CONCEP TS & SYNTHESIS

EMPHASIZING NEW IDEAS TO STIMULATE RESEARCH IN ECOLOGY

Ecology, 85(11), 2004, pp. 2917–2926

q 2004 by the Ecological Society of America

A PREDICTIVE MODEL OF EDGE EFFECTS

LESLIE RIES1 AND THOMAS D. SISK
Northern Arizona University, Department of Biological Sciences and Center for Environmental Sciences and Education,
Flagstaff, Arizona 86011 USA

Abstract. Edge effects are among the most extensively studied ecological phenomena,
yet we lack a general, predictive framework to understand the patterns and variability
observed. We present a conceptual model, based on resource distribution, that predicts
whether organismal abundances near edges are expected to increase, decrease, or remain
unchanged for any species at any edge type. Predictions are based on whether resources
are found predominantly in one habitat (decreased abundance in preferred habitat, increase
in non-preferred), divided between habitats (predicts an increase near both edges), spread
equally among habitats (predicts a neutral edge response), or concentrated along the edge
(increase). There are several implications of this model that can explain much of the
variability reported in the edge literature. For instance, our model predicts that a species
may show positive, negative, and neutral responses, depending on the edge type encoun-
tered, which explains some intraspecific variability observed in the literature. In addition,
any predictable change in resource use (for example, by region or season) may explain
temporal or spatial variability in responses even for the same species at the same edge
type. We offer a preliminary test of our model by making predictions for 52 bird species
from three published studies of abundance responses near forest edges. Predictions are
based solely on general information about each species’ habitat associations and resource
use. Our model correctly predicted the direction of 25 out of 29 observed edge responses,
although it tended to under-predict increases and over-predict decreases. This model is
important because it helps make sense of a largely descriptive literature and allows future
studies to be carried out under a predictive framework.
Key words: ecological boundary; ecotone; edge responses; habitat edge; predictive model; re-
source distribution.

INTRODUCTION negative view of edge effects and brought critical at-

tention to the issue of habitat fragmentation (Brittingh-
Changes in species’ distributions near habitat edges
am and Temple 1983). Since these seminal reports,
are among the most extensively studied phenomena in
ecology because edge responses are a key component there have been hundreds of studies describing edge
to understanding the influence of landscape structure responses for many taxa, with much of the focus re-
on habitat quality. Edges can be defined as the bound- maining on forest edges (for reviews, see Paton 1994,
ary between patches with differing qualities; thus, the Murcia 1995, Risser 1995, McCollin 1998, Lidiker
identification of edges will depend on how researchers 1999, Lahti 2001, Chalfoun et al. 2002, Sisk and Battin
define patches (Strayer et al. 2003). Increases in avian 2002).
richness and abundance at forest edges have been noted Despite this extensive interest, the field has remained
for several decades (Lay 1938, Johnston 1947) and led largely descriptive, with no underlying framework to
to early claims that edge habitat was beneficial for wild- make sense of the variability reported, giving the im-
life. However, the discovery that many songbirds ex- pression that general patterns of edge responses are
perience higher predation and parasitism rates near for- elusive (Murcia 1995, Sisk and Battin 2002). Most edge
est edges (Gates and Gysel 1978, Chasko and Gates response studies are observational, have low replica-
1982) led to a fundamental shift from a positive to tion, do not control for factors known to interact with
edge dynamics (Murcia 1995), and are carried out with
1 Present address: Department of Biology, University of no a priori predictions, so it is difficult to interpret the
Maryland, College Park, Maryland 20742 USA. meaning of the patterns and variability reported in a
E-mail: lries@umd.edu rigorous fashion. However, the importance of this topic
2917
 2918 LESLIE RIES AND THOMAS D. SISK Ecology, Vol. 85, No. 11

to ecology and its applications to conservation call for habitat near forest edges experiences increased shad-
a synthesis of proposed mechanisms into a conceptual ing, resulting in lower temperatures and higher humid-
model that can make sense of previously reported pat- ity (Cadenasso et al. 1997). In both cases, the envi-
terns and allow future studies to be conducted within ronment near the edge is likely to be more hospitable
a theoretical framework. This will lead to a greater to organisms adapted to conditions of the adjacent
understanding of the factors that influence edge re- patch interiors.
sponses and allow for predictions, even for poorly stud- Another way edges may be enhanced is by containing
ied species in a variety of landscapes. Therefore, our resources absent or rare in both adjoining patches. This
objectives were to (1) summarize the major patterns concentration of resources near edges may support in-
and proposed mechanisms reported in the literature; creased abundances of species that rely on those re-
(2) present a conceptual model based on those mech- sources. One common example is shrub-dependent
anisms that predicts edge abundance responses for any birds being attracted to forest edges that have devel-
species at any edge type; (3) explore the variability in oped a shrub layer rare or absent in either bordering
the literature on edges and its potential underlying habitat (Mills et al. 1991, Berg and Part 1994). In this
causes within the framework of our model; and (4) offer case, if habitats were mapped finely enough, the edge
a preliminary test of our model by determining how might be identified as a unique habitat type and the
well it predicts the nature of edge responses for over observed response would not be considered a true
50 bird species, as reported in three published studies. ‘‘edge effect.’’ However, most vegetation maps cannot
capture such fine distinctions in habitat, and in many
EDGE RESPONSE PATTERNS AND MECHANISMS cases the increase in resource availability near edges
Results from dozens of field studies confirm that may not constitute a unique vegetation class. This may
many species respond to habitat edges in a variety of be especially true when an organism that is responding
ways. Species may show increases, decreases, or no to the presence of an edge provides the resource base
Concepts & Synthesis

change in abundance, depending on the specific edge for another organism, which may then also show an
type encountered. These changes may be due to abiotic edge effect. Spotted Owls (Strix occidentalis) present
or biotic changes in the environment (Murcia 1995) a good example of this phenomenon. When their prey
caused by ecological flows across edges (Cadenasso et base is dominated by wood rats (Neotoma spp.), a spe-
al. 2003), changes in interspecific interactions (Fagan cies that shows an increased abundance near edges, the
et al. 1999), or a combination of these and other factors. owls also show an increase near edges. On the other
In the avian literature, increased abundances near edges hand, when their prey base is dominated by flying
(also called positive edge responses) are generally more squirrels (Glaucomys sabrinus), a species that shows
common than decreases or negative edge responses no edge effect, the owl also shows no response to edges
(Villard 1998, Sisk and Battin 2002). There currently (Zabel et al. 1995). Another example is the butterfly
is insufficient evidence to determine whether this pat- Lopinga achine, an edge-associated species whose host
tern extends to other taxa. Neutral edge responses (no plant is found in highest concentrations near forest edg-
change in abundance near the edge) are probably under- es, while shading from shrubs at these edges provides
reported due to publishing bias and have received little the most suitable microclimatic conditions for larval
attention, despite their potential importance in under- growth (Bergman 1999). These types of cascading edge
standing general underlying mechanisms. effects may be very common.
Three mechanisms have been cited most commonly The third mechanism, complementary resource dis-
to explain increased abundances near edges: (1) spill- tribution, occurs when two bordering patches contain
over, (2) edges as enhanced habitat, and (3) comple- different resources, and being at the edge allows the
mentary resource distribution. Increased abundances most convenient access to both (Dunning et al. 1992,
near edges have often been attributed simply to spill- McCollin 1998, Fagan et al. 1999). In this case, re-
over or ‘‘mass effects’’ (Shmida and Wilson 1985), sources available only in one patch ‘‘complement’’ the
which occur when individuals disperse into non-habitat resources available in the adjacent one. In comple-
by crossing the boundary from their preferred habitat. mentary resource distribution, no particular resource is
This results in elevated abundances near edges (within concentrated at the edge, but the juxtaposition of re-
non-habitat), and is due solely to proximity and the fact sources results in higher quality habitat at edges by
that organisms are not likely to penetrate very deeply offering greater access. One classic example is the
into a patch of non-habitat. In addition, the quality of Brown-headed Cowbird (Moluthrus ater), which for-
the edge in non-habitat patches may also be enhanced ages in open pastures but parasitizes forest-dwelling
by its adjaceny to higher quality habitat, which also songbirds (Brittingham and Temple 1983). Many other
may lead to increased abundances near edges within taxa that are associated with forest edges are assumed
non-habitat. For instance, forest habitat near open edg- to be foraging in the open, yet obtaining other resources
es tends to be more similar to the bordering open habitat from the forest, including deer (Alverson et al. 1988)
(hotter, drier, and with more light) compared to the and numerous bird species (Gates and Gysel 1978,
forest interior (Chen et al. 1999). Conversely, open McCollin 1998). In contrast, we refer to resource dis-
November 2004 A PREDICTIVE MODEL OF EDGE EFFECTS 2919

2000, Fletcher and Koford 2003). Although these spe-

cies may avoid less-preferred habitat, they may still
spill over into bordering patches. Therefore, when a
patch of suitable habitat borders a patch of lower qual-
ity habitat, a gradual transition from the highest den-
sities in the interior of the preferred habitat patch to
the lowest densities in the interior of the adjoining
patch is expected (Sisk and Margules 1993, Lidiker
1999). This transition in abundance is assumed to re-
flect a gradient in habitat quality that may ultimately
be based on resource availability and abiotic factors
such as microclimatic shifts across the edge zone (Mur-
cia 1995). All of the mechanisms presented relate either
to the availability or proximity of resources, and led
us to develop a conceptual model that uses resource
distribution as a basis for predicting general edge re-
sponses.
A RESOURCE-BASED MODEL OF EDGE EFFECTS
For this model, we assume a simple landscape com-
posed of two adjacent patches. Habitat quality in each
patch is determined by the relative amount of available
FIG. 1. A predictive model of edge effects. This general resources, so ‘‘lower quality’’ habitat refers to a patch

Concepts & Synthesis

model predicts changes in population abundance near habitat that has fewer resources than the adjacent patch. When
edges based on resource distribution. When (a) resources are no resources are available in a patch, it is defined as
concentrated in one patch and those in the lower quality patch
are supplementary, then a transitional edge response is pre- non-habitat. Resources may include provisions such as
dicted. However, when (b) resources in the lower quality food or nest sites, service-providers such as pollinators
patch are complementary (different), then being on the edge and seed dispersers, or abiotic resources such as light.
allows increased access and a positive response is predicted Density levels in patch interiors are assumed to reflect
in both patches. When resource availability is relatively equal
between patches, a neutral response is predicted when (c)
relative differences in habitat quality, so low-quality
resources are supplementary, and a positive one when (d) and non-habitat patches will have lower or zero den-
resources are complementary. When (e) resources are con- sities in those patch interiors. This model is therefore
centrated along the edge, a positive response is once again a patch-based model with edges defined as the bound-
predicted. aries between patches. While patch definition may vary
among researchers, the fact that patch quality is defined
here by the relative amount of resources means that
tribution as ‘‘supplementary’’ when two adjacent our model should be broadly applicable even when
patches both contain resources, but there are no re- landscapes are classified under different schemes. Our
sources in one patch not also available in the other. In model predicts the expected change in abundance near
that case, being near the edge offers no benefit with edges, based on patterns of resource distribution be-
respect to access to resources. tween those two patches, as illustrated in Fig. 1.
Edge avoidance (decreased abundance near edges) When habitat borders lower quality or non-habitat
is most commonly reported for habitat-specific species, and resources in the lower quality habitat are supple-
usually forest ‘‘interior’’ species. Examples include the mentary (so offer nothing not already found in the high-
Ovenbird, Seiurus aurocapillus (Burke and Nol 1998), er quality patch), then individuals are predicted to show
the Red-eyed Vireo, Vireo olivaceus (King et al. 1997), a transitional response across the edge. This transitional
the red-backed vole, Clethrionomys gapperi (Mills response is characterized by a gradual decrease in den-
1995), and the plant Trillium ovatum (Jules 1998). sity from a maximum in the interior of the higher qual-
These species are generally assumed to be avoiding ity habitat patch to a minimum in the interior of the
changes in the environment near edges that make them lower quality or non-habitat patch (Fig. 1a). It is im-
hostile to species adapted to interior conditions. This portant to note that most empirical studies report re-
has been well documented for the Ovenbird (S. auro- sponses within only one patch type (on one side of the
capillus) where the hotter, drier conditions near forest edge), so a transitional response will appear to be either
edges are associated with lower densities of their prey positive or negative, depending upon the reference
items (Burke and Nol 1998). Species associated with point of the observer (see responses on either side of
open habitat have also shown decreased abundances the edge in Fig. 1a). In contrast, when resources in the
near forest edges, including butterflies (Haddad and lower quality patch are complementary (offer some-
Baum 1999) and grassland birds (O’Leary and Nyberg thing not found in the higher quality patch), then an
 2920 LESLIE RIES AND THOMAS D. SISK Ecology, Vol. 85, No. 11

increase in abundance is predicted on both sides of the same edge type. When these changes are predictable,
edge (Fig. 1b). This is because, in both patches, being more refined edge response predictions are possible.
near the edge allows access to additional resources only For example, avian edge responses have been shown
available in the adjacent patch. to vary between seasons (Noss 1991, Hansson 1994),
For situations in which both patches provide rela- and this may be due to predictable changes in resource
tively equal resource availability, responses are again use throughout the year. Many birds are known to show
expected to vary depending on how those resources are different habitat associations during winter and breed-
distributed. When resources are supplementary (not di- ing seasons (which is intuitive based on the fact that
vided) between patches, no edge response is predicted nesting resources are not needed during the nonbreed-
(Fig. 1c). However, when resources are complementary ing season) and in those cases, our model will predict
(divided between patches), then being near the edge different edge responses during summer and winter,
offers increased access to both sets of resources, so the even at the same edge type. Likewise, regional varia-
species in question is again predicted to increase in tion in edge responses has been suggested for birds in
abundance near edges (Fig. 1d). Predictions of positive the eastern vs. western U.S. (Sisk and Battin 2002).
abundance responses, based on complementary re- While this is difficult to test due to a paucity of studies
source distribution (Fig. 1b, d), are most applicable to in the west (Sisk and Battin 2002), such differences
mobile organisms because they can most easily gain would be predicted by our model for any species show-
access to resources in two patches. However, some ses- ing regional differences in resource use.
sile organisms could also demonstrate such responses One consequence of conducting research under this
if advantages at the edge could be realized via, for model framework is that characterization and compar-
example, root or branch growth. Finally, when resourc- ison of edge responses requires investigators to account
es are concentrated along the edge, then a positive edge for habitat quality on both sides of the edge in their
response is again predicted (Fig. 1e). In this case, the study design. Our model assumes that the relative avail-
Concepts & Synthesis

concentration of resources along the edge distinguishes ability of resources between patches is one of the main
this prediction from those resulting from adjacent re- drivers of edge responses (Fig. 1). While a general
sources that may be distributed evenly within each classification of habitat, such as ‘‘forest’’ or ‘‘open,’’
patch (Fig. 1b, d). may often be a good proxy for resource availability,
that need not be the case. Many published studies in-
VARIABILITY EXPLAINED BY THE MODEL clude different habitat types under a single, broad clas-
By synthesizing many of the mechanisms that have sification such as ‘‘open,’’ pooling, for example, grass-
been proposed in the edge literature into a single con- land, crops, roads, or development, all of which may
ceptual framework (Fig. 1), we suggest that many of present very different resource availability for different
the patterns and much of the variability reported in the species. When using general vegetation classifications
edge literature may be explained. For instance, this to represent habitats (a common practice that is prob-
model predicts that all species may show positive, neg- ably the most sensible approach in most cases), it is
ative, and neutral edge responses, depending on the necessary to know to what extent resource availability
specific edge type encountered. This may explain re- is associated with each habitat class. Unfortunately, this
ports of variable edge responses for the same species information is not often provided in the literature, hin-
at different edge types (Murcia 1995, Lidiker 1999). dering attempts to understand variability in edge re-
Thus, the claim that certain species or groups are in- sponses reported in many studies. We suggest that fu-
trinsically edge avoiding (such as ‘‘forest interior’’ spe- ture edge studies include information on relative hab-
cies) or edge exploiting (such as predators), may be an itat quality and resource distribution on both sides of
artifact of a focus on a single edge type (edges between the edge. Only when that information is available are
forest and open patches). This may also explain the a priori predictions possible.
lack of congruence between edge responses and area
sensitivity that has been noted in some studies (Villard VARIABILITY NOT EXPLAINED BY THE MODEL
1998) because patches may be surrounded by a variety Despite the potential for our model to explain much
of different habitat types; although in general edge re- of the inter- and intraspecific variability that has been
sponses do correlate with changes in density found in reported in the literature, it is clear that even when
different patch sizes (Bender et al. 1998, George and factors such as habitat quality, resource distribution,
Brand 2002). As future field studies target different and seasonal or regional variation in resource use are
taxa and more edge types, we expect that most species controlled for, some variability will remain. However,
will show a variety of edge responses, although there we suggest that unexplained variability is largely re-
may be groups of species that are particularly insen- stricted to finding both a consistent unidirectional edge
sitive to edges. response (either positive or negative) and neutral re-
Another implication of this model is that changes in sponses. For instance, Sisk and Battin (2002) reviewed
the use or distribution of resources may lead to changes edge responses for 12 bird species whose results were
in edge responses, even for the same species at the reported in multiple studies, all at forest edges and all
November 2004 A PREDICTIVE MODEL OF EDGE EFFECTS 2921

located in the eastern U.S. No species showed both to show positive, neutral, and negative edge responses
positive and negative responses, but most were re- depending on the edge type encountered. Therefore, to
ported to show neutral responses in some studies, as truly gauge ‘‘intrinsic’’ edge sensitivity, it is necessary
well as significant, unidirectional responses in others. to determine whether there are certain species or groups
For instance, the Ovenbird (S. aurocapillus) showed of species that either consistently show edge responses
negative responses in two studies, with two additional where they are predicted (edge-sensitive species) or
studies reporting neutral responses. Similarly, the Red- never show edge responses, regardless of predictions
eyed Vireo (V. olivaceus) had negative responses re- (edge-insensitive species). This is currently difficult,
ported in three studies and a neutral response reported because most studies have taken place at a single edge
once. type (forest edges), and have not been carried out in a
One reason for this type of intraspecific variability way that allows the separation of neutral responses into
is that there are several ecological factors that are those that are predicted and those that are not. Only
known to influence the pattern of resource distribution species that fail to show edge responses where pre-
relative to edges, as well as a species’ response to that dicted should be considered edge insensitive. However,
pattern. As these different ecological factors interact, there are reasons to suspect that certain species or
realized edge responses will range along a continuum groups may be differentially sensitive to edges, and
from strong to weak, and in some cases the effects may several authors have suggested that specific life-history
disappear altogether. Although there are likely several or ecological traits should be associated with this sen-
ecological factors that interact to change the strength sitivity, including body size, mobility, and defenses
of a species’ edge response, those that have received against predation (Wiens et al. 1985, Lidiker 1999).
the most attention are edge orientation and edge con- Other factors may include the scale at which organisms
trast (Murcia 1995). Edge orientation has been most perceive the landscape or the cues they use to assess
rigorously explored within the plant literature. Several habitat quality. By using our model to separate neutral

Concepts & Synthesis

studies have shown how the directional orientation of responses into those that are predicted and those that
the edge within the landscape may influence both the are not, it will be possible to determine if edge-insen-
strength and depth of penetration of edge effects, but sitive species truly exist. If they do, it would be useful
not the direction of the response (Wales 1972, Ranney to determine if there are life-history or ecological traits
et al. 1981, Palik and Murphy 1990, Fraver 1994). Edge that are predictably associated with intrinsic insensi-
contrast describes the degree to which bordering patch- tivity to edges.
es differ structurally from each other. Unfortunately, Ultimately, one of the difficulties of grappling with
most studies have not controlled for habitat quality on the underlying causes of variation in the nature and
both sides of the edge while varying edge contrast, strength of edge responses is the limitation of many
making it difficult to separate the influence of edge field studies. Most empirical studies of edge effects
contrast from habitat quality. However, Fletcher and have low site replication and limited statistical power
Koford (2003) showed that the magnitude of negative (Murcia 1995) and are unlikely to detect any but the
edge responses for a grassland bird was stronger at strongest patterns. Therefore, it is difficult to determine
forest (high-contrast) compared to agricultural (low- if reported neutral edge responses are actually describ-
contrast) edges, even though both constituted equally ing situations where a species is distributed evenly
poor habitat. Landscape composition has also been across an edge gradient (a truly neutral response) or if
shown to influence edge responses, with more highly it is due to a lack of power to detect responses which
fragmented landscapes showing stronger edge effects may, in fact, be operative. Nevertheless, it is clear that
in some situations (Donovan et al. 1997) and weaker there are many potential causes of the neutral responses
in others (Kremsater and Bunnell 1992). Another factor that underlie much of the variability reported in the
suggested to impact the magnitude of edge responses edge literature. Therefore, observing a neutral response
is internal patch heterogeneity (Noss 1991), and there when a positive or negative one is predicted may not
are likely other factors that underlie the variable indicate factors operating that conflict with the under-
strength of some edge responses. Identifying these fac- lying framework of our model. Instead, the separation
tors and determining how they predictably interact with of predicted from unpredicted neutral responses should
resource distribution will allow for additional variation assist in future model development through the iden-
to be accounted for. tification of the factors that underlie this variability
Another factor that may explain some unpredicted (assuming a study had sufficient power to detect edge
neutral edge responses is that certain species may be responses).
intrinsically less sensitive to the presence of habitat In contrast, we consider the observation of a signif-
edges. However, there is currently little evidence to icant response that was not predicted (for example,
suggest that any species is particularly edge sensitive observing a positive response when a negative one was
or insensitive, although Brand (2004) found that birds predicted or observing a positive or negative response
with smaller body sizes are less likely to show edge when a neutral one was predicted) to be indicative of
responses. Based on our model, all species are expected an incomplete knowledge of the distribution of critical
 2922 LESLIE RIES AND THOMAS D. SISK Ecology, Vol. 85, No. 11

resources for the focal organism, which may have led 200 m) from forest edges. June surveys were conducted
to a spurious prediction, or the operation of a dynamic in 1991 and 1992, and data on the edge responses of
not captured by our model. For instance, some mam- 25 species were presented. Sisk et al. (1997) reported
mals have been shown to avoid edges to escape pre- edge responses for 26 birds at oak woodland–grassland
dation (Bowers and Dooley 1993, Jacob and Brown edges in central, coastal California (detailed statistics
2000), although we found no evidence of this for birds. were reported in Sisk 1992). Four transects were sur-
There are also examples of interspecific competition veyed during the 1988 and 1989 breeding season, with
driving edge responses that may not be predicted by plots placed at the edge, 100, and 200 m into the oak
our model (Suarez et al. 1998, Piper and Catterall woodland and grassland habitats. Brand and George
2003). In these cases, unpredicted responses may be (2001) studied edge effects of 14 species within red-
used to identify situations where more complex species wood forest patches bordered by open habitat, includ-
interactions are occurring. ing flood plains, prairies, and human-altered habitat
including developments, roads, and power line corri-
A PRELIMINARY TEST OF THE MODEL dors. They established 12 rectangular plots extending
Rigorous testing of this model will involve deter- 400 m into the forest (no surveys were conducted in
mining the distribution of critical resources throughout open habitat) and performed surveys during the 1996
the landscape for each species of interest, predicting and 1997 breeding seasons. For all three studies, we
edge responses based on that information, and col- examined edge responses within the forest patches.
lecting independent verification data to test predictions. Because all three of our focal studies took place with-
Such detailed data on habitat quality are not usually in forest patches bordered by openings of various sizes
reported in the edge literature, and obtaining them will and types, we needed to classify each species relative
require directed field efforts, which we suggest should to their associations with forest and open habitat. For
become standard information reported in future edge this test of the model, we assumed that habitat asso-
Concepts & Synthesis

studies. However, habitat associations and general re- ciation, as reported independently in the bird literature,
source use are well described for some taxa, particu- relative to patch type (forest vs. open) was a suitable
larly birds, and it is possible to apply the model absent proxy for resource distribution. However, this assump-
local information on resource use and distribution, al- tion did not seem appropriate for shrub-dependent spe-
though predictions are likely to be affected by the rel- cies because shrubs are often associated with both for-
ative coarseness of this information. In order to per- est and open habitat. Therefore, shrub-dependent spe-
form a preliminary test of our model, we made pre- cies were excluded from this test. We used detailed
dictions for bird species whose edge-abundance re- accounts from the Birds of North America series (in-
sponses had been reported in the recent literature. We dividual references given in the Appendix) to classify
then compared model predictions with observed re- each species. When these accounts were unavailable,
sponses to determine how well our model performed. we used less-detailed information found in Ehrlich et
We focused on North American birds because habitat al. (1988). All information on edge associations was
associations are well described for most species. We ignored when making predictions. Each of the 59 spe-
limited our search to studies of multiple species at cies represented in the three studies was placed into
abrupt edges between forest and open habitats (because one of the four following categories. (1) Forest: species
habitat associations are well described relative to both was associated solely with forest. (2) Open: species
of those habitat types). In order to allow the most robust was associated solely with open habitat, which included
comparisons of predicted and actual responses, we se- any habitat with no overstory (including scrub). (3)
lected studies where quantitative data on edge respons- Both: species was associated with both forest and open
es were presented, with statistics, for multiple species. habitats. This included any species that was identified
We restricted our search to studies of multiple species as being associated with openings in forests or solely
to avoid publication biases that may lead studies to with open woodlands (thus, habitat associations were
remain unpublished if no significant effect was found, defined at a finer scale than the patch). Species that
an outcome that is most likely for single-species stud- were classified as both were further classified as to
ies. In addition, we required at least three replicate sets whether resource distribution was complementary or
of sampling points to increase the likelihood that edge supplementary. When resource distribution was de-
responses, if present, were detected. scribed as divided between habitats (always in refer-
Three edge response studies met our criteria. Ger- ence to nesting and foraging), resource distribution was
maine et al. (1997) studied edge effects at small open- listed as complementary. Absent this information, spe-
ings (0.4 ha) created by timber cuts in a hardwood cies accounts simply did not give any information on
forest in Vermont. These cuts had .95% of trees re- resource use, so we classified resource distribution as
moved and contained few shrubs (Germaine et al. unknown. (4) Shrub-dependent: These species were ex-
1997). Five independent study areas were established, cluded from the analysis.
with surveys being conducted within patch cuts, and Of the 59 species classified, seven were shrub-de-
inside the forest at three distance classes (50, 100, and pendent and so were excluded from the model test. Five
November 2004 A PREDICTIVE MODEL OF EDGE EFFECTS 2923

Concepts & Synthesis

FIG. 2. This flowchart demonstrates how edge response predictions were generated for 52 bird species, and how those
predictions compare with observed edge responses, as reported in three published studies. Model predictions were based on
the habitat associations of each species and the distribution of resources on both sides of the habitat edges (see Appendix
for species-by-species details). From this information, a positive (POS), neutral (NEUT), or negative (NEG) edge response
was predicted based on our model (see Fig. 1 for details). In some cases, resource distribution was not known (UNK), so
we predicted either a positive or neutral response, while excluding the possibility of a negative response. Results are shown
for each species reported in three studies (Germaine et al. 1997, Sisk et al. 1997, Brand and George 2002). The superscript
numbers indicate the study reporting the observed responses. Species are grouped where model predictions were correct and
incorrect. Incorrect predictions were further divided into cases where the model failed to predict a positive or negative edge
response (wrong) and where an unpredicted neutral response was observed (neutral).

of the remaining 52 species were represented in two therefore predicted to show a negative response at for-
studies, so there were 57 separate opportunities to test est edges. No species was identified as being solely
the predictions of our model. Fig. 2 shows the classi- associated with open habitat. The remaining 23 species
fication of each of the 52 species relative to habitat were classified as being associated with both forest and
associations, resource distribution, the resulting pre- open habitat, and, of those, five were shown to have
diction, and the response observed in each study. De- complementary resource distribution and therefore pre-
tails on each of these species predictions, including dicted to show a positive edge response. For the re-
common names, references for all habitat information, maining 18 species, we lacked the information to de-
a brief habitat description, and the categories assigned termine if resource use was complementary or supple-
to each species, are found in the Appendix. Of those mentary, so we predicted either a positive or neutral
52 species, 29 were classified as forest-associated, and response, but excluded the possibility of a negative one.
 2924 LESLIE RIES AND THOMAS D. SISK Ecology, Vol. 85, No. 11

Our model did well in predicting edge responses for ability of our model to account for intraspecific vari-
the 57 cases from these three empirical studies (Fig. ability. However, a recent test of this model for 15
2). With only the most basic information on habitat butterfly species at 12 edge types of varying structures
associations and resource use, we were able to correctly found that the model was successful in explaining dif-
predict 25 out of 29 cases (86%) when positive or neg- ferent observed edge responses for most species even
ative edge responses were reported, a significantly bet- at different edge types (Ries 2003). Further tests of the
ter result than would be expected if predictions were model, especially through directed field efforts, will
made at random (x2 5 15.21, df 5 1, P , 0.0001). continue to test the different mechanisms proposed in
Our model did best when predicting positive responses. Fig. 1 and evaluate the ability of this model to account
In four of the six cases when a positive response was for both inter- and intraspecific variability, as well as
predicted, it was observed with neutral responses oc- highlight other ecological factors that are important in
curring in the remaining two cases (Fig. 2). When we edge responses and may be used to explain additional
lacked information to differentiate between neutral and variability.
positive responses, but were able to exclude the pos-
CONCLUSIONS
sibility of predicting a negative response (18 cases),
only neutral or positive responses were observed (Fig. Our predictive model of changes in abundance near
2). Finally, our model was least successful in predicting edges presents a framework for understanding the
negative edge responses. Of the 33 cases where neg- broad patterns and much of the variability reported in
ative responses were predicted, they occurred only 11 a large, mostly descriptive literature. This literature
times, with 18 neutral and four positive responses ob- reports variable edge responses for many species, sug-
served (Fig. 2). As explained above, the observed neu- gesting complex mechanisms and few general patterns.
tral responses may be due to lack of statistical power, However, when viewed in the light of this relatively
insufficient detail regarding habitat quality, or intrinsic simple model, it is clear that variability in edge re-
Concepts & Synthesis

edge insensitivity. However, the four positive responses sponses should be expected, and that most of these
directly contradict the predictions of the model. One responses are predictable based on the patterns of re-
species (the Dark-eyed Junco, Junco hyemalis) is source distribution and use by each species. We also
known to be associated with open-canopy forests and present a framework for investigating variation in edge
was listed as an edge-exploiter in its species accounts, responses that is not explained by our model, through
information that we ignored when generating predic- the exploration of ecological factors that may underlie
tions. Another species (Swainson’s Thrush, Catharus the variable strength of edge responses, the search for
ustulatus) may be responding to an increase in shrubs life-history or ecological traits associated with intrinsic
along edges in the study area (T. L. George, personal edge sensitivity, and the possibility of higher-order spe-
communication). However, we have no explanation for cies interactions. By examining previous studies under
the responses of the remaining two species (Wood the umbrella of this predictive framework, and incor-
Thrush, Hylocichlla mestelina, and Black-throated porating modest habitat characterizations into future
Blue Warbler, Dendroica caerulescens), both of which edge studies, a more mechanistic understanding of edge
have strong forest associations. It is possible that there effects will emerge. As habitats become increasingly
was a complementary resource in the bordering open fragmented, conservation decisions will necessarily
habitat that may have caused the increase in density rely on predictive models of how multiple species are
near the edge (Fig. 1b). As better site-specific infor- expected to respond to complex and continuously
mation on resource use and distribution becomes rou- changing landscapes. This model of edge responses fills
tinely reported within the literature, cases such as these, a gap in a larger conceptual framework that attempts
where observed edge responses are in direct contra- to explain how habitat heterogeneity and the spatial
diction of predictions, can be more rigorously explored. patterning of landscapes impact the abundance and dis-
Although this preliminary test was successful in tribution of a broad range of organisms.
making predictions for most observed edge responses, ACKNOWLEDGMENTS
we tested only a subset of the several mechanisms in- The development of this model would not have been pos-
corporated into our model (Fig. 1). There were no spe- sible without the input of several people, including Bill Fa-
gan, James Battin, Nick Haddad, Barry Noon, Arriana Briand,
cies in our three studies associated with open habitat, and other members of the Sisk lab group. Helpful comments
so the increase predicted in less-preferred habitat for on the manuscript were made by James Battin, Brent Dan-
the transitional edge response (Fig. 1a) was not tested. ielson, Robert J. Fletcher, T. Luke George, Nick Haddad,
However, another study that measured the response of Mike Kearsley, Matthew Loeser, Thomas G. Whitham, and
an anonymous reviewer. Funding for this project was pro-
an open-habitat bird within forest edges (the Southern vided by the Strategic Environmental Research and Devel-
Emu-wren, Malurus lamberti) found the increase with- opment Program (Project CS-1100).
in forest edges (Baker et al. 2002) that is predicted by LITERATURE CITED
our model. Also, because all three studies in our pre- Alverson, W. S., D. M. Waller, and S. L. Solheim. 1988.
liminary test took place at only one edge type, all dur- Forests too deer: edge effects in northern Wisconsin. Con-
ing the breeding season, it was not possible to test the servation Biology 2:348–358.
November 2004 A PREDICTIVE MODEL OF EDGE EFFECTS 2925

Baker, J., K. French, and R. J. Whelan. 2002. The edge effect Germaine, S. S., S. H. Vessey, and D. E. Capen. 1997. Effects
and ecotonal species: bird communities across a natural of small forest openings on the breeding bird community
edge in southeastern Australia. Ecology 83:3048–3059. in a Vermont hardwood forest. Condor 99:708–718.
Bender, D. J., T. A. Contreras, and L. Fahrig. 1998. Habitat Haddad, N. M., and K. A. Baum. 1999. An experimental test
loss and population decline: a meta-analysis of the patch of corridor effects on butterfly densities. Ecological Ap-
size effect. Ecology 79:517–533. plications 9:623–633.
Berg, A., and T. Part. 1994. Abundance of breeding farmland Hansson, L. 1994. Vertebrate distributions relative to clear-
birds on arable and set-aside fields at forest edges. Ecog- cut edges in a boreal forest landscape. Landscape Ecology
raphy 17:147–152. 9:105–115.
Bergman, K. O. 1999. Habitat utilization by Lopinga achine Jacob, J., and J. S. Brown. 2000. Microhabitat use, giving-
(Nymphalidae: Satyrinae) larvae and ovipositing females: up densities and temporal activity as short- and long-term
implications for conservation. Biological Conservation 88: anti-predator behaviors in common voles. Oikos 91:131–
69–74. 138.
Bowers, M. A., and J. L. Dooley. 1993. Predation hazard and Johnston, V. R. 1947. Breeding birds of the forest edge in
seed removal by small mammals: microhabitat versus patch Illinois. Condor 49:45–53.
scale effects. Oecologia 94:247–254. Jules, E. S. 1998. Habitat fragmentation and demographic
Brand, L. A. 2004. Prediction and assessment of edge re- change for a common plant: trillium in old-growth forest.
sponse and abundance for desert riparian birds in south- Ecology 79:1645–1656.
eastern Arizona. Dissertation. Colorado State University, King, D. I., C. R. Griffin, and R. M. DeGraaf. 1997. Effect
Fort Collins, Colorado, USA. of clearcut borders on distribution and abundance of forest
Brand, L. A., and T. L. George. 2001. Response of passerine birds in northern New Hampshire. Wilson Bulletin 109:
birds to forest edge in coast redwood forest fragments. Auk 239–245.
118:678–686. Kremsater, L. L., and F. L. Bunnell. 1992. Testing responses
Brittingham, M. C., and S. A. Temple. 1983. Have cowbirds to forest edges: the example of black-tailed deer. Canadian
caused forest songbirds to decline? Bioscience 33:31–35. Journal of Zoology 70:2426–2435.
Burke, D. M., and E. Nol. 1998. Influence of food abundance, Lahti, D. C. 2001. The ‘‘edge effect on nest predation’’ hy-
nest-site habitat, and forest fragmentation on breeding ov- pothesis after twenty years. Biological Conservation 99:
enbirds. Auk 115:96–104.

Concepts & Synthesis

365–374.
Cadenasso, M. L., S. T. A. Pickett, K. C. Weathers, and C. Lay, D. W. 1938. How valuable are woodland clearings to
G. Jones. 2003. A framework for a theory of ecological birdlife? Wilson Bulletin 50:254–256.
boundaries. Bioscience 53:750–758. Lidicker, W. Z. 1999. Responses of mammals to habitat edg-
Cadenasso, M. L., M. M. Traynor, and S. T. A. Pickett. 1997. es: an overview. Landscape Ecology 14:333–343.
Functional location of forest edges: gradients of multiple McCollin, D. 1998. Forest edges and habitat selection in
physical factors. Canadian Journal of Forest Research 27: birds: a functional approach. Ecography 21:247–260.
774–782. Mills, G. S., J. B. Dunning, and J. M. Bates. 1991. The
Chalfoun, A. D., F. R. Thompson, and M. J. Ratnaswamy. relationship between breeding bird density and vegetation
2002. Nest predators and fragmentation: a review and volume. Wilson Bulletin 103:468–479.
meta-analysis. Conservation Biology 16:306–318. Mills, L. S. 1995. Edge effects and isolation: red-backed
Chasko, G. G., and J. E. Gates. 1982. Avian habitat suitability
voles on forest remnants. Conservation Biology 9:395–
along a transmission-line corridor in an oak–hickory forest
402.
region. Wildlife Monographs 82:1–41.
Murcia, C. 1995. Edge effects in fragmented forests: impli-
Chen, J. Q., S. C. Saunders, T. R. Crow, R. J. Naiman, K. D.
cations for conservation. Trends in Ecology and Evolution
Brosofske, G. D. Mroz, B. L. Brookshire, and J. F. Franklin.
10:58–62.
1999. Microclimate in forest ecosystem and landscape
ecology. Bioscience 49:288–297. Noss, R. F. 1991. Effects of edge and internal patchiness on
Donovan, T. M., P. W. Jones, E. M. Annand, and F. R. Thomp- avian habitat use in an old-growth Florida hammock. Nat-
son. 1997. Variation in local-scale edge effects: mecha- ural Areas Journal 11:34–47.
nisms and landscape context. Ecology 78:2064–2075. O’Leary, C. H., and D. W. Nyberg. 2000. Treelines between
Dunning, J. B., B. J. Danielson, and H. R. Pulliam. 1992. fields reduce the density of grassland birds. Natural Areas
Ecological processes that affect populations in complex Journal 20:243–249.
landscapes. Oikos 65:169–175. Palik, B. J., and P. G. Murphy. 1990. Disturbance versus edge
Ehrlich, P. R., D. S. Dobkin, and D. Wheye. 1988. The effects in sugar-maple/beech forest fragments. Forest Ecol-
birder’s handbook. Simon and Schuster, New York, New ogy and Management 32:187–202.
York, USA. Paton, P. W. C. 1994. The effect of edge on avian nest success:
Fagan, W. E., R. S. Cantrell, and C. Cosner. 1999. How how strong is the evidence. Conservation Biology 8:17–
habitat edges change species interactions. American Nat- 26.
uralist 153:165–182. Piper, S. D., and C. P. Catterall. 2003. A particular case and
Fletcher, Jr., R. J., and R. R. Koford. 2003. Spatial responses a general pattern: hyperaggressive behaviour by one spe-
of bobolinks (Dolichonyx oryzivorus) near different types cies may mediate avifaunal decreases in fragmented Aus-
of edges in northern Iowa. Auk 120:799–810. tralian forests. Oikos 101:602–614.
Fraver, S. 1994. Vegetation responses along edge-to-interior Ranney, J. W., M. C. Bruner, and J. B. Levenson. 1981. The
gradients in the mixed hardwood forests of the Roanoke importance of edge in the structure and dynamics of forest
River basin, North Carolina. Conservation Biology 8:822– islands. Pages 67–96 in R. L. Burgess and D. M. Sharpe,
832. editors. Forest island dynamics in man-dominated land-
Gates, J. E., and L. W. Gysel. 1978. Avian nest dispersion scapes. Springer-Verlag, New York, New York, USA.
and fledging success in field–forest ecotones. Ecology 59: Ries, L. 2003. Placing edge responses into a predictive frame-
871–883. work. Dissertation. Northern Arizona University, Flagstaff,
George, T. L., and L. A. Brand. 2002. The effects of habitat Arizona, USA.
fragmentation on birds in coast redwood forest. Studies in Risser, P. G. 1995. The status of the science examining eco-
Avian Biology 25:92–102. tones. Bioscience 45:318–325.
 2926 LESLIE RIES AND THOMAS D. SISK Ecology, Vol. 85, No. 11

Shmida, A., and M. V. Wilson. 1985. Biological determinants Strayer, D. L., M. E. Power, and W. F. Fagan. 2003. A clas-
of species diversity. Journal of Biogeography 12:1–20. sification of ecological boundaries. BioScience 53:723–
Sisk, T. D. 1992. Distributions of birds and butterflies in 729.
heterogeneous landscapes. Dissertation. Stanford Univer- Suarez, A. V., D. T. Bolger, and T. J. Case. 1998. Effects of
sity, Stanford, California, USA. fragmentation and invasion on native ant communities in
Sisk, T. D., and J. Battin. 2002. Habitat edges and avian coastal southern California. Ecology 79:2041–2056.
ecology: geographic patterns and insights for western land- Villard, M. A. 1998. On forest-interior species, edge avoid-
scapes. Studies in Avian Biology 25:30–48. ance, area sensitivity, and dogmas in avian conservation.
Sisk, T. D., N. M. Haddad, and P. R. Ehrlich. 1997. Bird Auk 115:801–805.
assemblages in patchy woodlands: modeling the effects of Wales, B. A. 1972. Vegetation analysis of north and south
edges in a mature oak–hickory forest. Ecological Mono-
edge and matrix habitats. Ecological Applications 7:1170–
graphs 42:451–471.
1180. Wiens, J. A., C. S. Crawford, and J. R. Gosz. 1985. Boundary
Sisk, T. D., and C. R. Margules. 1993. Habitat edges and dynamics: a conceptual framework for studying landscape
restoration: methods for quantifying edge effects and pre- ecosystems. Oikos 45:421–427.
dicting the results of restoration efforts. Pages 57–69 in D. Zabel, C. J., K. McKelvey, and J. P. Ward, Jr. 1995. Influence
A. Saunders, R. J. Hobbs, and P. R. Ehrlich, editors. Nature of primary prey on home-range size and habitat-use pat-
conservation 3: reconstruction of fragmented ecosystems. terns of northern spotted owls (Strix occidentalis caurina).
Surrey, Beatty and Sons, Sydney, North Wales, Australia. Canadian Journal of Zoology 73:433–439.

APPENDIX
Information used to generate predictions for 52 bird species whose edge responses were reported in the recent literature
is available in ESA’s Electronic Data Archive: Ecological Archives E085-093-A1.
Concepts & Synthesis