East Asia (Whitmore 1984), with Delineation of tropical vegetation types 129

these ‘savanna forests’ merely being considered to con- Acrisol)’. Unusual vegetation formations can also be easily stitute the driest end of a long continuum (Blasco 1983). included – for example, the ‘bana’ vegetation of Venezuela However, the dipterocarp floristic similarity argument is not (a low forest typically found on podsolised sands with convincing, as there are numerous examples of genera con- severe and rapid fluctuations of the water table, see Cuevas taining species generally associated with both forest and and Medina, 1986) would be described as a ‘stunted ever- savanna in South America (Hoffmann et al. 2004, 2005a), green forest (c, tropical pluvial; l, ‘bana’; d, none; s, Umbric Australia (Adam 1992; Byrne et al. 2011) and Africa Podzol (Oxyaquic))’ – and with the word ‘bana’ simply (Plana 2004). Moreover, on the basis of numerous savanna- changed to ‘campinarana’ then describing an almost identi- like characteristics of the woody species dominating such cal vegetation formation sometimes occurring along sandy ‘forests’, including fire-tolerant traits, Ratnam et al. (2011) river terraces, such as near Manaus in Brazil (Anderson have argued that, dipterocarp or not, they should be consid- 1981), and with the simple substitution ‘l, kerangas’ also ered as savannas. We do not disagree with this, especially applying then to many forests of similar structure and soil as there may be virtually no species overlap between ‘dry type in Borneo (Whitmore 1984). dipterocarp’ forests and moister evergreen forests, even We also note that where the primary interest of the when located in close proximity (Lamotte et al. 1998). investigator is floristics and/or evolutionary history (e.g. Werneck 2011), then there is no problem in turning things around. For example, where phylogenetic linkages are 6.4. Refining the classification through the use of paramount but structure is still of interest, then taking additional prefix and suffix modifiers TUC-01 as an example, an appropriate description might Although we believe that the scheme proposed in Figure 9 be ‘seasonally dry tropical forest (deciduous, stunted and provides a good basis for the definition of tropical vegeta- axylale-rich)’. Although we also note that we do not per- tion types, as for many other schemes discussed we also sonally advocate the use of ‘deductive’ terms such as propose that not only could leaf habit be included as a ‘seasonally dry’, even for floristic associations, such addi- prefixed modifier, as suggested in Section 2.2, but that con- tions of the proposed clearly defined vegetation formation textual information could sometimes also be included in types as ancillary information in such studies may be brackets afterwards, this containing one or more of (as con- useful, especially as within any given ‘biome’ observed sidered necessary by the classifier) the following ‘clods’ vegetation formation types may vary widely, depending on scheme: local climatic conditions and rain fall patterns (Section 2.5 and Section 2.6). c = climatic information l = ‘local term’ o = any other relevant information d = dominant species, genus or family 6.5. Adequacy of coverage and application of the scheme s = soil type to other tropical vegetation types The range of vegetation formation types sampled as part For example, noting that NXV-02 (Figure 6(m)), clas- of this study can be seen through an examination of sified according to Figure 9 as a ‘tall closed woodland’ the labelled points on Figure 1 (a)–(c). This shows that characterised by an appreciable number of species usually for Australia, as our sampling was limited to around the associated with forest (Supplementary Information Table forest/savanna transition zone, our dataset does not include E2), is locally known as cerradão and was found on a any open woodlands or shrublands. Likewise, the low open typic Ferralsol soil growing under a tropical pluviseasonal forest and low closed forest formations were not sampled, Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 climate (as defined by Rivas-Martínez (1997) and with as is also the case for the open and closed scrub. Some of about half of its species deciduous, then this stand could be these formations do, however, occur in tropical Australia; described as ‘semi-evergreen tall closed woodland (c, trop- for example, the heathlands and nearby low forests of the ical pluviseasonal; l, cerradão; o, forest elements; d, none; sandier soils on the northern tip of Cape York Peninsula s, Haplic Ferralsol)’. (AUSLIG 1990). The advantage in taking such an approach for the For West Africa, from close to the southern margin provision of extra information is that a large amount of of the Sahara Desert to the tropical forest region proper contextual background about a site can be included but (see Figure 2) we managed to sample a wider range of without any notions of implied causality. Other examples heights and covers, more or less covering a diagonal from would be TUC-01 (Figure 6(f)) as a ‘stunted deciduous left to right on the structurogram but, with the exception axylale-rich forest (c, tropical pluviseasonal; l, chiquitano; of BBI-02 in Burkina Faso, failed to sample the scrub o, none, d, none; s, Cambisol)’ and HOM-02 (Figure 6(b)) or bushland/thicket vegetation types. The latter are most being a ‘deciduous grassland savanna (c, tropical xeric prevalent in Eastern Africa, especially in areas of intermit- pluviserotin; l, Sahelian savanna; d, Acacia sahelensis;s, tent rainfall (White 1983), also occurring in southern Africa Arenosol)’. As an example for South-East Asia, a Sabah south of the Tropic of Capricorn (Walter 1975). Similarly, forest might well be ‘evergreen tall forest (c, tropical plu- the positioning of the sample plots on the structurogram for vial; l, terra firme forest; o, none; d, Dipterocarpaceae; s, South America shows our sampling scheme covered a wide 130 M. Torello-Raventos et al.

range of vegetation types, but again predominantly a left- a vegetation formation type, then the ‘o = anthropogenic’ to-right diagonal on the structurogram and with no mea- suffixed modifier be employed. surements in the relatively low-stature Caatinga vegetation The basic measurement/quantification requirements of type which often has considerable canopy cover. our categorisation being quite modest and the grouping pro- One obvious defect in our study thus is that we did cess unambiguous, we hope that the scheme developed here not sample short vegetation types characterised by a rea- will prove to be of general use for both vegetation scientists sonably dense woody canopy cover, these generally being as well as biospheric modellers and/or remote-sensing sci- shrublands, thornscrub or thickets such as caatinga. These entists interested in applying a common methodology for vegetation formation types which normally do not have describing the many different tropical vegetation forms in a any appreciable grass layer present (Stamp and Lord 1923; simple and relatively euphonious manner. There often being Andrade-Lima 1981; White 1983) are, however, easily a considerable similarity between tropical and subtropical accounted for in the scheme of Figure 9 where descrip- forest, woodland, savanna and grassland formations (Walter tions for vegetation formation types without an appreciable 1975), we also see no reason why the scheme proposed herbaceous layer (ϕG ≤ 0.1) are of a more or less symmet- here could not be applied to areas beyond the equato- ric syntax with the savanna formations. Thus, for example, rial zone and perhaps even into the temperate region. Yet an African thorn scrub might be a ‘closed deciduous scrub- such an extension would clearly require some modifications land’ and, floristically being most closely associated with and/or extensions; for example to encompass tundra-like the (semi-)deciduous forest flora of South America (Daly formations. and Mitchell 2000), then an average statured caatinga stand of 5 m high would have its base classification ‘deciduous scrubland’ or ‘closed deciduous shrubland’. Other exam- 7. Conclusions ples would be grass-free areas of ‘mulga’ in Australia (e.g. We provide a new tropical vegetation classification scheme Burrows and Beale 1969) as typically ‘evergreen scrub- based on the structure (canopy cover and vegetation height, land’ or ‘evergreen shrub-rich woodland’. In addition, the specifically three parameters: total canopy area index of maritime and estuary vegetation referred to universally as all woody vegetation, fractional ground cover (herbs + ‘mangroves’ we would place under the forest or scrub grasses); and 0.95 quantile height of trees) and species categorisations, depending upon structural characteristics. composition. This scheme has been developed because nei- Although we included over 10 tropical forests in our ther structure nor stand-level floristics were found to be sampling scheme, it could be argued that this is in no way adequate on their own to provide an unequivocal categorisa- representative of global tropical forest diversity as a whole tion of tropical vegetation, especially in the forest/savanna and that the general classification scheme is through that ecotone where stands consisting predominantly of savanna defect somehow flawed. Yet, through the ‘clods’ scheme species can assume a forest-like stature. above, our classification gives at least as many structural- In this new scheme, developed through the separate physiognomic categories as have generally been proposed ordination of structural versus floristic characteristics with for the lowland forests of South-East Asia (Whitmore data taken from over 60 forest, savanna and grassland 1984), Africa (White 1983) and South America (UNESCO stands from three continents, the words ‘woodland’ and 1981), and with the general scheme having a flexibility ‘forest’ are used to denote specifically different floristic for any additional factors to be included (for example, a associations. The word ‘savanna’ is also employed in this high liana abundance, or the existence of isolated emergent context to define vegetation formation types where the trees). fractional herbaceous cover is greater than 0.1. Our trop- The extent to which the proposed scheme will be found ical vegetation scheme is also readily applicable to other Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 useful for other tropical forest types remains to be ascer- vegetation formation types of the tropics, such as thorn- tained, but we see no reason why, for example, it could not scrub. It also has potential applications outside the tropical be applied to montane forests. As illustration, a montane regions, but its utility in this respect has not yet tested. forest at 1700 m in Ecuador described along with its soils Future work should be directed towards assessing the utility by Grubb et al. (1963) would be classified as a ‘stunted of the scheme for tropical and subtropical vegetation for- shrub-rich forest (c, tropical pluvial; l, montane forest; mation types for which a ground layer is virtually absent, a s, Andosol)’. Likewise, ‘swamp forests’ such as are cur- general testing of its applicability in regions not covered by rently found in many part of South-East Asia, or seasonally this study, and in an examination of the feasibility of inte- flooded forests such as those that constitute about 10% of gration of the proposed vegetation formation type classifi- the Amazon Basin, should all be easily classified with the cation scheme to help place results from remote sensing and proposed scheme. As illustration, the Sumatran ‘alluvial regional/global modelling studies into a unified framework. swamp forest’ of CW = 2.2 described by Laumonier (1997) would classify as forest (c, pluvial; o, seasonally flooded, isolated emergents; s, Ombric Fluvisol). In the scheme Acknowledgments / of Figure 9 incorporating all possible tree grass combina- This work was funded by the UK Natural Environment Research tions, we would also suggest that where human influences Council through a TROBIT Consortium grant administered by have been clearly identified as the main factor structuring the University of Leeds. Simon Lewis was funded by a Royal Delineation of tropical vegetation types 131

Society University Research Fellowship and Elmar Veenendaal Borcard D, Gillet F, Legendre P. 2011. Numerical Ecology with received additional funding from the EU funded Geocarbon R. Berlin (Germany): Springer. project (nr. 283080). Part of the work in Mato Grosso, Brazil, was Bucher EH. 1982. Chaco and caatinga: South American arid funded by PROCAD/CAPES and we also acknowledge the sup- savanna, woodlands and thickets. In: Huntley BJ, Walker port and assistance of CSIR-Forestry Research Institute of Ghana BH, editors. Ecology of tropical savannas. New York (NY): (CSIR-FORIG) and Resource Management Support Center of the Springer-Verlag. p. 48–79. Ghana Forestry Commission (FC-RMSC). WCS-Cameroon and Bunyavejchewin S. 1983. Canopy structure of the dry dipterocarp J. Sonké provided logistical assistance in Cameroon and Annette forest in Thailand. Thai Forest Bulletin 14:1–132. den Holander provided fieldwork assistance in both Bolivia and Burrows WH, Beale IF. 1969, Structure and association in the Cameroon. Shiela Lloyd assisted with anger management as well Mulga (Acacia aneura) lands of south-western Queensland. as manuscript and figure preparation. Australian Journal of Botany 17:539–552. Burtt BD. 1942. Some East African vegetation communities. Journal of Ecology 30:65–146. Notes on contributors Burtt-Davy J. 1938. The classification of tropical woody vege- tation types. Institute Paper No. 13. Oxford (UK): Imperial Mireia Torello-Raventos was a Master of Science candidate at the Forestry Institute. 85 p. Centre for Ecosystem Studies at the University of Wageningen in Byrne M, Steane DA, Joseph L, Yeates DK, Jordan GJ, Crayn D, The Netherlands with the work described here starting as part of Aplin K, Cantrill DJ, Cook LG, Crisp MD, et al. 2011. her dissertation. She is currently a Ph.D. student at James Cook Decline of a biome: evolution, contraction, fragmentation, University in Cairns, Australia investigating soils and climate as extinction and invasion of the Australian mesic zone biota. modulators of tropical tree wood characteristics. Journal of Biogeography 38:1635–1656. All other authors form part of the Tropical Biomes in Campbell NA, Reece JB. 2002. Biology. 6th ed. San Francisco Transition (TROBIT) Consortium funded through the UK (CA): Benjamin Cummings. National Environment Research Council. A multidisciplinary Chabot BF, Hicks DJ. 1982. The ecology of leaf life spans. Annual consortium, TROBIT is concerned with developing through mix- Review of Ecology and Systematics 13:229–259. ture of fieldwork and modelling, a means to better predict any Champion HG, Seth SK. 1968. A revised survey of the forest types tropical biome shifts likely as a consequence of climate change. of India. Delhi (India): Manager of Publications. Champion HG. 1936. A preliminary survey of the forest types of India. Indian Forest Records Vol. 1. Chao A, Chazdon RL, Colwell RK, Shen T-J. 2005. A new References statistical approach for assessing similarity of species com- Ackerly D. 1996. Canopy structure and dynamics: integra- position with incidence and abundance data. Ecology Letters tion of growth processes in tropical pioneer trees. In: 8:148–159. Mulkey SS, Chazdon RL, Smith AP, editors. Tropical for- Cochrane TT, Cochrane TA. 2010. Amazon forest and savanna est plant ecophysiology. New York (NY): Chapman and Hall. lands: a guide to the climates, vegetation, landscapes, and p. 619–658. soils of central tropical South America. Seattle (WA): Acocks JPH. 1988. Veld Types of South Africa. 3rd Edition. CreateSpace. Memoirs of the Botanical Survey of South Africa 57:1–146. Cochrane TT, Sánchez LG, de Azevedo LG, Porras JA, Carver C. Pretoria (Sout Africa): Government Printer. 1985. Land in tropical America. Cali (Colombia): Centro Adam P. 1992. Australian Rainforests. Oxford (UK): Clarendon Internacional de Agricultura Tropical. Press. Cole MM. 1960. Cerrado, Caatinga and Pantanal: the distribution Anderson AB. 1981. White-sand vegetation of Brazilian and original of the savanna vegetation of Brazil. Geographical Amazonia. Biotropica 13:199–210. Journal 126:168–179. Anderson GD, Talbot LM. 1965. Soil factors affecting the distri- Colwell RK. 2009. EstimateS: statistical estimation of species bution of the grassland types and their utilization by wild ani- richness and shared species from samples. Available mals on the Serengeti Plains, Tanganyika. Journal of Ecology online at http://viceroy.eeb.uconn.edu/EstimateS (accessed 53:33–56. 20 February 2013). Andrade-Lima D. 1981. The caatinga dominium. Revista Coomes DA, Grubb PJ. 1996. Amazonian caatinga and related Brasileira de Botanica 4:149–153. communities at La Esmeralda. Venezuela: forest structure.

Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 AUSLIG (Australian Surveying and Land Information Group). physiognomy and floristics, and control by soil factors. 1990. Atlas of Australian resources. Canberra (Australia): Vegetatio 122:167–191. Australian Surveying and Land Information Group. Corlett RT. 2009. The ecology of tropical East Asia. Oxford (UK): Beard JS. 1953. The savanna vegetation of northern tropical Oxford University Press. America. Ecological Monographs 23:149–215. Coutinho LM. 1978. O conceito de cerrado. [The cerrado concept] Beard JS. 1955. The classification of tropical American Revista Brasileira de Botânica 1:17–23. vegetation-types. Ecology 36:89–100. Cronje HP, Panagos MD, Reilly BK. 2008. The plant commu- Berry SL, Roderick ML. 2002. Estimating mixtures of leaf func- nities of the Andover Game Reserve, South Africa. Koedoe tional types using continental-scale satellite and climatic data. 50:184–201. Global Ecology and Biogeography 11:23–39. Cuevas E, Medina E. 1986. Nutrient dynamics within Amazonian Blasco F. 1983. The transition from open forest to savanna in forests. I. Nutrient flux in fine litter fall and efficiency of continental Southeast Asia. In: Bourlière F, editor. Tropical nutrient utilisation. Oecologia 68:466–472. savannas. Amsterdam (Netherlands): Elsevier. p. 167–181. Daly DC, Mitchell JD. 2000. Tropical lowland vegetation of Blasco F, Whitmore T, Gers C. 2000. A framework for the South America – an overview. In: Lentz D, editor. Imperfect worldwide comparison of tropical woody vegetation types. balance: Landscape transformations in the pre-Columbian Biological Conservation 95:175–189. Americas. New York (NY): Columbia University Press. p. Boland DJ, Brooker MIH, Chippendale GM, Hall N, Hyland 391–453. BPM, Johnston RD, Kleinig DA, McDonald MW, Turner JD. Dansereau P. 1951. Description and recording of vegetation upon 2006. Forest trees of Australia. Canberra (Australia): CSIRO. a structural basis. Ecology 32:172–229. 132 M. Torello-Raventos et al.

De Amorim IL, Sampaio EVSB, de Lima Araújo E. 2005. and natural history of a neotropical savanna. New York (NY): Flora e estrutura da vegetação arbustivo-arbórea de uma área Columbia University Press. p. 178–197. de caatinga do Seridó, RN, Brasil. Acta Botanica Brasilica Franczak DD, Marimon BS, Marimon-Junior BH, Mews HA, 19:615–623. Maracahipes L, Oliveira EA. 2011. Changes in the structure De Laubenfels DJ. 1975. Mapping the world’s vegetation: region- of a savanna forest over a six-year period in the Amazon- alization of formations and flora. Syracuse (NY): Syracuse Cerrado transition, Mato Grosso state, Brazil. Rodriguésia University Press. 62:425–436. DeFries RS, Townshend JRG. 1994a. Global land cover: compar- Furley PA, Ratter JA. 1988. Soil resources and plant communi- ison of ground based data sets to classifications with AVHRR ties of the Central Brazilian cerrado and their development. data. In: Foody GM, Curran PJ, editors. Environmental Journal of Biogeography 15:97–108. Remote Sensing from Regional to Global Scales. New York Gentry A. 1995. Diversity and floristic composition of neotropical (NY): Wiley. p. 84–110. dry forests. In: Bullock SH, Mooney HA, Medina E, editors. DeFries RS, Townshend JRG. 1994b. NDVI-derived land cover Seasonally dry tropical forests. Cambridge (UK): Cambridge classifications at a global scale. International Journal of University Press. p. 146–194. Remote Sensing 15:3567–3586. Gentry AH. 1988. Changes in plant community diversity and Domingues TF, Meir P, Feldpausch TR, Saiz G, Veenendaal EM, floristic composition on environmental and geographical Schrodt F, Bird M, Djagbletey G, Hien F, Compaore H, et al. gradients. Annals of the Missouri Botanic Garden 75: 2010. Co-limitation of photosynthetic capacity by nitrogen 1–34. and phosphorus in West Africa woodlands. Plant Cell and Gillison AN. 1994. Woodlands. In: Groves RH, editor. Australian Environment 33:959–980. vegetation. Cambridge (UK): Cambridge University Press. Du Rietz GE. 1931. Life-forms of terrestrial flowering plants. I. p. 227–255. Acta Phytogeographica Suecica 3:95. Goodland R, Pollard R. 1973. The Brazilian cerrado vegetation: a Dyksterhuis EJ. 1957. The savannah concept and its use. Ecology fertility gradient. Journal of Ecology 61:219–224. 38:435–442. Gossweiler J. 1939. Carta Fitogeográphica de Angola. [Phyto- Eiten G. 1968, Vegetation Forms. A Classification of stands of geographical Map of Angola] Luanda (Angola): Gov. Geral vegetation based on structure, growth forms of the com- de Angola. 242 p. ponents and vegetative periodicity. Boletim do Instituto Gower JC. 1971. A general coefficient of similarity and some of de Botânoci, vol. 4., São Paulo (Brazil): Secretaria da its properties. Biometrics 27:857–874. Agricultura. 88 p. Grace J, Lloyd J, Miranda AC, Miranda H, Gash JHC. 1998. Eiten G. 1972. The cerrado vegetation of Brazil. Botanical Review Fluxes of carbon dioxide and water vapour over a C4 pas- 38:201–341. ture in southwestern Amazonia (Brazil). Australian Journal Eiten G. 1983. Classificação da Vegetação do Brasil. Brasilia of Plant Physiology 25:519–530. (Brazil): CNPq. Greenway PJ. 1973. A classification of the vegetation of East Eiten G. 1986. The use of the term “savanna”. Tropical Ecology Africa. Kirkia 9:1–68. 27:10–23. Grubb PJ. 1977. Control of forest growth and distribution on Eiten G. 1992. How names are used for vegetation. Journal of wet tropical mountains: with special reference to mineral Vegetation Science 3:419–424. nutrition. Annual Review of Ecology and Systematics 8: Ellenberg H, Mueller-Dombois D. 1966. Tentative physiognomic- 83–107. ecological classification of plant formations of the earth. Grubb PJ, Lloyd JR, Pennington TD, Whitmore TC. 1963. A com- Bericht uber das Geobotanische Forschungsinstitut Rubel parison of montane and lowland rain forest in Ecuador. I. 37:21–55. The forest structure, physiognomy, and floristics. Journal of Eva HD, de Miranda EE, Di Bella CM, Gond V, Huber O, Ecology 51:567–601. Sprenzaroli M, Jones S, Coutinho A, Dorado A, Guimararães Haase R, Beck SG. 1989. Structure and composition of savanna M, et al. 2002. A vegetation map of South America. vegetation in northern Bolivia: a preliminary report. Brittonia European Commission EUR 20159 EN. Luxembourg: Office 41:80–100. for Official Publications of the European Communities. Hall-Martin AJ. 1975. Classification and ordination of forest and Eyre SR. 1963. Vegetation and soils: a world picture. London thicket vegetation of the Lengwe National Park, Malawi. (UK): Edward Arnold. Kirkia 10:131–184. FAO. 2010. Global forest resources assessment 2010: main report. Hansen MC, DeFries RS, Towshend JRG, Marfu L, Sohlberg R.

Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 Rome (Italy): FAO. 2008. Development of MODIS tree cover validation data Feldpausch TR, Banin L, Phillips OL, Baker TR, Lewis SL, set for Western Province, Zambia. Remote Sensing of Quesada CA, Affum-Baffoe K, Arets EJMM, Berry NJ, Environment 83:320–335. Bird M, et al. 2011. Height-diameter allometry of tropical Hansen MC, DeFries RS, Townshend JRG, Sohlberg R. 2000. forest trees. Biogeosciences 8:1081–1106. Global land cover classification at 1 km spatial resolution Fensham RJ. 1995. Floristics and environmental relations of using a classification tree approach. International Journal of inland dry rainforest in North Queensland, Australia. Journal Remote Sensing 21:1331–1364. of Biogeography 22:1047–1063. Harrison MN, Jackson JK. 1958. Ecological classification of the Fisher R, McDowell N, Purves D, Moorcroft P, Sitch S, Cox P, vegetation of the Sudan. Forests Bulletin of Sudan (New Huntingford C, Meir P, Woodward FI. 2010. Assessing uncer- series) 2:1–45. tainties in a second-generation dynamic vegetation model Herlocker DJ. 1975. Woody vegetation of the Serengeti National caused by ecological scale limitations. New Phytologist Park. College Station (TX): Texas Agricultural Experiment 187:666–681. Station, Texas A&M University. Fosberg RF. 1961. A classification of vegetation for general Hiernaux P, Lassine D, Trichon V, Mougin E, Baup F. 2009. purposes. Tropical Ecology 2:1–28. Woody plant population dynamics in response to climate Fox ID, Neldner VJ, Wilson GW, Bannik PJ. 2001. Vegetation changes from 1984 to 2006 in Sahel (Gourma, Mali). Journal of the Australian tropical savannas. Brisbane (Australia): of Hydrology 375:103–113. Environmental Protection Agency. Hoffmann WA, Orthen B, Franco AC. 2004. Constraints to Franco AC. 2002. Ecophysiology of woody plants. In: Oliveira seedling success of savanna and forest trees across the PS, Marquis RJ, editors. The Cerrados of Brazil. Ecology savanna-forest boundary. Oecologia 140:252–260. Delineation of tropical vegetation types 133

Hoffmann WA, da Silva Jr ER, Machado GC, Bucci SJ, Scholz Killeen TJ, Hinz PN. 1992. Grasses of the Precambrian FG, Goldstein G, Meinzer FC. 2005b. Seasonal leaf dynam- shield region in eastern lowland Bolivia. II. Life-form and ics across a tree density gradient in a Brazilian savanna. C3-C4 photosynthetic types. Journal of Tropical Ecology 8: Oecologia 145:307–316. 389–407. Hoffmann WA, Franco AC, Moreira MZ, Haridasan M. 2005a. Killeen TJ, Jardim A, Mamaini F, Rojas N. 1998. Diversity, com- Specific leaf area explains differences in leaf traits between position and structure of a tropical semideciduous forest in the congeneric savanna and forest trees. Functional Ecology Chiquitanía region of Santa Cruz, Bolivia. Journal of Tropical 19:932–940. Ecology 14:803–827. Holdridge LR. 1947. Determination of world plant formations Killeen TJ, Siles TM, Grimwood T, Tieszen LL, Steininger MK, from simple climatic data. Science 105:367–368. Tucker CJ, Panfil SN. 2001. Habitat heterogeneity on a forest- Holdridge LS, Crenke WC, Hatheway WH, Liang TM, Tosi JA. savanna ecotone in Noel Kempff Mercado National Park 1971. Forest environments in tropical life zones. A pilot (Santa Cruz, Bolivia); implications for the long-term conser- Study. Oxford (UK): Permagon Press. vation of biodiversity in a changing climate. In: Bradshaw Holm T. 1896. On the prevailing ombrophilous character GA, Marquet P, editors. How landscapes change: human of the foliage of tropical plants. Botanical Gazette 21: disturbance and ecosystem disruptions in the Americas. 163–164. Ecological Studies, Vol. 162. Berlin (Germany): Springer- Honorio Coronado EN, Baker TR, Phillips OL, Pitman NCA, Verlag. p. 285–312. Pennington RT, Vásquez Martínez R, Monteagudo A, Koy K, McShea WJ, Leimbruber P, Haack BN, Aung M. 2005. Mogollón H, Dávila Cardozo N, Rios M, et al. 2009. Multi- Percentage canopy cover – using Landsat imagery to delineate scale comparisons of tree composition in Amazonian terra habitat for Myanmar’s endangered Eld’s deer (Cervus eldi). firme forests, Biogeosciences 6:2719–2731. Animal Conservation 8:289–296. Hopkins B. 1966. A field key to the savanna trees of Nigeria. Küchler AW. 1949. A physiognomic classification of vegeta- Nairobi (Kenya): Ibadan University Press. tion. Annals of the Association of American Geographers Huber O. 1995. Vegetation. In: Berry PE, Holst BK, 39:201–210. Yatskeievych K, editors. Flora of the Venezuelan Guyana, Kurz S. 1875. Preliminary report on the forest and other vege- Vol. 1 Introduction. St Louis (MO): Missouri Botanic Garden tation of Pegu. Calcutta (India): CB Lewis, Baptist Mission Press. p. 97–160. Press. Hult R. 1881. ’Försök till analytisk behandling av vaxtforma- Lamotte S, Gajaseni J, Malaisse F. 1998. Structure diversity tionerna. Societas pro Fauna et Flora Fennica Meddelanden in three forest types on north-eastern Thailand (Sakaerat 8:1–155 [cited in Nicholson 1996]. Reserve, Pak Tong Chai). Biotechnology, Agronomy, Society Humboldt A von. 1805. Essai sur la géographié des plantes; and Environment 2:192–202. accompagné d’un tableau physique des régions équinoxiales. Laumonier Y. 1997. The vegetation and physiography of Sumatra Par Al. de Humboldt et A. Bonpland, redigé par Al. de (Geobotany 22). Kluwer. 222 p. Humboldt. Paris (France): Levrault, Schoell et Cie. 155 pp. Lawesson JE. 1994. Some comments on the classification of Hundley HG. 1961. The forest types of Burma. Tropical Ecology African vegetation. Journal of Vegetation Science 5:441–444. 2:48–76. Legendre P, Gallagher ED. 2001. Ecologically meaningful Hutley LB, O’Grady AP, Eamus D. 2000. Evapotranspiration transformations for ordination of species data. Oecologia from Eucalypt open-forest savanna of Northern Australia. 129:271–280. Functional Ecology 14:183–194. Leemans R, Cramer W, van Minnen JG. 1997. Prediction of the Ingrouille MJ, Eddie B. 2006. Plants: evolution and diversity. global biome distribution using bioclimatic equilibrium mod- Cambridge (UK): Cambridge University Press. els. In: Mellilo JM, Breymeyer A, editors. Carbon cycling in IUSS (International Union of Soil Science). 2006. World ref- grassland and forest ecosystems. New York (NY): Wiley. erence base for soil resources: a framework for interna- Lehmann CER, Archibald SA, Hoffmann WA, Bond WJ. 2011. tional classification, correlation and communication. Working Deciphering the distribution of the savanna biome. New Group WRB World Soil Resources Report 103. Rome (Italy): Phytologist 191:197–209. FAO. Lewis SL, Lopez-Gonzalez G, Sonké B, Affum-Baffoe K, Baker Johnson HS, Hatch MD. 1968. Distribution of the C4-dicarboxylic TR, Ojo LO, Phillips OL, Reitsma J, White L, Comiskey J, acid pathway of photosynthesis and its occurrence in dicotyle- et al. 2009. Increasing carbon storage in intact African tropi- donous plants. Phytochemistry 7:375–380. cal forests. Nature 457:1003–1006.

Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 Josse C, Navarro G, Encarnación F, Tovar A, Comer P, Ferreira Lloyd J, Bird MI, Vellen L, Miranda AC, Veenendaal EM, W, Rodríguez F, Saito J, Sanjurjo J, Dyson J, et al. 2007. Djagbletey G, Miranda HS, Cook G, Farquhar GD. 2008. Ecological systems of the Amazon Basin of Peru and Bolivia. The relative contributions of woody and herbaceous vegeta- Classification and mapping. Arlington (VA): NatureServe. tion to tropical savanna ecosystem productivity: field studies Kaufman L, Rousseeuw PJ. 2008. Introduction. In: Finding groups in Australia, Brazil and Ghana and a quasi-global estimate. in data. John Wiley & Sons, Inc. p. 1–67. Available online at: Tree Physiology 28:451–468. http://dx.doi.org/10.1002/9780470316801.ch1 (accessed 20 LloydJ,GouldenM,OmettoJP,FyllasNM,QuesadaCA, February 2013). Patino S. 2009. Ecophysiology of forest and savanna vege- Keay RWJ. 1949. An example of Sudan zone vegetation in tation In: Amazonia and climate change. Washington (DC): Nigeria. Journal of Ecology 37:335–364. American Geophysical Union. p. 463–484. Kikuzawa K. 1991. A cost benefit analysis of leaf habit and leaf Loveland TR, Belward AS. 1997, The IGBP-DIS global 1 km land longevity of trees and their geographic pattern. American cover data set, DISCover: first results. International Journal of Naturalist 138:1250–1263. Remote Sensing 18:3289–3295. Killeen TJ, Chavez E, Pena-Claros M, Toledo M, Arroyo L, Loveland TR, Reed BC, Brown JF, Ohlen DO, Zhu Z, Yang L, Caballero J, Correa L, Guillen R, Quevedo R, Saldias M, et al. Merchant JW. 2000. Development of a global land cover char- 2006. The Chiquitano dry forest, the transition between humid acteristics database and IGBP DISCover from 1 km AVHRR and dry forest in Eastern Lowland Bolivia. In: Pennington data. International Journal of Remote Sensing 21:1303–1330. RT, Ratter JA, editors. Neotropical savannas and dry forests: Low AB, Rebelo AG. 1998. Vegetation of South Africa, Lesotho diversity, biogeography and conservation. Boca Raton (FL): and Swaziland. Pretoria (South Africa): Department of CRC Press. p. 205–225. Environmental Affairs and Tourism. 85 p. 134 M. Torello-Raventos et al.

McKenzie N, Jacquier D, Isbell R, Brown K. 2004. Australian Paijmans K. 1975. Explanatory notes to the vegetation map of soils and landscapes: an illustrated compendium. Papua New Guinea. Melbourne (Australia): CSIRO. Collingwood (Australia): CSIRO Publishing. Pennington RT, Lavin M, Prado DE, Pendry CA, Pell SK, MacKinnon K, Hatta G, Halim H, Mangalik A. 1996. The ecology Butterworth CA. 2004. Historical climate change and speci- of Kalimantan. Oxford (UK): Oxford University Press. ation: neotropical seasonally dry forest plants show patterns MacNaughton-Smith PNM. 1965. Some statistical and other of both Tertiary and Quaternary diversification. Philosophical numerical techniques for classifying individuals. London transactions of the Royal Society of London. Series B: (UK): HMSO. Biological Sciences 359:515–538. Maechler M, Rousseeuw P, Struyf A, Hubert M. 2005. Cluster Penridge LK, Walker J. 1988. The crown gap ratio (C) and cover analysis basics and extensions. Unpublished. percent. 2. Studies of the properties of C. Australian Journal Malhi Y, Aragao LEOC, Galbraith D, Huntingford C, Fisher R, of Ecology 13:109–120. Zelazowski P, Sitch S, McSweeney C, Meir P. 2009. Phillips OL, Baker TR, Brienen R, Feldpausch TR. 2010. Exploring the likelihood and mechanism of a climate-change- Field manual for plot establishment and remeasurement. induced dieback of the Amazon rainforest. Proceedings of the Available online at:http://www.geog.leeds.ac.uk/projects/ National Academy of Sciences 106:20610–20615. rainfor/ (accessed 20 February 2013). Marimon-Junior BH, Haridasan M. 2005. Comparação da veg- Pinder L, Rosso S. 1998. Classification and ordination of etação arbórea e características edáficas de um cerradão e plant formations in the Pantanal of Brazil. Plant Ecology um cerrado sensu stricto em áreas adjacentes sobre solo 136:151–165. distrófico no leste de Mato Grosso, Brasil. [Comparison of Plana V.2004. Mechanisms and tempo of evolution in the African woody vegetation and soil characteristics for adjacent cer- Guineo-Congolian rainforest Philosophical Transactions of radão and a cerrado sensu stricto . . . .] Acta Botanica Brasilica the Royal Society of London Series B 359:1585–1594. 19:913–926. Prance GT, Schaller GB. 1982. Preliminary study of some vege- Martin B, Cossalter C. 1976. Les eucalyptus des Iles de la Sonde. tation types of the Pantanal. Brittonia 34:228–251. Bios et Forêts des Tropiques 166:3–22. Quesada CA, Lloyd J, Anderson LO, Fyllas NM, Schwarz M, Matthews E. 1983. Global vegetation and land use: new high res- Czimczik CI. 2011. Soils of Amazonia with particular refer- olution data bases for climate studies. Journal of Climate and ence to the RAINFOR sites. Biogeosciences 8:1415–1440. Applied Meteorology 22:474–487. Quesada CA, Lloyd J, Schwarz M, Patiño S, Baker TR, Czimczik Maxwell JF. 2004. A synopsis of the vegetation of Thailand. C, Fyllas NM, Martinelli L, Nardoto GB, Schmerler J, et al. Natural History Journal Chulalongkorn 4:19–29. 2010. Chemical and physical properties of Amazon forest McQuitty LL. 1960. Hierarchical linkage analysis for the isola- soils in relation to their genesis. Biogeosciences 7:1515–1541. tion of types. Educational and Psychological Measurement Ratnam J, Bond WJ, Fensham RJ, Hoffmann WA, Archibald S, 20:55–67. Lehmann CER, Anderson MT, Higgins SI, Sankaran M. Méio BB, Freitas CV, Jatobá L, Silva M, Ribiero JF, Henriques 2011. When is a ‘forest’ a savanna, and why does it matter? RPB. 2003. Influência da flora das florestas Amazônica e Global Ecology and Biogeography 20:653–660. Atlântica na vegetação do cerrado sensu stricto. [Influence Ratter JA. 1992. Transition between cerrado and forest vegeta- of the flora of the Atlantic and Amazon forests on the veg- tion in Brazil. In: Furley PA, Proctor J, Ratter JA, editors. etation of the cerrado] Revista Brasileira de Botânica 26: Nature and dynamics of forest-savanna boundaries. London 437–444. (UK): Chapman and Hall. p. 417–429. Monk KA, de Fretes Y, Reksodiharjo-Lilley G. 1997. The Ratter JA, Askew GP, Mpntgomery RF, Gifford DR. 1977. Ecology of Nusa Tenggara and Maluku (Ecology of Indonesia Observacões adicionais sobre o cerradão de solos mesotró- Series, Vol 5). Hong Kong: Periplus. 966 p. ficeos no Brasil Central [Additional observations on the Moss CE. 1910. The fundamental unit of vegetation. New cerrado vegetation on mesotrophic soils in central Brazil]. In: Phytologist 9:18–53. Ferri MG (co-ord). IV Simpósio sobre o Cerrado. São Paulo Mougin E, Hiernaux P, Kergoat L, Grippa M, de Rosnay P, (Brazil): Ed. USP, p. 303–316. Timouk F, Le Dantec V, Demarez V, Lavenu F, Arjounin A, Ratter JA, Bridgewater S, Ribeiro JF. 2003. Analysis of the et al. 2009. The AMMA Gourma observatory site in Mali: floristic composition of the Brazilian cerrados vegetation III: Relating climatic variations to changes in vegetation, surface comparison of the woody vegetation of 376 areas. Edinburgh hydrology, fluxes and natural resources. Journal of Hydrology Journal of Botany 60:57–109. 375:14–33. Ratter JA, Bridgewater S, Ribeiro JF. 2006. Biodiversity patterns

Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 Mueller-Dombois D, Ellenberg H. 1974. Aims and methods of of the woody vegetation of the Brazilian cerrados. In: vegetation ecology. New York (NY): John Wiley and Sons. Pennington RT, Lewis GP, Ratter JA, editors. Neotropical Nichols GE. 1923. A working basis for the ecological classifica- savannas and seasonally dry forests: plant diversity, tion of plant communities. Ecology 4:11–23, 154–179. biogeography and conservation. Boca Raton (FL): CRC Nicolson M. 1996. Humboldtian plant geography after Humboldt: Press. the link to ecology. The British Journal for the Hiostory of Ratter JA, Bridgewater S, Ribeiro JF, Fonsésca-Filho J, Rodrigues Science 29:289–310. da Silva M, Milliekn W, Pullman M, Pott A, Olivviera-Filho Oliveira-Filho AT, Ratter JA. 2002. Vegetation physiognomies and AT, Durigan G, et al. 2011. Analysis of the floristic com- woody flora of the cerrado biome. In: Oliveira PS, Ratter position of the Brazilian Cerrado vegetation IV: revision of JA, editors. The Cerrados of Brazil. Ecology and natural his- woody vegetation of 376 areas and presentation of a revised tory of a neotropical savanna. New York (NY): Columbia database of 367 areas. Available online at http://cerrado.rbge. University Press. p. 91–120. org.uk (accessed 20 February 2013). Oliver J. 1979. A study of geographical imprecision: the tropics. Ratter JA, Richards PW, Argent G, Gifford DR. 1973. Australian Geographical Studies 17:3–17. Observations on the vegetation of north eastern Mato Grosso, Olson JS, Watts J, Allison L. 1983, Carbon in live vegeta- 1. The woody vegetation types of the Xavantina-Cachimbo tion of major world ecosystems.ReportW-7405-ENG-26. Oak Expedition Area. Philosophical Transactions of the Royal Ridge (TN): US Department of Energy, Oak Ridge National Society of London, Series B 226:449–492. Laboratory. Raunkiaer C. 1934. The life forms of plants and statistical plant Orr DM. 1975. A review of Astrelba (Mitchell Grass) pastures in geography, being the collected papers of C. Raunkiær. Oxford Australia. Tropical Grasslands 9:21–36. (UK): Oxford University Press. Delineation of tropical vegetation types 135

Reich PB, Uhl C, Walters MB, Prugh, L. Ellsworth DS. 2004. Veenendaal EM, Mantlana KB, Pammenter NW, Weber P, Leaf demography and phenology in Amazonian rain forest: Huntsman-Mapila P, Lloyd J. 2008. Growth form and sea- a census of 40,000 leaves of 23 tree species. Ecological sonal variation in leaf gas exchange of Colophospermum Monographs 74:3–23. mopane savanna trees in northwest Botswana. Tree Ribeiro JF, Walter BMT. 1998. Fitofisionomias do bioma cerrado Physiology 28:417–424. [Physiognomic forms of the cerrado biome] In: Sano SM, Veldman JW, Putz FE. 2011. Grass-dominated vegetation, not de Almeida SP, editors. Cerrado: ambiente e flora. Planaltina species-diverse natural savanna, replaces degraded tropical (Brazil): EMBRAPA-CPAC, p. 89–166. forests on the southern edge of the Amazon Basin. Biological Richards PW.1957. Ecological Notes on West African Vegetation: Conservation 144:1419–1429. I. The plant communities of the Idanre Hills, Nigeria. Journal Veloso HP, Rangel Filho ALR, Lima JCA. 1991. Classificação of Ecology 45:563–577. da vegetação brasiliera, adapta a um sistems universal Richards PW. 1996. The tropical rain forest: an ecological study. [Classification of the vegetation of Brazil ...]. Rio de Janiero Cambridge (UK): Cambridge University Press. (Brazil): MEFP/IBGE/DRNEA. Richards PW, Tansley AG, Watt AS. 1940. The recording of Vidal J. 1960a. Le Végétation du Laos, 2ème Partie: groupements structure, life form and flora of tropical forest communities végétaux et flore. Toulouse (France): Travaux du Laboratoire as a basis for their classification. Journal of Ecology 28: Forestier de Toulouse 1–455. 224–239. Vidal J. 1960b. Les forêts du Laos. Bios et Forêts des Tropiques Rivas-Martínez S. 1997. Clasificación bioclimática de la Tierra. 70:5–21. [Bioclimatic classification of the earth] Itinera Geobotanica. Walker J, Gillison AN 1982. Australian savannas. In: Huntley BJ, 10. León. Walker BH, editors. Dynamics of savanna ecosystems. New Sage RF, Kubien DS. 2003. Quo vadis C4? An ecophysiologi- York (NY): Springer-Verlag. p. 5–24. cal perspective on global change and the future of C4 plants. Walker J, Hopkins MS. 1990. Vegetation. In: McDonald RC, Photosynthesis Research 77:209–225. Isbell RF, Speight JG, Walker J, Hopkins MS, editors. Saiz G, Bird MI, Domingues T, Schrodt F, Schwarz M, Australian soil and land survey field handbook. Melbourne Feldpausch TR, Veenendaal EM, Djagbletey G, Hien F, (Australia): Inkata Press. p. 58–86. Compaore H, et al. 2012. Variation in soil carbon stocks Walter H. 1975. Vegetation of the earth in relation to climate and their determinants across a precipitation gradient in West and the eco-physiological conditions. London (UK): English Africa. Global Change Biology 18:1670–1683. Universities Press. Santisuk T. 1988. An account of the vegetation of north- Watson L, Dallwitz MJ. 1992 onwards. The grass genera ern Thailand (Geoecological Reserach Vol. 5). Stuttgart of the world: descriptions, illustrations, identification, and (Germany): Steiner-Verlag. 101 p + 74 photographs. information retrieval; including synonyms, morphology, Särkinen T, Ignaci JRV, Linares-Palomino R, Simon MF, Prado anatomy, physiology, phytochemistry, cytology, classifica- DE. 2011. Forgotten forests – issues and prospects in biome tion, pathogens, world and local distribution, and references. mapping using seasonally dry tropical forests as a case study. Version: 23rd April 2010. Available online at http://delta- BMC Ecology 11:27. intkey.com (accessed 20 February 2013). Sarmiento G. 1984. The ecology of neotropical savannas. Webb LJ. 1959. A physiognomic classification of Australian rain Cambridge (MA): Harvard University Press. forests. Journal of Ecology 47:551–570. Schimper AFW.1903. Plant geography upon a physiological basis. Werneck FP. 2011. The diversification of eastern South American Jena (Germany): Gustaf Fischer. open vegetation biomes: historical biogeography and perspec- Schnell R. 1952. Vegetation et flore de la region montagneuse du tives. Quaternary Science Reviews 30:1630–1648. Nimba. Mem. IFAN 22:1–604. White F. 1983. The vegetation of Africa: a descriptive memoir to Sellers PJ, Los SO, Ticker CJ, Justice CO, Dazlich DA, Collatz accompany the UNESCO/AETFAT/UNSO vegetation map GJ, Randall DA. 1996. A revised land surface paramater- of Africa. Paris (France): UNESCO. 356 p. ization (SiB2) for atmospheric GCMs II. The generation Whitmore TC. 1984. Tropical rain forests of the Far East. Oxford of global fields of terrestrial biophysical parameters from (UK): Clarendon Press. 352 p. satellite data. Journal of Climate 9:706–737. Whittaker RH. 1962. Classification of natural communities. Smith TB, Wayne RK, Girman DJ, Bruford MW. 1997 A role Botanical Review 28:1–239. for ecotones in generating rainforest biodiversity. Science Whitten T, Henderson GS, Mustafa M. 2001. The Ecology of 276:1855–1857. Sulawesi. Tokyo (Japan): Tuttle Publishing.

Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 Sokal R, Sneath P. 1963. Numerical taxonomy. San Francisco Whitten T, Soeriaatmadja RE, Afiff SA. 1997. The Ecology of (CA): W.H. Freeman. Java and Bali. Oxford (UK): Oxford University Press. Specht RL. 1972. Water use by perennial evergreen plant commu- Wiesner J. 1894. Pflanzenphysiol. Mittheilungen aus Buitenzorg. nities in Australia and Papua New Guinea. Australian Journal III: Ueber den vorherrschend ombrophilen Charakter des of Botany 20:273–299. Laubes der Tropengewoechse. Sitzungsber. Denkschriften der Stamp LD, Lord L. 1923. The ecology of part of the riverine tract Kaiserlichen Akademie der Wissenschaften, Mathematisch- of Burma. Journal of Ecology 11:129–159. Naturwissenschaftliche Klasse 103:169–191. Stott P. 1992. The savanna forests of mainland southeast Wikramanayake E, Dinerstein E, Loucks CJ, Olson DM, Morrison Asia: an ecological survey. Progress in Physical Geography J, Lamoreux J, McKnight M, Hedao P. 2002. Terrestrial 8:315–334. ecoregions of the Indo-Pacific. A conservation assessment. Tani A, Ito E, Kanzaki M, Ohta S, Khorn S, Pith P, Tith B, Washington (DC): Island Press. Pol S, Lim S. 2007. Principal forest types of three regions Wild H, Fernandes A. 1967. Vegetation map of the Flora of Cambodia: Kampong Thom, Kratie, and Mondolkiri. In: Zambesiaca area. Salisbury (Rhodesia): MO Collins. With a Sawada H, Araki M, Chappell NA, LaFrankie JV, Shimizu 71-page supplement. A, editors. Forest environments in the Mekong River Basin. Williams L. 1965. Vegetation of Southeast Asia. Studies of for- Tokyo (Japan): Springer-Verlag. p. 201–213. est types. 1963–1965. Washington (DC): US Department of Tansley AG. 1913. A universal classification of plant- Agriculture, Agricultural Research Service. 302 p. communities. Journal of Ecology 1:27–42. Williams LJ, Bunyavejchewin S, Baker PJ. 2008. Deciduousness UNESCO 1981. Vegetation map of South America: explanatory in a seasonal tropical forest in western Thailand: interannual notes. Paris (France): United Nations Educational, Scientific and intraspecific variation in timing, duration and environ- and Cultural Organisations. mental cues. Oecologia 155:571–582. 136 M. Torello-Raventos et al.

Wilson KL, Bruhl JJ. 2007. Towards a comprehensive survey α is the average proportion of skylight passing through each tree. of C3 and C4 photosynthetic pathways in Cyperaceae. Aliso Across a range of savanna sites in Australia, Brazil and Ghana, 23:99–148. Lloyd et al. (2008) visually estimated α to range from 0.39–0.60. Wilson MF, Henderson-Sellers A. 1985. A global archive of land On the other hand, looking only at vegetation taller than 5 m high cover and soils data for use in general circulation models. across a range of sites in Zambia, Hansen et al. (2008) suggest Journal of Climatology 5:119–143. a general value of 0.76. It seems intuitive that α should increase Wu Y, Strahler AH. 1994. Remote estimation of crown size, with ς as can be seen from the bottom axes in Figure A1. Note stand density and biomass on the Oregon transect. Ecological also that canopy area index as defined here differs from crown Applications 4:299–312. area index as recently referred to by Bohlman and Pacala (2012), the latter being defined there as the number of crowns per unit ground area. Appendix 1. Relationship between foliar projective cover, canopy cover, crown cover and canopy area index Various measures of foliar cover have been presented in the liter- ature, the relationships between which are presented here. Firstly, we consider the projected area of tree and shrub crowns (includ- ing within-crown light gaps) per unit ground area as CW which, as it is expressed as per unit ground area we denote as canopy area index. This is a measure sometimes used by ecologists, for example, being equivalent to what De Amorim et al. (2005) refer to as ‘total crown area’. Foresters, however, often prefer to use themeasureofcrown cover which can be defined as the fraction of ground covered by crowns (including within-crown light gaps), ς. As a general result: for a number of overlapping figures with total area a, overlapping freely within area A, the binomial theo- rem gives the covered proportion as being equal to1–e−a/A (e.g. Wu and Strahler 1994) and it therefore simply follows that

ς = 1 − exp−CW ,(A1)

and with the relationship between ς and CW so predicted shown in Figure A1 below. Some remote-sensing products claim to detect canopy cover rather than crown cover, the former being defined as the portion of the skylight orthogonal to the surface which is intercepted by trees (e.g. Hansen et al. 2002). Defining then α as the proportion of light intercepted on average by the tree crowns, and noting that canopy cover as so defined above is essentially equivalent to the fractional foliage cover or projective foliar cover of the stratum Figure A1.1. Relationship between crown cover and canopy area in question, ζ (Lloyd et al. 2008) it then follows that ζ = ας where index as predicted by Equation A1. Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013 Delineation of tropical vegetation types 137

Appendix 2. Plot sampling layout

Figure A2.1. Distribution of the sampling locations within the plot and the pit location outside the plot for the sampled sites. Each line represents a ‘Gentry transect’ where measurements were made of subordinate stratum crown cover and species composition and axylale cover as detailed in Sections 2.3 and 2.4 of the main text. Solid squares: sampling for soil physical and chemical properties and axylale cover; Hatched squares: sampling for axylale cover only. Downloaded by [JAMES COOK UNIVERSITY] at 17:27 27 October 2013