, Volume 149, Issue 4, pp 620–634

Reducing complexity when studying seed dispersal at community scales: a functional classification of vertebrate seed dispersers in tropical forests


    • CSIRO Sustainable Ecosystems and the Rainforest CRC
  • David A. Westcott
    • CSIRO Sustainable Ecosystems and the Rainforest CRC
Plant Animal Interactions

DOI: 10.1007/s00442-006-0475-3

Cite this article as:
Dennis, A.J. & Westcott, D.A. Oecologia (2006) 149: 620. doi:10.1007/s00442-006-0475-3


The process of seed dispersal has a profound effect on vegetation structure and diversity in tropical forests. However, our understanding of the process and our ability to predict its outcomes at a community scale are limited by the frequently large number of interactions associated with it. Here, we outline an approach to dealing with this complexity that reduces the number of unique interactions considered by classifying the participants according to their functional similarity. We derived a classification of dispersers based on the nature of the dispersal service they provide to plants. We described the quantities of fruit handled, the quality of handling and the diversity of plants to which the service is provided. We used ten broad disperser traits to group 26 detailed measures for each disperser. We then applied this approach to vertebrate dispersers in Australia’s tropical forests. Using this we also develop a classification that may be more generally applicable. For each disperser, data relating to each trait was obtained either from the field or published literature. First, we identified dispersers whose service outcomes were so distinct that statistical analysis was not required and assigned them to functional groups. The remaining dispersers were assigned to functional groups using cluster analysis. The combined processes created 15 functional groups from 65 vertebrate dispersers in Australian tropical forests. Our approach—grouping dispersers on the basis of the type of dispersal service provided and the fruit types it is provided to—represents a means of reducing the complexity encountered in tropical seed dispersal systems and could be effectively applied in community level studies. It also represents a useful tool for exploring changes in dispersal services when the distribution and abundance of animal populations change due to human impacts.




Seed dispersal is a process that is fundamental to the structuring and functioning of rainforest communities (e.g. Terborgh et al. 2002) and can directly involve individuals of hundreds of species (e.g. Snow 1981). For example, in tropical , 70–90% of woody plant species are dispersed by animals (Willson et al. 1989; Cooper and Cooper 1994; Hyland et al. 2003) and these animals can represent the majority of the vertebrate biomass (Terborgh 1986). While some tight mutualisms between dispersers and particular plant species have been suggested (e.g. Rick and Bowman 1961; Temple 1977; Bennett 2005), such relationships are generally diffuse (Herrera 1986) with most dispersers consuming the fruits of a number of plant species and a variety of dispersers servicing any one species of plant (Snow 1981; Witmer and Cheke 1991).

While any specific plant–animal interaction is important and interesting in its own right, in complex and diffuse systems the task of understanding the system by describing each interaction very quickly becomes logistically intractable and uninterpretable. One means of penetrating this fog of diversity is suggested by the observation that not only do many frugivores feed on the same resources, but that many may also provide a similar dispersal service to the plants on which they feed (Clark et al. 2001). Traditionally, ecologists have approached such situations by assigning species to functional groups, guilds or classes based on their similarity in terms of traits relevant to the issue of interest (Terborgh and Robinson 1986; Clark et al. 2001).

Current classification systems for seed dispersers tend to focus on the taxonomic level, grouping dispersers into major taxonomic groups: birds, bats, monkeys, etc. (e.g. Fleming 1979; Gautier-Hion 1990). This approach is to some extent appropriate, as the behavioural and morphological abilities and constraints that reflect a shared evolutionary heritage often mean that they handle seeds and move in similar ways, thus potentially providing similar dispersal services. However, this approach ignores the variation within these groups (Lord et al. 2002) and the fact that species within very different taxonomic groups may also provide very similar services. Such variation within and similarities across taxa have the potential to render a taxonomic classification ineffectual (Bahr 1982). This is because taxonomic groupings are not defined on behavioural or functional traits and the taxonomic relationships of dispersers are relevant only in so far as they determine the dispersal service provided. Consequently, a functional classification based primarily on behavioural or functional traits relevant to the process will be best suited to investigating the outcomes of seed dispersal processes.

We advocate an approach to classifying dispersers that summarises the essential elements of the service they provide to plants. The approach must separate significant variation in dispersal service but combine species that provide similar services. Sufficient attributes must be incorporated to encompass the broad range of mechanisms through which dispersal operates and impacts on seed survival and seedling recruitment. This means that small details of differences between species will be relinquished in favour of being able to simplify the interactions between plants and animals sufficiently to be able to measure the community-wide process.

In this paper we use a combination of conceptual work and field data to (i) identify a set of traits for the classification of dispersers on the basis of their functional role in dispersal (“Classification traits” below); (ii) develop a functional classification of seed dispersers for an Australian community (“Classifying Australian tropical seed dispersers” below); and (iii) using (i) and (ii) above, to develop a more generally applicable classification system (“A generally applicable functional classification?” below). We take a mechanistic approach, using attributes of dispersers or their behaviour that measure or predict aspects of the dispersal service they provide. Our conceptual framework is based largely on Schupp’s (1993) disperser effectiveness criteria, but is supported by additional field data from Australian rainforests to identify functionally important traits of dispersers. Our aim is to produce a classification that reduces the number of interactions involved in seed dispersal in tropical rainforest by recognising aspects that species’ roles within the process have in common (Pizo 2002; Renne et al. 2002; Stansbury and Vivian-Smith 2003; Moran et al. 2004). We believe that such a classification makes the description of seed dispersal at a community level in these complex ecosystems feasible and interpretable, but it also maintains the level of detail necessary to realistically describe the entire process.

Data collection

This paper combines a conceptual approach to identifying dispersal services provided by vertebrates with considerable field data from Australian tropical forests. In this section we describe the collection of field data.

Study area

The data for this study were collected between 2000 and 2005 in various locations in the wet tropics of north-eastern Queensland, the largest area of continuous rainforest in Australia, covering 6,300 km2 (Tracey 1982). Rainfall on areas covered by rainforest varies from 1,500 mm to more than 9,000 mm annually with over 70% falling in the period December to March. Our study sites receive 2,000–4,000 mm of rainfall annually. Our study sites included locations on the Atherton Tableland and surrounding ranges and the coastal lowlands.

Observations at focal trees

We observed the behaviour of 43 species of disperser during ∼3000 h of observation at 65 species of fruiting trees, shrubs, palms and vines. We aimed for observations to be conducted by pairs of people for multiple periods of 4 h per species, but in practice they varied from 30 min to 11 h 40 mins (mean 17 h ± 0.2 SE per species). One person observed all activity within the tree and recorded the animal species, number of individuals and their behaviour at 5 min intervals. The other person recorded the behaviour of an individual animal from its entry or first sighting in the tree until its departure or when it was lost to sight. Data recorded were the number of fruit swallowed, chewed at or dropped, the number of defecations, the duration of the observation and whether it incorporated the entire visit or only part thereof. Additional data on the time of day, weather, crop size, diameter at breast height and location of the focal tree were also recorded. Observers used 10 × 40 binoculars to aid observation.

Interaction data

We recorded a total of 6,109 interactions and 2,463 unique interactions between a species of consumer and a species of plant. Interactions were recorded by observations at fruiting trees (above), from literature sources (Barker and Vestjens 1984a, 1984b; Cooper and Cooper 1994; Higgins and Davies 1996; Bentrupperbäumer 1998; Higgins 1999) and from reliable contributions from other observers (see “Acknowledgements”). These data were linked to a database of fruit characters from 1,141 species native to Australia’s wet tropical rainforests in north-east Queensland. This included 314 species for which we directly measured a sample of 12 or more fruits usually from two or more individual plants (N=1–10). For the remaining species, similar data were gleaned from the literature (primarily Hyland et al. 2003 and Cooper and Cooper 2004). The measurement relevant to this paper was the width (nearest 0.1 mm) of the entire fruit. Fruit width was taken as the midpoint of the reported range for each species for those we had not directly measured and the mean for those we had.

Captive feeding trials

Captive feeding trials were conducted with 116 individual birds and mammals of 17 species handling the fruits of 74 species of plant to examine feeding behaviour and retention of seeds. For murid rodents that did not swallow seeds, we conducted cafeteria trials to determine species consumed and handling techniques (see Dennis et al. 2005). To obtain gut passage rate data for species that swallow seeds, we brought 56 wild-caught individuals of 12 species into individual cages and fed them the fruits of 45 species of plant. Animals were given an ad libitum maintenance diet (without seeds) throughout the trials. To determine transit times we recorded the time of ingestion for each fruit and the time of egestion for each seed. For fruits with a known number of seeds we continued to collect droppings until all seeds were accounted for. For fruits with large or unpredictable numbers of seeds, we continued to monitor until at least four consecutive droppings were free of seed or fruit materials.

Studies of animal movement

To determine average foraging movements, fourteen species (13 birds, one bat), with a mean of 3.4 individuals per species (±2.23 SD) were radio-tracked for a mean of 4.6 days each (±1.9 SD) from stations of known location. We determined the locations of animals, using triangulation of simultaneous bearings from three to eight stations, every 2–6 min throughout a tracking session. Tracking sessions lasted for at least the maximum gut passage time of a seed through that particular animal, but generally covered at least 75% of an activity period; defined as the period of the day or night in which the animal usually foraged for food. How these data were used is described in the “Quality of handling” section. The rate and pattern of caching by scatter-hoarding marsupials is described by Dennis (2003) and was also applied to murid rodents.

Classification traits

We used a combination of conceptual work and field data in a process similar to that described by Boutin and Keddy (1993) to identify functionally important traits of dispersers. We selected functional traits of frugivores that influence the seed dispersal service they provide based on the quantity of seeds moved, the quality of handling and the diversity of seeds moved. The quantity and quality of seed handling were based loosely around the effectiveness parameters described by Schupp (1993). While Schupp’s (1993) traits were designed to measure effectiveness, they are also a good basis for a comparative measure of the type of service provided to a plant by a disperser.

We identified ten broad traits, and within these, 56 potential measures with which to define functional groups. We then screened these potential measures, choosing between correlated traits and rejecting those that were poor measures of a dispersal service, reducing them to three continuous, three ordinal and 20 nominal variables (Table 1). The use of nominal and ordinal variables was to allow species with few data to be assigned to a category on the basis of some understanding of their behaviour. We developed categories that aim to be generally applicable to tropical communities, assuming that frugivores show similar sets of behaviour that translate into similar outcomes for seeds regardless of location.
Table 1

Functional traits of seed dispersers used in the classification process

Seed dispersal component

Trait measured

Variable or class



Quantity of fruits removed

No. swallowed per visit

*Body weight

Affects number of fruit that can be carried for a given fruit size


*Frequency of visitation


A surrogate measure of the relative abundance or frequency of visitation for dispersers, it affects quantity of fruit likely to be moved from a crop of fruit on a plant


Regular individuals or pairs


Regular groups


Reliability of visitation


Describes seasonality of visitation and species that are not frugivorous but occasionally eat fruit; affects probability of removing fruit from a crop


Rarely frugivorous


Quality of fruit handling


Through gut

Affects number dispersed together, subsequent germination and subsequent likelihood of predation








Affects likelihood of a handled seed being killed; e.g. crushing predators are more likely pass small hard seeds intact than chewing or grinding predators








Affects subsequent seedling competition and likelihood of predation or secondary dispersal







*Movement range

Each variable has a strong influence on dispersal distance for swallowed seeds


*Movement rate


*Gut passage rate




*No. families eaten

Affects the taxonomic breadth of service provision in the ecosystem



*Fruit size range dispersed

Affects the range of plants serviced


*Foraging locations


Affects the range of plants serviced and the presentation of fruit taken




*Understorey to canopy


*Predominantly canopy


Within forest

Impacts on the locations at which seeds may be deposited


Across landscape


Those traits used in a cluster analysis for the volant birds are marked with *. The data column indicates whether each is continuous (C), nominal (N) or ordinal (O)

Identification of appropriate variables for inclusion was not always obvious, as some seemingly important variables had poor predictive power. For example, gape width initially seemed an appropriate index for the maximum size of fruits dispersed by a species. We tested this with data on Australian birds, and while measured gape widths of the species (specimens from the Australian National Wildlife Collection, CSIRO) correlated with the actual fruit sizes consumed (r=0.53, P<0.05), the relationship was not strong enough to warrant its use. This was due to two factors. First, some species, particularly the fruit-doves (Columbidae; see Table 2 for examples), had extremely elastic gapes and swallowed fruits much larger than their gape measurements would suggest were possible (Fig. 1). Second, some species with wide gapes did not appear to use those wide gapes for the fruit components of their mixed diets (e.g. Campephagidae; see Table 2 for examples). Instead, we suggest using records of the range of fruit sizes dispersed as the key trait where the sample size is large enough. This parameter also gives an indication of the largest size dispersed (all species eating large fruit also ate small fruit in the Australian data), including fruits that were carried rather than just those that were swallowed.
Table 2

Disperser functional groups in the wet tropics of Australia: hierarchical divisions and representation

Division 1

Division 2

 Division 3

Division 4

Australian wet tropics representatives


Poor dispersers

Generally not significant

Digestive predators

Columbidae: Columba leucomela, Chalcophaps indica, Macropygia amboinensis; Megapodiidae: Alectura lathami

Chewing predators

Psittacidae: Platycercus elegans, Alisterus scapularis, Cyclopsitta diopthalma, Trichoglossus haematodus, T. chlorolepidotus (rarely Glossopsitta pusilla); Cacatuidae: Cacatua galerita; Macropodidae – (rarely Thylogale stigmatica, Dendrolagus lumholtzii, D. bennetti); Psuedocheiridae: (rarely Psuedochirops archeri, Psuedochirulus herbertensis); Phalangeridae: (rarely Trichosurus johnsonii)

Occasional dispersers

Significant for some species


Suidae: Sus scrofa

Predatory rodents

Muridae: Melomys cervinipes, Rattus leucopus, R. fuscipes, Uromys caudimaculata, U. hadrourus




Small (A1.1)

Ptilonorhynchidae: Prionodura newtoniana, Ptilonorynchus violaceus, Scenopoeetes dentirostris, Ailuroedus melanotus; Paradisaeidae: Ptiloris victoriae

Large (A1.2.1)

Columbidae: Ptilinopus regina, P. superba, P. magnificus; Oriolidae: Oriolus sagittatus, O. flavocinctus


Hypsiprymnodontidae: Hypsiprymnodon moschatus

Throughout landscape

Frugivores (A1.2.2)

Meliphagidae: Meliphaga lewinii; Dicaeidae: Dicaeum hirundinaceum; Zosteropidae: Zosterops lateralis

Facultative (A2)

Meliophagidae: Lichenostomus frenatus, Meliphaga notata, Xanthotis macleayana; Campephagidae: Lalage leucomela


Meliphagidae: Philemon buceroides, Philemon corniculatus, Myzomela obscura, Meliphaga gracilis, Lichenostomus flavus, L. versicolor, L. Chrysops; Cracticidae: Cracticus quoyi,; Campephagidae: Coracina papuensis; Dicruridae: Dicrurus bracteatus; Pteropodidae: Syconycteris australis



Wide-ranging slow gut (B1.1)

Cuculidae: Scythrops novaehollandiae; Columbidae: Lopholaimus antarcticus; Campephagidae: Coracina novaehollandiae, C. lineata

Wide-ranging rapid gut (B1.2)

Oriolidae: Sphecotheres viridis; Sturnidae: Aplonis metallica

Wide-ranging large fruit (B2)

Cracticidae: Strepura graculina; Cuculidae: Eudynamys scolopacea; Columbidae: Ducula bicolor

Wide-ranging small seeds

Pteropodidae: Pteropus conspicillatus, P. alecto, Nyctimene robinsoni, (rarely P. scapulatus)


Mega-terrestrial frugivores

Casuaridae: Casuarius casuarius

Fig. 1

The relationship between gape width and the maximum size of fruit swallowed for 24 species of fruit-eating bird. While gape width and the maximum fruit size known to be swallowed are correlated (r=0.53), the discrepancies shown in this figure indicate that gape width is not a good measure of the size of fruits or seeds dispersed by a fruit-eating animal

Next we identified species consuming fruits (potential dispersers) and documented the variables and classes for each. This was applied to tropical Australian fauna. Once furnished with a species by trait matrix, the final step was to group the dispersers. We did this in two ways: by using logical decisions where species or sets of species had dispersal characters that were obviously distinct and did not require separation using statistical analysis (e.g. mega-terrestrial frugivores; Table 2); and by using a statistical approach, cluster analysis (Lavorel et al. 1999), where distinctions were continuous and multivariate. Below we outline the traits and their variables and classes.

Quantity removed

Disperser abundance and size can have a significant impact on the quantity of seeds dispersed within a community (e.g. Stevenson 2000; McConkey 2000). However, the quantity of fruits removed from a tree can be a function of a variety of disperser attributes. For example, (1) large dispersers are likely to move more fruits per visit than smaller dispersers; (2) species that visit more often are likely to move more fruits than species that visit less often; (3) those that visit as groups tend to remove more than those that visit individually; (4) abundant species tend to visit more than rare species; (5) species that are primarily frugivorous are likely to remove more fruit than those that are less frugivorous; and (6) species that are migratory or nomadic provide services to only a subset of plants or only at certain times.

As abundance data are rarely available for entire animal communities, we use five classes and one variable within three traits to reflect these aspects of a disperser’s behaviour and service (Table 1, Quantity of fruits removed). Body weight (taken from Baker et al. 1997, Strahan 1995, Dennis 1997 and from our own measurements of captured animals) was used as a relative measure of the weight of fruit that can be consumed in a feeding bout.

Frequency of visitation was used as a partial surrogate for measuring relative abundance. It is a more direct measure than the abundance of an animal, as many species have mixed diets and their abundance does not necessarily reflect their visitation to fruiting trees. Frequency of visitation fell into three classes: occasional; regular individuals or pairs; and regular groups. Occasional visitors appear infrequently (defined as < 1/100 h in our observations) and include partial frugivores and rare species. Species foraging as groups have significantly more individuals visit a tree than those foraging as individuals or pairs (230% more in our data; ANOVA, number of individuals visiting per tree for each category in Frequency of visitation [Table 1]: F(2,1755)=68.4, P<0.001).

Reliability of visitation was measured using two classes. Migratory and nomadic species show significant seasonal movements within or out of a bioregion (identified from literature: Higgins and Davies 1996; Higgins 1999; Pizzey 1999). Although not providing a reliable year-round service to the community, this group is likely to provide very long distance dispersal at times. Species that are rarely frugivorous visit as little as occasional frugivores (<1/100 h of observation) but have fewer plant species in their diet (≤5). Unlike occasional visitors, fruit is not a regular component of their diet.

Quality of handling

As with the quantity of fruits dispersed, the quality of seed dispersal can be a function of a variety of fruit handling attributes (Table 1). These include: (1) the likelihood of seeds being destroyed or dispersed intact; (2) how they are deposited; and (3) how far from the source they are likely to be dispersed. Schupp (1993) describes these as the quality of treatment (1) and the quality of deposition (2 and 3).

Species that treat seeds gently (Table 1, Gentle), and are therefore likely to disperse them intact, can pass them through the gut, carry them or carry and cache them. Each of these types of handling can have different outcomes for the handled seeds. For example, those that pass through a gut can have a different speed of germination and proportion that germinate (Traveset and Verdu 2002) compared to those that are carried (which are more or less equivalent to hand-cleaned controls in germination experiments), those that remain entire or those that are cached into specific microsites. For these reasons we separate these types of handling into three categories (Table 1).

Animals that tend to destroy seeds may also disperse some (e.g. Forget et al. 2002). Like species that treat seeds gently, they use a range of handling techniques and these may have different outcomes for the seeds (Table 1, Predatory). Seed predators may chew individual seeds (e.g. Psittacidae and Muridae; e.g. Dennis et al. 2005), grind or digest them in the alimentary canal (e.g. some Columbidae, Megapodidae; e.g. Crome 1975) or crush them in the mouth (e.g. Suidae; J. Mitchell, personal communication, 2002) (Table 1). Seeds that are chewed individually are unlikely to escape death regardless of size or hardness, whereas hard seeds are more likely to escape grinding or digestion than soft seeds, and fruits with numerous small, hard seeds are likely to have a proportion survive crushing in the mouth.

A third trait separates species that tend to deposit seeds in clumps; as scattered singletons or small clusters; or buried (Table 1, Deposition). The first two categories may affect rates of secondary dispersal or predation and root competition; those in clumps being more likely to be effected by secondary dispersal, predation and competition than those scattered (Howe 1989). Seeds that are buried are also affected by secondary dispersal and predation but also experience quite different microsite conditions for germination (Vander Wall 1990).

Finally, the distance seeds are moved from the source can result in different probabilities of survival (Janzen 1970). We use three aspects of disperser movement behaviour to indicate the distance seeds might be moved: movement range; movement rate and gut passage rate (Table 1, Distance). The overall distances a species usually moves during a foraging day determines the potential range of distances a seed might be dispersed. The rates of movement within that range in combination with the time that seeds are retained interact to determine the actual distances seeds are moved.

Movement ranges and rates were both calculated from displacement curves. We estimated displacement curves by choosing 20 random starting points in an animal’s movement path and then measuring how far the animal had moved away from each point during each subsequent 2 min interval for the rest of that activity period. The number of starting points in a given quarter of an activity period was weighted by the frequency of foraging for a species during each quarter as measured at fruiting trees (see Westcott et al. 2005). We then calculated mean displacement for each species from data for all of the individuals of a species and all of the days for each individual. To define movement range classes, we used data from our Australian species and estimated distance moved during the mean gut passage time of a seed. The classes were: (i) short <100 m; (ii) moderate 100–200 m; (iii) wide 200–800 m; and (iv) very wide >800 m.

For movement rate data we divided the slopes of the displacement curves from the intercept to the asymptote or inflection point into classes on the basis of groupings in the data. Four classes describe: (i) slow; (ii) moderate; (iii) fast; and (iv) very fast rates of movement. In our Australian data these classes were defined by the slope of the displacement curve: (i) slope <1; (ii) slope 1–2; (iii) slope 2–10; and (iv) slope 10–20.

We divided mean gut passage times (retention times) for species with gentle handling through the gut into three classes (N=11 species sampled): (i) short <30 min; (ii) moderate 30–60 min; and (iii) long >60 min (species that retain seeds by carrying tend to fall into the short retention category). There was a significant correlation between the log10 body size and the mean gut passage time for the five species that were almost entirely frugivorous (log10 body size = 1.57 + 0.012 × gut passage, r=0.96, P<0.05), which we used as a guide when assigning other species. Specialist frugivores <100 g had mean gut passage times of less than 30 min, those between 100 and 200 g had mean times falling between 30 and 60 min, while larger species had mean gut passage times >60 min. When seeds were passed through small (<50 g) species with mixed diets, they took relatively longer (mean 44±27.3 SD) than seeds passing through small (<50 g) specialist frugivores (27.9±12.8 SD; t4=−4.97, P=0.007). This places species of mixed diet <50 g into the 30–60 min category, including birds as small as the 10 g Silvereye, Zosterops lateralis. Species of mixed diet that were >50 g had a mean gut passage time >60 min (t-test small and large mixed diet: t4=3.47, P=0.02).

Diversity handled

The final component of the dispersal service provided by an animal relates to the diversity of species and locations in which the animal operates. Schupp (1993) incorporates these components into quality of seed dispersal as: habitat and microsite selection, and; seed (diet) mixing. We felt these components warranted separate consideration and incorporated taxonomic diversity, the diversity of fruit sizes and foraging locations into two variables and six classes (Table 1, Diversity). These components of animal behaviour determine different aspects of the breadth of service provision of a disperser. Some may provide services to a taxonomically restricted set of plants (e.g. Crome 1975); others may feed on a particular size range of fruits (e.g. Gautier-Hion et al.1985); or from particular strata (e.g. Emmons 1980); or may deposit seeds in particular habitats or a range of habitats (Schupp 1993).

Taxonomic diversity was measured as the number of plant families incorporated into an animal’s diet. Size range was the range of fruit sizes consumed (largest minus smallest), using fruit width as the measure of size. We also classified frugivores on the basis of their dominant foraging strata from patterns seen in our observations of behaviour at fruiting trees. Species that foraged on the ground (terrestrial), from the ground to the canopy (scansorial), from the understorey to canopy or predominantly (>90%) in the canopy were separated on the basis of the frequency of observations in different strata. Two final classes separate habitat specialists (within forest) from generalists (across landscape).

Classifying Australian tropical seed dispersers

To apply this classification system to tropical Australian frugivores and granivores, and subsequently present a general classification, we populated the trait matrix with data for 65 Australian vertebrates. Where species could not be assigned to classes on the basis of directly measured data, we used a general understanding of behaviour and natural history from literature and our own observations to assign them. In the case of gut passage categories, we assigned species using the relationships described above. We then made divisions that we considered did not require statistical analysis. First, we separated species that were primarily granivorous from those that were primarily frugivorous (Table 1, Quality of fruit handling—predatory or gentle; Tables 2, 3, Division 1).
Table 3

A general functional classification of frugivores and granivores describing attributes that differentiate each group

Division 1

Division 2

 Division 3

Division 4

Summary of service provided


Poor dispersers

Generally not significant

Digestive Predators

Kill most seeds but can disperse over long distancesa. Sometimes secondary dispersers

Chewing predators

Kill most seeds but can disperse over short distancesa

Occasional dispersers

Significant for some species


Kill most seeds but can disperse hard seeds, potentially long distancesa

Caching predators

Kill most seeds but also scatter or larder-hoard. Sometimes secondary dispersers.





Scatter or small clump-disperse over short distancesa


Scatter or small clump-disperse wide range of fruit sizes over short-to-long distancesa


Short-distancea dispersers and/or scatter-hoarders; kill few seeds

Throughout landscapec


Short-to-moderate-distance dispersala of primarily small fruits and seeds throughout the landscape


Regular, short-distance dispersala of few species with small fruit


Rare, short-distance dispersala of very few species


Cross landscaped

Wide-ranging long-retention

Very long distancea scatter or small clump-dispersal, primarily from canopy

Wide-ranging short-retention

Moderate-to-long-distancea, scatter or small clump-dispersal from canopy

Wide-ranging large fruit

Moderate-to-very-long-distancea, scatter or small clump-dispersal of high wide range of fruits (nomadic or migratory species)

Wide-ranging small seeds

Long dispersal distances for seeds <5 mm, short distancesa for larger seeds, high levels of seed wastage and frequent clumping of some seeds


Megab-terrestrial frugivores

Moderate-to-long-distancea, clump disperseral

aSee definition for movement range in “Classification traits and variables”

bSmall body size,  <50 g; medium body size, 50–200 g; large body size, 201–5,000 g; mega body size >5,000 g; At division 4, “Small” and “Large” both include medium

cSpecies occurs throughout the landscape matrix

dIndividuals cross the landscape matrix


Next, we divide granivores into poor dispersers and occasional dispersers (Tables 2, 3, Division 2). Digestive and chewing predators form two subgroups within the poor dispersers (Tables 2, 3, Division 4). A small proportion of seeds, particularly very hard seeds, may pass through the guts of digestive predators (Table 1, Class—grinding/digesting), remaining viable. For example, during gut passage trials, Macropygia amboinensis passed 0.015% of an estimated 4,000 ingested seeds from five species of plant in a viable state. Chewing predators (Table 1, Class—chewing) rarely disperse seeds through accidental epizoochory or by carrying fruits or infructescences away from the source (AJD, personal observation). In either case, numbers are likely to be small but distances are likely to be shorter for seeds carried externally (chewing predators) than for those carried internally (digestive predators).

Granivores that act as occasional dispersers were subdivided into mega-predators and predatory rodents. Mega-predators (Table 1, Class—crushing; Tables 2, 3, Division 4) may pass intact, variable proportions of seeds, particularly from large fruit with multiple small and hard seeds. They also have slow gut passage rates and move over kilometres, so intact seeds may be dispersed over long distances (Fragoso 1999; J. Mitchell, personal communication, 2002). Predatory rodents cache some seeds and may leave some unretrieved. Most of this caching is done less than 50 m from the source (e.g. Theimer 2001). The digestive predators and scatter-hoarding rodents also contain members that are, on occasion, secondary dispersers.


Within frugivores we separate species that are rarely frugivorous from those that are more frequently frugivorous (Table 1, Quantity of fruits removed, reliability of visitation—rarely frugivorous; Tables 2, 3, Division 4—Opportunists). These are a mixed bag of species that feed primarily on non-fruit resources but have been recorded as taking fruit from a few species on rare occasions. They tend to move over short-to-moderate distances (Tables 2, 3, Division 2) and are frequently habitat generalists (Tables 2, 3, Division 3).

We separate three further groups of frugivores that provide clearly unique services from those that required statistical analysis to define classes. First, mega-terrestrial frugivores (Tables 2, 3, Division 1—frugivores, 2—long-distance, 3—terrestrial) have wide-ranging (Class iii), moderate rate (Class ii) movements, are terrestrial foragers, are large, visit regularly as individuals or pairs, pass seeds intact through the gut, deposit seeds in clumps, have a slow gut passage rate (Class iii) and eat a high diversity of families and sizes of fruits (e.g. Casuarius casuarius). Second, terrestrial within-forest frugivores (Tables 2, 3, Division 1—frugivores, 2—Short-distance, 3—within-forest, 4—terrestrial) have short-ranging (Class i), slow rate (Class i) movements, remain within rainforest, are terrestrial foragers, are small, visit regularly as individuals or pairs, carry fruits and seeds, scatter and/or bury them and eat a high diversity of families and sizes of fruits (e.g. Hypsiprymnodon moschatus). The third group is that of the fruit bats (Tables 2, 3, Division 1—frugivores, 2—long-distance, 3—cross landscape, 4—wide-ranging small seeds), which display different dispersal services to plants with either small or large seeds. Due to the size of their oesophageal lumen, this group is capable of dispersing only small seeds through the gut (generally <3 mm but up to 5 mm) and have wide-ranging (Classes iii and iv), fast to very fast rate (Class iii and iv) movements, forage throughout the landscape and have a slow gut passage rate (Class iii). In addition, they disperse medium and large seeds by carrying them, generally over short distances (e.g. Pteropus conspicillatus, Nyctimene robinsoni).

We created classes for the remainder of the dispersers using cluster analysis on data for each trait or variable marked with * in Table 1 for Australia’s volant birds. Birds were used in the cluster analysis only because the mammals were easily classified without statistical analysis. The traits included continuous, nominal and ordinal data. Nominal and ordinal data were standardised for the analysis. Continuous data were log-transformed. The cluster analysis used Ward’s linkage method and Pearson’s r as a distance measure (Statsoft Inc. 2004). To determine which traits were responsible for the divisions in the cluster tree, we compared the trait values of clusters at each level of division using t-tests for continuous variables and Mann–Whitney U-tests for other data.

The initial division in the cluster analysis (Fig. 2, line 1) splits the species into two main clusters (A and B). Cluster A species tend to move over smaller distances at a slower pace, may be rainforest specialists or occur throughout the landscape, disperse a smaller size range of fruit, are smaller, have more rapid gut transit and have a wider range of foraging locations than cluster B species; they also tend to be sedentary and visit occasionally or as individuals or pairs (Table 4, A & B). Within this cluster and at the next division (Fig. 1, line 2; Table 4, A1 & 2), a group of four small frugivores is separated off from the rest (A2). They tend to have very short range movements, eat small fruit from only a few families and tend to be only partially frugivorous. The A1 group further subdivide (Fig. 1, line 3; Table 4, A1.1 & 1.2) into a group that have rapid gut passage rates (A1.1), which also tend to have slow, short-range movements, and are reliable visitors. The A1.2 cluster splits further (Fig. 1, line 4; Table 4, A1.2.1 & 1.2.2) into a group of small species with slow movement rates (A1.2.2) and a group of larger frugivores with faster movements (A1.2.1). Line 4 cluster codes from Fig. 1 are also listed in Table 2 for reference.
Fig. 2

Cluster tree describing functional groups of volant frugivores after a priori divisions separated 39 species. Lines 1–4 refer to divisions within the community (see text for details)

Table 4

Analysis of traits separating groups in cluster tree (see Fig. 2)


Tests for difference between class divisions







Fruit size range







Body size







No. of families







Movement range







Movement rate







Gut passage time







Foraging locations







Clump dispersal







Frequency of visitation














Continuous variables were compared using t tests and ordinal and categorical data using Mann–Whitney U tests. The numbers are t-values with subscript degrees of freedom or Z-values with subscript N for each group. Those marked with * have P<0.05, while those marked with ** have P<0.01

Cluster B species move over wider ranges at a faster pace, are able to cross the landscape, disperse a wider size range of fruit, are larger, have slower gut transit and have a narrower range of foraging locations than cluster A species, they also tend to be migratory or nomadic and visit as flocks (Table 4, A & B). Within cluster B, a group of three species split off at level 3 (Fig. 1, line 3; Table 4, B1 & 2). These tend to move over moderate-to-wide rather than wide-to-very-wide ranges, eat larger fruit than other group B species and forage throughout the strata (B2). The remainder of the wide-ranging cluster divide further into a group with short retention times and a group with long retention times (Fig. 1, line 4; Table 4, B1.1 & 1.2). In addition, the short retention group tend to be highly frugivorous species that forage in rapidly moving groups (B1.2). The long retention group (B1.1) contains the widest ranging species, which may be either partially or highly frugivorous.

We tested the stability of the groups defined by the classification analysis using a jackknife procedure. Each species was removed from the analysis, one at a time, and the resulting cluster tree compared with that derived with all species included. The groupings outlined in Fig. 1 were remarkably resilient to changes in the observations included. Of the seven functional groups created by the cluster analysis, six retained their members in all but one of the 27 jack-knife iterations, although on four additional occasions two closely related groups merged with the removal of a species from the analysis. The remaining group, B2 (wide-ranging large fruit in Table 2) proved less stable. It changed its membership on three occasions with its species splitting into other groups. B2 also changed its location across the major division in the classification in 11 iterations, although its membership remained constant in each of these instances. That this particular group should be unstable is not surprising. The group includes large-bodied frugivores that share a variety of traits with the fruit-doves in the large, within-forest frugivores (A1.2.1), including the fruits they feed on and their foraging zones. However, the wide-ranging migratory species (B2) all fly long distances over the canopy and therefore provide the potential for very long distance dispersal. Despite their affinities with large, within-forest frugivores (A1.2.1), we believe they are appropriately placed within division B.

A generally applicable functional classification?

Having created this classification we are tempted to think it might have general application in tropical forests around the world. This remains to be tested, particularly in the absence of data on some groups that do not occur in Australia during its creation (e.g. primates and squirrels). The reason we believe the classification might have more general application is that it was created using functional traits relating to services (Table 1) rather than taxon-specific traits. Table 3 provides a generic framework of functional classes and briefly describes the services provided by each class. The classification is designed to be usable at several levels depending on the detail required for each community scale investigation.

In “Appendix 1” in the Electronic Supplementary Material, we offer a key to assigning species to these functional classes in any area. As part of making the classification broadly applicable to communities without detailed data, we have generalised some of the decision parameters. These generalisations are made on the basis that some of the detailed differences in classes we used initially were not necessary for assigning species into functional groups. For example, species in our long-distance groups (Table 2, Division 2) all fell into movement range classes iii and iv, whereas those in our short-distance groups fell into movement range classes i and ii.

On the basis of our data and observations, in order to assign species to movement range classes we suggest that species with generally short-to-moderate movements, regardless of gut passage rates, tend to have daily ranges with a radius falling into classes i and ii (<200 m). Species with generally moderate-to-long-distance movements, regardless of gut passage rates, fall into classes iii and iv (>200 m). For the purposes of the key, retention times were simplified to short (Class i <30 min) and long (Classes ii and iii >30 min). Seed retention can be calculated from the relationships described above for birds that swallow seeds. Species that retain seeds through carrying can generally be assigned to the shortest class (i <30 min) on the basis that few species are likely to regularly carry fruit for longer than this. Mammals that swallow seeds can probably be assigned to moderate and long classes (ii and iii >30 min) on the basis that their gut passage times are generally longer than birds (e.g. Nogales et al. 2005). Insufficient studies have been conducted on reptiles to make generalisations at this stage (e.g. Lord et al. 2002).


The process of seed dispersal is widely recognised as having a profound effect on vegetation structure and diversity in tropical forests (Harms et al. 2000; Terborgh et al. 2002). Thus, much research has focussed on unravelling the mechanisms by which this impact is realised. However, most studies working at a community scale focus on describing patterns or the outcomes of dispersal (e.g. Foster 1982; Harms et al. 2000; Wright 2002), whereas those that tackle mechanisms tend to focus at small scales, studying a component of the community rather than the entire community (e.g. Coates-Estrada and Estrada 1988; Westcott and Graham 2000; Forget et al. 2001; Dennis 2003; Clark et al. 2005). Consequently, a disjunction exists between these two levels of investigation, frustrating attempts to move from empirical descriptions of process to empirical descriptions of community level pattern and vice versa. This has led to the suggestion that seed dispersal research has reached a bottleneck requiring methodological advances for any significant further progress (Levey 1999; Wang and Smith 2002).

Our approach has been to take a step back from the detail of species-level studies but to retain the capacity to capture the essence of the mechanisms of that level. While species-level processes determine community-level outcomes, successfully translating these processes to community and landscape scales and to appropriate timeframes requires forgoing a level of detail and generalising to the broader community (Schupp 1992). Simplifying plant and frugivore communities through the use of functional classification achieves this goal and has enabled us to reduce an array of 65 vertebrate dispersers into 15 distinct functional groups. This aggregation of species providing similar services to plants simplifies the interaction network being considered, making it logistically possible to consider the mechanisms (i.e. dispersal and predation) and to predict their outcomes at a community and a landscape scale (i.e. seed shadows and predation shadows).

Taxonomic versus functional groups

In most seed dispersal literature, dispersers are assigned to taxonomic groups on the implicit assumption that they provide different dispersal services to plants (e.g. Gautier-Hion et al. 1985). Similarly, fruits are assigned to different syndromes based on the identity of the animals or mechanisms that are perceived or known to be their primary dispersers (e.g. van der Pijl 1972; Clark et al. 2005). Few such studies focus on the functional outcomes of the service provided but instead deal primarily with which fruit characters dispersers choose or the patterns of seed rain (as measured in seed traps) created for the different fruit syndromes (monkey-dispersed, bird-dispersed, wind-dispersed, etc.). Several authors have discussed the inadequacy of the fruit syndromes based on vertebrate taxonomic groups (e.g. Howe 1986; Lord et al. 2002). This begs the question: how does a classification based on functional outcomes compare to one based on the taxonomic identity of consumers? Firstly, we can reiterate the same general message given by Lord et al. (2002) when discussing fruit syndromes in relation to animal choice. At the high taxonomic level used to describe fruit syndromes, for example within birds or mammals, there is a huge range in behaviour and abilities for selecting, handling and depositing fruits and seeds. In our classification, a “bird-fruit”, when it is dispersed by a bird, may receive one of 11 distinct types of service that birds provide (11/15 functional classes; Table 2). These range from very rare, short-distance dispersal with high levels of predation (e.g. chewing predators) through moderate-to-long-distance, clump dispersal (e.g. mega-terrestrial frugivores) to moderate-to-very-long-distance scatter-dispersal (e.g. wide-ranging large fruit). In Australia, which is relatively depauperate, mammals provide six distinct dispersal services (6/15 functional classes; Table 2) ranging from occasional (predatory rodents) to regular, short-range, scatter-hoarding and burial (terrestrial within-forest frugivores) to very long range scatter and clump dispersal of small seeds (wide-ranging small seeds). Overall, this suggests that a taxonomic classification provides little understanding of the outcomes of dispersal.

Mammals and birds overlap in the services provided in only two classes for the Australian fauna. This is primarily due to only a few mammals being frugivorous or granivorous in Australia. In countries with more diverse frugivorous or granivorous mammals (e.g. primates, sciurids, viverrids, perissodactyls, proboscids, etc.) the overlap between the services provided by mammals and birds would be considerably higher according to our classification. Despite the overlap, two groups of mammals stand out as providing distinct services within taxonomic groups. These are the wide-ranging small seeds and predatory rodent groups. The fruit bats (wide-ranging small seeds) exhibit two distinct services for fruits with seeds of different sizes (see above). While the predatory rodents kill most of the seeds they handle, for many species seeds are regularly cached, either through scatter or larder-hoarding (Muridae in Australia but including Sciuridae and Dasyproctidae elsewhere; e.g. Emmons 1980; Becker and Wong 1985; Galetti et al. 1992; Forget 1996; Kitamura et al. 2002). While the functional relevance of larder-hoarding for seed dispersal is still little-understood, the relevance of scatter-hoarding is well demonstrated (Vander Wall 1990; Forget 1996; Jansen et al. 2002). A third group appears restricted to mammals but less taxonomically constrained. The mega-predator group contains pigs (Suidae) in Australia but includes peccaries (Tayassuidae) and probably some ungulates and ruminants in other places (e.g. Kiltie 1981; Dubost 1984; Bodmer 1989; Fragoso 1999).

At a finer scale, some classes remain taxonomically restricted. Those represented by single species in Australia are obvious examples (e.g. mega-terrestrial and terrestrial within forest frugivores). In addition, the small within-forest frugivore group contains all of the Ptilinorhynchidae and Paradisaeidae in the study area (five spp.). However, other taxonomic groups become separated. For example, members of the Columbidae occupy four functional groups. Overall, our findings indicate that the use of broad or restricted taxonomic groupings to indicate potential dispersal services is unwarranted due to the high levels of variation in service provision within taxonomic groups.


We derived a functional classification for seed dispersers resulting in 15 functional groups at the finest level of division that describe the services provided to an entire rainforest plant community. We believe that this approach and the classification is transferable to other locations and may aid research into seed dispersal at community scales by reducing the complexity of interaction networks. Our functional classification is designed to be used in conjunction with estimations of the dispersal kernels produced by each functional group of dispersers. We can then apply this to estimate the total dispersal kernels for plants based on the contributions of the different functional groups and gain a better understanding of the patterns and mechanisms involved in the seed dispersal component of plant recruitment. We can also examine changes in total dispersal kernels brought about by changes in the animal communities creating them. This may be used to compare different species of plant, communities in different locations (see Bleher and Bohning-Gaese 2001) and disturbed or fragmented forests with undisturbed forests.

One of the potential uses of the functional classification system we are advocating is in translating changes in species abundance or composition in habitat fragments into a measure of the health of the process of seed dispersal. By assigning species to functional groups of service providers, it is possible to more accurately indicate which services are declining or becoming extinct or whether the inherent “redundancy” in the system is sufficient that the loss of a few species does not mean the complete loss of a service to that system (Wright 2003). In Australian systems it is clear that some services (functional classes) have no redundancy; they are performed by single species. Others may be occupied by a range of disperser species and be more resilient to changes in animal distribution and abundance. At the same time, not all plants receive or require all services, and the loss or decline in one may have significant impacts on a plant’s competitive ability in complex tropical forests if it already receives a restricted subset of services.


We thank Matt Bradford and Adam McKeown for excellent technical assistance. Earthwatch Institute volunteers contributed to various aspects of data collection and management. Peter Green, Dan Metcalfe, John Ludwig and two anonymous referees gave helpful comments on the manuscript. James Cook University, the Earthwatch Institute and Queensland Parks and Wildlife Service provided additional financial support. This research was conducted under permits WITK02096404 and WISP02096504 and animal ethics approval OB15/12. The following people contributed observations of frugivore fruit interactions to our database: John Grant, Glen Holmes, Dawn Maggary and Margaret Thorsborne.

Supplementary material

442_2006_475_MOESM1_ESM.doc (30 kb)
Supplementary material (DOC 30 KB)

Copyright information

© Springer-Verlag 2006