Linking forest edge structure to edge function: mediation of herbivore damage


M. L. Cadenasso (fax 914 6775976; e-mail


1 Forest edges, which are prominent features in the north-eastern United States landscape, may control the flux of organisms between forest and non-forest habitats. Previous studies have described edge structure rather than function, as determined by interaction with such fluxes.

2 The function of the forest edge may be linked to the structure of its vegetation. We tested this hypothesis by experimentally altering the structure of the vegetation at two deciduous forest edges in Millbrook, New York, USA. Intact and thinned plots were established at each edge and we determined whether the structure of the edges influenced the flux of herbivores, as measured by herbivore damage to transplanted tree seedlings.

3 Herbivore damage to seedlings at site 1 was affected by edge vegetation structure and by distance from the edge. The edge structure effect was due to herbivory by voles, which was significantly greater in the intact than in the thinned treatment. Regardless of treatment, voles damaged seedlings only on the edge and 30–40 m from the edge and did no damage in the forest interior (90–100 m), whereas deer damaged significantly more seedlings in the forest interior than on the edge. At site 2, where vole damage was concentrated on the edge, damage to seedlings was affected only by distance from the edge, not edge structure.

4 The two dominant herbivores, white-tailed deer and meadow voles, preferentially damaged different seedling species. In addition, tree seedlings browsed by deer resprouted more frequently than those clipped by voles. Our results suggest that both edge structure and distance from the edge influence herbivore activity and, as a result, influence the spatial arrangement, density and composition of populations of tree seedlings during regeneration in forest fragments.


In the north-eastern United States, as elsewhere, fragmentation of landscape structure has increased both the prominence of forest edges and the proportion of forest area that lies close to an edge. While processes characteristic of the forest interior, such as tree regeneration, plant–plant interactions and plant–animal interactions, still occur in forest patches, their dynamics may be influenced by factors originating from outside the patch boundary (Saunders et al. 1991; Angelstam 1992). The impact of exogenous factors on internal forest dynamics may depend on the permeability of the edge to those factors (Wiens et al. 1985; Stamps et al. 1987; Forman & Moore 1992), and this study represents the first experimental test of whether the flux of one such factor, tree seedling herbivores, is indeed affected by the structure of the forest edge.

The importance of forest edges has been acknowledged and investigated by ecologists for a long time, although the empirical approach has been predominantly static and descriptive (Murcia 1995). Leopold (1936) first suggested that edges benefit game management. Later, ecologists recognized that there are both positive and negative effects of edges, depending on the organism or scale of study (Harris 1988). In the context of the plant community, the focus of edge research has been on quantifying abiotic gradients (Geiger 1965; Wales 1967, 1972; Matlack 1993; Weathers et al. 1995; Cadenasso et al. 1997) and biotic structure (Ranney et al. 1981; Williams-Linera 1990; Brothers & Spingarn 1992) across edges, allowing comparison between edges that differ in age (Matlack 1994; Camargo & Kapos 1995), aspect (Wales 1972; Matlack 1994) or adjacent land use (Moran 1984; Chen et al. 1992). Functional significance is often inferred from these largely descriptive patterns of plant community structure: for example, the use of edge to interior ratios assumes that zones within a certain distance from the edge have a particular function (Levenson 1981; Laurance & Yensen 1991). The link is, however, largely untested except for the effect of edges on nest predation and parasitism in birds (Andrén & Angelstam 1988; Andrén 1995), where results are often contradictory (Paton 1994). This research tests the assumption by examining how edge structure mediates herbivore damage to tree seedlings, and thus assesses the functional impact of the edge on the interior forest community and plant population dynamics.

The call for more research on the functioning of forest edges has been justified by their potential to mediate flows between forest and non-forest habitats (Wiens et al. 1985; Forman & Moore 1992; Hansson 1992; Wiens 1992), depending on the characteristics of both the two habitats and the edge zone and on the nature of the flow (Wiens 1992; Wiens et al. 1993). There are several externally originating processes that can flow across the landscape and through forest edges (Saunders et al. 1991; Forman & Moore 1992; Wiens 1992). These flows of organisms, material and energy link habitat patches in the landscape and may affect landscape pattern, population dynamics and nutrient cycling (Wiens 1992). The further hypothesis that the dynamics within each patch will be affected by the existence of edges (Wiens et al. 1993) has not been tested experimentally.

We adopted a conceptual model that integrates structure and function and illustrates the potential functions of forest edges (Pickett & Cadenasso 1995). This model incorporates ideas from a variety of sources (Forman & Moore 1992; Wiens 1992; Wiens et al. 1993; Forman 1995) and comprises three distinct zones – the forest interior, the forest edge and the surrounding non-forest habitat. Most organisms, material and energy moving from the forest exterior into the interior must pass through the forest edge (exceptions include birds or seeds that can travel over the edge and deposit directly in the interior, and fluxes in ground water). Depending on the situation, an edge may inhibit or enhance movement across the boundary or it may have no effect.

Our experimental test of the link between structure and function of forest edges considered this model for the particular case of the flux of tree seedling herbivores between deciduous forests and adjacent old fields released from agriculture. Edges between these contrasting land covers are structurally and compositionally abrupt, and biotic and abiotic changes across these edges are steep.

The meadow vole (Microtus pennsylvanicus, Ord 1815) is the primary herbivore in the old field, and the white-tailed deer (Odocoileus virginianus, Zimmermann 1780) is the main forest herbivore. Herbivory is important in determining tree seedling composition and density in both forests (Alverson et al. 1988; Tilghman 1989; Gill 1992) and old fields (Rankin & Pickett 1989; Gill & Marks 1991; Ostfeld & Canham 1993). For example, proximity to the forest edge is critical to the establishment of woody species in adjacent old fields because it provides a seed source (Gill & Marks 1991) and influences herbivore activity (Williamson & Hirth 1985; Myster & McCarthy 1989; Ostfeld et al. 1997). Here, we considered the effect of edge on tree regeneration in the forest, which has been less well studied.

The structure of the edge is hypothesized to influence how it mediates any flux. Edge structure can be viewed from a geographical perspective: a forest edge represents a physical disjunction in the landscape at a scale of kilometres. We therefore considered how herbivore activity differed with distance from the forest edge. Alternatively, the ecological structure of an edge is represented by the complex of vegetation layers that make up the edge zone and varies over a scale of metres. We tested the function of the edge at this scale by experimentally altering the structure of the vegetation on the edge and measuring how herbivore activity differed with vegetation structure.

Two questions were posed. First, how does the structure of the vegetation on the forest edge mediate herbivory in the forest? To determine the direct effects of edge structure on a key herbivore, small mammals were trapped. Secondly, what is the net effect of that mediation on tree regeneration in the forest?

Site description

Two 1-ha sites were located at the Institute of Ecosystem Studies (IES) in Millbrook, Dutchess County, New York, USA (41°50′ N, 73°45′ W). Both were embedded in large (> 40 ha) upland mesic deciduous forests that are approximately 60 years old, and had similar canopy composition (dominated by several species including Quercus rubra, Q. alba, Q. prinus, Acer rubrum, A. saccharum, Fagus grandifolia, Betula lenta and Carya spp., with Ostrya virginiana and Carpinus caroliniana as mid-canopy trees and Hammemalis virginiana and Viburnum acerifolium in the understorey). The fields adjacent to each site had been released from hay production or grazing in the 1960s. The field at site 1 had been mowed periodically, there was no woody establishment and it was dominated by several species of Solidago, Schizachyrium scoparium, Poa spp. and other grass and forb species (Armesto et al. 1991). The forest at site 2 was separated from the field by a 3-m wide paved road, accessible to only the IES. This field had not been mowed except along the road margins and, in addition to the herbaceous species found at site 1, had small patches of Rhus glabra and Cornus racemosa and two individuals of Acer rubrum approximately 5 m tall. Both sites had relatively straight and structurally abrupt edges. Site 1 faced north-east (21° from true north) and site 2 had a west aspect (275° from true north) (Fig. 1). (Taxonomy is according to Gleason & Cronquist 1991.)

Figure 1.

The two field sites used in the experiment. Each site occupied 100 m of forest edge and extended 100 m into the forest interior and contained a thinned and a control (intact) plot allocated at random.

Experimental manipulation of edge structure

Plots were established at each stand such that the external boundary approximately followed the canopy dripline and the sites extended 100 m into the forest interior (Fig. 1). Each of the sites was divided into two 40-m plots separated by a 20-m buffer zone, and intact and thinned treatments were allocated at random (Fig. 2). In autumn 1994 the woody vegetation lower than half canopy height was removed from the outermost 20 m of the thinned plot at each site. This included the lateral branches of the canopy trees that were lower than half canopy height, complete trees where they were shorter than half the canopy height, and shrubs; the herbaceous layer was not cut. Although we altered the physical structure of the edges of the thinned plots, but no canopy trees were removed, the shape and location of these abrupt edges, as drawn on a map, were not therefore altered. In other words, we altered the ecological but not the geographical structure of the edges. Our manipulation produced a structure similar to a newly created forest edge before colonization by shrubs and saplings or development of lateral branches by the newly exposed canopy trees.

Figure 2.

Experimental design. Sites were embedded within deciduous forests (light shading) and a 20-m buffer zone (hatched) separated the two experimental plots. At three distances (0–10, 30–40 and 90–100 m) on each plot, five pairs of caged (▪) and open (□) quadrats were established randomly.


Herbivory experimental design

Five pairs of caged and uncaged quadrats were established in a stratified random design in each of three distance bands from the edge of each treatment plot (0–10 m, 30–40 m and 90–100 m) (60 quadrats per site) (Fig. 2). Caging was accomplished by constructing a four-sided exclosure measuring 1.5 × 0.75 m and made of 61-cm wide square-weave ‘hardware cloth’ with 1.27-cm mesh. The perimeter was buried 4–6 cm to discourage animals from burrowing under, and the cages were regularly checked for signs of disturbance. During burial, damage to litter and vegetation was avoided as much as possible. Twenty-centimetre wide aluminium flashing was riveted to the top of the hardware cloth around the entire cage perimeter, to prevent small mammals from using the mesh to climb up the sides of the cage and enter the cage through the top. Hexagonal-weave ‘chicken wire’ (2.54-cm mesh) was secured over the top of the cage to prevent deer browsing inside the exclosure. An uncaged quadrat of the same size was located approximately 3 m away from the caged quadrat with which it was paired.

In the spring of 1995, all (caged and uncaged) quadrats were planted with 12 seedlings of each of Betula lenta, Quercus rubra and Ailanthus altissima. All three species are found in these forests and represent an early and a late successional species and a species invasive in the Hudson Valley, respectively. The seedlings were grown from seed either collected on the IES grounds or bought from F.W. Schumacher Co. (Sandwich, MA). Seeds were germinated in February and March and seedlings were maintained in the IES greenhouse until planted in the field in early June. Immediately after planting seedlings were watered generously. The three species were located randomly in the quadrats.

Seedlings were monitored weekly for the first 10 weeks then every 2 weeks until leaf drop in the autumn. They were resurveyed in the spring of 1996. Seedlings not surviving the first 2 weeks after planting due to shock or stress were replaced only once. Each seedling was scored as undamaged, damaged by herbivores, mechanically damaged or dead from other causes. Each herbivore damages seedlings in a unique way, allowing for easy field scoring of damage by a specific herbivore. Voles clip seedlings at the base, approximately 2 cm from the soil, cleanly severing the stem. Deer remove biomass from the top of the seedling and, occasionally, remove only the apical bud. Due to their lack of upper incisors, deer leave a very jagged cut on the remaining seedling stem. Slugs were observed on the plants and their slime trails were apparent on the seedling stem. Insect defoliation was incremental and the defoliation could be tracked until seedling mortality. Slug and insect damage were combined into an invertebrate class for the analyses. All seedling damage observed could be attributed to one of these classes of herbivores.

Small mammal trapping

Small mammals were trapped using Sherman live traps placed in a regular grid extending from 20 m beyond the edge to 50 m into the forest. Eight traps were placed 10 m apart along three transects in each intact and thinned plot. Each site was trapped every third week for three consecutive nights. Because traps were moved between the two sites, traps were placed on the trap station, left closed, and baited at least 3 days prior to the actual trapping period, to ensure that animals were familiar with the traps. This strategy was used to prevent inaccuracies in the capture data that may have occurred if a lag time existed for the small mammals to ‘find’ the traps.

Each site was trapped on seven occasions between 13 June and 26 October 1995, the period during which herbivory was measured. A site was not trapped in the week immediately following seedling planting so that the herbivores would have access to the seedlings used in the experiment. Traps were set in the late afternoon, baited with rolled oats, and checked the following morning within 2 h of sunrise.

Statistical analyses

All statistical analyses were done separately for data from both site 1 and site 2. Because parametric and non-parametric statistical approaches yielded the same results, we only present the parametric analyses.

We used a three-factor nested multivariate analysis of variance (manova) together with univariate three-factor nested anovas to analyse the effects of edge vegetation structure, distance from the edge and herbivore exclusion cages on the response variable, and damage to seedlings due to herbivory vs. damage due to all other causes. Distance from the edge was nested within edge structure and within cage effect. Damage was calculated as the percentage of the 36 individual seedlings planted in each caged or uncaged quadrat that was damaged either by a herbivore or by other means (i.e. mechanical damage and other causes combined). anova results for a particular factor (edge structure, cage, distance nested within edge structure or distance nested within cage) were not considered unless manova results for that factor were significant.

The effects of edge vegetation structure and distance from the edge nested within edge structure on damage to seedlings caused by each of the three herbivore classes (deer, voles, invertebrates) were analysed with a two-factor nested manova. If the manova results for a factor were significant, then two-factor nested anovas were used to evaluate which variable (herbivore class) contributed to the significant differences detected by the multivariate tests. Damage was calculated for each open quadrat as a percentage of the 36 planted seedlings that had been subjected to herbivory by each of the three classes. Invertebrates did not damage seedlings at site 2 and were excluded from the analyses for this site.

We also tested for patterns in damage among seedling species. A two-factor nested manova was used to analyse the effects of edge vegetation structure and distance from edge nested within edge structure on damage to tree seedlings caused by herbivory for each of the three tree species used. Damage was calculated as a percentage of the 12 individuals of each species in each open quadrat that had been affected. Again, two-factor nested anova results were considered for each factor significant in the manova tests to evaluate which of the response variables (seedling species) contributed to the significant differences detected by the multivariate tests.

To test the net effect of herbivore damage on (i) the mortality and (ii) the composition of the seedling community, we analysed (i) overall seedling response to damage caused by each of the herbivores and (ii) damage caused by each of the herbivores to each seedling species to determine if a herbivore damaged seedling species differentially. A chi-square frequency analysis was used to test the effect of herbivory by deer and voles on the resprouting frequency of seedlings following damage. The assumption of this analysis was that the ability of the damaged seedling to resprout would be determined by the identity of the herbivore because of the characteristic manner in which each herbivore damages seedlings. Edge structure and distance from the edge may influence which herbivore encounters the quadrat, but were not expected to influence the frequency of resprout by seedlings following damage. Therefore, the location of the seedling was not considered in this analysis. Damage was calculated as the percentage of seedlings damaged by each herbivore that resprouted in each open quadrat. For example, if deer and voles damaged seedlings in a quadrat and all those damaged by deer resprouted and none damaged by voles resprouted, the response value for deer was 100% and for voles 0%. Because the analysis was done to test the net effect of herbivory on seedling community mortality, quadrats with no herbivore damage were not used in the analysis. The chi-square test determined whether the frequency of resprout of seedlings damaged by deer and damaged by voles differed from an expected equal frequency.

A similar test with a separate analysis for each herbivore class was used to determine whether each species of seedlings was damaged equally by each herbivore. Location of the seedling with respect to edge structure and distance from the edge was again assumed not to influence the preference of the herbivore. Twelve individuals of each seedling species were exposed to herbivory in each open quadrat. The percentage of each species in each quadrat damaged by a particular herbivore was calculated. Because the analysis was to test the net effect of herbivory on the composition of the seedling community, quadrats in which the particular herbivore caused no damage were excluded from the analysis. A chi-square test was performed for each herbivore to test whether the herbivore damaged seedling species differentially. Because invertebrates only damaged seedlings at site 1, the analysis of invertebrate preference was performed only for this site.


Herbivore exclusion cages and damage to seedlings

Herbivore exclusion cages affected damage to tree seedlings at both site 1 and site 2 (Fig. 3 and Table 1). The significant effect of cages in the manova permitted univariate analyses conducted separately on each of two damage classes to be used to determine which damage class contributed to the significant cage effect. At both site 1 and site 2, herbivore damage was significantly greater in the open quadrats than within the exclusion cages, and univariate analyses showed that this accounted for the significant cage effect in the manova (Table 1); cages had no effect on mortality due to all causes other than herbivory (Table 1). Damage due to causes other than herbivory was similar for all combinations of edge structure, distance from edge and caging (Fig. 3).

Figure 3.

Damage to tree seedlings planted in herbivore exclusion cages and open quadrats in two edge structure treatments, intact and thinned, at two sites. Damage to seedlings due to herbivores and to other causes in caged and open quadrats are represented by different shading patterns. Each value is the mean of five replicates ± 1 SE.

Table 1. manova and univariate anovas for effects of edge vegetation structure, herbivore exclusion cages and distance from edge (nested within each) on tree seedling damage
Site 1
(a) Multivariate analysis
Edge2, 500.90992.4750.0944
Cage2, 500.448930.6890.0001
Distance (edge)8, 1000.76891.7550.0950
Distance (cage)4, 1000.88891.5170.2030
(b) Univariate analyses
Herbivory damageEdge1383.95282.940.0926
Distance (edge)4868.76331.660.1733
Distance (cage)2318.20131.220.3045
Other damageEdge1108.24582.090.1547
Distance (edge)4425.19372.050.1014
Distance (cage)2201.66131.940.1537
Site 2
(a) Multivariate analysis
Edge2, 500.93601.70980.1913
Cage2, 500.593017.15920.0001
Distance (edge)8, 1000.84011.13790.3449
Distance (cage)4, 1000.83592.34320.0599
(b) Univariate analyses
Herbivory damageEdge162.28131.400.2424
Distance (edge)4338.44491.900.1246
Distance (cage)2427.62274.800.0123
Other damageEdge1157.56122.720.1051
Distance (edge)4182.59610.790.5378
Distance (cage)23.34900.030.9715

Herbivore damage

At site 1, both edge structure and distance from the edge nested within edge structure affected herbivore damage to seedlings. The two edge treatments elicited different patterns of herbivory. Voles were the dominant herbivore on the intact plot, where they clipped 40% of the planted seedlings and caused 96% of all observed herbivore damage; deer browse was minimal (Fig. 4). In contrast, in the thinned plot, browse by deer and defoliation by invertebrates were the dominant herbivore activities, while voles clipped only 2.8% of the planted seedlings (Fig. 4). Univariate analyses showed that the response of voles was the predominant driver of the significant effect of edge structure on seedling damage (Table 2), although invertebrate damage also tended to be higher on the thinned plot (Fig. 4 and Table 2). The effect of distance from the edge was due to both voles and deer (Table 2). Regardless of edge structure, voles damaged seedlings on the edge and 30–40 m into the forest but not in the forest interior (Fig. 4). In the intact plot, deer damage to seedlings increased markedly with distance from the edge and was sufficient to give an overall distance effect despite similar browsing throughout the forest in the thinned plot (Fig. 4).

Figure 4.

Herbivore damage to tree seedlings at three distances from intact and thinned treatment edges at two sites separated by herbivore class. Each value is the mean of five replicates ± 1 SE.

Table 2. manova and univariate anovas for effects of edge vegetation structure and distance from edge nested within edge structure on tree seedling damage by herbivores
Site 1
(a) Multivariate analysis
Edge3, 220.45158.90900.0005
Distance (edge)12, 58.50.14075.35650.0001
(b) Univariate analyses
Distance (edge)42371.70835.340.0032
Distance (edge)44618.046518.860.0001
Distance (edge)4176.91570.940.4579
Site 2
(a) Multivariate analysis
Edge2, 230.91031.13390.3391
Distance (edge)8, 460.44142.90480.0104
(b) Univariate analyses
Distance (edge)4516.40951.460.2455
Distance (edge)471.93914.670.0063

There was no effect of edge structure on herbivory (Table 2) at site 2, but herbivory was significantly greater on the edge than at other distances for both plot types (Fig. 4). The significant effect of herbivory was due to voles being limited to the edge and 30–40 m into the forest; deer browsed throughout this site, but there was no invertebrate damage.

Damage to seedlings according to tree species

The patterns in herbivore damage to Q. rubra seedlings accounted for the significant effect of distance from the edge on damage to tree seedlings (Table 3). Herbivory was significantly greater on the edge than at any other distance on the intact plot, and was concentrated on the edge and 30–40 m from the edge of the thinned plot (Fig. 5). The trends towards greater damage to both Q. rubra and B. lenta seedlings on the intact plot compared with the thinned plot were responsible for the nearly significant effect of edge type in this manova (Table 3). At site 2, neither edge structure nor distance from edge affected the amount of damage to seedlings when separated by tree species (Table 3).

Table 3. manova and univariate anovas for effects of edge vegetation structure and distance from edge nested within edge structure on herbivore damage to seedlings by seedling species
Site 1
(a) Multivariate analysis
Edge3, 70.50003.35820.0847
Distance (edge)12, 18.80.06612.81010.0219
(b) Univariate analyses
Ailanthus altissimaEdge1477.52331.020.3389
Distance (edge)42652.11381.420.3042
Betula lentaEdge13175.79634.170.0717
Distance (edge)49247.56483.030.0770
Quercus rubraEdge11131.81254.610.0604
Distance (edge)45380.08575.470.0163
Site 2
(a) Multivariate analysis
Edge3, 70.82330.50060.6937
Distance (edge)12, 18.80.34960.76440.6777
Figure 5.

Herbivore damage to each of three species of tree seedlings in different edge treatments and distances. Each value is the mean of five replicates ± 1 SE.

Seedling resprout following damage

Seedlings damaged by deer and voles did not resprout with equal frequency (Fig. 6 andTable 4). At both sites, significantly more seedlings resprouted following browse by deer than after clipping by voles.

Figure 6.

The percentage of seedlings damaged by deer and voles that resprouted following damage at site 1 and site 2. Damage represents the mean value for those open quadrats that experienced damage by deer or voles ± 1 SE. The number of quadrats used in the analysis can be found in Table 4.

Table 4.  Chi-square analysis for departure from equal effect of deer and vole on seedling resprout after damage
 nFrequencyχ2 statisticd.f.P
Site 1
Site 2

Herbivore preference for tree seedling species

None of the herbivore classes damaged seedling species with equal frequency (Fig. 7 and Table 5). At both sites, deer damaged seedlings of B. lenta at a higher frequency than either Q. rubra or A. altissima, while voles exhibited a preference for Q. rubra and B. lenta. Defoliation of seedlings by invertebrates, which only occurred at site 1, was concentrated on A. altissima seedlings (Fig. 7 and Table 5).

Figure 7.

The percentage of each species of seedlings that (a) deer, (b) voles and (c) invertebrates damaged at site 1 and site 2. Values shown are the mean of those for open quadrats in which each herbivore damaged a particular seedling species, ± 1 SE. Species that were damaged in only one quadrat were not included in the figure. The number of quadrats used in the analyses can be found in Table 5.

Table 5.  Chi-square analyses for whether the frequency of damage of a particular herbivore to each of the three seedling species is significantly different from the expected equal frequency
 nFrequencyχ2 statisticd.f.F
Site 1
A. altissima13158.3944.4920.001
B. lenta17883.3   
Q. rubra7125.0   
A. altissima333.3305.4520.001
B. lenta7358.3   
Q. rubra16400.0   
A. altissima9208.3236.3920.001
B. lenta18.3   
Q. rubra358.3   
Site 2
A. altissima466.7592.3620.001
B. lenta16558.3   
Q. rubra6116.7   
A. altissima18.3107.2720.001
B. lenta225.0   
Q. rubra9100.0   

Vole captures

The species trapped included meadow voles, white-footed mice (Peromyscus leucopus), shrews (Blarina brevicauda), eastern chipmunk (Tamias striatus) and meadow jumping mice (Zapus hudsonius) but, of these species, only meadow voles are herbivorous (Ostfeld & Canham 1993). As expected, all vole captures were in the adjacent field (−20, −10 m) and the forest edge (0 m) rather than the forest interior. To illustrate the data, we calculated the number of captures at each of the three distances in each plot over a three-night trapping period, resulting in the number of captures per 27 trap nights per week. The data were temporally pseudoreplicated (Hurlbert 1984), and were not therefore subjected to analysis. However, at site 1, vole captures were consistently greater on the thinned than on the intact plot; no differences were apparent between treatments at site 2 (Fig. 8 and Table 6).

Figure 8.

Frequency of vole captures in the field adjacent to and in the intact and thinned treatment edges at site 1 and site 2.

Table 6. t-test analyses of the effect of edge structure on vole captures. Variance are equal at both sites
Site 1
Intact edge74.42860.6494−5.5018120.0001
Thinned edge712.85711.3875   
Site 2
Intact edge74.00001.0911−0.2222120.8279
Thinned edge74.28570.6801   


Is the structure of the vegetation on the edge linked to its function as reflected by herbivore flux? Herbivore damage to seedlings inside the cages was minimal, showing that exclusion was successful, while seedling mortality due to causes other than herbivory was similar in the caged and open quadrats, indicating that cages did not introduce any additional stresses. Further analyses of the responses to herbivores were limited to the open quadrats.

Ecological edge structure and herbivore damage

At site 1, the ecological structure of the edge had a significant effect on the amount of herbivore damage to tree seedlings. This was primarily due to voles, which were the dominant herbivore on the intact plot but damaged significantly fewer seedlings on the thinned plot. We had expected vole activity to be less on the thinned plot because the experimental manipulation removed vegetation cover and might therefore make voles more vulnerable to predation. However, although voles reach higher densities in herbaceous communities when low-statured vegetation cover is dense, and tend to avoid areas of decreased vegetation cover (Birney et al. 1976; Peles & Barrett 1996), the experimental thinning did not alter ground layer vegetation. In addition, more voles were captured outside the thinned treatment but damage was greater in the intact plot. This apparent contradiction may be explained by a compensatory response of the ground layer to the experimental manipulation. Removal of lateral branches of the canopy trees and of small trees and shrubs may have increased light in the thinned plot. Although the resulting increase in herbaceous cover may have encouraged voles, the production of more herbaceous stems may have made tree seedlings on the thinned plot less obvious to predators (Birney et al. 1976). Further tests are necessary to confirm the detailed mechanism by which the ecological structure of the edge mediates the impact of voles as herbivores on tree seedlings.

At site 2, the ecological structure of the edge vegetation did not affect herbivore damage. The number of seedlings damaged may have been too small relative to the number of experimental treatment combinations for a significant effect of the ecological structure to be detected by the multivariate analysis. We did not expect the road bordering this site to impact vole activity. The road is infrequently used by vehicle traffic and is only 3 m wide, and Oxley et al. (1974) concluded that roads inhibited the movement of small mammals from forest habitats but not those characteristic of more open environments. However, voles were trapped only on the field side of the road, suggesting that avoidance of the road may prevent voles from responding to different edge treatments. Although there was an ecological effect only at site 1 and herbivore damage was low at site 2, both sites showed a strong geographical effect.

Geographical edge structure and herbivore damage

The response of herbivores to the geographical structure of the edge was addressed by testing for patterns in herbivore damage to seedlings with distance from the plot edge. Voles are restricted to herbaceous habitats but do use forest edges (Grant 1971), and the decrease of their damage with distance from edge at both sites was therefore expected. At site 1, the geographical structure also affected deer damage to seedlings. On the intact plot they caused more damage farther from the edge, and the shift from voles as the primary herbivore close to the field to deer in the forest was as expected from their habitat preference (Grant 1971; Tilghman 1989). In the thinned treatment, however, deer damaged seedlings consistently over distance from the edge. The thinned structure was maintained throughout the growing season and therefore deer were not attracted to the edge by resprouting stems, as these were periodically removed. However, a greater number of seedlings may have been available to deer for a longer period of time because of the reduced vole herbivory on the thinned plot. The geographical-scale response of deer damage to edge structure may therefore be dependent on the ecological-scale response of voles.

Net effects on tree regeneration

The spatial patterns of damage caused by each herbivore will contribute to the net effect of herbivory on the composition and density of the seedling community, but it is also necessary to consider which species each herbivore damages and how seedlings respond to the damage.

Distance from the edge affected damage to the three seedling species differentially. Greater damage to Q. rubra seedlings close to the edge could be explained by the preference for this species shown by voles, whose habitat is limited to the edge. Distance from the edge did not influence damage to either B. lenta or A. altissima seedlings. Both deer and voles showed some preference for B. lenta and damage therefore occurred throughout the forest. Although invertebrates preferred A. altissima seedlings, they caused minimal damage and any influence of either edge structure or distance from the edge was masked by damage to this species by the other herbivores.

Both deer and voles are known to be seedling herbivores, and their differences in damage pattern over space in the forest and in species preference will result in the seedling community being differentially affected by herbivory through space. Thus regeneration of the forest may depend on patterns of herbivore activity as well as on other, more commonly recognized, spatially differentiated factors such as seed source (Bjorkbom 1979; Sork 1983; Houle 1994) and the availability of light (Oliver & Larson 1990; Macdougall & Kellman 1992).

The identity of the herbivore may also affect the density of seedlings of each tree species because seedlings respond differently to damage by each herbivore. The mode of damage may explain why more seedlings resprout following deer browse than resprout following clipping by voles. Regardless of species, deer typically leave several leaves and, occasionally, only the apical bud is removed, whereas voles clip seedlings very close to the ground and only a small portion of the stem remains. The effect of damage by specific herbivores in addition to their preferences must therefore be considered when investigating the mechanism by which herbivory affects composition and mortality of seedling populations during forest regeneration.

Understanding how forest edges influence herbivore damage to seedlings is important because an increasing proportion of the forested landscape is in close proximity to edges. Data from sightings of deer in spotlights has provided a conservative population estimate for white-tailed deer on the IES property at 18 km−2 (Winchcombe 1993), and because this is considerably higher than that suggested to inhibit forest regeneration in Wisconsin (4 km−2) (Alverson et al. 1988), deer may strongly influence seedling density and composition at our sites. Deer traverse forest edges every dawn and dusk (Montgomery 1963) and their interaction with edge structure and its influence on forest dynamics may therefore be much greater than expected from the proportion of space occupied by edges.

This is the first experimental test of the function of forest edges on plant community regeneration. It confirms the suggested importance of forest edge function as a mediator of herbivore impacts (Alverson et al. 1988) and suggests hypotheses for more detailed mechanistic studies. In addition, the technique of altering the ecological structure of a forest edge while keeping the geographical scale configuration of the edge intact may be a useful strategy for studying different landscape fluxes and how they interact with different types of edges to determine edge functions.


We would like to thank Julie E. Hart and M. Michael Traynor for extensive field assistance in cage construction, seedling planting and monitoring. Dave Bulkeley and the IES Greenhouse staff helped grow and maintain the seedlings until they were planted in the field. The cutting of vegetation on the forest edge was done by Brad Roeller, Alan Kling and the IES Grounds crew. Peter Morin provided valuable statistical advice. The manuscript has been greatly improved by comments from Charlie Nilon, Rick Ostfeld, John Wiens and Richard Forman. This research was funded by the NSF, grant DEB 9307252.

Received 29 October 1998revision accepted 13 July 1999