Classic metapopulations are rare among common beetle species from a naturally fragmented landscape

Summary

1. The general importance of metacommunity and metapopulation theories is poorly understood because few studies have examined responses of the suite of species that occupy the same fragmented landscape. In this study, we examined the importance of spatial ecological theories using a large-scale, naturally fragmented landscape.

2. We measured the occurrence and abundance of 44 common beetle species in 31 natural rainforest fragments in Tasmania, Australia. We tested for an effect on beetle distribution of geographic variables (patch area, patch isolation and amount of surrounding habitat) and of environmental variables based on plant species, after first accounting for spatial autocorrelation using principal coordinates of neighbour matrices. The environmental variables described a productivity gradient and a post-fire succession from eucalypt-dominated forest to late-successional rainforest.

3. Few species had distributions consistent with a metapopulation. However, the amount of surrounding habitat and patch isolation influenced the occurrence or abundance of 30% of beetle species, implying that dispersal into or out of patches was an important process.

4. Three species showed a distribution that could arise by interactions with dominant competitors or predators with higher occurrence in small patches.

5. Environmental effects were more commonly observed than spatial effects. Twenty-three per cent of species showed evidence of habitat-driven, deterministic metapopulations. Furthermore, almost half of the species were influenced by the plant succession or productivity gradient, including effects at the within-patch, patch and regional scales. The beetle succession involved an increase in the frequency of many species, and the addition of new species, with little evidence of species turnover. Niche-related ecological theory such as the species-sorting metacommunity theory was therefore the most broadly applicable concept.

6. We conclude that classic and source-sink metapopulations are probably rare in this large-scale, naturally fragmented system, although dispersal processes like those occurring in metapopulations may have a substantial influence on community composition. However, deterministic processes (niche specialisation, species-sorting metacommunities and deterministic metapopulations) drive the occurrence or frequency of the majority of species. We urge further research into the prevalence of spatial ecological processes in large-scale natural ecosystems to expand our understanding of the processes that may be important in nature.

Introduction

Substantial emphasis has been placed on the random nature of extinction, dispersal and colonisation in ecological theory (MacArthur & Wilson 1967; Hanski 1998, 1999). Stochastic colonisation and extinction are fundamental to the classic metapopulation concept (Levins 1970; Hanski 1998) and neutral metacommunity theory (Bell 2001; Hubbell 2001). However, other manifestations of metapopulation and metacommunity theory recognize that the distributions of species can be influenced by the spatial and temporal distribution of suitable habitat (Thomas & Hanski 1997; Leibold et al. 2004; Cottenie 2005). Despite this very rich theoretical arena, and supporting evidence from a range of empirical studies for most of these concepts (e.g. Thomas, Singer & Boughton 1996; Elmhagen & Angerbjorn 2001; Muneepeerakul et al. 2008), the relative importance of any of these concepts in explaining spatial patterns in nature is unknown. There is very little information about the proportion of species that conform to particular spatial ecological theories (Driscoll 2007; Driscoll & Lindenmayer 2009). We aim to help address this knowledge gap by examining the occurrence and abundance of beetle species in a large-scale naturally fragmented landscape.

Correlates of a species’ distribution among a set of habitat patches can reveal the most important processes influencing spatial dynamics. For example, extinction and colonisation in a classic metapopulation can lead to lower occupancy in smaller patches due to higher rates of extinction, and lower occupancy in the most isolated sites due to reduced chances of recolonisation (Etienne, Ter Braak & Vos 2004). Therefore, if the occurrence of a species is negatively related to patch isolation and positively related to patch size, classic metapopulation dynamics would be supported (Hanski & Simberloff 1997; Etienne et al. 2004; Matthies et al. 2004), although without population turnover data, these cannot be distinguished from source-sink metapopulations (Driscoll 2008). Contrasting with the emphasis that theoretical ecology places on habitat isolation, there is evidence that the total amount of surrounding habitat can strongly influence species’ distributions, regardless of the amount of fragmentation (Fahrig 1997, 2002; Harrison et al. 2006; Radford & Bennett 2007).

However, for many species, the geography of patches may be less important than patch quality (Harrison & Bruna 1999; Jellinek, Driscoll & Kirkpatrick 2004). Spatial dynamics within habitat archipelagos can be deterministic, with colonisation and extinction driven by the state of the habitat patch (Thomas 1994). For example, habitat degradation (Harrison & Bruna 1999), succession after a disturbance (Stelter et al. 1997) or after habitat creation (Sjogren-Gulve 1994) can be major drivers of spatial population dynamics. If occurrence is influenced by patch quality, a deterministic metapopulation is supported. A deterministic metapopulation differs from a classic, or source-sink metapopulation, because the latter assume that patches are always available to a species and that stochastic extinction and colonisation account for the dynamics (Thomas 1994).

Niche theory not only encompasses the occurrence pattern described by the deterministic metapopulation concept, but also describes differences in density that arise from habitat specialisation (Hutchinson 1957; Leibold et al. 2004; Richter-Boix, Llorente & Montori 2007). If abundance but not occurrence of a species is dependent on habitat characteristics, metapopulation dynamics are not supported, though niche-partitioning would be inferred.

Interactions among species can also influence occurrence patterns (Taylor 1990; Leibold et al. 2004). When dominant competitors or predators have lower dispersal ability than subordinate competitors or prey, the subordinate species can find refuge in more isolated sites (Tilman 1994; Yu et al. 2004). Subordinate species may therefore have higher occurrence in the most isolated sites, a pattern opposite to that expected under single-species metapopulation theory (Driscoll 2008).

We examined the occurrence and abundance of beetles in 31 natural cool-temperate rainforest patches which ranged in size and isolation. Floristic composition also varied among sites in response to a productivity gradient and succession after fire. Our aim was to determine how many species showed evidence that was consistent with theories describing species and community responses in fragmented landscapes. We took advantage of the information harboured in species’ distributions (Etienne et al. 2004) to test for evidence of:

• 1
   Classic and source-sink metapopulations (negative relationship between occurrence and patch isolation, positive relationship with patch area, only a proportion of patches occupied, occurrence not influenced by environmental variables) (Levins 1970; Pulliam 1988; Hanski 1998).
• 2
   Deterministic metapopulations (occurrence related to environmental variables, not geographic variables, only a proportion of patches occupied) (Thomas 1994).
• 3
   Habitat-amount theory (positive relationship of occurrence or abundance with amount of surrounding forest, not with patch isolation or size) (Fahrig 1997, 2002).
• 4
   Dominant-species/dispersal tradeoffs (higher occurrence in most isolated patches) (Taylor 1990; Leibold et al. 2004).
• 5
   Niche theory (in addition to evidence from point 2 above, abundance related to environmental parameters) (Hutchinson 1957; Leibold et al. 2004; Richter-Boix et al. 2007).

We examine a large number of species from the same fragmented landscape to assess the likely general importance of these contrasting ecological theories.

Materials and methods

Study area

In south-west Tasmania, Australia, cool-temperate rainforest occurs naturally in discrete patches surrounded by sedgeland (Read 1999). The patches range in size from a few hectares up to several square kilometres (Jarman, Brown & Kantvilas 1984; Kirkpatrick & Dickinson 1984). The rainforest varies in plant species composition (Jarman, Kantvilas & Brown 1999), and forest patches can occur in a range of post-fire successional stages (Jackson 1968). The forest beetle communities are distinct from those in the surrounding sedgeland matrix (Driscoll 2005).

We have deliberately eschewed selecting an ‘ideal’ study landscape, with patches of a prescribed size, isolation and with uniform habitat. We want to know how important spatial ecological theory is in a large-scale naturally fragmented landscape where there is substantial variation in patch geography and, as in most landscapes, the plant communities are heterogeneous.

Beetle collection

Beetles were sampled from 31 rainforest patches in two adjacent regions (Fig. 1). Region one has slightly higher precipitation with infertile soils formed on quartzites and phyllites whereas many sites in region two are underlain by fine-textured sedimentary rocks. Site productivity is driven by rainfall, topography, altitude and soil fertility (Kirkpatrick 1984), which results in a bias towards more productive sites in region 2.

Figure 1.  Each of 31 rainforest patches (10 in region 1 from the north-west, 21 in region 2 from the south-east) were sampled using 48 pairs of pit-fall traps in south-west Tasmania, Australia. Hashed area = hydro-electric dams, dark grey = forest, light grey = sedgeland, dashed line = road, white circles = sampled patch. The white arrow in the top centre of the right hand panel indicates the outlier population (B13).

Each patch was sampled using three trap grids separated by c. 100 m (Driscoll, 2009). Equal sampling effort was used regardless of patch size to ensure that detection rates were proportional to population density and not sampling effort. Grid locations were stratified to sample different successional ages in the patch. In the absence of fire, forest patches expand into the buttongrass matrix so usually have a fringe of early successional habitat. Each grid consisted of 16 pairs (32 cups in total) of 225-mL plastic cups used as pit-fall traps. We used plastic container lids suspended on wooden skewers to prevent the cups from filling with rain, hail or snow. Pairs were spaced at 5-m intervals, and cups within pairs separated by 1 m. Each cup had 50 mL of Gault’s solution as a preservative (Walker & Crosby 1988). The traps were set in January 2002 and left open for 8 weeks. We used this very large sampling effort (96 pit-fall traps per site) to ensure that a reasonable number of species would be detected reliably, if they were present (Driscoll, in press). Trapped animals were preserved in ethanol, and all identifications were completed by the same person (KB). Vouchers are housed in the Tasmanian Forestry Insect Collection, Hobart, code numbers FT44089 to FT44145.

Plant surveys

Percentage cover of all plant species within each of the pit-fall grids (three grids of 15 × 15 m) was estimated in three categories: <20%, 20–50% and >50%. The size of tree species was estimated in three categories: (i) young [<15 cm diameter at breast height (DHB)]; (ii) mid (not reaching canopy, 15–50 cm DBH); and (iii) mature and old-growth (DBH > 50 cm, canopy tree). Nomenclature follows Buchanan (2006).

Amount of habitat and isolation measures

Geographic parameters were the area of forest within 300 m of grid B (usually located between the other two) and the area of forest within 300 m of the patch boundary (isolation). The distance of 300 m was chosen to maximize dispersion of values across the range of possible values. We estimated the area of forest around the points or patches at 100-m intervals (100–1000 m) and 250-m intervals (1000–2000 m). We then counted the number of patches or points in 10% interval categories of per cent forest cover. The buffer distance that best separated the sites was where the maximum number of the 10 categories included sites and in approximately equal numbers. The average proportion of forested area within 300 m of a sample point was 0·49 (SD 0·24), and the average proportion of forested area in a 300 m buffer around patches was 0·21 (SD 0·21). The buffer values therefore spanned a broad range of possible values. We examined the influence of using a 1000 m buffer and found this made no difference to the apparent prevalence of contrasting processes.

Environmental variables

To summarize plant community differences between patches we used non-metric multidimensional scaling (NMDS) with a Bray–Curtis distance matrix. The distance matrix was based on the plant species per cent cover data, with values for each species averaged across grids within sites. NMDS using two dimensions was completed using functions initMDS and isoMDS of r version 2.5.0 (R Development Core Team, 2007). Multiple starts of the NMDS procedure were undertaken to avoid being trapped in local minima. New starts were continued until the stress had not been reduced for 100 successive iterations. The maximum iteration within isoMDS was 200, and convergence tolerance 10⁻⁷. Procrustes rotation ensured that most of the variation was captured in axis 1 (vegan 1.8, Oksanen et al. 2007). To assist with interpretation of the two NMDS axis, we used anova to examine the relationship of each plant species, region and forest succession stage with each NMDS axis.

We also estimated the stage of post-fire succession. Eucalypts dominate the early stages of succession, but a dense canopy of the rainforest species Nothofagus cunninghamii, Eucryphia lucida and Atherosperma moschatum prevents eucalypt regeneration (Jackson 1968). After the eucalypts die (100–450 years), rainforest remains (Jackson 1968; Read 1999). The size and abundance of eucalypts compared with rainforest trees indicates successional status. Tree growth rate is influenced by nutrient availability (Read 2001), so tree size cannot be used to accurately date the stand (Koch, Driscoll & Kirkpatrick 2008). We devised a succession scale to represent the gradient from (1) pure rainforest through to (3) eucalypt stands:

• 1
   No eucalypts and either Nothofagus of size category 2 (150–500 mm) or 3 (>500 mm), or Atherosperma at size 2 or Eucryphia at size 3 or other rainforest understorey species predominant (two sites only).
• 2
   Eucalypts at size 3 and Nothofagus, Eucryphia, or Phyllocladus at size 2 or 3.
• 3
   Eucalypts at size 1 (<150 mm) or 2 and Nothofagus or Eucryphia present, or eucalypts at size 2 and 3 with no Nothofagus, Eucryphia or Phyllocladus.

The minimum score from the three plant grids was used to represent the latest stage of succession present in the patch.

Step-wise generalized linear models

We analysed separately presence/absence data and frequency data (number of trap-pairs in which a species was captured in a patch; a surrogate for abundance).

To account for spatial autocorrelation (Gonzalez-Megias, Gomez & Sanchez-Pinero 2005) prior to fitting a step-wise model, six principal coordinates of neighbour matrices (PCNM) were calculated (Borcard & Legendre 2002; Dray, Legendre & Peres-Neto 2006). For each analysis, we fitted the PCNM variables in a binomial generalized linear model (GLM) with a logit link-function (McCullagh & Nelder 1989). We permuted the beetle frequency data 1000 times to obtain a distribution of r² (deviance explained by factor/total deviance) and this was used to estimate P-values for each of the six variables (proportion of random r²-values ≥ actual value). We adjusted the P-values to control the false discovery rate (Benjamini & Hochberg 1995) using p.adjust in the r package stats 2.5.0 (R Development Core Team, 2007). Each species and analysis type (presence/absence, frequency) was considered a separate family. PCNM variables with a q-value of ≤0·1 were used as the base model for the step-wise procedure. An alternative approach of using analysis type alone to designate family was extremely conservative with all P-values above zero having q-values >0·1. This did not capture obvious spatial structure, such as the near absence from one region but not the other, and so the less conservative approach was used.

Using a simultaneous forwards and backwards step-wise approach (McCullagh & Nelder 1989) with AIC as a stopping rule, we fitted three vegetation variables (MDS axis 1, MDS axis 2, succession score) and three geographic variables (patch area, patch isolation, forest area) to the base model containing any significant PCNMs. Variable selection procedures produce over-optimistic P-values (Maindonald & Braun 2007). We therefore tested the significance of fitted parameters using a permutation method (Arditi 1989; Good 1994). We permuted the beetle species data among sites 1000 times and with each permutation re-fitted the parameters selected by the step-wise procedure then calculated r² as the test statistic. We adjusted P-values (Benjamini & Hochberg 1995) for all variables included in the models, based on all variables available for selection. We report results where the adjusted P is <0·1 to reduce the risk that low power leading to type 2 errors may obscure support for metapopulation processes.

When using step-wise models, the inclusion of a variable that explains the most variation may obscure slightly weaker relationships with other variables if they are correlated with the first variable. To determine if this was a problem in our analyses, we examined the amount of independent and joint variation explained by the environmental, geographic and spatial PCNM variables, using hierarchical partitioning (Chevan & Sutherland 1991; MacNally 1996). We only included PCNM variables that were significantly correlated with a species to simplify the model before analysis. Significance of the independent variation explained was ascertained by permutation.

Species analysed separately occurred in at least 20% of patches (six) and had a frequency of at least 0·049 within patches in which they were trapped (probability of detection with 48 traps >90%McArdle 1990). Species’ frequency is a reasonable guide to the probability of detection (Driscoll, in press). Six patches were used as an arbitrary cut off below which we did not expect reliable models to arise. Species that occurred in 20–80% of patches were used in the presence/absence analyses. Verheyen et al. (2004) also used a 20–80% occupancy range, beyond which they regarded metapopulation dynamics as unlikely or undetectable.

We applied a similar step-wise approach, but using an anova model, for the number of species in a patch, and the number of rare species. Rare species were those with a frequency of <0·049.

Species–environment relationships within patches

Pooling data from the three grids has the potential to obscure relationships between beetle abundance and environmental parameters. We therefore performed similar analyses to those described for patch-level data, but analysed beetle frequency data and plant data from each grid within a patch. We first fitted patch-name to remove patch-level variation. We then fitted three environmental parameters in the step-wise binomial GLMs, including succession score for each grid and two NMDS axes from analyses of plant species data at the grid level. The grid-level NMDS axes had the same biological interpretation as the patch-level axes (see Results).

Examining species traits such as flight capacity can provide valuable additional insight into the response of species to geographic and environmental variables (Driscoll & Weir 2005), but our analyses in this respect were not very informative (Appendix S1).

Results

Although 168 morphospecies were identified from 26 532 individuals, only 44 species satisfied the criteria for species-level analysis (Appendix S2). Eighteen of 44 species occurred on more than 80% of patches. One site (B13) had a unique beetle and plant community compared with the rest of the sites. It was the only site on the floodplain of a large river and has been excluded from further analyses (Appendix S3).

Environmental axis

MDS axis 1 represented the forest succession gradient with early succession eucalypt communities at low values and late-successional rainforest at high values (mean axis 1 value for succession scores 1 vs. 2 + 3 were 0·138, −0·2415, P = 0·003). Plant species that were most negatively related to MDS axis 1 included Melaleuca squarrosa (subcanopy tree, r² = 0·69, P < 0·0001); Eucalyptus nitida (canopy tree, r² = 0·60, P < 0·0001); Gleichenia microcarpa (fern, r² = 0·61, P < 0·0001); Bauera rubioides (shrub, r² = 0·60, P < 0·0001); Nematolepis squamea (shrub, r² = 0·57, P < 0·0001); and Pteridium esculentum (fern, r² = 0·69, P < 0·0001). These species are typical of sclerophyllous plant communities in SW Tasmania. Plant species most positively correlated with MDS axis 1 included Grammitis billardierei (fern, r² = 0·45, P = 0·0001); Atherosperma moschatum (canopy tree, r² = 0·28, P = 0·0026); Eucryphia lucida (canopy tree, r² = 0·24, P = 0·0051); Nothofagus cunninghamii (canopy tree, r² = 0·18, P = 0·019) and are typical rainforest species.

Mean scores on MDS axis 2 were significantly differentiated by region (region 1 vs. 2, −0·1229, 0·0668, P = 0·040). The scores on this axis had a positive relationship with Anodopetalum biglandulosum (usually subcanopy tree, r² = 0·32, P = 0·0012), a species characteristic of the more tangled rainforests on low-nutrient sites. The scores also had negative relationships with Nothofagus cunninghamii (canopy tree, r² = 0·35, P = 0·0006), Phyllocladus aspleniifolius (canopy tree, r² = 0·38, P = 0·0003), Blechnum wattsii (fern, r² = 0·43, P = 0·0001) and Trochocarpa gunnii (shrub, r² = 0·41, P = 0·0001), species that are widespread in tall rainforests with sparse understorey. We therefore interpret low values of MDS axis 2 as an indicator of the more productive sites and callidendrous to thamnic rainforest (sensuJarman et al. 1984) with or without emergent eucalypts, and high values as an indicator of less productive sites and thamnic to implicate rainforests (Jarman et al. 1984) with or without emergent eucalypts (later referred to as low-nutrient patches).

Generalized linear models

The hierarchical partitioning analyses usually either supported the GLM results, or were not significant. Only two GLM results were contradicted by the hierarchical partitioning analyses (Roptoperus tasmaniensis, Myrmicholeva ligulata, Appendices S3 and S4). These made no difference to our interpretation or conclusions. The key message from the hierarchical partitioning analyses is that relationships between patch geography or the environment and beetle species were not obscured in the step-wise models by covariation of the explanatory variables.

Spatial parameters derived from the PCNM analysis were included in occurrence models of nine species and in frequency models for 13 species. PCNM1 separated patches from the northern and southern regions. Four species had lower occurrence or lower frequency in the southern region while nine species had the opposite pattern (Appendix S5). PCNM2, 3, 4 and 5 each influenced occurrence or frequency of six, four, one and three species respectively. Although difficult to interpret, their inclusion in species’ models indicates the removal of some spatial autocorrelation effects prior to examining environmental and patch geography relationships.

Next, we focus on the environmental and geographic results of the GLM. The results are presented in order of the five processes listed at the end of the introduction. We not only draw attention to species that show evidence in support of each process, but also note when those species appear to have other processes acting simultaneously.

Classic and source-sink metapopulations

Of the 26 species occurring on 20–80% of patches, only one species had a pattern consistent with classic or source-sink metapopulation predictions. Catoposchema tasmaniae occurred more frequently in larger patches (Appendices S2 and S5). However, three other species had higher frequency on the least isolated patches (Lissotes curvicornis; Palimbolus victoriae; Telura vitticollis; Appendices S2 and S5).

Deterministic metapopulations

Environmental parameters influenced the occurrence of 10 species suggesting they have a deterministic metapopulation, although three of these species also had their occurrence influenced by geographic parameters. Four of these species had higher occurrence on low-nutrient sites (Aridius nodifer, Aspidiphorus humeralis, Palimbolus victoriae, Pselaphinae TFIC sp. 29) and occurrence of the latter species was also higher with more surrounding forest (Appendix S2). Six species had higher occurrence in late successional rainforest (Austronemadus TFIC sp. 03, Brycopia hexagona, Dinichus terreus, Pedilophorus gemmatus, Pterocyrtus globosus, Pterocyrtus tasmanicus), although occurrence of Dinichus terreus was also higher in smaller patches, and P. gemmatus occurred more often with more surrounding forest (Appendices S2 and S5).

Amount of surrounding habitat

The prediction that the amount of surrounding habitat should influence occurrence was supported by two species (Pselaphinae TFIC sp. 29, P. gemmatus), which had higher occurrence in sites with more surrounding forest. The latter species also had higher frequency and occurrence on low-nutrient sites while P. gemmatus also had higher occurrence in rainforest. Seven additional species had higher frequency in sites with more surrounding forest (Aspidiphorus humeralis, Austronemadus TFIC sp. 03, Coripera deplanata, Decilaus striatus, Roptoperus tasmaniensis, Sogdini ANIC Gen B. TFIC sp. 01, Zeadolopus TFIC sp. 02), though the first two and last species also had frequency and/or occurrence influenced by environmental parameters (Appendices S2 and S5).

Dominant-species/dispersal tradeoffs

Three flightless species had higher occurrence in small patches (Sloaneana tasmaniae, Trechimorphus diemenensis, Dinichus terreus, the latter also with occurrence higher in late-successional rainforest, Appendices S2 and S5).

Niche effects on abundance

The frequency but not the occurrence of 10 species was only influenced by environmental parameters (though three of these included PCNM variables in their models). Four of those species were more frequent in late-successional rainforest (Myrmicholeva ligulata; Nargomorphus globulus; Pselaphaulax CHANDLER Tasmania 1; Rybaxis parvidens), four on low-nutrient sites (Microchaetes hystricosus; Thalycrodes cylindricum; Scydmaenidae TFIC sp. 04; Thalycrodes pulchrum), one in patches with eucalypts and in low-nutrient sites (Heteronyx tasmanicus), and one in high-nutrient sites (Decilaus TFIC sp. 19, Appendices S2 and S5). Two additional species had higher frequency in low-nutrient sites, but these also had frequency influenced by amount of surrounding forest (Zeadolopus TFIC sp. 02) or patch isolation (Telura vitticollis).

Within patches, at the grid level, environmental parameters explained from 4% to 12% of model variation for 14 species (at P < 0·1, Appendix S6). Two of these reflected the same response as observed at the patch level of analysis, with higher frequency or occurrence towards the rainforest end of the succession (Pterocyrtus tasmanicus and Pterocyrtus globosus). Nine species that had not shown a relationship with environmental parameters at the patch scale, did so at the grid scale, including eight species that were more frequent in rainforest (Agonica simsoni, Choleva TFIC sp 01, Decilaus striatus, Decilaus TFIC sp 04, Pseudomicrocara TFIC sp 02, Roptoperus tasmaniensis, Stichonotus leai, Trechimorphus diemenensis), and one in low-nutrient sites (Pseudomicrocara TFIC sp 01). Three species that had higher frequency in low-nutrient sites at the patch scale had higher frequency in eucalypts at the grid scale (Aridius nodifer, Thalycrodes cylindricum, Telura vitticollis, Appendix S6). There were 17 species with environmental relationships at the patch level but without a relationship with the environment at the grid level (Appendices S2, S5 and S6).

Number of species

The number of beetle species in a patch and the number of rare species were significantly positively correlated with MDS axis 2, implying more species in low-nutrient sites (Appendix S5). The number of species and the number of rare species were also significantly negatively correlated with succession score, with more species in late-successional rainforest (Appendix S5).

Summary of patterns that were consistent with theoretical expectations

Of 44 common species, only 2·3% of species showed evidence to support a classic metapopulation model, whereas 22·7% supported a deterministic metapopulation, with patch occupancy influenced by the stage of forest succession and rainforest type (Fig. 2). The occurrence or frequency of 20·5% of species was strongly influenced by the amount of habitat surrounding the sample site, regardless of patch size or isolation. Three common beetle species (6·8%) had higher frequency or occurrence in small patches, which may reflect negative interactions with dominant-species in large patches. Twelve species (27·3%) showed evidence of weak niche-partitioning at the patch scale, with habitat type influencing only their frequency. Taking into account within-patch environmental relationships, 21 species (47·7%) showed weak niche-partitioning (Appendix S6).

Figure 2.  Summary of the number of species showing support for each of the theories. The boxes for each row indicate the evidence supporting each theory and the number of species showing that evidence. Species with higher frequency on the least isolated sites in the metapopulation row do not support a metapopulation, but imply that a similar process of dispersal limitation without population turnover. Double headed arrows indicate where one species had more than one response. Species with occurrence and frequency showing the same response are only counted under occurrence and the arrow links are omitted.

Discussion

Classic and source-sink metapopulations

Only one beetle species had the positive relationship between occurrence and patch size that was consistent with classic and source-sink metapopulation predictions (Hanski & Simberloff 1997; Etienne et al. 2004). Three more species had a lower frequency on isolated patches, consistent with a boost in beetle density by higher immigration into the least isolated sites (Brown & Kodric-Brown 1977; Pulliam 1988; Leibold et al. 2004). If this pattern was generated by dispersal limitation, metapopulation dynamics may become possible for those species in landscapes with more isolated patches. The rarity of evidence for classic or source-sink metapopulation dynamics is not consistent with the results of most other studies (Table 1), but is consistent with the results of Driscoll (2008), despite that work taking place at a much smaller spatial scale.

Table 1.   Examples of studies that have assessed the prevalence of metapopulation dynamics

Proportion of community that may have a metapopulation	Study system	References
15% of plant species 75% of specialist plants	Rosemary scrub in Florida USA	Quintana-Ascencio & Menges (1996)
37% of forest-specialist plant species	Regenerating English forests	Verheyen et al. (2004)
58–66% of moths	Rocky islands in Finland	Nieminen & Hanski (1998)
54% of common water plants	Swedish lakes	Dahlgren & Ehrlen (2005)
7·5% of beetle species	Eucalyptus-dominated fragments in Tasmania, Australia	Driscoll (2008)

Eighteen species occurred on >86% of patches and so probably do not form metapopulations. We suggest that high rates of migration between patches may prevent populations of these species from ever becoming extinct (Harrison 1991). Alternatively, the populations may be isolated, but are so large that the time to extinction exceeds the time over which the patches change configuration in response to changing climatic conditions (on the order of thousands of years, Leigh 1981; Jackson 1999).

A key assumption of our interpretation is that dispersal ability limits access to patches (Hanski & Simberloff 1997). However, if establishment was limiting, isolation effects would not be observed (Eber & Brandl 1996; Verbeylen, De Bruyn & Matthysen 2003). If establishment limitation was stochastic, the process would be consistent with Levins’ simple metapopulation model (Levins 1970). However, deterministic processes could also limit colonisation (such as natural enemies: Ryall & Fahrig 2006). Conceivably, seven species that occurred on 20–80% of sites and that did not have significant model effects at the patch scale, could have establishment-limited metapopulations. However, we can find few examples of this kind of metapopulation in the literature (Werth et al. 2006) and dispersal limitation is commonly assumed to be more realistic (Hanski 1998). Population turnover data, genetic data and knowledge of competitor or predator distributions could help distinguish dispersal from establishment-limited processes (Driscoll 2007).

Deterministic metapopulations

Deterministic processes influenced the occurrence of more species (10) than did isolation and size (1). Deterministic metapopulations may be more common in nature than classic metapopulations in which dynamics are driven by stochastic processes (Thomas 1994). We note also that niche concepts (Hutchinson 1957), including the species-sorting metacommunity concept (Leibold et al. 2004) can provide an equally valid construct for describing (presumed) spatial dynamics that are driven by differences in the environment.

Habitat amount

The amount of surrounding habitat can have a strong influence on species’ distributions (Fahrig 1997, 2003; MacDonald & Kirkpatrick 2003; Driscoll 2004; Radford & Bennett 2007). Habitat amount influenced the occurrence of two species, implying they may have spatial dynamics that are similar to the classic metapopulation. However, direct connectivity, which defines the patches of a metapopulation, appears to be less important than the area of near-by habitat. Occupancy may be lower with less surrounding habitat through lower immigration (Brown & Kodric-Brown 1977), higher emigration (Hill, Thomas & Lewis 1996; Fahrig 2001) or increased immigration of dominant matrix species (Turner 1996; Harrison & Bruna 1999). We think the latter explanation is less likely because there was very little overlap of matrix and forest beetle fauna (Driscoll 2005, 2008). Similar processes may account for the influence of amount of surrounding habitat on the frequency of seven additional species. These findings emphasize that the patch-based model of a landscape, with the assumption that a separate panmictic population resides within each patch, does not always apply because spatial population dynamics occurs at a different, in this case smaller, scale (Lindenmayer, McIntyre & Fischer 2003).

Dominant-species/dispersal tradeoffs

Three species had higher occurrence on small patches. They were all flightless, and therefore had a distribution inconsistent with the dispersal-limited species-sorting metacommunity model. The model suggested that species that would otherwise be out-competed or devoured may survive in remote patches because of their superior dispersal ability (Tilman 1994; Driscoll 2008). These three species may be finding refuge on patches too small to support persistent populations of their predators and competitors. However, we have only examined spatial patterns and more direct approaches to identifying competitive or predatory relationships are now needed (e.g. Yu et al. 2004). Small or isolated patches may be important refuges for a suite of presumably subordinate, forest-specialist beetles (Tscharntke et al. 2002; Driscoll 2008).

Niche theory

Environmental parameters influenced the frequency of 21 species but not their occurrence and so we did not discuss them in the context of deterministic metapopulations. The influence of environmental parameters on frequency could reflect differential reproduction and survival in different habitats (Barker & Mayhill 1999) or habitat selection (Fletcher 2007). Our results are consistent with recent evidence that species-sorting metacommunity processes have the strongest influence on community composition (Cottenie 2005; Parris 2006; Brooks et al. 2008).

Species with an environmental relationship at the patch scale, but not within patches at the grid scale imply that habitat type within a patch is important for reproduction, but that animals can disperse throughout the patch, eliminating any differences between habitats within patches (mass effects, Shmida & Ellner 1984; Leibold et al. 2004). On the other hand, eight species with higher frequency in rainforest at the grid but not the patch scale probably reflect methodological limitations. Patch-level environmental measures may not reflect habitat availability within patches, preventing us from detecting an effect at the patch scale (Whittingham et al. 2005).

In support of niche theory, post-fire succession was characterized by increased frequency and addition of species, rather than species turnover. This additive effect contrasts with beetle succession in boreal forests, which has more distinct communities (Paquin 2008). However, evidence from eucalypt forests younger than those that we sampled suggests there is a suite of early successional species in Tasmania (Baker 2006), similar to that observed in the boreal forest (Paquin 2008). A clear implication of our results for management is that burning or logging old-growth forests and rainforest will reduce local beetle diversity.

Although most of the PCNM spatial variables could not be interpreted, the main spatial pattern was the difference between the northern and southern regions (PCNM1). This pattern could partly be driven by environmental differences because region was correlated with MDS axis 2, with a tendency for more low-nutrient sites to occur in the north. Environmental filters therefore seem to act at multiple spatial scales, with species distributions influenced by factors that vary within patches at the grid level, between patches and between regions.

Conclusions

Within a large-scale, naturally fragmented ecosystem, we have shown that beetles have a broad range of responses, exhibiting patterns that are consistent with a diverse range of theory (Fig. 2). Metapopulation theory widely permeates the conservation and ecological literature (Hanski & Gaggiotti 2004), but in our study landscape, evidence for classic or source-sink metapopulations was rare. Part of the reason for this was that a proportion of species responded to the amount of surrounding habitat at a scale smaller than the patches. Habitat patches did not determine the structural basis of dynamics for many species, contrary to metapopulation assumptions. Nevertheless, one-fifth of common beetles may have been subject to the same dispersal limitation that can lead to metapopulation dynamics. Dispersal limitation may have influenced community development, even though metapopulations were rare.

In contrast to the limited evidence supporting spatial processes, niche concepts were widely supported. Vegetation type more frequently influenced occurrence than habitat size, amount or isolation, suggesting that deterministic spatial dynamics are common relative to stochastic dynamics (Hanski & Simberloff 1997). If this is generally true, spatial dynamic models may need to incorporate environmentally driven population turnover rather than just stochastic processes (Verheyen et al. 2004; Wilcox, Cairns & Possingham 2006).

We hope that more studies take the approach of examining multiple species in large-scale fragmented landscapes. Turnover data and genetic analyses would be valuable additions to our approach to enable recognition of establishment-limited rather than just dispersal-limited metapopulations. Although individual species studies will continue to be important for exploring mechanisms (e.g. Lindenmayer 2000), taking a multi-species approach in a natural ecosystem is essential for assessing the importance of competing theories, and thereby for understanding how communities work.

Acknowledgements

We thank Madelaine Hanson for assistance in the field, Amelia Koch for her feedback on an earlier draft and Simon Grove for access to Forestry Tasmania’s beetle collections. Jeff Wood and Hwan-Jin Yoon provided statistics advice. This research was completed under Tasmanian Parks and Wildlife Service permit number FA 01030, complying with relevant State and Federal laws. The field work was funded through an Australian Research Council Post Doctoral Fellowship, which DD held in the School of Geography and Environmental Studies at the University of Tasmania. Beetle collections were sorted by KB with funding to DD from Flinders University of South Australia. The project was written up at The Australian National University.