Modelling edge effects of mature forest plantations on peatland waders informs landscape-scale conservation

Authors


Summary

  1. Edge effects of native forest fragmentation have been well studied, but there are few studies of open-ground habitats fragmented by plantation forests. We measure forestry edge effects on open-ground breeding birds, following one of Europe's biggest and most controversial land-use transformations.
  2. The ‘Flow Country’ of northern Scotland is one of the world's greatest expanses of blanket bog. It became fragmented by conifer forests planted in the late 20th century, and these now adjoin open peatlands protected under European conservation legislation. Detrimental edge effects on breeding birds were anticipated, but not apparent shortly after planting.
  3. Using survey data collected in 2003–2006, and logistic regression modelling, we tested whether breeding distributions of three wader species of international conservation concern, dunlin, European golden plover and common greenshank, were influenced by distance to forest edge, controlling for habitat and topography.
  4. All three species were more likely to occupy flatter, more exposed ground close to bog pools and were influenced by peatland vegetation structure. There was an additive and adverse effect of proximity to forest edge for dunlin and European golden plover, but not common greenshank. This effect was strongest within 700 m of forest edges. We used these results to predict which areas should benefit most from removal of adjacent forestry and so guide maintenance and restoration of the bird interests of the protected areas.
  5. Synthesis and applications. Edge effects of mature forestry on dunlin and golden plover are apparent over several hundred metres and are now being used to guide forest planning in northern Scotland. The scale of edge effect is broadly consistent with other avian studies in open-ground habitats across Eurasia and North America, so buffer zones of this order are consistent with possible impacts of plantation forestry on open-ground habitats of bird conservation interest. Given renewed interest in conifer afforestation as a climate change mitigation measure, an improved understanding of edge effects and the mechanisms through which they operate is vital to managing plantation forestry in ways that maintain open-ground landscapes of high conservation value.

Introduction

Understanding how habitat edges influence distribution, abundance and behaviour of wildlife is critical to large-scale conservation management (Murcia 1995; Ries et al. 2004; Fuller 2012). The pivotal role of predation and reproductive failure in bird demography (Martin 1995) underlies strong interest in edge effects in avian conservation. Reviews show that effects are variable and context-specific (Batary & Baldi 2004) because land-use composition and history determine the nest predator fauna, the hard or soft nature of habitat edges, and whether some species are adapted to young-growth habitat at ecotones between forest and open habitats (Stephens et al. 2003; Thompson 2007; Dolman 2012; Fuller 2012).

Most evidence of edge effects comes from studies of forest habitats fragmented by agricultural land uses. These have found reduced occupancy and nest success close to edges in highly fragmented forest landscapes in both tropical and temperate systems across a broad spectrum, including gamebirds (e.g. capercaillie Tetrao urogallus – Kurki et al. 2000), tree-nesting seabirds (e.g. marbled murrelet Brachyramphus marmoratus – Malt & Lank 2007) and songbirds (Deng & Gao 2005; Poulin & Villard 2011; Zurita et al. 2012). These effects are usually driven by nest predators (e.g. corvids and foxes) and brood parasites (e.g. brown-headed cowbird Molothrus ater) which persist at higher abundances in fragmented landscapes and may show behavioural associations with habitat edges (Andren 1995; Howell, Dijak & Thompson 2007; Cervinka et al. 2011). Conversely, edge effects of plantation forestry on bird populations in open habitats of high conservation interest remain little studied (Reino et al. 2009). This is despite plantation forestry now occupying 7% of global forest cover (140 million hectares), and increasing rates of afforestation or reforestation, driven by production, climate change adaptation, ecosystem service protection and biodiversity conservation goals (Bauhaus, van der Meer & Kanninen 2010).

In the UK, increasing forest cover was established policy through the 20th century, responding to an historical reduction in woodland cover and lack of a strategic timber reserve (Foot 2003; Oosthoek 2013). Available land has often been at higher latitudes and altitudes where land values were lower, but landscape and conservation interests greater (Avery & Leslie 1990). The ‘Flow Country’ of northern Scotland is a globally important example. It is one of the world's most extensive, contiguous areas of blanket bog, a landscape dominated by deep peat, and vegetation cover rich in peat-forming Sphagnum mosses, along with grasses, sedges and ericaceous dwarf shrubs. This landscape is now fragmented by plantations of non-native lodgepole pine Pinus contorta and Sitka spruce Picea sitchensis, in areas that had been treeless for millennia (Stroud et al. 1987). This was one of the most controversial European land-use transformations of the late 20th century (Avery & Leslie 1990; Oosthoek 2013). Much remaining open peatland (146 000 ha) is now designated as special protection areas (SPAs) and special areas of conservation (SACs) under European Union (EU) Birds and Habitats Directives. This recognizes a globally unique assemblage of habitats and breeding birds that is unusually diverse amongst northern temperate peatland systems. This included 66% of common greenshank Tringa nebularia (Gunnerus, 1767), 35% of dunlin Calidris alpina (L.) and 17% of European golden plover Pluvialis apricaria (L.) in the European Community at the time of designation (Stroud et al. 1987; Lindsay et al. 1988).

Forest planting resulted in an estimated loss of 17–19% of the Flow Country populations of each of these species (Stroud et al. 1987; Lindsay et al. 1988), but with little initial evidence that forest edge effects were causing further losses (Avery 1989; Stroud, Reed & Harding 1990). However, such effects were predicted as plantations grew and became more suitable habitat for generalist predators such as hooded crows Corvus cornix and red foxes Vulpes vulpes which hunt over surrounding areas (Stroud et al. 1987; Lindsay et al. 1988). Since then, as forestry has matured, there have been declines of European golden plover and dunlin in some areas (e.g. Sim et al. 2005) and weak evidence of edge effects from studies at coarse spatial grain (Hancock, Grant & Wilson 2009).

We used annual bird survey data from the Forsinard Flows Nature Reserve to measure associations between habitat, land cover and topography, and occurrence of dunlin, European golden plover and greenshank in the Caithness and Sutherland Peatland SPA (Fig. 1) in order to test whether there is now stronger evidence of edge effects. We controlled for both habitat (vegetation and soil reflectance and proximity to bog pool systems) and topographical variation (slope and elevation) because these are known to influence distribution of breeding waders on peatlands (Lavers & Haines-Young 1996; Lavers, Haines-Young & Avery 1996), and more widely across upland landscapes (Whittingham, Percival & Brown 2002; Pearce-Higgins & Grant 2006). In this sense, we followed the recommendations of Burnham and Anderson (2002) in defining a biologically informed model set as our starting point. Where an additive association between distance to forest edge and grid square occupancy was found, we then used a simple scenario of forest removal to predict areas where benefit of forestry removal through mitigation of edge effects is likely to be greatest. This can then be used to help determine whether forestry areas should be replanted or restored as blanket bog, adding to other considerations such as local timber and wood fuel markets, and Government policies on woodland expansion and greenhouse gas (GHG) consequences of growing trees on deep peat soils (e.g. Reynolds 2007; Morison et al. 2010; Lohila et al. 2011).

Figure 1.

Grid squares (200 m) making up study area (dots). Map covers 49 × 33 km and is centred on Forsinard Flows Nature Reserve (58°21′N 3°53′W). Forestry cover (1998) shaded pale grey. Caithness and Sutherland Peatland Special Protection Area (SPA) shaded dark grey. Specific locations are: 1 = Blar nam Faoileag, 2 = Cross Lochs, 3 = Imriche, 4 = Sletill, 5 = Talaheel.

Materials and methods

Breeding Bird Surveys

Survey data were based on two visits to over 250 km2 of peatland (6288 grid squares of 200 × 200 m; Fig. 1) in one (24% of squares), two (2%), three (36%) or four (38%) years between 2003 and 2006. Territorial individuals of all bird species were mapped by walking transects that covered the whole square to within 100 m (Brown & Shepherd 1993). These data resolved a species as either present (1) or absent (0) in each square, pooled over the four survey years. This measure was the response variable in subsequent occupancy modelling.

Explanatory Variables – Habitat

Peatland vegetation cover was first characterized from four collinear reflectance bands (Landsat 7; March 2000) covering the visible red to far infra-red range and with little effect from atmospheric interference (Avery & Haines-Young 1990; Campbell 2002). The value for each square was derived as the mean of all 30-m pixels within or overlapping the square. The first principal component of these bands explained 86% of variation and was used as a measure of vegetation cover. We also included the Normalised Difference Vegetation Index (NDVI), values of which range from −1 to +1, as a measure of density of photosynthetically active vegetation (Campbell 2002).

Explanatory Variables – Topography

We measured distance from each grid cell centre to the nearest bog pool and nearest point of forestry cover (see Appendix S1, Supporting information), using Ordnance Survey digital 1 : 10 000 maps and the National Inventory of Woodland and Trees (NIWT), respectively. We used the UK national digital terrain model (DTM) to measure mean slope, altitude and topographical exposure of each grid square, derived as the mean of all points on a 10-m grid within each 200-m square. Topographical exposure was defined as the vertical angle to the horizon summed across the eight prime compass points (Quine & White 1993).

Finally, we multiplied the number of visits by the proportion of squares surveyed to create a nuisance variable to control for variation in survey effort and the fact that some squares at forestry, lake or study area edges could only be partially surveyed. Descriptive statistics for all explanatory variables are summarized in Table 1.

Table 1. Descriptive statistics for explanatory variables measured in each 200-m square in the Flow Country study area
VariableAbbreviation (Table 2)Units, range and mean
First principal component of reflectance bands b3, b4, b5 and b7 – an index of vegetation coverRPrincipal component(−8·92 to 8·77; mean = 0)
Photosynthetic intensity (NDVI) = (b4−b3)/(b3 + b4)nIndex (−0·53 to +0·53; mean = 0·15)
SlopesDegrees (0 to 32°, mean = 2°)
Elevationekm (0·043 to 0·525, mean = 0·167)
Topographical exposure (inverse)tDegrees (1·3 to 94·9°, mean = 11·2)
Loge distance to forest edgefMetres (0 to 5684, mean = 1100)
Loge distance to pool systempMetres (0 to 2202, mean = 421)
Effort (Number of years surveyed × proportion of square surveyed)

Data Analysis

Preliminary analyses

Collinearity of explanatory variables was low. Variance inflation factors ranged from 1·08 to 2·70, and Pearson correlation coefficients for all pairs of explanatory variables were <0·7, except for that between slope and topographical exposure. For each species, we therefore conducted preliminary univariate analyses including either slope or topographical exposure (correcting for survey effort) in order to select one of these two variables (Wald chi-square tests). The variables selected were topographical exposure (dunlin, European golden plover) and slope (common greenshank).

In squares where a species was detected, we tested whether probability of detection on both (1) or just one survey visit (0) in a year was related to any explanatory variable, using univariate logistic regression models. In total, we fit 77 models (three species × four years × seven explanatory variables; too few common greenshanks were recorded in one year to allow models to be fit). Only five models showed a significant relationship at < 0·05, and none at < 0·01 (Bonferroni-adjusted = 0·0006), suggesting that modelling correlates of occupancy is unlikely to be biased by associations between explanatory variables and bird detectability.

Occupancy modelling

We used generalized linear models (proc genmod; SAS 9.2, SAS Institute Inc., Cary, NC, USA), specifying a binomial error distribution. The response variable was presence (1) or absence (0) of a species in each 200-m square. For each species, all combinations of the six explanatory variables were fitted, also including the effort variable in all cases, to give a candidate set of 63 models. Models were compared using an information theoretic approach, with lower values of Akaike's information criterion (AIC) indicating better models, and we considered any model with an Akaike weight (wi) of at least 10% of that of the best approximating (lowest AIC) model to be plausible (Burnham & Anderson 2002). Where several models were plausible, we used multimodel inference, with parameters estimated as the mean of those models, weighted by the Akaike weight of each model (Burnham & Anderson 2002). The weight of evidence for each explanatory variable over all models was estimated by summing Akaike weights of all models including that variable.

The data come from an array of grid squares; therefore, inference may be influenced by spatial covariance in the data, but formal modelling of spatial structure in model residuals risks under-emphasizing the importance of explanatory variables that are themselves spatially autocorrelated (Diniz-Filho, Bini & Hawkins 2003). We therefore calculated Moran's I statistics for the residuals of these final models to quantify any residual spatial covariance that may affect model interpretation.

Assessing goodness-of-fit

The SAS LOGISTIC procedure was used to calculate the concordance index (c) for the best approximating model for each species. This index indicates the probability that the fitted values from the model will place all pairs of squares (one occupied, the other unoccupied in each pair) in the correct rank order of likelihood of predicted occupancy. Thus, chance performance is clearly defined (c = 0·5) and perfect discrimination is indicated by c = 1 (Vaughan & Ormerod 2005). Model calibration (accuracy of predicted probabilities) for best approximating models was tested by logistic regression of observed presences and absences in grid squares against the logit-transformed probabilities of occurrence from the model (Pearce & Ferrier 2000). Model fits are well calibrated if intercepts are not significantly different from zero, and slopes not significantly different from one.

Predicting Benefits of Forest Removal

For any species showing an additive effect of distance to forest edge, we calculated fitted probabilities of occupancy first from model-averaged parameter estimates (Pocc), and then having set distance to forest for each square to the maximum observed value of 5·7 km (Pmax). Values of PmaxPocc were then mapped to show those parts of the study area where probability of occupancy is predicted to increase most for the species concerned. Finally, we compared these values between squares that had experienced an increase in distance to forest between 1998 and 2003 as a result of early forest removals (see Appendix S1, Supporting information) and squares whose distance to forest was unaffected, using a t-test and assuming unequal variances.

Results

Dunlin were recorded in 8·1%, European golden plover in 9·6% and common greenshank in 3·7% of squares over the four years of bird surveys.

For dunlin (wi = 0·666) and European golden plover (wi = 0·804), the global model was 3·6 and 4·3 times better supported, respectively, than any other. For common greenshank, there was no clear best model, with ten having at least 10% of the maximum wi of 0·186 (Table 2). Model-averaged parameter estimates and variable weights (Table 3) show that all three species were strongly associated with flat ground and pool systems, but showed weaker associations with elevation and photosynthetic intensity of vegetation (positive), and vegetation cover (negative). In addition, there was a strong positive association with distance to forest edge for dunlin and European golden plover, but not common greenshank, indicating avoidance of locations closer to forest edges for the first two species. The best approximating models for each species were well-fitting. Concordance indices were 0·817 (dunlin), 0·737 (European golden plover) and 0·813 (common greenshank), indicating ‘very good’ (dunlin, greenshank) and ‘good’ (golden plover) discriminatory ability (Vaughan & Ormerod 2005), especially given low prevalence (Liu et al. 2005). The best approximating models were also well calibrated; neither the slope nor the intercept of the relationship between observed presences and absences (y) and logit-transformed predicted occupancy likelihoods were significantly different from 1 or zero, respectively (P-values from 0·27 to 1 for Wald chi-square test of slope and intercept for all three species). Moran's I values were 0·136 (dunlin), 0·106 (golden plover) and 0·030 (greenshank) over lag distances of one grid square, declining to 0·009, 0·015 and 0·002, respectively, over lag distances of 10 grid squares. All these values were statistically significant due to the very large sample sizes, but values of the order of 0·1 or below suggest that spatial autocorrelation effects in model residuals are so small as not to compromise inference (Ryan et al. 2004; Kraan et al. 2009).

Table 2. Measures of performance of all plausible models (those within 10% of best model, for which ∆AIC = 0). Abbreviations in first column from Table 1. ‘All’ = global model containing all explanatory variables. Sixty-three models were considered for each species
Variables in modelAIC∆AICAkaike weight
Dunlin
All2904·200·666
r,n,t,f,p2906·82·60·186
European golden plover
All3568·600·804
n,e,t,f,p3571·62·90·186
Common greenshank
n,e,s,p1656·700·186
n,s,p1656·80·10·178
r,n,e,s,p1657·71·00·112
r,n,s,p1657·71·10·110
n,e,s,f,p1658·51·80·076
n,s,f,p1658·61·90·073
All1659·42·70·049
r,n,s,f,p1659·42·70·048
e,s,p1660·03·40·035
s,p1660·94·20·022
Table 3. Model-averaged parameter estimates (β) and standard errors (SE) for each explanatory variable. The last column gives the probability that each variable is present in the best model (pbest)
VariableDunlinEuropean golden ploverCommon greenshank
βSE p best βSE p best βSE p best
Intercept−4·8870·580 −6·4950·516 −2·9770·551 
Effort0·5040·05910·5320·05110·5130·0861
Reflectance−0·0950·0370·893−0·0710·0320·808−0·0420·0490·379
NDVI2·2260·8430·9162·6440·7980·9902·1710·9260·881
Elevation2·3241·0590·7935·1710·91212·5221·7430·509
Exposure−0·1570·0161−0·0850·0121
Slope−0·2330·0910·946
Forest distance0·5080·05510·4480·04910·0320·0650·295
Pool distance−0·3420·0351−0·1540·0341−0·4460·0461

Based on model-averaged parameter estimates, Fig. 2 shows mean fitted probability of occupation of 200-m squares (Pocc) by dunlin and European golden plover in bands of increasing distance to forest edge. Incremental changes in distance to forest edge are highly influential close to forest edges, but at greater distances from forested areas, they have less impact that is additive to the combined influence of other variables. The peak in predicted probability of occurrence in the most distant band arises because this 10% of squares is concentrated in one area of high-quality blanket bog habitat and is one of very few large contiguous areas far from forestry (Blar nam Faoileag – area 1, Fig. 1). In other distance bands, grid squares were widely distributed across the study area. Variants on the final models in which distance to forest edge is reduced to a binary variable (‘close’ vs. ‘far’), with the threshold set at 100 m increments from 100 m to 1500 m, all have much poorer goodness-of-fit (ΔAIC ≥ 16·2 for dunlin, and ≥23·1 for European golden plover) than the continuous effect. However, the minimum ΔAIC summed over both species is achieved when the threshold is set at 700 m. This may be a useful ‘rule-of-thumb’ minimum distance to set forest edges from areas of otherwise high-quality habitat for these species in the Flow Country.

Figure 2.

Mean predicted probability of occupation of 200-m squares by dunlin (filled bars) and European golden plover (open bars) (Pocc) in increasing distance bands from forest edge. Each distance band encompasses 10% of all squares. Predictions assume survey visits in all 4 years to remove effects of variation in survey effort.

Figure 3 plots values of PmaxPocc to illustrate where forest removal is predicted to result in greatest increases in probability of occupancy by dunlin and European golden plover. Squares for which distance to forest increased between 1998 and 2003 following early forest removals showed greater predicted increases in probability of occupancy than others (dunlin 0·14 ± 0·10SE vs. 0·08 ± 0·08, t = 22·3, < 0·0001; European golden plover 0·14 ± 0·08 vs. 0·08 ± 0·06, t = 26·8, < 0·0001).

Figure 3.

Predicted absolute increase in probability of occupancy of 200-m grid squares by (a) dunlin, and (b) European golden plover if distance to forest edge is set for all squares to the maximum value in the data set (5·7 km). Colour scales continuously from minimum value of zero at blue end of spectrum to maximum of 0·53 (dunlin) and 0·46 (European golden plover) at red end of spectrum. Forestry cover (1998) shaded pale grey. Boundary of SPA shown as black line.

Discussion

Twenty-five years after breeding habitat associations of peatland waders in the Flow Country were first studied (Stroud et al. 1987; Stroud, Reed & Harding 1990; Lindsay et al. 1988; Lavers & Haines-Young 1996; Lavers, Haines-Young & Avery 1996); this study found similar associations with topography, vegetation cover and proximity to pool systems. However, with maturation of plantation forests, there is now an additive effect of distance to forest edge for dunlin and European golden plover, with reduced occupancy within several hundred metres from forest edges.

We cannot be certain that reduced habitat occupation close to mature forests is a demographic or behavioural response to perceived or real predation risk. However, nest predators such as hooded crows, red foxes and pine martens Martes martes occupy plantation conifer forests across Scotland (Avery & Leslie 1990; Chadwick, Hodge & Ratcliffe 1997; Ratcliffe 2007). Studies elsewhere have shown increased nest predation on breeding waders close to forest edges, and a preference for nesting on flat ground may be associated with the need for clear views of approaching predators (e.g. Valkama, Currie & Korpimaki 1999; Whittingham, Percival & Brown 2002; Finney, Pearce-Higgins & Yalden 2005). Equally, the lack of edge effect on common greenshank habitat occupancy is consistent with the fact that this species (like other Tringa species, but unlike dunlin and European golden plover) vigorously mobs predators and occupies bog habitats within native forests (Nethersole-Thompson & Nethersole-Thompson 1979). However, explanations based on forest influences on bog habitat condition (e.g. through alteration of hydrology, grazing patterns, disease or fire risk) also need to be considered. For example, hydrological and habitat effects of forestry on adjacent blanket bog in the Flow Country extend up to 40 m from plantation boundaries, at least in the first few decades (Shotbolt, Anderson & Townend 1998), and could affect prey availability for breeding birds. Forestry also holds high densities of ticks Ixodes ricinus relative to blanket bog (Gilbert 2013), and although the form of the relationship between tick densities and distance to forest edges remains unknown, ticks can infest chicks of moorland-breeding waders (Newborn et al. 2009).

Conservation and Management Implications and Wider Policy Relevance

Edge effects of forests on occupancy or nest success of open-ground breeding birds are detected consistently over distances of up to hundreds of metres (and occasionally beyond) across a wide range of upland grassland, lowland farmland and shrubland habitats in Eurasia and North America (Table 4). This suggests that landscape planning should consider buffer zones hundreds of metres wide in addition to direct habitat loss when assessing impacts of plantation forestry on open-ground habitats of bird conservation interest, or when prioritizing plantations for removal and restoration of open-ground habitat. Nonetheless, such consideration should be cautious and species-dependent. For example, this study found evidence of edge effects for two breeding wader species, but not a third, whilst evidence of absence of forest edge effects on habitat occupancy by corncrakes Crex crex (Berg & Hiron 2012) may be explained by this species' exceptionally low susceptibility to nest predation (Green et al. 1997). Equally, some species, amongst which black grouse Tetrao tetrix and common cuckoo Cuculus canorus are European examples, may benefit directly from landscapes comprising both forest and open-ground or the specific resources offered by ecotone habitats (e.g. young-growth) at edges (Fuller 2012; Grant & Pearce-Higgins 2012). In such cases, the mix of habitats might be described positively as a ‘mosaic’, rather than negatively as a ‘fragmentation’.

Table 4. Summary of studies of relationships between occupancy or breeding success of open-ground bird species and distance to forest edge. Previous studies in the Flow Country are discussed in main text and not detailed again here. Studies dependent entirely on artificial nest experiments are not included due to the difficulties in interpreting results (Dolman 2012)
StudyFocal open-ground speciesLocationOpen-ground habitat affectedKey findings
Parr (1992)European golden ploverNE ScotlandUpland heathlandPoor breeding success associated with severe egg predation and coincident with afforestation of study area, bringing all territories within 2 km of forestry
Valkama, Currie & Korpimaki (1999)Eurasian curlewW FinlandTilled farmland20% of nests hatched young in a landscape where random points on average 180 m from forestry, but 91% where random points on average 600 m from forestry
Finney, Pearce-Higgins & Yalden (2005)European golden ploverN EnglandBlanket bogProbability of use of 100 m squares by adult birds and probability of nests fledging young increased with distance from conifer forestry in landscape with forestry at 0–5 km from focal squares and nests
Pearce-Higgins & Grant (2006)Red grouse Lagopus l. scoticus, European golden plover, Eurasian curlew, common snipe Gallinago gallinago, skylark Alauda arvensis, meadow pipit Anthus pratensis, northern wheatear Oenanthe oenanthe, whinchat Saxicola rubetra, European stonechat S.rubicolaS Scotland & N EnglandUpland grassland, heathland and blanket bogNo correlations between bird abundance and forestry cover after controlling for other habitat and landscape variables, but in a study where survey plots had been chosen to have no forestry within 200 m of their boundaries
Amar et al. (2011)European golden plover, northern lapwing, dunlin, Eurasian curlew, common snipeUKUpland grassland, heathland and blanket bog50% forest cover within 1 km of survey plot associated with 85-95% decline of common snipe, dunlin and European golden plover. Absence of forest cover associated with population stability
Douglas et al. (2013)Eurasian curlewS Scotland & N EnglandUpland grassland, heathland and blanket bogPopulation changes and fledging success negatively associated with forestry cover within 1 km of edge of study plots
Berg & Hiron (2012)CorncrakeS SwedenMixed farmlandNo avoidance of forest edges, with 42% of territories within 100 m
Shochat, Abramsky & Pinshow (2001)Scrub specialists including desert lark Ammomanes deserti, long-billed pipit Anthus similis, spectacled warbler Sylvia conspicillataCentral IsraelSemi-desert scrubScrubland specialists lost from afforested landscapes where scrub patch size falls below 50 ha (equivalent to a circle of radius 400 m)
Buchanan et al. (2003)Ring ouzel Turdus torquatusScotlandUpland grassland2-km squares with up to 30% forest cover more likely to experience declines over a decade than squares with no forest cover, despite almost no change in forest cover, and after accounting for other habitat variables
Rodewald & Vitz (2005)Shrubland specialists including blue-winged warbler Vermivora cyanoptera, prairie warbler Dendroica discolor, yellow-breasted chat Icteria virens, indigo bunting Passerina cyanea and field sparrow Spizella pusillaS Ohio, United StatesShrublandTwice as many individuals of 7/8 shrubland specialist species mist-netted 80 m from forest edges than 20 m from edges
Reino et al. (2009)Calandra lark Melanocorypha calandra and short-toed lark Calandrella brachydactylaS PortugalTilled farmlandPoint count abundances increased 8- to 14-fold up to 300 m from forest edges for these two open-ground species

Forestry advisers and managers in our study area have begun to use maps of the kind illustrated by Fig. 3, in conjunction with recent advice on the negative GHG implications of forestry planting on deep peat (Morison et al. 2010), to identify areas that, once harvested, should not be brought into a second tree-planting rotation. This will add to the 2200 ha of forestry previously removed in support of blanket bog restoration under EU LIFE Nature programmes (Wilkie & Mayhew 2003). Our analyses also show that these early forest removals were well located with respect to predicted long-term benefits from mitigation of edge effects (areas 2–5 in Fig. 1 compared with Fig. 3). These initiatives are predicted to improve habitat quality for dunlin and European golden plover, and, in the long term, there may be additional benefits as the cleared areas become suitable for re-occupation by breeding waders. In a recent study of effects of removal of tree and shrub rows on breeding grassland birds in Wisconsin, increases in nesting density of eastern meadowlark Sturnella magna, Henslow's Sparrow Ammodramus henslowii and bobolink Dolichonyx oryzivorus were recorded within just three years, alongside reductions in activity of woodland-associated predators such as raccoon Procyon lotor (Ellison et al. 2013). In our peatland study areas, changes may be more gradual because deep ploughing at planting and persistence of felling brash make bog vegetation recovery a slow process, and felled areas may favour predators for several years until the brash breaks down and is incorporated by re-expanding bog vegetation. This may explain why removal of some forest blocks between 1998 and 2003–2006 (Fig. 1) had little impact on model goodness-of-fit (see Appendix S1, Supporting information). Recent developments in harvesting machinery and markets for whole-tree harvesting for wood fuel are likely to mean that much less brash remains after future fellings. We propose to continue monitoring of bird populations and vegetation change as restoration of formerly forested areas continues, to test whether edge effects reduce over time, as predicted, and whether this varies with harvesting treatment.

It is important to improve the evidence on impacts of plantation forest expansion on birds of high conservation importance, and the mechanisms underlying these effects. Forest species may benefit from the creation of wooded habitats and increased connectivity of forest cover, whilst open-ground species may suffer from habitat loss and fragmentation. It is also timely to consider how forests should be sited in the landscape, given the general support for woodland expansion in support of climate change mitigation and other policy goals (Read et al. 2009; Quine et al. 2011), set alongside increasing concern for declining wader populations (e.g. Newton 2004). In Scotland, government policy aspires to increase woodland cover from 17% to 25% of the land, an addition of some 650 000 ha over the 21st century (Scottish Government 2009) – 10 000 ha per annum of new afforestation over the next ten years (WEAG 2012). Because of the high agricultural value of lowland farmland (WEAG 2012), marginal upland areas are likely to continue to be favoured for such expansion. Many such areas are also of conservation value (Thompson et al. 1995), so where expansion does take place, the emerging edge effects on breeding waders should be considered. Studies of predator populations and behaviour, and bird distribution and demography on adjacent open-ground habitats are clearly important.

Acknowledgements

Simon Hodge, Gordon Patterson, Ian Bainbridge and Stuart Housden supported initiation of this research. Mark Brewer, Paul Britten, Graeme Buchanan, David Fouracre, Cynthia Moore, Emma Teuten and Ellen Wilson provided GIS and statistical support. Andrew Coupar, Pete Mayhew, John Risby and Lesley Cranna advised on conservation and management of peatlands. Comments from Chris Elphick, Nils Warnock and an anonymous reviewer greatly improved earlier drafts. We thank staff and volunteers at Forsinard Flows Nature Reserve for bird survey data collection, especially Alasdair Boulton, Ian Dillon, Paul Jacques, Ewen Munro and James Plowman.

Ancillary