Patch‐scale edge effects do not indicate landscape‐scale fragmentation effects

Negative landscape‐scale fragmentation effects are often inferred from negative patch‐scale edge effects. I tested this cross‐scale extrapolation using two evaluations. First, I searched for studies that estimated the direction of both a patch‐scale edge effect and a landscape‐scale fragmentation effect. The directions were concordant and discordant in 55% and 45% of cases, respectively. Second, I extracted from the literature a sample of landscape‐scale fragmentation effects on individual species. Then, for each species I searched for studies from which I could calculate the slope of its patch‐scale edge effect. Species showing negative patch‐scale edge effects were nearly equally likely to show negative or positive landscape‐scale fragmentation effects, and likewise for species showing positive patch‐scale edge effects. The results mean that the efficacy of policies related to habitat fragmentation cannot be inferred from observed patch‐scale edge effects. Such policies require landscape‐scale evidence, comparing species' responses in landscapes with different levels of fragmentation.

An example of such cross-scale extrapolation is the inference of landscape-scale fragmentation effects from patch-scale edge effects (Betts et al., 2019;Haddad et al., 2015;Ries et al., 2004;Willmer et al., 2022).A "patchscale edge effect" is a change in an ecological response with proximity to habitat edge (Ries et al., 2004).Such edge effects are documented either by sampling the ecological response along transects from the edge to interior of habitat patches-contiguous areas of a given habitat type such as forest or grassland-or by comparing the F I G U R E 1 For a given total amount of habitat in a landscape (sum of green areas within the large squares), the total length of habitat edge (in darker green) increases with increasing habitat fragmentation.
ecological response at the edge versus the interior of habitat patches.A higher value of the ecological response at edge sites than interior sites is then a positive patchscale edge effect, while the opposite pattern is a negative patch-scale edge effect.
Inferences about landscape-scale fragmentation effects from patch-scale edge effects are intuitive, because a landscape with more-fragmented habitat usually contains more edge than a landscape with less-fragmented habitat.This is necessarily so when comparing landscapes with the same total amount of habitat but different levels of fragmentation, i.e., differing in fragmentation per se (sensu Fahrig, 2003; Figure 1).Habitat fragmentation per se (hereafter simply "fragmentation") is typically measured as the number of habitat patches or the edge density (edge length per area) in a landscape, where a landscape is scaled to the process and species of interest (Jackson & Fahrig, 2012, 2015).A higher value of the ecological response in landscapes with more-fragmented habitat than in landscapes with less-fragmented habitat is a positive fragmentation effect, while the reverse is a negative fragmentation effect.
The cross-scale extrapolation from patch-scale edge effects to landscape-scale fragmentation effects assumes that if a species has higher (or lower) density at habitat edges than interiors, its density in the landscape should increase (or decrease) with habitat fragmentation (Figure 2).This is because landscapes with morefragmented habitat have more edge than landscapes with less-fragmented habitat (Figure 1).Although this seems self-evident, it is logically possible that the density of an interior species could increase with landscape-scale fragmentation, and of an edge species could decrease with landscape-scale fragmentation (Figure 3).For example, dispersing individuals could tend to settle in edges, causing a positive patch-scale edge effect, but predation rates might be higher in landscapes with more-fragmented habitat, causing a negative landscape-scale fragmentation effect.Thus, the cross-scale extrapolation from edge effects to fragmentation effects cannot be assumed, but must be empirically demonstrated.
Here I ask, can the direction of a landscape-scale fragmentation effect be reliably inferred from the direction of a patch-scale edge effect?To address this question, I conducted two evaluations using published information.First, I searched for studies that estimated both a patch-scale edge effect and a landscape-scale fragmentation effect on the same response, to determine whether their directions of effect were the same.As I found few such studies, I then conducted a second evaluation in which the patchscale edge effect and landscape-scale fragmentation effect could come from different studies.For this, I extracted from the literature a sample of the direction of fragmentation effects on individual species and, for each of these species, I conducted a literature search for studies documenting a gradient in density/occurrence from interior to edge habitat.I then tested whether the slope of the edge effect is a reliable indicator of the direction of the fragmentation effect, across species.This second evaluation makes the common assumption that patch-scale edge effects are species attributes (Villard, 1998).

Edge and fragmentation effects from the same study
On April 13, 2023, I searched in Web of Science and Scopus for papers containing both local-scale edge effects and landscape-scale fragmentation effects.For edge effects, I used the search term "edge effect."For fragmentation effects, as I was interested in fragmentation per se, I searched for (i) studies using landscape-scale measures of fragmentation that are often uncorrelated with habitat amount (e.g., edge density, number of patches) and (ii) studies using landscape-scale measures of fragmentation that are typically correlated with habitat amount (e.g., mean patch size) but that controlled for habitat amount in their analyses.Note that landscape-scale fragmentation studies measure the species response at one or more points in each of multiple landscapes that vary along a gradient of habitat fragmentation (Figure 2b, right side).Thus, I used the following search string: "edge effect" AND ("fragmentation per se" or "SLOSS" or "edge density" or "edge length" or "number of patches" or "mean patch size" or "boundary length" or "patch density" or "median patch size" or "clumping index" or "splitting index" or "aggregation index" or "like adjacencies" or "fractal dimension" or "IJI" or "mean circumscribing circle" or "largest patch index" or "shape index" or "mean core area" or "proportion core area" or "mean nearestneighbor" or "mean perimeter to area" or "mean edge to area") AND ("habitat" or "forest" or "grassland" or "wetland" or "coral" or "landscapes" or "watersheds" or F I G U R E 2 Inferring landscape-scale fragmentation effects from patch-scale edge effects is a cross-scale extrapolation.Green rectangles are habitat patches.(a) Illustration of patch-scale edge effects.Dots represent individuals of two species: the edge species is more likely to occur and has higher densities near the edge than in the interior of the patch, while the interior species is less likely to occur and has lower densities near the edge than in the interior of the patch.Note in an empirical study transects are usually placed on multiple patches.(b) Illustration of cross-scale extrapolation from patch-scale edge effects to the direction of landscape-scale fragmentation effects.Red x's are sample sites.For the same total amount of habitat, a landscape containing more-fragmented habitat has more edge and less interior than a landscape containing less-fragmented habitat.If patch-scale edge effects determine landscape-scale fragmentation effects, then species with negative edge effects (interior species) should be less likely to occur and have lower densities at sites in more-fragmented landscapes, and species with positive edge effects (edge species) should be more likely to occur and have higher densities at sites in more-fragmented landscapes.Note in empirical studies often only one site is sampled in each landscape, typically at its center.
F I G U R E 3 Illustration of how a species could simultaneously show a negative patch-scale edge effect and a positive landscape-scale fragmentation effect (a) or a positive patch-scale edge effect and a negative landscape-scale fragmentation effect (b).Large squares are landscapes, green rectangles are patches, rectangle outlines within patches indicate the approximate distance dividing edge from interior habitat for these species, and dots represent individuals of an interior species (a) and an edge species (b).All four landscapes contain the same amount of habitat.
"catchments"), within research areas "environmental sciences/ecology" and "biodiversity conservation" for Web of Science, and subject areas "agricultural and biological sciences" and "environmental science" for Scopus.
For any study containing quantitative estimates of a patch-scale edge effect and a landscape-scale fragmentation effect, I recorded the direction of these effects, irrespective of statistical significance, taken from tables, figures, or appendices.I did not attempt to extract effect sizes for two reasons.First, the diversity of responses and study designs would have led to essentially no quantitatively comparable effects across studies.Second, my goal was specifically to determine whether the direction of a fragmentation effect can be inferred from the direction of an edge effect.Due to the paucity and diversity of studies containing both patch-scale edge effects and landscapescale fragmentation effects, I did not analyze the results but simply summarized them in a table.

Edge and fragmentation effects as species attributes
For the second evaluation, I first searched in Web of Science for studies of landscape-scale fragmentation effects on density or occurrence of individual species published within a 6-year period from January 1, 2016 to December 2, 2021.The 6-year period was selected as a trade-off between screening a manageable number of studies and finding an adequate sample of single-species fragmentation effects.I then searched the patch-scale edge-effect literature, without any date restrictions, to try to match up as many of the landscape-scale fragmentation effects as possible to a patch-scale edge effect on the same species.

Landscape-scale fragmentation effects
To find papers estimating landscape-scale fragmentation effects on individual species, I used the same search string as above, excluding the term "edge effect."I extracted the direction of the fragmentation effect (irrespective of statistical significance) on species density/occurrence from tables, figures, or appendices.I also noted whether the effect was statistically significant (or in the top model in multimodel inference).I only recorded the direction of the fragmentation effects because the coefficients were not quantitatively comparable; fragmentation metrics varied among and within studies, and coefficients were almost always taken from models containing multiple predictor variables.

Local-scale edge effects
For each species for which I recorded a direction of fragmentation effect, I then conducted a search for studies from which I could estimate a patch-scale edge-effect slope (as in Figure 2b, left side).I did not limit the year range of these searches and I found relevant papers dating back to the 1980s.I conducted each species' search in Web of Science using the search string "[species name]" AND "edge." Where this led to a large number of clearly irrelevant papers, I added the string AND ("distance" or "gradient" or "near" or "far" or "transect" or "proximity" or "interior").
For each paper where sufficient information was provided, I calculated a scaled slope of density or (probability of) occurrence versus distance from the edge (see below).Note I did not include other ecological responses to edges such as nest survival, because my goal was to test the extrapolation from patch-scale edge effects on density/occurrence to the direction of landscape-scale fragmentation effects on density/occurrence.Including measures such as predation rate or nest survival would have entailed a second extrapolation from those outcomes to density/occurrence.
To estimate the scaled slope of the edge effect, I required values of density, occurrence, or proportion/probability of occurrence with distance from the edge.I extracted these from tables, figures, or appendices.I then converted the value at each distance (summed per distance) to a proportion of total observations (summed across distances), and I converted each distance to its proportion from the maximum distance from the edge in that study.In a few cases, the transect was longer than the apparent edge effect.In those cases, instead of the maximum distance from the edge, I used the distance at which the observed density/occurrence values appeared to level off, to get the slope of the edge effect over the relevant portion of the edge gradient.For example, from figure 2 in Kroodsma (1984), the density of Towhee (Pipilo erythrophthalmus) was level from 90 to 240 m, so I estimated the edge-effect slope between 90 and 0 m.I used the proportion of distance from the interior rather than from the edge so that negative slopes would represent negative edge effects and positive slopes would represent positive edge effects (Figure 2b, left side).Note I scaled both the species observations and the distances to create comparable slope values so that I could conduct a cross-species evaluation of the relationship between the edge-effect slope and the direction of the fragmentation effect (below).Thus, I assumed that authors of the edge-effect studies selected relevant distances from the edge for their species.

Relationship between edge effect and direction of fragmentation effect
To determine whether edge effects reliably indicated the direction of fragmentation effects, I fit a mixed effects model with a binomial response (positive or negative fragmentation effect) on the scaled slope of the patch-scale edge effect (fixed effect), and fragmentation study i.d. and species as two random effects, using the lme4 package (Bates et al., 2015) in R (R Core Team, 2020).

Edge and fragmentation effects from the same study
The search yielded 175 titles, of which 10 studies contained 31 cases of both local-scale edge effects and landscape-scale fragmentation effects on the same ecological response (Table 1).When single studies contained multiple cases, these were for different taxa; different edge types for the edge effect; responses of different trait-defined groups; or responses of different biodiversity measures.Taxa were mainly invertebrates and plants.Habitat type for the invertebrates was mainly crop; for other taxa, it was forest.
Given the small sample size and the high diversity of study conditions and nonindependence across them (e.g., confounding of habitat type and taxon), it was not possible to build a robust statistical model from the information in Table 1.However, the results did not support the idea that the direction of a local-scale edge effect reliably indicates the direction of a landscape-scale fragmentation effect: in 17 cases, the two directions were concordant, but in 14 cases, they were discordant.Studies with multiple cases often found both concordant and discordant cases.There was no obvious pattern of concordance/discordance with taxon: there were 10 concordant and 7 discordant invertebrate cases, 5 concordant and 4 discordant plant cases, and 2 of each for mammals.There was a hint of higher concordance in community-level responses (10 concordant vs. 5 discordant) than in single-species responses (7 vs. 9), but a G-test of independence for this comparison was nonsignificant (p = 0.63).

Edge and fragmentation effects as species attributes
The search for fragmentation studies yielded 839 titles, 62 containing a direction of landscape-scale fragmentation effect on density/occurrence of at least one species.

F I G U R E 4
Relationship between the scaled slope of the patch-scale edge effect and the direction of the landscape-scale fragmentation effect.Each point represents the direction of a fragmentation effect on a given species taken from a single fragmentation study, matched to a patch-scale edge effect from an edge-effect study on that same species.Negative edge-effect slopes indicate the species density/occurrence is lower at the edge than in the interior of habitat (as in Figure 2).If edge effects reliably predict the direction of fragmentation effects then the points should be clustered in the bottom left and top right quadrants of the figure.Points are vertically jittered to allow distinguishing individual points.
These 62 papers yielded 678 values of the direction of fragmentation effect, for 425 different species.Fragmentation was measured using a variety of metrics.Grouped into general categories these included 299 cases of edge density, 253 of patch density, 177 of aggregation, 35 of principal components combining fragmentation metrics, and 8 of mean patch size.These numbers do not add to 678 because 94 studies contained fragmentation effects estimated using multiple types of fragmentation metrics.If the direction of fragmentation effect was the same for two (or more) metrics, I counted these as a single positive or negative fragmentation effect.Of these 425 species, my search for studies from which I could calculate patchscale edge-effect slopes yielded 115 species with at least one edge-effect slope (42 birds, 28 plants, 24 mammals, 18 arthropods, 2 microorganisms, and 1 amphibian), and 254 edge-effect slope values altogether.Negative and positive edge-effect slopes were evenly distributed across year of publication (Figure S1).When there were multiple fragmentation directions or multiple edge-effect slopes for a single species, I included each combination as an observation in the dataset.
There was essentially no relationship between the slope of the edge effect and the direction of the fragmentation effect (Figure 4).Although the coefficient was positive as expected, the relationship was very weak and nonsignificant (p = 0.758; Table S1).Species with stronger TA B L E 1 Comparison of patch-scale edge effects with landscape-scale fragmentation effects from studies measuring both effects on the same ecological response.Inferring that "edge length" refers to edge length of wheat fields in the landscape, but it is not explicit.

Study
d one of the 10 species that are located distinctly away from the center of the ordination space (species at the center not included).
e Positive effect of mean field size.f Standardized differences.
negative patch-scale edge effects (interior species) were not more likely to show negative than positive fragmentation effects, and species with stronger positive patch-scale edge effects (edge species) were not more likely to show positive than negative fragmentation effects, irrespective of taxon, biome, or whether or not the fragmentation effect was statistically significant (Figures S2-S4).

DISCUSSION
The results suggest that patch-scale edge effects do not reliably indicate the direction of landscape-scale fragmentation effects.It is possible that edge effects cause some observed fragmentation effects, that is, some cases of concordance in Table 1, and/or some points in the lower left and upper right quadrants of Figure 4.However, the result that 14 of 31 cases (45%) in Table 1 were discordant, and the lack of overall relationship in Figure 4 invalidate extrapolation from local-scale edge effects to landscape-scale fragmentation effects.Knowledge of a patch-scale edge effect cannot be used as a reliable indicator of the direction of a landscape-scale fragmentation effect, and so landscape-scale policies for habitat fragmentation should not be informed by patch-scale edge effects (e.g., Pierce et al., 2011).Such cross-scale recommendations remain current.For example, using their finding that negative edge effects on species richness are more common in tropical than temperate regions, Willmer et al. (2022) suggested that the landscape design recommended by Arroyo-Rodriguez et al. (2020) is not suitable in tropical regions.This and other such inferences rely on extrapolation from patch-scale edge effects to landscape-scale fragmentation effects, which we cannot assume to be correct.
Two previous studies that evaluated the extrapolation from patch-scale edge to landscape-scale fragmentation effects in particular contexts found results similar to mine.Mendes and Prevedello (2020) tested the extrapolation from higher temperatures at forest edges to higher temperatures in landscapes with more-fragmented forest.They found the opposite: landscapes with higher forest fragmentation per se are cooler than landscapes with lower fragmentation per se.The second study, a meta-analysis by Yarnall et al. (2022), found opposite responses to patch-scale edge effects and landscape-scale fragmentation effects in seagrass for total faunal density and two out of four guild-specific fish and invertebrate densities.They suggest that this results from landscape-scale processes, and "caution investigators against assumptions that edge and fragmentation effects are functionally equivalent." How is it possible for a species or species group to show a negative edge effect while at the same time a positive effect of fragmentation, or vice versa, as illustrated in Figure 3? I can only speculate on this, as my study was designed to test the extrapolation, not identify mechanisms.Similar to Mendes and Prevedello (2020) and Yarnall et al. (2022), I speculate that the answer lies in landscapescale processes that operate across multiple patches and that cannot be inferred from observations of individual patches.Patch-scale edge effects and landscape-scale fragmentation effects can arise from various processes at their respective scales.For example, some species respond to higher productivity and insect prey density at forest edges (Wirth et al., 2008), and others require cooler, damper conditions in the forest interior (Ries et al., 2004).At a landscape scale, habitat fragmentation per se can influence species interactions, spread the risk of extinction, and increase habitat connectivity (e.g., Hammill & Clements, 2020;Roland, 1993;Tischendorf & Fahrig, 2000, respectively;reviewed in Fahrig, 2017;Fahrig et al., 2019).The concordance or discordance between local-scale edge effects and landscape-scale fragmentation effects will then depend on the relative strengths of such processes operating at the two scales and on how they interact with each other.If the landscape-scale processes are strong, population density could be higher (or lower) across a more-fragmented landscape than a less-fragmented landscape, despite a negative (or positive) patch-scale edge effect, resulting in discordance of local-scale edge effects and landscape-scale fragmentation effects.Future work is needed to (i) determine whether the lack of reliable cross-scale inference holds over a larger sample size, (ii) elucidate the cause(s) for the lack of strong predictive relationship, and (iii) determine whether there are situations where edge effects reliably indicate fragmentation effects.
That patch-scale edge effects do not reliably indicate the direction of landscape-scale fragmentation effects is an example of a general problem that research is often conducted at spatial extents smaller than those relevant to conservation decision-making (Estes et al., 2018).For example, a similar problem occurs when extrapolating from patch size effects to fragmentation effects (Riva & Fahrig, 2023).Using the "FragSAD" database of species on multiple patches across multiple studies, Chase et al. (2020) showed a disproportionate patch-scale increase in species richness, with increasing patch size.However, using the same database, Riva and Fahrig (2023) showed declining species richness with increasing mean patch size at a landscape scale, indicating that the patch-scale results could not be extrapolated to a landscape scale.
If extrapolation from patch-scale patterns to landscapescale conservation implications is not valid, then we should base landscape-scale conservation recommendations on studies that compare ecological responses across multiple landscapes.Such studies are challenging as they typically must be conducted over very large spatial extents to capture landscape gradient(s) (e.g., Herse et al., 2020;Morante-Filho et al., 2021).However, there is a growing number of such studies.In some cases, large-scale coordinated data-collection efforts are needed (e.g., Sirami et al., 2019).Recent increases in citizen science databases (e.g., eBird, NatureServe) may alleviate this, but these have sampling biases that must be accounted for (Dickinson et al., 2010;e.g., Riva et al., 2023).
To conclude, the results of this study suggest that the direction of fragmentation effects cannot be reliably inferred from patch-scale edge effects.Species or communities that are more prevalent at the edge than the interior may show either positive or negative landscape-scale fragmentation effects, and likewise for species or communities that are more prevalent in the interior than at the edge.Thus, negative (or positive) edge effects should not be used as evidence of negative (or positive) fragmentation effects.Patch-scale edge effects can inform policy regarding single patches, but fragmentation effects, and the efficacy of policies related to them, require evidence at a landscape scale, where the response is measured and compared across landscapes with different levels of fragmentation.

A U T H O R C O N T R I B U T I O N
L. Fahrig conducted all components of the study.

A C K N O W L E D G M E N T S
I am grateful to Christine Anderson, Allison Binley, Brandon Edwards, Charles Francis, Joan Freeman, Carmen Galán Acedo, Ana Hernandez Martinez de la Riva, Emma Hudgins, Sahebeh Karimi, Matt Keevil, Amanda Martin, Greg Mitchell, Iman Momeni, Samantha Morin, Dave Omond, Federico Riva, Adam Smith, and the GLEL Friday discussion group for their comments and suggestions on an earlier draft of this paper.I thank three reviewers for their insightful comments and suggestions.

F U N D I N G
Funding was provided by the John Simon Guggenheim Foundation and the Natural Sciences and Engineering Research Council of Canada.

C O N F L I C T O F I N T E R E S T S TAT E M E N T
No conflicts of interest.

D ATA AVA I L A B I L I T Y S TAT E M E N T
All data used in this paper are included in the Supporting Information.

O R C I D
Lenore Fahrig https://orcid.org/0000-0002-3841-0342 Figure 1, years combined Text bottom left p. 212 b Negative InDijak and Thompson (2000), the three responses are distinguished due to different edge types for the local-scale edge effects (bold text).InMatos et al. (2017), the pairs of responses are distinguished due to different metrics of landscape-scale fragmentation (bold text).Blue fill indicates negative and yellow fill indicates positive responses to edges or fragmentation.Green fill indicates cases where both the local-scale edge effect and the landscape-scale fragmentation effect are in the same direction (concordant), while orange fill indicates cases where they are in different directions (discordant).1 Akter S, Rizvi SZM, Haque A, Reynolds OL, Furlong MJ, Melo MC, Osborne T, Mo J, McDonald S, Johnson AC, Gurr GM. (2023) Continent-wide evidence that landscape context can mediate the effects of local habitats on in-field abundance of pests and natural enemies.Ecology and Evolution, 13, e9737. 2 Dijak WJ, Thompson FR.(2000) Landscape and edge effects on the distribution of mammalian predators in Missouri.The Journal of Wildlife Management 64, 209-216.3 Filgueiras BKC, Peres CA, Iannuzzi L, Tabarelli M, Leal IR. (2023) Functional reorganization of dung beetle assemblages in forest-replacing sugarcane plantations Journal of Insect Conservation 26, 683−695.4 Fischer C, Riesch F, Tscharntke T, Batáry P. (2021) Large carabids enhance weed seed removal in organic fields and in large-scale, but not small-scale agriculture.Landscape Ecology 36, 427−438.5 Gallé R, Geppert C, Földesi R, Tscharntke T, Batáry P. (2020) Arthropod functional traits shaped by landscape-scale field size, local agri-environment schemes and edge effects.Basic and Applied Ecology 48, 102-111.6 Jin Y, Didham RK, Yuan J, Hu G, Yu J, Zheng S, Yu M. (2020) Cross-scale drivers of plant trait distributions in a fragmented forest landscape.Ecography 43: 467−479.7 Matos FAR, Magnago LFS, Gastauer M, Carreiras JMB, Simonelli M, Meira-Neto JAA, Edwards DP. (2017) Effects of landscape configuration and composition on phylogenetic diversity of trees in a highly fragmented tropical forest.Journal of Ecology 105, 265−276.8 Meyer CFJ, Kalko EKV.(2008) Assemblage-level responses of phyllostomid bats to tropical forest fragmentation: land-bridge islands as a model system.Journal of Biogeography 35, 1711−1726.9 Reidy JL, Thompson FR, Peak RG. (2009) Factors affecting Golden-cheeked Warbler nest survival in urban and rural landscapes.Journal of Wildlife Management 73, 407-413.10 Vinter T, Dinnétz P, Danzer U, Lehtilä K. (2016).The relationship between landscape configuration and plant species richness in forests is dependent on habitat preferences of species.European Journal of Forest Research 135, 1071−1082.a Very slight.b Negative effect of landscape contagion (landscapes mostly forest and crop) in a model controlling for forest mean nearest neighbor distance.c