Rainforest Collembola (Hexapoda: Collembola) and the insularity of epiphyte microhabitats


Denis J. Rodgers, Science, Environment, Engineering and Technology Group, Environmental Futures Centre, Griffith University, Nathan Campus, Nathan, Queensland 4111, Australia. E-mail: d.rodgers@griffith.edu.au


Abstract.  1. This study compares species composition of collembolan (Collembola: Hexapoda) assemblages associated with rainforest soil, forest floor leaf litter and epiphyte-associated leaf litter deposits, canopy foliage and bark surfaces in a subtropical rainforest in southeast Queensland.

2. The results of analyses of similarity show that in both winter and summer, the species composition of collembolan assemblages differs significantly between each of the microhabitats studied. These results also confirm earlier work showing vertical stratification of species composition in collembolan assemblages associated with leaf litter suspended in epiphytes.

3. Many of the species occurring in leaf litter suspended in epiphytes were not found on intervening bark surfaces. This supports a hypothesis suggesting that epiphytes within rainforest canopies can be thought of as habitat islands for arthropods.


Collembola are often the most abundant hexapods collected in studies of arthropods in rainforest canopies (Kitching et al., 1993; Palacios-Vargas & Gómez-Anaya, 1993; Guilbert et al., 1994, 1995; Palacios-Vargas et al., 1998, 1999; Yanoviak et al., 2003, 2004), and have long been known to occur in a wide range of rainforest microhabitats, including soils, forest floor leaf litter and humic soils and leaf litter suspended in epiphytes (i.e. Delamare Deboutteville, 1951). Collembola are also the frequent prey items of mites, ants, micro-spiders, pseudoscorpions, small lizards, frogs and insectivorous birds (Hopkin, 1997). It thus seems likely that in addition to their significant role in decomposition processes, the Collembola may be important in sustaining the diversity of these predator groups within rainforest canopies.

Many authors have suggested that the presence of epiphytes in rainforest canopies may play an important role in sustaining the abundance and diversity of Collembola and other arthropods in rainforest canopies (Palacios-Vargas & Gómez-Anaya, 1993; Guilbert et al., 1994, 1995; Kitching et al., 1997; Palacios-Vargas et al., 1998, 1999). The humic soils associated with rainforest epiphytes have also been likened to habitat islands within the matrix of the rainforest canopy (Paoletti et al., 1991). Although speculative, this suggestion is tantalising since it may link studies of epiphyte-associated arthropod assemblages to a well-established field of ecological theory, i.e. MacArthur and Wilson’s (1967) theory of island biogeography.

Rodgers and Kitching (1998) have shown that the relationship between collembolan assemblages in forest floor leaf litter and epiphyte-associated litter deposits (hereafter, referred to as suspended litter) is vertically stratified and accords with some predictions of island biogeographic theory (assuming that forest floor leaf litter is a source of species colonising leaf litter deposits suspended within the canopy). This relationship was such that the similarity of species composition was significantly greater between forest floor and lower canopy suspended litter assemblages than between the forest floor and suspended litter in the upper canopy (Rodgers & Kitching, 1998). The remoteness of the island (i.e. lower canopy cf upper canopy) was suggested to provide some explanation for these results (Rodgers & Kitching, 1998).

In this article, studies of this epiphyte island hypothesis are extended. We suggest that if epiphytes are indeed insular habitats, this will be reflected in the comparative species composition of assemblages in suspended litter and other microhabitats within the surrounding matrix of the rainforest canopy. The similarities of the species composition of collembolan assemblages from forest floor soils and leaf litter, suspended litter and the bark surfaces and canopy of foliage of trees are compared.


Site description

This study was carried out on a 1-ha rainforest site within Lamington National Park (28°13′S 153°07′E) in South-east Queensland which has been established as a reference location for a range of biodiversity studies (Kitching et al., 1999; Laidlaw et al., 2000). Winter at the site is cool and dry (July mean maximum 16 °C, mean minimum 8 °C, rainfall ca. 110 mm month−1) and summers are warm and wet (January mean maximum 25 °C, mean minimum 16 °C, rainfall ca. 500 mm month−1) (Laidlaw et al., 2000). The forest type and the tree flora of the site has been described in detail by Laidlaw et al. (2000).


Samples were collected in a series of expeditions in 1995 and 1996 from surface soil, leaf litter from the forest floor and suspended in epiphytes, the bark surface of trees, and the canopy foliage. All microhabitats other than soils were sampled in both winter and summer to assess seasonal variation in the relationships between assemblages. Soil samples were collected only during winter. Arthropods were extracted from all soil and leaf litter samples using Berlese–Tullgren funnels which were run for a period of 24 h.

Surface soil.  Eighteen 1 l soil cores (0–75 mm depth) were collected at random locations within the site.

Forest floor leaf litter.  Twelve 1 l samples of forest floor leaf litter were collected at random locations within the site in both sampling occasions.

Suspended litter.  Although there were several large, litter-accumulating species of epiphyte at the site, our sampling was restricted to collections from Asplenium australasicum (Hook). Canopy access for the collection of these samples was achieved using single rope techniques (e.g. Barker & Standridge, 2002). In all cases, 1 l samples of leaf litter were collected from the deposit within the crown of the plant. Since small epiphytes (i.e. <300 mm basal diameter) often contain <1 l of litter, these were excluded from sampling. Samples were partitioned into those from the upper canopy (14–21 m) and those from the lower canopy (0–7 m). Six samples were collected from the lower canopy and six from the upper canopy in both summer and winter. Selection of sampling locations within the site was randomised to the greatest extent possible within the constraints of safety of access.

Canopy foliage sampling.  Canopy foliage was sampled using pyrethrin knockdown techniques (Kitching et al., 2002) conducted in three randomly selected 10 × 10 m quadrats within the site boundaries. Within each quadrat 20 circular collecting funnels (each 0.5 m2 in area) were hung at around 1.5 m height. A 5-min burst of pyrethrin-based insecticide was then sprayed into the canopy above the collecting funnels using a petrol-driven misting device hauled into the canopy using a rope and pulley. Approximately 3 h after the application of the insecticide, the catch was brushed into an ethanol-filled collecting vial fitted to the base of each funnel. Samples containing no Collembola were excluded from subsequent analyses. Canopy foliage sampling was carried out once in winter and once in summer.

Bark sampling.  The bark surfaces of 30 randomly selected trees greater than 500 mm dbh were sampled by pinning a collecting apron around a 0.5 m length of the circumference of the tree, and then spraying an area above the tray (0.5 m2) with a pyrethrin-based insecticide. The bark surface in the sprayed area was then gently brushed to knock the catch down into the collecting tray and thence into a vial of ethanol. Again sampling was carried out in both summer and winter.

All Collembola were stored in 95% ethanol and sorted to the level of morphospecies using a stereo-microscope. Several (i.e. 1–10) specimens of each morphospecies were slide-mounted in Hoyer’s medium for further examination and identification to genus and, in some cases, to species level.


Our aim was to test the hypothesis that the species composition of collembolan assemblages was significantly different in different rainforest microhabitats. We were concerned therefore with analysing species composition. Although abundance data is essential in detecting functional differences among samples within an assemblage, it is less helpful in discriminating between assemblages which are defined by their species composition (Krebs, 1985). We therefore used only presence/absence data in our analyses. Samples containing no Collembola were excluded from subsequent analyses because such data cause unreasonable distortions in multivariate analyses (i.e. two samples containing no data have absolute similarity whereas they are also absolutely dissimilar to all other samples that contain non-zero data).

Data analyses

Similarity matrices were generated for samples from the summer and winter data sets using the Bray–Curtis similarity measure as recommended by Faith et al. (1987). Analyses of similarity (anosim) (Clarke & Green, 1988; Clarke & Warwick, 2001; Clarke & Gorley, 2006) were carried out using PRIMER statistical software (PRIMER-E Ltd., UK) to test the hypothesis that assemblages associated with different microhabitats differed significantly in terms of species composition. We follow the recommendation of Anderson et al. (2008) for these analyses in reporting exact permutation probability values directly. Additionally, since we make a total of 24 statistical comparisons, we note that an overall α = 0.05 is achieved at P = 0.002 for each of the results reported.

To confirm our original observations in Rodgers and Kitching (1998) concerning vertical stratification, we also compared the degree of multivariate dissimilarity between leaf litter assemblages from the forest floor and lower canopy suspended litter versus that between assemblages from the forest floor and upper canopy suspended litter. This was carried out by manually (using a text editor) extracting the appropriate blocks of dissimilarity values from the overall dissimilarity matrices and converting these blocks to vectors. The mean values of these vectors were then compared using a randomisation-based analogue of the t-test in the RT package (Manly, 1996).

We also used the similarity percentage (simper) program (PRIMER-E Ltd., UK) within the PRIMER package to provide a quantitative assessment of the contributions of individual species to patterns of similarity and difference between assemblages associated with particular microhabitats. The results produced by these analyses are extensive (i.e. 1768 individual statistics). Rather than reporting all of these, we mention only those simper results which are especially strong and which facilitate the interpretation of the anosim results.

Graphical displays

Direct gradient plots (sensuGauch, 1982) displaying the distribution of species among sample groups were produced as an aid to interpreting the results of anosim analyses. To produce these plots, we arranged sample groups on the abscissa along an assumed gradient from forest floor to forest canopy. We then proceeded to rearrange the data rows such that species which occur predominantly in soil samples are grouped towards the bottom of the ordinand axis, those that occur in forest floor leaf litter come next, and so forth, until species occurring predominantly in forest canopy foliage are displayed at the top of the plot. The data within the columns remains unchanged in this process. To improve the visual clarity of the plots, we transformed ones in the data matrix (representing the presence of a species in a sample) to black symbols, and zeros (representing the absence of a species from a sample) to blank symbols. The order in which species are displayed on these axes are consistent in both the winter and summer plots.


Summary data

The sampling program produced a total of 52 collembolan species from the 1-ha site. Twenty-eight of the 52 species were recorded in soil samples, 40 in forest floor litter samples, 32 in suspended litter deposits, 19 on bark surfaces and 13 from canopy fogging samples (Figs 1 and 2). Among these species, there was little evidence of strict habitat specialisation since only six species were recorded from just one of the microhabitats. Of these only one, Temeritas sp. occurred in just a single sample and was collected from the bark surface. The remaining five of these specialist species were all found in soil samples. Two of these species, Tomoceridae gen. sp. and Triacanthella sp. were found in only two samples and Cryptopygus sp.1 in five samples. Only Dinaphorura harrisoni Bagnall was found in all 18 soil samples and Cryptopygus sp.2 in 13 of the soil samples. Of the remaining 47 species, 10 were found in two of the microhabitats (soil and forest floor litter in all cases) and the remainder of species in at least three microhabitats.

Figure 1.

 Direct gradient plot of species distribution by microhabitat (winter).

Figure 2.

 Direct gradient plot of species distribution by microhabitat (summer).

anosim and simper results

The simple summary data presented above suggest that, with the possible exception of the soil assemblage, there was relatively little difference in the assemblages associated with the microhabitats studied. However, the anosim results presented in Table 1 show that in both summer and winter, significant differences in species composition occurred between all microhabitat assemblages (marginal differences in the species composition of assemblages in suspended litter are discussed below). The contrast in these results indicates that the patterns of species composition underlying the anosim results are complex multivariate relationships and are ideally suited to this form of analysis.

Table 1. anosim results: similarity of species composition by microhabitat (10 000 permutations were performed for each comparison except for those indicated by * in the Table).
Sample groups comparedRPermuted statistics, ≥RSignificance level, P
  1. *For these comparisons, the number of possible permutations was constrained to 462 because of the small number of data points available.

 Global test0.6360<0.001
 Leaf litter: forest floor × lower and upper canopy0.8240<0.001
 Leaf litter: forest floor × lower canopy0.8811<0.001
 Leaf litter: forest floor × upper canopy0.9921<0.001
 Leaf litter: lower canopy × upper canopy*0.2919<0.019
 Leaf litter × canopy fogging0.7370<0.001
 Leaf litter × bark spraying0.9480<0.001
 Canopy fogging × bark spraying0.5190<0.001
 Global test0.5320<0.001
 Leaf litter: forest floor × lower and upper canopy0.9500<0.001
 Leaf litter: forest floor × lower canopy0.9691<0.001
 Leaf litter: forest floor × upper canopy0.9970<0.001
 Leaf litter: lower canopy × upper canopy*0.4941<0.020
 Soil cores × leaf litter0.7190<0.001
 Soil cores × canopy fogging0.5200<0.001
 Soil cores × bark spraying0.8450<0.001
 Leaf litter × canopy fogging0.4460<0.001
 Leaf litter × bark spraying0.6130<0.001
 Canopy fogging × bark spraying0.2340<0.001

The contributions of individual species to the similarity between samples within that group can be illustrated by D. harrisoni and Folsomina onychiurina Denis. Both of these species occurred in all soil samples and each contributed 16.1% to the overall similarity between soil samples. In the summer data set (Fig. 2), Epimetrura rostrata Greenslade & Sutrisno contributed 98.1% of the within-group similarity for canopy foliage samples.

Examples of the contributions of individual species to overall levels of dissimilarity between groups of samples are: in the winter comparison between the soil and forest floor leaf litter assemblages, D. harrisoni was consistently present in soil and absent from forest floor litter samples and contributed 6.2% to the overall dissimilarity between these groups. Similarly, Pseudoparonella queenslandica (Schött) was consistently present in forest floor litter samples and present in only one soil sample and contributed 5.9% to the overall dissimilarity between these groups.

The anosim results can thus be interpreted as a consequence of the distribution of species among microhabitats as illustrated in Figs 1 and 2. In some of the pairwise microhabitat comparisons, just one or two species have a particularly strong effect on the analysis and this is intuitively obvious when looking at Figs 1 and 2. A clear example of this is the distribution of P. queenslandica which contributed more than any other species to the dissimilarity between forest floor litter and suspended litter in the winter data set. Less obvious are the cases in which the overall dissimilarity between groups is an accumulated consequence of many small differences in the distributions of individual species. An example of this kind of relationship is that between forest floor litter and bark surface samples from the summer data set. In this case, no species contributed more than 4.5% to the overall dissimilarity between these groups and 16 species contributed between 3% and 4.5%.

Seasonal contrasts

There is a strong trend in the results such that more species were collected in summer than in winter. From Figs 1 and 2, it can be seen that in summer there were 44 species overall, 34 in forest floor litter, 28 in lower canopy suspended litter, 18 in upper canopy suspended litter, 15 on the bark surface and 9 in canopy fogging samples. In winter, there were 36 species overall, 28 in forest floor litter, 13 in lower canopy suspended litter, 10 in upper canopy suspended litter, 12 on the bark surface and again, 9 collected in canopy fogging samples. Another seasonal difference is that in summer, 22 of the 30 bark surface samples and 44 of the 60 canopy fogging samples contained Collembola. In winter, 29 of the 30 bark surface samples and only 28 of the 60 canopy fogging samples contained Collembola. Some clear seasonal changes in the distribution of individual species are also shown in Figs 1 and 2. In particular, Ceratrimeria maxima (Schött) and Lepidosira sp.3 occur in all samples from suspended litter in summer as well as in forest floor leaf litter, but in winter C. maxima is absent from suspended litter in the upper canopy and occurs in only two samples from the lower canopy, whereas Lepidosira sp.3 remains present in all lower canopy samples but occurs in only two samples from the upper canopy.

Vertical stratification

The anosim results presented in Table 1 show that the differences in species composition between suspended litter in the upper versus lower canopy were only marginally significant (i.e. P = 0.019 in summer and P = 0.02 in winter). This relationship is, however, sufficiently strong to warrant further examination. Comparing multivariate dissimilarities between forest floor and suspended litter assemblages, we found that for the summer data set, dissimilarities between forest floor and the lower canopy [hereafter as mean ± SD (n): 0.5732 ± 0.096 (72)] were significantly lower than those between forest floor and upper canopy (0.6325 ± 0.081 (72); P < 0.001). In winter, the dissimilarities between forest floor and lower canopy (0.6330 ± 0.098 (72)) were also significantly lower than those between forest floor and upper canopy (0.7657 ± 0.100 (72)) (P < 0.001). These results confirm the existence of vertical stratification among these assemblages in both summer and winter.


The anosim results show that soil, forest floor leaf litter, suspended litter, the bark surfaces of trees and the foliage of the forest canopy are all microhabitats which host significantly different assemblages of collembolan species. These results were consistent for both the winter and summer data sets. Furthermore, these patterns are especially robust given the substantial seasonal changes in the distributions of individual species. These results confirm the suggestion of Paoletti et al. (1991) that epiphytes can be thought of as habitat islands within the matrix of the rainforest canopy. The results concerning vertical stratification in suspended litter collembolan assemblages are also important since they confirm the existence of the pattern this study seeks to explain.

The significance of these results is that they provide a link between future work and the rich lode of ideas and hypotheses associated with the theory of island biogeography (MacArthur & Wilson, 1967). Ideas relating the species composition and species richness of collembolan assemblages in epiphyte habitats to the size of the island, its remoteness and the presence, absence or abundance of intermediate stepping stone islands suggest themselves and can readily be tested in rainforests with a similar epiphyte flora. The same conceptual framework could potentially be applied to studies of assemblages of other apterous arthropod groups associated with similar epiphyte microhabitats.

The results provide no evidence of epiphyte specialists among the Collembola on this site. This is consistent with the conclusions of Palacios-Vargas (1981) with regard to the Collembola associated with bromeliads in Mexico (Palacios-Vargas & Gómez-Anaya, 1993). A number of species from other locations have been suggested to have a particular affinity for epiphyte habitats. For example, in their description of Deuterosminthurus delatorrei, Palacios-Vargas and Gonázlez (1995) note that this species is associated with epiphytes in Mexico. Palacios-Vargas and Castaño-Meneses (2002) also observe that Pseudoisotoma sensibilis (Tullberg) and Sminthurinus quadrimaculatus (Ryder) are abundant in bromeliads in Mexico. To our knowledge, however, there is no data concerning the occurrence of these species from other adjacent canopy microhabitats (i.e. canopy foliage or bark surfaces). Their status as epiphyte specialists therefore remains uncertain.

A stronger argument can be made for the existence of canopy specialists (as distinct from epiphyte specialists) among the Collembola. In the description of E. rostrata which was collected by canopy fogging, Greenslade and Sutrisno (1994) state that this species was the first record of a collembolan canopy specialist. Our own observations on the distribution of E. rostrata among microhabitats are consistent with this suggestion and are also remarkably similar to the distribution of Salina banksi MacGillivray. This species occurs frequently in canopy fogging samples, rarely on the forest floor and is more common in epiphytes during seasonally dry periods (Palacios-Vargas et al., 1998). Although strictly epiphyte-associated collembolans may not exist, the presence of epiphytes may represent an important resource for species that do have a strong association with forest canopy habitats more broadly.

Although canopy specialist collembolans may interact significantly with epiphytes, our results show that the majority of species found in these microhabitats are forest floor species for whom epiphytes represent insular habitat patches. It is the distribution of these species which gives the distinctive vertical stratification we observe in the epiphyte associated collembolan assemblage. To understand how this vertical stratification is produced, the movement of Collembola between forest floor and epiphyte islands is a key process requiring further study. The results show that several species occur in forest floor and suspended litter but do not occur on bark surfaces or in the canopy foliage. However, this group of species must traverse the bark surface at some time to colonise the suspended litter. Consequently, a number of questions present themselves for further study. Since our sampling occurred only during the day, is the failure to detect these species on bark surfaces because their movements are crepuscular or nocturnal? Does this dispersal occur as low-level but consistent traffic or as less frequent but synchronised mass movements similar to those reported by Lyford (1975), Farrow and Greenslade (1992), Hågvar (1995) and Gauer (1997)? It may also be the case that the species observed in the forest canopy on this site have life-history links with forest floor habitats. This has been recorded for Xenylla brevispina Kinoshita by Itoh (1991) and for Sminthurus arborealis Itoh by Itoh (1994), both of which live as adults in forest canopies but lay eggs in forest floor soils.

Vertical stratifications of a variety of forms have been recorded among collembolan assemblages in other forest types. In a comparison between forest floor leaf litter and leaf litter collected from the forest canopy, Castaño-Meneses et al. (2006) recorded a distinctly different composition of collembolan families in association with these habitats. Palacios-Vargas et al. (1998) observed that there was greater similarity between collembolan assemblages in forest floor leaf litter and the shrub layer than between the shrub and canopy layers in Mexico and suggest that an abundance of juveniles of some species in the shrub layer may indicate vertical migration between this layer and the canopy.

Another well-documented vertical distribution of forest Collembola is that recorded by Yoshida and Hijii (2005) who have been able to attribute the vertical structure of collembolan assemblages in cedar plantations to the complex interactions of the behaviour and ecology of individual species within the assemblage. This is a level of understanding to which workers in more complex native forests in the tropics and sub-tropics can only aspire. We hope, however, that having identified the essential structural characteristics of the collembolan assemblages on our study site we are a step closer to developing such an understanding.


Our results have confirmed that the species composition of collembolan assemblages associated with leaf litter suspended in rainforest epiphytes are vertically stratified. We have also shown that for many collembolan species, leaf litter suspended in epiphytes represents a microhabitat resource that occurs as patches isolated within a matrix of very different habitat types. In this sense, epiphytes can be thought of as habitat islands. This provides a partial explanation for the pattern of vertical stratification we have observed. If the forest floor is a source of species colonising habitat islands represented by epiphytes then epiphytes in the upper canopy can be seen as more isolated than those in the lower canopy.

Now that the nature of these structural patterns is established a number of functional questions emerge. These relate to the processes by which epiphytes are colonised and especially to the movement of collembolans within rainforest canopies and between the forest floor and canopy. Improvements in our knowledge of the life history and population ecology of individual species will also be required if the dynamic function of these systems is to be properly understood.


An Australian Postgraduate Research Award Scholarship, Earthwatch International and the Cooperative Research Centre for Tropical Rainforest Ecology and Management provided financial support for this research. We acknowledge the support of the many volunteers who participated in the fieldwork associated with this study.