We investigated edge-related changes in the abundance of the 78 most common beetle species collected (see
Methods), of which 13% exhibited no significant gradient in abundance across habitat edges, 32% had significant but weak responses to edges [
supporting information (SI) Appendix], and 5% were abundant deep in the grassland matrix but not present in the forest (
SI Appendix). For the remaining species with strong edge responses we could calculate edge penetration distances (the distance over which habitat edges influenced species abundances within forest fragments). Edge effects penetrated >250 m inside the habitat fragments for 20% of all common species (
S = 15), and as many as one in eight species had edge penetration distances that appeared to extend as far as 1 km inside forest fragments ( and ). This was a significantly greater proportion than expected by chance alone, based on a comparison with a null model that was constructed by sampling variation in species' abundances along a comparable 2-km gradient within continuous habitat (
SI Appendix; binomial proportion test:
Z = 3.03,
P < 0.005). The null model suggested that one in 19 species would show kilometer-scale edge effects by chance, indicating that the true proportion of species with such large-scale edge responses may be as low as one in 13 species. This accords well with visual inspection of the figures, which indicate that, despite strong statistical fits, the scale of edge effects may have been overestimated for some species (
e). Sensitivity analysis showed that this result was robust to variation in the power threshold for determining common species (see
Methods), down to a lower threshold of
n ≥ 10 at which 12% of 199 common species exhibited large-scale edge responses (SI Fig. 7 in
SI Appendix). Of the species with kilometer-scale edge effects, all but one (
d) were forest-interior species that declined in abundance near forest edges (SI Table 1 in
SI Appendix). Interestingly, for some of the species with large-scale edge responses, abundance values still did not converge on those recorded in deep forest (
f), suggesting that the true edge effect might extend much farther than the 1-km scale over which we sampled. This raises the obvious but unanswered question of what scale one might have to sample at to measure the full spatial extent of edge effects on invertebrates and other organisms.
We stress that most of the common forest species investigated in this study were also present at low numbers closer than 1 km to the edge. However, the relative abundances of those species were atypical of a forest-interior community, and such changes to community biostructure may have important implications for population persistence (
6) and ecosystem functioning and stability (
13,
14) in fragmented landscapes. To test edge responses across different functional groups involved in a range of ecosystem processes, we classified all species into six feeding guilds (
SI Appendix). When analyzed at the guild level, the combined abundance of the 80 saprophagous species 1 km within forests was almost four times greater than at 250 m from the nearest forest edge (SI Fig. 8 in
SI Appendix; 4.3 ± 2.5 SE and 1.1 ± 0.2 individuals per m
2 per day, respectively). The abundance and biomass of saprophages are thought to limit rates of energy flow in many ecosystems (
15), so such large reductions in abundance near habitat edges likely have important implications for nutrient cycling processes and ecosystem productivity. No other guilds were comparable in edge sensitivity to saprophages, and only herbivorous beetles were more abundant in the matrix and declined in abundance inside the forest.
The high proportion of common species responding to habitat edges over large spatial scales resulted in the composition of deep-forest beetle communities differing from that within even 1 km of the forest edge. We quantified these changes in community structure with respect to edge distance using a variety of indices (
SI Appendix and ), which exhibited a clear peak in species diversity at the forest edge (
a) that was caused by a mixing of forest- and grassland-inhabiting species (
16). The high α-diversity (within sample diversity) near edges largely resulted from an increased incidence of singleton species (
b), many of which were likely vagrants from adjoining habitats. This led to elevated spatial turnover (β-diversity) among samples within edge habitats relative to turnover among samples within deep-forest interior habitats (
c). Finally, an analysis of community composition (
SI Appendix) showed that these striking large-scale gradients in diversity across the forest edge were mirrored by a change in the structure of beetle assemblages that also appeared to penetrate >1 km into the forest (
d). Moreover, 43 of the 691 rarer species (<40 individuals) were found only in deep forest 2 km from the nearest edge in the largest remaining forest tract in New Zealand. Although this was not significantly more than expected by chance (
SI Appendix), these rare species did include one of the largest ground beetles (Carabidae) in New Zealand—a 60-mm undescribed species of the endemic genus
Mecodema that also includes four species that are endangered or in serious decline, and a further 16 that are classified as sparse or range-restricted.
Taken together, the patterns of α- and β-diversity, singletons, and community composition all indicate that edge habitats might be hyperdynamic (
17,
18), exhibiting increased variability and less stable community structure than in forest interiors. Importantly, this suggests that ecological processes operating <1 km from forest edges might also be altered relative to those in the deep forest. One reason for this could be that microhabitats closer to the edge may be more heterogeneous and more strongly affected by stochastic dispersal processes. Structuring processes near edges may be dominated by local disturbance effects, such as tree damage and mortality, which are known to be elevated at up to several hundred meters inside forest fragments (
19,
20), and by the relative dispersal abilities of individual species (
21).
Our results differ markedly from prior studies of edge-effect impacts on invertebrates. Quantitative summaries of the edge-effects literature have suggested that changes in invertebrate abundance and species richness typically penetrate just 30–100 m into forest fragments and that changes in community composition penetrate just 200–300 m (
3,
19). By contrast, large-scale edge effects occurring over gradients of >1 km have only been proposed for some highly dispersive mammal species (
11) but never empirically measured, and the validity of such deep-penetrating effects has been contested (
22). Our data indicate that large-scale edge effects may be more common than previously realized and that they can influence even small organisms such as invertebrates.
We identify two reasons for the lack of congruence between our results and previous estimates of the scale of edge effects. First, and self-evidently, the tendency to restrict sampling designs to relatively short edge gradients prevents large-scale edge effects from being detected (
3) (
SI Appendix and SI Fig. 9 in
SI Appendix). Second, it is possible that edge effects in other regions do not occur over large spatial scales because the species that are most sensitive to edges may have been the first to be lost by an extinction filtering process (
23), leaving behind only those species that are relatively insensitive to habitat edges.
In the present study we measured 74 environmental variables to quantify changes in vegetation composition, microclimatic conditions, landscape structure, and the topographic and geological features underlying our sampling sites to eliminate the possibility that the large-scale beetle species responses we report are a trivial response to large-scale variation in the environment that may covary with distance to edge (SI Table 2 in
SI Appendix). No environmental variables exhibited kilometer-scale edge effects, and nearly all variables (
n = 71; 93%) exhibited either no edge effect or effects that penetrated only short distances inside the forest (SI Fig. 10 in
SI Appendix), suggesting that large-scale edge responses may be driven more by variation in biotic interactions than by environmental conditions.
The population-level impacts of edge effects scale exponentially with the distance over which edge effects penetrate habitat fragments (
6). A recent model of edge effects considered kilometer-scale edge effects to be a “worst-case” scenario and showed that they may reduce the population size of core-dwelling species by up to 80% (
6). Large-scale edge effects can also lead to the local extinction of species that are restricted to habitat interiors—the most threatened species in fragmented landscapes (
24)—from both habitat fragments (
6) and protected natural areas (
5). One widespread approach to conserving species in fragmented landscapes is to create habitat corridors to unite habitat fragments (
25). However, corridors consist almost exclusively of edge habitat, making it unlikely that interior-dwelling species will benefit from their construction (
6). The surprisingly high proportion of species that appear to have large-scale edge effects indicates that, at landscape scales, the conservation value of corridors may be considerably lower than expected. We suggest that a more appropriate conservation strategy may be to direct habitat restoration programs toward enlarging core areas, which can be achieved by widening the narrowest sections of large fragments. However, we recognize that at larger spatial and temporal scales the mitigation of global conservation challenges such as climate change may require corridor networks that transcend national and geographic boundaries.
Kilometer-scale edge effects such as those suggested by this study imply that isolated nature reserves would have to be at least 11,500–14,000 ha in area, depending on their shape, to retain at least half of their area as core habitat for forest-interior beetle assemblages (
6,
26) (
SI Appendix). Because more than three-quarters of the world's forested reserves are smaller than this (
SI Appendix), the global conservation potential of protected areas may be overestimated. Large reserves are essential to prevent the extinction of flagship species, such as large carnivores, from protected areas (
5), but the pervasive impacts of large-scale edge effects suggest that the shape of reserves may be equally important. If edge effects in other ecosystems operate at spatial scales as large as those encountered in this study, the relatively undisturbed, frontier forest areas of the world (
27) may be considerably smaller than previously imagined.
