Maturation of forest edges is constrained by neighbouring agricultural land management


(corresponding author,



(i) Do species richness and turnover across forest edges change with edge age and management intensity of adjacent lands? (ii) Does edge species composition respond to aging and landscape management and what are the environmental factors explaining this response?


Agricultural landscapes of the Picardy region, N France.


We sampled forest edges differing in age (from a few decades to several centuries) and embedding landscape matrix (from slightly managed ‘bocages’ to intensively cultivated open fields). We recorded vascular plant species and a set of environment, landscape and historical variables along transects oriented perpendicularly to forest edges. We used mixed models to assess the impact of edge age and landscape type on edge species richness and turnover. We investigated the relationship between edge community composition and explanatory variables using redundancy analyses and a split-plot design.


Species richness decreased with both increased edge age and increased landscape management intensity, while species turnover was not influenced by any of these factors. Edge maturation was characterized by a specialization of the flora over the entire transect, which is likely a response to increased shade, litter layer thickness and soil acidity. As landscape management was more intensive, true forest species were replaced by nitrophilous and/or calciphilous non-forest species, which might be more tolerant of agrochemical and lime drift and are able to disperse through a hostile matrix.


Although edge aging was associated with the progressive development of environmental gradients, especially light availability and litter thickness, plant communities poorly reflect these gradients under the constraint of neighbouring landscape management. On the contrary, the stronger the management intensity, the sharper the edge–interior gradient.

Lambinon et al.



In most temperate regions of the world, forest habitats have a long history of fragmentation due to human activities, making edges a ubiquitous feature of modern landscapes (Riitters et al. 2000). In Western Europe, landscapes have become patterned through at least two millennia of human management. Clearance of forests for agriculture and recovery of forests on abandoned lands have made the present-day forests patchy, consisting of patches of different quality, age, size and isolation, which are embedded in a more or less intensively managed agricultural matrix (Honnay et al. 2005; Jamoneau et al. 2011). Forest fragmentation is associated with a decrease of the forest interior habitat but, to some extent, an increase in the edge length. Hence fragmentation is expected to reduce the available habitat for forest specialists, but provide a unique habitat for a number of edge species (Jacquemyn et al. 2001; Honnay et al. 2002a; Kolb & Diekmann 2004; Brunet 2007), making the boundary plant community of special interest in biological conservation. Forest edges are not only important structural components of the modern rural landscape, but also major functional features (Pickett & Cadenasso 1995; Ries et al. 2004), as they regulate many biological processes, such as dispersal, competition, species invasibility and speciation (Fagan et al. 1999; Ries et al. 2004), provide services to neighbouring crop fields, such as the regulation of phytophagous insect populations (Sarthou et al. 2005), and maintain biodiversity in forest interiors (Matlack 1993).

Vegetation structure and composition change along edge-associated environmental gradients (Matlack 1993; Meiners & Pickett 1999; Dutoit et al. 2007; Hamberg et al. 2009), defining the so-called edge effect (Laurance & Yensen 1991; Murcia 1995). Edge influence varies in space and time (Harper et al. 2005; Ewers & Didham 2008). Their depth can vary from a few meters (Ranney et al. 1981) to more than a kilometre (Ewers & Didham 2008), although it generally does not exceed 50 m in temperate forests in terms of plant species (Honnay et al. 2002b).

A first set of factors determining edge influence relates to the age of forest edges (Matlack 1993; Cadenasso et al. 2003). Gradients in local biotic and abiotic factors are likely to develop as a given forest patch ages: towards the forest interior light availability strongly decreases, while litter thickness and soil acidity increase (Hermy et al. 1999; Brunet et al. 2000). A limited number of habitat specialists are able to survive in such environmental conditions and tend to replace generalists towards the forest interior. Edge maturation should thus result in niche partitioning along the created gradient (Matlack 1994; Mikk & Mander 1995; Murcia 1995), hence leading to a high species turnover along the forest edge–interior gradient (Honnay et al. 2002a; Marchand & Houle 2006).

A second set of factors is associated with the surrounding landscape, as adjacent non-forest habitats greatly influence the edge structure and the degree of edge influence, and hence the species composition (Ries et al. 2004; Harper et al. 2005; Jamoneau et al. 2011). Edges are particularly vulnerable to disturbances. In agricultural landscapes edges are exposed to operations conducted in neighbouring farmland. They can be destroyed directly by agricultural vehicles (e.g. ploughing), and/or indirectly through the drift of agrochemicals. Forest edges have been shown to receive both high atmospheric nitrogen deposits (Draaijers et al. 1988; Weathers et al. 2001) and high input of nutrients by drifting fertilizer from adjacent farmlands (Kleijn & Snoeijing 1997). The resulting eutrophication leads to the dominance of competitive-ruderal species, such as Rubus spp. and Urtica dioica L., which in turn competitively exclude habitat specialists (Marrs 1993; Van der Veken et al. 2004). Biocides, especially herbicides, have been shown to directly reduce species diversity (Kleijn & Snoeijing 1997). All of the above factors can dramatically alter environmental gradients at the edge or prevent them from developing, making the forest–agricultural land boundary sharper.

As the boundary between forests and other components of the landscape, edges are often the first filter for organisms moving from one forest patch to another (Wiens 1992; Honnay et al. 2002a). The permeability of this filter determines how edges mediate fluxes of organisms, especially plant propagules and their vectors (Stamps et al. 1987), and thus influences both the edge and forest interior composition. Landscape fluxes of species are particularly critical to forest patches embedded in agricultural landscapes, as they are the basis of forest metacommunity functioning (i.e. local community exchanging species through dispersion; Jamoneau et al. 2011, 2012). For example, it is likely a focal forest patch accumulates more forest specialists over time as the forest cover increases within a given radius around it (Jamoneau et al. 2011), because these species are well known for their dispersal limitations (Honnay et al. 1999; Verheyen & Hermy 2001).

The objective of this study was to elucidate the circumstances under which a forest edge is structured as a sharp or as a gradual transition zone between agricultural land and the forest interior. We hypothesized that (i) an edge–forest interior environmental gradient progressively develops during edge maturation and that species accumulate over time so that the edge becomes a gradual transition zone with increasing species richness and turnover as an edge ages (H1); (ii) disturbances originating from agricultural landscape matrices alter forest edges so that the edge structure shifts from a gradual transition zone to a sharp boundary as disturbance intensity increases, together with a decrease in both species richness and turnover (H2). To test these hypotheses, we addressed the following research questions:

  1. Do species richness (α- and γ-diversity) and turnover (β-diversity) along the edge–forest interior gradient increase with increasing edge age but decrease with increasing agricultural intensity?
  2. How does plant species composition respond to edge gradients and what are the environmental factors explaining this response?


Study area

The study was conducted in N France, Picardy region (49°25′–50°11′ N, 1°52′–3°55′ E, 19 400 km2, 60–220 m a.s.l.). The climate is sub-oceanic, with annual rainfall and temperature averaging 700 mm and 10 °C, respectively. The geological substrate mainly consists of Cretaceous chalk usually covered by Quaternary loess. Rural landscapes have been largely patterned by agricultural activities for at least two millennia; open fields dominate, which are intensively cultivated mainly for cereals and sugar beet. However, several small natural areas still support less intensive activities, mainly cattle rearing, because of specific climate (cold, wind and/or high precipitation) or edaphic (acidic and/or waterlogged soils) conditions. In the latter case, fields and grasslands are smaller and generally enclosed by hedgerows, characterizing the so-called ‘bocage’ landscape. Both landscape types contain woodland elements, mainly small private woodlands that are managed by their owners for recreational purposes (hunting) and/or fuel production. A few large forest tracts are also present, which are managed for wood production as regular (state owned) or irregular (privately owned) high forests. In the study area, the total forest area has remained stable since the 18th century, but the number of patches has regularly increased, indicating a concomitant contraction of ancient forests and reforestation of abandoned farmlands (Jamoneau et al. 2010). As a result, new forest boundaries have been created while the existing boundaries have experienced different episodes of forest expansion and contraction in relation to neighbouring agricultural practices. The general trend is towards old, stable edges for state and former royal forests, but younger, fluctuating edges for privately owned woodlands.

Site inclusion and data collection

Based on the results of an earlier study (Jamoneau et al. 2010) we followed a stratified sampling design so as to include forest edges of various ages (from ca. 26 yr to several centuries), embedded in contrasting agricultural landscapes (from bocage to intensively cultivated open fields). All sampled edges were composed of a mixture of deciduous canopy trees and showed no evidence of (recent) management. The most frequent tree species along the sampled forest edges were Carpinus betulus, Fraxinus excelsior, Quercus robur, Fagus sylvatica and Acer pseudoplatanus. A total of 54 transects were randomly disposed perpendicular to the forest edge, at least 50 m from large clearings to reduce confounding multiple edge influences. We disposed one transect per forest fragment, except for three very large forest tracts (>4000 ha), where several transects could be placed on different edges (i.e. with a minimum distance of 500 m between two transects). Each transect consisted of 15 contiguous 1 m × 2-m plots, forming a 15-m long, 2-m wide sampling strip, extending from the edge outer limit (defined as the stem base of the first canopy tree with a DBH ≥10 cm) towards the forest interior.

Within each plot, total vegetation cover was estimated for each vegetation layer: tree layer (>6 m), shrub layer (2–6 m), lower shrub layer (0.5–2.0 m, herbs excluded) and herb layer (<0.5 m). Vascular plant species were counted in late Apr–Jun 2009, and their cover–abundance was scored using the following scale: i = one individual with low cover (1%), + = few individuals (2%), 1 = many individuals but cover <5%, 2a = 5–10%, 2b = 10–25%, 3 = 25–50%, 4 = 50–75% and 5 = 75–100%. The following variables were measured in the field at plot scale (plot variables): litter thickness (cm), bare soil cover (%), branch cover (%) and light intensity (LUX) 1 m above the ground using a light-meter (Digital Light Meter, Lutron® LX-105, Taipe, Taiwan). We estimated the relative light intensity in each plot i as %LUX i  = (LUX i )/((LUXextB + LUXextA)/2)*100; where LUXi is the absolute light intensity measured in plot i, LUXextB and LUXextA are the light intensities measured between 10:00 and 17:00 h in full light conditions outside the forest, immediately before the first measurement (B) and immediately after the last measurement (A) within a given transect.

At the transect level, we recorded a set of local, landscape and historical variables. We recorded transect slope (between the first and last plot of the transect using a clinometer) and aspect, and subsequently computed a south orientation index (Chabrerie & Alard 2005) using the formula: OSOUTH = 180° – Absolute Value (measured orientation – 180°). This index ranged from 0 (N-facing edge) to 180 (S-facing edge). Forest patches and other landscape elements (hedgerows, croplands and grasslands) were digitized at the 1:5000 scale from aerial photographs (2002) using a geographic information system (GIS; ArcGis®. v.8.3, ESRI, France). Scanned maps at the 1:25 000 scale (BD-Carto® database, French National Geography Institute) were used to refine mapping and georeferencing. For each forest fragment, we extracted patch area and the number of connected hedgerows. The total area (polygons: croplands, grasslands, forest and built areas) or length (linear elements: hedgerows, roads, rivers) of the different landscape elements were then computed at a radius of 1000 m around each transect. This radius was retained because it provides a good picture of landscape heterogeneity (Rescia et al. 1995) and represents a relevant scale to detect potential seed dispersal sources (Butaye et al. 2001).

To determine edge age, historical changes in forest cover were reconstructed using three sets of old maps: Cassini maps (drawn ca. 1780), French military ‘Etat Major’ maps (ca. 1880) and French National Geography Institute maps (ca. 1955), and recent aerial photographs (2002). Edge age was estimated using the median time between the map on which it was first represented and the immediately preceding map; hence, an edge could be 26 (appeared between 1955 and 2002), 87 (1880–1955), 172 (1780–1880) or 300 (<1780) yr old. As a single forest fragment can consist of a mosaic of patches with different ages, we computed an age index for each fragment i as follows:

display math

where p t is the proportion of forest area that was continuously present from time t to today, and age t is the median time between the map on which a given forest patch was first represented and the immediately preceding map. This age index was calculated within a 50-m radius around each transect to take into account the georeferencing error of old maps.

Data analyses

To reduce collinearity among landscape variables, we first performed a principal components analysis (PCA) with Varimax rotation of the landscape variables that were the most correlated with diversity indices: number of hedgerows connected to the forest fragment of the transect, total length of hedgerow, grassland area, cropland area and forest area. The first PCA axis accounted for 69.0% of the variability in landscape variables and correlated positively with the number of hedgerows connected to forest fragments (Pearson correlation: r = 0.968, P < 0.001), total length of hedgerow (= 0.981, < 0.001) and grassland area (= 0.976, < 0.001), and negatively with cropland area (= −0.760, < 0.001). This PCA axis was thus considered as reflecting the gradient of decreasing landscape management intensity, and transect scores on this first axis were subsequently used as a composite variable (LAND) to describe landscape management intensity in the vicinity of edges.

To examine the influence of edge age and neighbouring landscape management on plant community structure across forest edges we used several diversity metrics computed at the transect (= 54) and plot (= 810) scales. At the transect scale, we defined the γ-diversity as the number of species occurring in all 15 plots. Species turnover along the transect was described using the β-diversity metric proposed by Wilson & Shmida (1984): β = (l)/S m, where g and l are, respectively, the gain and loss of species along the transect, and S m is the average plot richness. The gain and loss of species were determined by counting the number of species appearing and disappearing, respectively, when one moved from the first to the last plot of a given transect. At the plot scale, we defined the α-diversity (α i ) as the number of species in plot i. We calculated these diversity indices for all species (αTot, βTot, γTot) and for species of the herb layer alone (αH, βH, γH).

The effect of landscape features (‘LAND’), edge age (‘IAGE’) and distance to forest edge (‘DIST’) on species diversity was tested using generalized linear models (GLMs) at the transect (γ and βturnover) scale and mixed models (GLMMs) at the plot (α) scale. In plot-scale models the transect was introduced as a random effect term to account for the autocorrelation between plots of a transect. We included an interaction term between LAND and IAGE in the models because the expected species accumulation over time (i.e. as an edge ages: IAGE) is likely to decrease as the permeability of the landscape matrix with respect to species decreases (i.e. as landscape management increases: LAND) (Jamoneau et al. 2011). The variables were checked for normality and box-cox transformed when necessary prior to analyses. All models were built using SPSS (version 17.0; IBM Corp., Somers, NY, USA).

The response of plant species composition in the herb layer to edge gradients and the environmental variables behind this response were analysed using a redundancy analysis (RDA) and CANOCO v. 4.5 software (Microcomputer Power, Ithaca, NY, USA), as the relationship between species cover–abundances and underlying environmental gradients was linear and the studied gradients were short. For each species, the median of the cover–abundance class was taken and log10+1-transformed. The plot x species matrix was centred and standardized prior to analysis. As plots were nested within transects, a split-plot design was applied (Lepš & Šmilauer 2003). The split-plot method is a hierarchical design that allows quantifying the spatial dimension of a phenomenon (here, the edge influence) studied by means of the whole plots (here, the transects) and the split plots (here, the plots of a transect). When testing for effects of transect-scale variables (i.e. LAND, IAGE, patch area, orientation index, slope, canopy height), the transects were freely permuted while keeping the split plots of each transect together. To test for contributions of plot-scale variables (i.e. litter thickness, bare soil cover, branch cover, light intensity, distance to forest edge, vegetation cover in herb, lower shrub, shrub and tree layers), the split plots were freely permuted within transects without permuting transects, using the ‘split plots restricted for time series or linear transect’ option in CANOCO. A forward selection method for entering variables was applied using a Monte Carlo permutation test (999 runs) for significance testing (< 0.05).


A total of 138 vascular plant species were present in the 810 plots of the 54 transects. The mean (±1SE) species richness was 20.2 ± 0.8 and 8.5 ± 0.3 per transect and per plot, respectively, with 18.1 ± 0.8, 2.7 ± 0.3, 2.8 ± 0.2 and 3.0 ± 0.2 species per transect in the herb, lower shrub, shrub and tree layer, respectively.

The species richness increased along the gradient of decreasing landscape management intensity (LAND) but decreased with patch age (IAGE) at both the transect (γ-diversity) and plot (α-diversity) scales (Table 1). The interaction term between the two variables was significant in all models. In contrast, species turnover along transects (β-diversity) was not significantly influenced by landscape management intensity or forest age. At the plot scale, the distance from the edge outer limit was the most influential factor on species richness, whatever the model considered. Along the edge–forest interior gradient αTot and αH decreased from 10.0 to 7.6 and from 7.8 to 6.0, respectively. This decrease became more and more pronounced and regular with increasing edge age, especially in the first three plots of the transects (Fig. 1). The light levels decreased along transects in the youngest edges (26-yr age class) but tended to slightly increase towards the forest interior in the oldest transects (300-yr age class). Open field and bocage landscapes exhibited slightly different patterns of species richness along transects, with more species per plot in edges for the latter landscape type compared to the former (Fig. 2). On average, light levels were more or less constant along the entire transect in edges in open fields, while they were lower and highly fluctuating in edges in bocage, where the first plots tended to receive less light than the inner ones (Fig. 2b). In the total set of transects, vegetation cover in the shrub and lower shrub layers decreased mainly in the first three plots of the transects, a pattern not observed for the tree and herb layers (results not shown).

Figure 1.

(a) Mean number of all species and (b) mean relative light intensity at the plot level as a function of plot position, according to edge age. n is number of transects per age class.

Figure 2.

(a) Mean number of all species and (b) mean relative light intensity at the plot level as a function of plot position and transect score on the first PCA axis defining landscape type.

Table 1. Effects of distance to forest edge, management intensity of adjacent landscape and edge age on species richness and turnover (a) at the transect scale (γ and β; = 54) and (b) at the plot scale (α; = 810). Transect was used as a random factor in mixed models
(a) Generalized linear models
Dependent variablesExplanatory variablesParameter estimateSE df χ2 P-value
  1. Significant P-values are shown in bold. Dependent variables: γTot: species richness in transects; γH: herb species richness in transects; βturnoverTot: species turnover in transects; βturnoverH: herb species turnover in transects; αTot: species richness in plots; αH: herb species richness in plots; Explanatory variables: DIST: distance to forest edge; LAND: composite variable, first axis of PCA performed on landscape variables defining the ‘bocage’ (hedgerow length, n. of hedgerows connected to forest fragment, grassland, crop and forest cover in a 1000-m radius around transects); IAGE: forest age index measured in a 50-m radius around transects.

γTotIntercept (n = 54)33.3164.5381.0053.90 <0.001
LAND15.0475.6671.007.05 0.008
IAGE−2.9780.9951.008.96 0.003
LAND × IAGE−3.0511.2641.005.83 0.016
γHIntercept (n = 54)29.8134.5371.0043.17 <0.001
LAND13.9385.6661.006.05 0.014
IAGE−2.6700.9951.007.21 0.007
LAND × IAGE−2.7791.2641.004.83 0.028
βturnoverTot (sqrt)Intercept (n = 54)2.5130.2771.0082.26 <0.001
LAND × IAGE−0.0070.0771.000.010.923
βturnoverHIntercept (n = 54)6.7661.9001.0012.69 <0.001
LAND × IAGE−0.1030.5291.000.040.846
(b) Mixed models
Dependent variablesExplanatory variablesParameter estimateSE df t-value P-value
αTot (sqrt)Model constant (n = 810)3.8880.32550.4911.95 <0.001
DIST−0.0230.003755.00−8.17 <0.001
LAND1.0060.40550.002.48 0.016
IAGE−0.1920.07150.00−2.70 0.009
LAND × IAGE−0.1980.09050.00−2.19 0.033
αH (box-cox)Model constant (n = 810)3.1070.30950.5310.05 <0.001
DIST−0.0150.003755.00−5.55 <0.001
LAND0.9810.38550.002.55 0.014
IAGE−0.1520.06850.00−2.25 0.029
LAND × IAGE−0.1900.08650.00−2.21 0.032

The RDA confirmed the importance of landscape management intensity and edge age on species composition (Fig. 3). Bare soil cover, branch cover, vegetation cover in the lower shrub layer, vegetation cover in the tree layer, light intensity and canopy height were not retained in the RDA after the forward selection of entering variables. The first RDA axis separated edges from old (IAGE) and/or large forest fragments (AREAFRAG), with a thick litter layer (LITTER) and facing south (OSOUTH) towards its positive part, from edges of recent and/or smaller fragments on slopes (SLOPE), which were characterized by dense herb and shrub layers (HERB, SHRUB), towards its negative part. The location of a plot within a transect (DIST) was of little importance. The oldest forest edges were associated with clonal, acidiphilous grasses (Holcus mollis, Deschampsia flexuosa) and ferns (Pteridium aquilinum), brambles (Rubus fruticosus coll.), mesotrophic vernal geophytes (Anemone nemorosa, Hyacinthoides non-scripta), tree seedlings (Fagus sylvatica, Quercus robur, Carpinus betulus) and the liana Lonicera periclymenum. In contrast, younger edges were associated with eutrophic vernal geophytes (Adoxa moschatellina, Ranunculus ficaria, Arum maculatum), ruderal species (e.g. Galium aparine, Geum urbanum, Urtica dioica, Anthriscus sylvestris, Sambucus nigra) and the creeping form of Hedera helix.

Figure 3.

Redundancy analysis of cover–abundance of all herb layer species (a) and explanatory variables collected at the transect (black arrows) and plot (grey arrows) scales (b). Empty circles are edges sampled in the least intensively managed landscapes (bocage type; PCA axis less than −0.3), while black squares indicated the most intensively managed landscapes (open field type; PCA axis less than or equal to −0.3). Species: Acer campestre (acecam); Acer platanoides (acepla); Acer pseudoplatanus (acepse); Adoxa moschatellina (adomos); Agrostis stolonifera (agrsto); Ajuga reptans (ajurep); Alliaria petiolata (allpet); Allium ursinum (allurs); Anemone nemorosa (anenem); Angelica sylvestris (angsyl); Anthriscus sylvestris (antsyl); Arum maculatum (arumac); Betula pendula (betpen); Brachypodium sylvaticum (brasyl); Bromus sterilis (broste); Carex sp. (caresp); Carex pendula (carpen); Carex sylvatica (carsyl); Carpinus betulus (carbet); Cardamine pratensis (carpra); Castanea sativa (cassat); Centaurea sp. (centsp); Clematis vitalba (clevit); Circaea lutetiana (cirlut); Cirsium oleraceum (cirole); Cirsium vulgare (cirvul); Cornus mas (cormas); Corylus avellana (corave); Crataegus laevigata (cralae); Crataegus monogyna (cramon); Dactylis glomerata (dacglo); Deschampsia cespitosa (desces); Deschampsia flexuosa (desfle); Dryopteris carthusiana (drycar); Dryopteris filix-mas (dryfil); Elymus caninus (elyrep); Epipactis helleborine (epihel); Epipactis sp. (epipsp); Equisetum telmateia (equtel); Euonymus europaeus (evoeur); Euphorbia amygdaloides (eupamy); Euphorbia cyparissias (eupcyp); Fagus sylvatica (fagsyl); Filipendula ulmaria (filpen); Fragaria vesca (fraves); Fraxinus excelsior (fraexc); Galeopsis tetrahit (galtet); Galium aparine (galapa); Galium odoratum (galodo); Geranium robertianum (gerrob); Geum urbanum (geuurb); Glechoma hederacea (glehed); Hedera helix (hedhel); Heracleum sphondylium (hersph); Holcus mollis (holmol); Hyacinthoides non-scripta (hyacin); Hypericum perforatum (hypper); Hypericum pulchrum (hyppul); Ilex aquifolium (ileaqu); Laburnum anagyroides (labana); Lamium galeobdolon (lamgal); Lapsana communis (lapcom); Ligustrum vulgare (ligvul); Listera ovata (lisova); Lysimachia vulgaris (lysvul); Lysimachia nummularia (lysnum); Lonicera periclymenum (lonper); Luzula forsteri (luzfor); Luzula pilosa (luzpil); Melampyrum pratense (melpra); Melica uniflora (meluni); Mercurialis perennis (merper); Milium effusum (mileff); Moehringia trinervia (moetri); Narcissus pseudonarcissus (narpse); Orchidacaeae sp. (orchsp); Orchis purpureum (orcpur); Ornithogalum umbellatum (ornumb); Oxalis acetosella (oxaace); Phalaris arundinacea (phaaru); Plantago lanceolata (plalan); Poa annua (poaann); Poa nemoralis (poanem); Poa trivialis (poatri); Polygonatum multiflorum (polmul); PoPulus nigra (popnig); Potentilla sterilis (potste); Primula elatior (priela); Primula veris (priver); Prunus avium (pruavi); Prunus spinosa (pruspi); Pteridium aquilinum (pteaqu); Quercus robur (querob); Ranunculus acris (ranacr); Ranunculus auricomus (ranaur); Ranunculus ficaria (ranfic); Ranunculus repens (ranrep); Ribes rubrum (ribrub); Ribes uva-crispa (ribuva); Rosa arvensis (rosarv); Rosa canina (roscan); Rubus idaeus (rubida); Rubus fruticosus coll. (rubusp); Rumex acetosa (rumace); Rumex sanguineus (rumsan); Salix caprea (salcap); Sambucus nigra (samnig); Sanicula europaea (saneur); Senecio ovatus (senova); Sonchus asper (sonasp); Sorbus aucuparia (sorauc); Stachys sylvatica (stasyl); Stellaria holostea (stehol); Stellaria media (stemed); Tamus communis (tamcom); Taraxacum sp. (tarax); Teucrium scorodonia (teusco); Tilia plathyphyllos (tilpla); Torilis japonica (torjap); Triticum sp. (tritsp); Ulmus glabra (ulmgla); Ulmus minor (ulmmin); Urtica dioica (urtdio); Veronica chamaedrys (vercha); Veronica hederifolia (verhed); Veronica montana (vermon); Viburnum lantana (viblan); Viburnum opulus (vibopu); Vicia sepium (vicsep); Vinca minor (vinmin); Viola reichenbachiana (viorei). Environmental variables: soil litter thickness (LITTER); vegetation cover in herb layer (HERB); vegetation cover in shrub layer (SHRUB); distance to forest border (DIST); slope (SLOPE); orientation index (OSOUTH); area of the forest fragment of the transect (AREAFRAG); landscape gradient from PCA axis 1 (LAND); age index of the forest (IAGE).

The second axis separated edges according to the surrounding landscape management intensity (LAND), with edges neighbouring intensively cultivated open fields having the lowest scores. These were associated with bird-dispersed, calciphilous shrub species (e.g. Ligustrum vulgare, Prunus spinosa, Viburnum lantana), nitrogen-demanding therophytes (Galium aparine, Alliaria petiolata) and geophytes (Allium ursinum, Arum maculatum, Ranunculus ficaria), the calciphilous grass Melica uniflora and the creeping forb Vinca minor. Edges from bocage landscapes were associated with species characterizing moist shaded areas (e.g. Deschampsia cespitosa, Veronica montana, Ranunculus auricomus, Circaea lutetiana, Cardamine pratensis) and true forest species (e.g. Lamium galeobdolon, Viola reichenbachiana, Oxalis acetosella, Polygonatum multiflorum, Carex sylvatica).


Response of edge communities to aging

We found a negative influence of edge age on species richness at both the plot (α) and transect (γ) scales, suggesting that few species are able to survive the environmental conditions that progressively develop as the edge ages, especially severe shading and leaf litter deposition (see Figs 1 and 2). This contrasts with the species accumulation predicted by the species–time relationship (STR; Rosenzweig 1995) and observed in the early stages of forest succession (Brunet 2007). The distance from the outer limit of the edge is the most important factor explaining species richness, with a clear gradient of decreasing richness from the edge to the forest interior, consistent with the well-known ‘edge effect’ (Saunders et al. 1991; Murcia 1995). As expected, this gradient becomes more pronounced as time since edge formation increases (Matlack 1993; Hermy et al. 1999; Honnay et al. 2002a; Brunet 2007). The strongest decrease in species richness was observed within the first 3 m. Similar or even lower values are reported in the literature (Brothers & Spingarn 1992; Honnay et al. 2002a); moreover, the few penetrating species have sharply decreasing cover values when expanding towards the forest interior (Honnay et al. 2002a).

Interestingly, the reverse trend was found for relative light intensity, with the outer plots receiving more light than the inner ones in recent edges (26 yr), and the inner plots receiving more light than the outer ones in ancient edges (300 yr; Fig. 1a). These light patterns contrast with those usually expected (e.g. Hermy et al. 1999; Brunet 2007), but are consistent with the competition-induced wave of biomass (Reichman et al. 1993): the high light availability at the beginning of the transect creates a peak of leaf density that in turn casts a shadow just behind it before light availability increases to intermediate values further behind. As a result, edges are likely to become less permeable to species coming from outside, which are predominantly light-demanding species, an effect already reported for ancient forest edges (Henry & Aarssen 1997; Honnay et al. 2002a; Marchand & Houle 2006); hence the penetration distance is expected to decrease as edge age increases (Meiners & Pickett 1999).

The increased litter thickness of aging edges represents a supplementary physical and chemical (low pH) barrier to many non-forest species. It is expected to promote a limited set of specialized species (Sydes & Grime 1981; Decocq & Hermy 2003), as well as tall competitors that use vegetative propagation, such as Pteridium aquilinum and Rubus fruticosus coll. (both associated with the oldest forests in the RDA), which may accelerated the loss of smaller species as the edge ages.

The RDA confirms that edge age is the most influential factor on species composition, with a clear shift along the first axis, which is associated with increasing litter thickness and decreasing shrub and herb layer cover: light- and/or nitrogen-demanding species (e.g. Geum urbanum, Poa trivialis, Galium aparine) are progressively replaced by shade-tolerant, mesotrophic species (e.g. Poa nemoralis, Hyacinthoides non-scripta, Anemone nemorosa). Remarkably, this species shift equally affects the plots of a given transect, since edge age did not affect species turnover along the transect (β-diversity), and plot position was marginally significant in the RDA. No niche partitioning effect was thus evidenced, contrary to our expectations and the theoretical models of edge structure (Matlack 1994; Mikk & Mander 1995; Murcia 1995; but see Alignier & Deconchat 2011). Instead, our results suggest a specialization of the flora, which primarily benefits to the so-called ‘ancient forest species’ (Peterken & Game 1984; Hermy et al. 1999). The latter are indeed known for not being suppressed by edge effects (Fraver 1994; Honnay et al. 2002a) and for accumulating over time as, due to their dispersal and recruitment limitations, they are poor colonizers (Verheyen & Hermy 2001). The positive correlation between edge age and patch area (see Fig. 3) suggests that larger forest patches may keep their edges supplied with more diaspores of ancient forest species compared to younger and smaller patches.

We conclude that an edge–forest interior environmental gradient progressively develops as an edge ages (H1); however it poorly translates into species richness and not at all into species composition (no age effect on β-diversity), consistent with the idea that the depth of influence of species richness differs from the depth of influence of microclimatic variables (Gehlhausen et al. 2000; Alignier & Deconchat 2011).

Response of edge communities to adjacent land management

Our findings highlight the key impact of matrix management on species richness and composition of forest edges. As management intensity (quantified by the variable LAND in this study) increased, species richness decreased at both the plot (α) and transect (γ) scales without influencing species turnover (β). At the same time, communities shifted from mesophilous, calciphilous (in edges of open fields) to more hygrophilous, acidiphilous (in edges in bocage) assemblages (see RDA axis 2). These differences in species composition may be partly explained by the fact that land use and management decisions depend on environmental conditions, especially soil quality. In our study area, substrates that were nutrient-poor and/or waterlogged were not suitable for intensive agriculture and thus were more prone to be patterned as ‘bocage-like’ landscapes. Hence, forest edge communities are expected to be more acidiphilous and/or hygrophilous in bocages than in open fields. However, three types of factor associated with agricultural practices can also explain the observed patterns.

First, mechanical disturbances are more common in open field landscapes. Ploughing and mowing operations conducted in croplands often expand to the field margins, including herbaceous fringes at the field–forest boundary, and may directly reduce species richness. Moreover, forest edges are maintained by thinning so that the outer branches of trees and shrubs bordering cultivated fields do not overhang the cropland, to minimize crop shading and allow agriculture vehicles to access the field outer limit. This is well reflected by the high relative light intensity in the first two plots; as a result, a few shrub and lower shrub species (e.g. Crataegus monogyna, Ulmus minor, Rubus fruticosus coll.) often form a dense curtain which shades out many other species (Honnay et al. 2002a; Alignier & Deconchat 2011), explaining the strong decrease in species richness in the first four plots. In contrast, in bocage landscapes forest patches are usually adjacent to pastured meadows and, due to grazing pressure, forest edges are typically cantilevered edges, characterized by an overhanging canopy of branches that grow above the adjacent meadow. Such a structure acts as a shelter that shades the edge understorey, buffering it from desiccation and light penetration (Murcia 1995), as reflected by the reduced relative light intensity in the first two plots of the transect. Highly competitive species such as Rubus fruticosus coll. and Urtica dioica may be hampered from becoming dominant, hence space and resources are likely available for smaller-sized, less light-demanding species (Endels et al. 2004), potentially explaining why species richness peaked 4 m inside the forest.

Second, agrochemicals are much more heavily used in open fields compared to bocage landscapes (Marshall & Moonen 2002). In open fields fertilizer input from neighbouring arable land is likely to favour nitrophilous species along edges (e.g. Alliaria petiolata, Ulmus minor), to lead to the dominance of competitive-ruderal species (e.g. Rubus fruticosus coll., Urtica dioica), and hence to reduce species diversity (Marrs 1993; Kleijn & Snoeijing 1997; Honnay et al. 2002a). Moreover, nutrient cycling processes may be affected in the first plots through increased solar radiation, which in turn increases the activity of soil microorganisms and invertebrates (Klein 1989; Parker 1989) and thus accelerates litter decomposition and nutrient release. Similarly, lime drift is likely to increase soil pH along forest edges, promoting calciphilous species such as e.g. Viburnum lantana, Orchis purpurea and Melica uniflora (Honnay et al. 2002a). The drift of other agrochemicals, particularly herbicides, may directly contribute to reduce species diversity (Kleijn & Snoeijing 1997).

Third, the composition of the landscape matrix can have a strong influence on the seed rain entering edges, as most species found in small forest patches are non-forest species (Jamoneau et al. 2011). As bocage landscapes consist of a mosaic of grasslands, small fields and hedgerows, both density and diversity of diaspore sources are expected to be much higher than for open fields. Moreover, hedgerows that connect forest patches may act as ecological corridors, allowing plant species and their animal vectors to migrate along them (Corbit et al. 1999; Damschen et al. 2006); the movements of these vectors (e.g. birds, rodents, ants) are also likely to be facilitated by the lighter use of biocides (Marshall & Moonen 2002; Murphy & Lovett-Doust 2004). This increasing permeability of the landscape matrix with decreasing management intensity is likely to increase propagule pressure on forest edges (Marshall & Moonen 2002; Jamoneau et al. 2011, 2012), and may explain both the higher species richness and the presence of typically weak dispersers (e.g. Viola reichenbachiana, Lamium galeobdolon) in forest edges of bocages compared to open fields.

It should be noted that landscape type and edge age are not fully independent in our study, as indicated by the significant interaction term in ANOVAs and collinearities among explanatory variables in the RDA. This was expected as forest fragments have been reported to be larger and older in open field landscapes than in bocages. Similarly, the negative correlation between slope and age reflects the fact that most slopes became afforested after the agriculture mechanization of the 1960s, when vehicles could no longer drive where animals were once used (Jamoneau et al. 2010, 2011, 2012).

We conclude that the edge–forest interior transition zone tends to be shortened as the management intensity of adjacent lands increases, with more abrupt changes in environmental conditions, species richness and floristic composition (H2); hence forest edges resemble more a sharp transition zone (i.e. an ecotone sensu Van der Maarel 1990) as proximal disturbances increase in intensity.


From our results, we propose that two interacting, opposite forces pattern edge communities of forest patches embedded in agricultural landscapes, namely local maturation and proximal disturbances. A gradual transition zone in environmental conditions clearly develops as the edge ages, which is associated with a more and more pronounced decrease in species richness along the edge–interior gradient. However, these environmental gradients poorly translate into species assemblages along the edge–forest interior gradient and no niche partitioning can be evidenced, reflecting the constraint imposed by neighbouring land management (Alignier & Deconchat 2011). Disturbance associated with landscape management indeed sharpens edges by reducing species richness and changing community composition over the entire gradient. Older and/or weaker disturbed edges are mainly characterized by a restricted set of ancient forest species able to cope with both shade and a thick litter layer, contributing to the conservation of threatened forest specialists in agricultural landscapes. In contrast, younger and/or strongly disturbed edges host ruderal, generalist species originating from open habitats and thus have little conservation value. We thus recommend putting most conservation effort into the oldest forest edges in contemporaneous agricultural landscapes, by limiting mechanical and chemical disturbances in adjacent farmlands.


We thank Jonathan Lenoir for assistance with statistical analyses; Céline Boissel and Robert Saguez for help during fieldwork; the two anonymous referees for helpful comments on an earlier version of the manuscript. This study was part of the METAFOR research project supported by the ‘Conseil Régional de Picardie’. This study was conducted while A.J. held a PhD fellowship from the ‘Conseil Régional de Picardie’.