Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Landscape matrix modifies richness of plants and insects in grassland fragments

Landscape matrix modifies richness of plants and insects in grassland fragments Traditionally, studies on habitat loss and fragmentation and effects on biodiversity have regarded habitat fragments as isolated islands surrounded by a hostile matrix, in analogy with the ocean in the theory of island biogeography ( MacArthur and Wilson 1967 ). There is, however, an increasing recognition that this may be an oversimplified assumption, and that the composition of the surrounding landscape can indeed influence species abundance or diversity in fragmented landscapes ( Ricketts 2001 , Cook et al. 2002 , Brotons et al. 2003 , Debinski 2006 , Ewers and Didham 2006 , Prugh et al. 2008 ). The composition and land use of the surrounding landscape might influence organisms through several mechanisms. Land use in the matrix can influence the quality of embedded habitat patches positively or negatively by altering conditions such as micro‐climate, nutrient status or biotic interactions near habitat edges ( Ries et al. 2004 ). The permeability of the matrix can influence dispersal between habitat fragments, and hence the functional connectivity of the landscape ( Roland et al. 2000 , Ricketts 2001 , Baum et al. 2004 ) and as a result, also the colonization‐extinction dynamics in the habitat fragments ( Vandermeer and Carvajal 2001 , Cronin and Haynes 2004 ). Mobile animals can benefit from complementary or supplementary resources available outside the habitat fragments, and hence suffer less from reduced habitat area ( Dunning et al. 1992 , Ås 1999 , Brotons et al. 2003 , Debinski 2006 ). A large species pool in the surrounding landscape, e.g. due to a high heterogeneity ( Benton et al. 2003 ), can also influence communities in embedded habitat fragments, by immigration of species for which the studied habitat type constitute sink habitat ( Leibold et al. 2004 , Rand et al. 2006 ). Hence, in general matrix land use can modify, positively or negatively, the negative effects of small patch area and high isolation on species richness remnant habitat patches ( Anderson and Wait 2001 , Watson et al. 2005 ). In central and northern Europe, extensive areas of semi‐natural grasslands have been transformed either into forest or arable land during the last century ( Cousins 2009 ). As a consequence, remnant patches of semi‐natural grassland typically occur as small and isolated patches surrounded either by arable land or forest or a mix of these. Until now, however, most empirical studies of the effect of matrix quality in fragmented landscapes have either focused on single species ( Roland et al. 2000 , Schooley and Wiens 2005 ), or examined diversity effects in only one replicate of each matrix type ( Watson et al. 2005 , Wethered and Lawes 2005 ). However, not all organisms are likely to be influenced by the mentioned mechanisms to the same extent, and we can expect diverging responses to matrix land use for different taxa ( Prevedello and Vieira 2010 ). In real landscapes, matrix quality is often confounded with patch area, quality, isolation or all of these ( Haynes and Cronin 2004 ). To separate the potential effects of patch area, isolation and matrix quality, we set up a study system consisting of semi‐natural grassland patches along gradients of patch area and isolation replicated in three standardized types of surrounding landscapes (hereafter referred to as ‘matrix types’): dominated by forest, arable land or a mixture of these. In each grassland patch we sampled species richness of four taxa, vascular plants, butterflies, bees and hoverflies. The four groups represent different life‐histories, trophic levels and feeding niches, where butterflies are herbivores (larvae) and nectar feeders (adults), bees are nectar and pollen feeders and the hoverflies are nectar feeders as adults but contain species that are herbivores, predators and detrivores as larvae. For all four groups, we hypothesize that species richness increases with increasing area and decreasing isolation, and in addition is influenced by matrix type. We also predict matrix type to interact with patch area and isolation, i.e. the matrix can either amplify or reduce the negative effects of small patch area and high isolation on species richness. Matrix land use can influence species richness patterns through several different, but not mutually exclusive mechanisms, which would probably result in different patterns in species richness ( Table 1 ). If matrix influences species richness in grassland patches by immigration from adjacent matrix habitats, we expect the highest species richness in patches surrounded by a mixed matrix which is more heterogeneous and likely to contain a wider range of source habitats for immigrants (cf. Benton et al. 2003 ). We would also expect the matrix effect to disappear if only habitat specialist species are included in the analysis ( Table 1 ). If matrix land use influences dispersal between habitat patches, we would expect a negative effect of forest matrix, because previous studies have found that forest inhibit the dispersal of grassland insects ( Roland et al. 2000 , Ricketts 2001 , Wratten et al. 2003 ). If matrix land use influences dispersal this would also result in matrix‐dependent effects of isolation (i.e. an interaction between matrix and isolation, Table 1 ). If the matrix influences species richness in the habitat patches through negative edge effects, we would expect stronger matrix effect in small grassland patches because these have a larger perimeter/area ratio. Often, matrix effects are more pronounced for species on higher trophic levels ( Ewers and Didham 2006 ), so we would expect a stronger matrix effect on insects than on plants. The insect groups could also benefit from complementary or supplementary resources in the matrix ( Dunning et al. 1992 ). This would reduce the negative effect of small patch area for the total species richness of insects, but probably not for the richness of grassland specialists. 1 Predictions on the responses of the studied plant and insect taxa depending on the mechanisms through which matrix land use influences species richness. Predictions Mechanism Matrix types Type of matrix effect (main effect, interaction with area or isolation) Total species richness vs richness of grassland specialists Taxa Immigration of matrix‐ inhabiting species Positive effect of mixed matrix Main + interaction with area Effect only on total richness No difference Dispersal Negative effect of forest Main + interaction with isolation No difference Strongest effect on plants (low mobility) Edge effects Mixed effects Main + interaction with area No difference Possibly stronger effect on higher trophic levels (i.e. insects) Complementation/supplementation Positive effect of mixed matrix Main + interaction with area Effect only on total richness Insects only Material and methods Study system Our study system consisted of 45 semi‐natural grassland patches, grazed by cattle or horses, in two regions (the counties of Östergötland and Uppland) in south‐eastern Sweden. We define semi‐natural grasslands as grasslands that have not been re‐sown or improved by inorganic fertilizers, but depend on continuous management by grazing or mowing. We selected grassland patches so that they were situated along similar gradients of area and isolation, but embedded in three contrasting matrix types: arable, forest and mixed. There was no correlation between patch area and isolation (log‐transformed values; r = 0.18, p = 0.22), and neither mean log(area) (F 2,45 = 2.16, p = 0.13) nor log(isolation) (F 2,45 = 1.66, p = 0.20) differed between landscape types. The study design was balanced, with 15 patches in each matrix type, and with 9 patches with each matrix type in Östergötland and 6 patches with each matrix type in Uppland ( Fig. 1 ). Based on a number of pre‐defined criteria the patches were selected from a national data base on semi‐natural grasslands ( ). We selected dry to mesic grassland patches with an area of 1 10 ha The minimum distance between any pair of patches was 3.5 km. Land use (%) within the landscapes surrounding each patch was measured from a topographic map (scale 1:50 000; Lantmäteriet, Gävle, Sweden) in a circle with 2 km radius around the focal grassland patch. Both insect and plant species richness has previously been found to respond to landscape composition at this spatial scale ( Lindborg and Eriksson 2004 , Öckinger et al. 2010 ). We arbitrarily set the maximum forest cover in landscapes categorized as arable to 20%. The forest landscapes were allowed to contain a maximum of 20% of arable land and the mixed landscape a minimum of 30% each of forest and arable land. In all landscapes, no more than 5% of the area was allowed to consist of water, urban area or land cover types other than arable land, forest and grassland. Aerial photos of examples of landscapes in each category are shown in Fig. 2 . Local grassland quality factors such as soil type, grazing pressure and cover of trees and shrubs were standardized as far as possible across matrix types. 1 Map showing the location of study sites of the three matrix types in Sweden and Europe. 2 Aerial photographs showing examples of landscapes classified as arable, mixed and forest. The studied sites are situated the centre of the circles, which have a radius of 2 km. Copyright Lantmäteriet Gävle 2010. Permission I 2010/0055. We used geographical data from a national data base on semi‐natural grasslands ( ) to identify all patches of semi‐natural grassland within 5 km from each studied habitat patch. We measured the isolation of each grassland patch i as I i =−Σ exp(−αd ij ) A j where d ij is the distance between patches i and j, A j is the area of patch j (in ha), and α is a variable describing the migration range. Since we are dealing with multiple species differing in dispersal capacity, we set α= 1 (cf. Hanski 1994 ). The rank order of isolation between patches is not sensitive to the value of α ( Moilanen and Nieminen 2002 ). The log‐transformed values were used in the analyses to improve normality. Species richness data In each grassland patch, we sampled the species richness of vascular plants, butterflies and burnet moths (hereafter collectively referred to as ‘butterflies’), bees and hoverflies. Plant species richness was estimated by recording all vascular plants in 10 randomly placed m 2 plots, and thereafter walking through the grassland patch at a standardized pace (30 min ha −1 ) recording all additional plants not found in the small scale inventory (cf. Sutherland 2000 ). Insects were sampled by standardized transect counts with a length proportional to patch area. For butterflies, we walked 200 m ha −1 , and the breadth of the transects were 5 m to each side and in front of the recorder. When necessary, butterflies were caught by a hand‐held net for identification in the field. Bees and hoverflies were sampled along transects with a length of 50 m ha −1 and a breadth of 1 m to each side of the recorder. Bumblebees could usually be identified in the field. All other bees and hoverflies were caught by a hand‐held net, killed, pinned and labeled for later identification in the lab. Transect walks were performed only on days with temperature ≥17°C, no precipitation, dry vegetation and low wind speeds. For each group of organisms, we identified a subset of all species as grassland specialists, based on independent data. Plants were regarded as specialists if they occur only in grasslands and disappear within 15 yr after ceased management ( Ekstam and Forshed 1997 ) or have a very high abundance in grasslands. For butterflies, the classification was done by an external expert (M. Franzén, Lund Univ.) based on data from Eliasson et al. (2005) . For bees, grassland specialists were defined as species recorded in the category ‘natural/semi‐natural grasslands’ and maximum three other major habitat types (out of 10) in the ALARM bee trait data base held by S. P. M. Roberts, Univ. of Reading, UK. For hoverflies we used the classification in the database ‘Syrph the Net’ ( Speight et al. 2008 ). Statistical analyses To test whether the response of species richness of each organism group to area and isolation differed depending on matrix type, we applied general linear models (GLM (SAS proc Mixed, Littell et al. 2006 )), separately for each organism group. Log‐transformed species richness of each group (plants, butterflies, bees, hoverflies) were included as response variables. Matrix type and log‐transformed patch area and isolation measures were predictor variables. To test whether the responses of species richness to patch area or isolation were affected by matrix type, we also included the interaction terms matrix × log(area) and matrix × log(isolation). Non‐significant interactions (> 0.1), but not main terms, were sequentially removed from the model. We also estimated the matrix‐specific responses to patch area and isolation, and ran a separate GLM for each matrix type. Mean values for the matrix categories were separated using the LSmeans statement and Pdiff option in SAS proc Mixed. P‐values are adjusted according to the Tukey method. To test the hypothesis that any matrix effects are not only due to immigration of matrix‐inhabiting species, we repeated all analyses with the richness of grassland specialists instead of total species richness for each group. Results Plants Matrix type was the only factor affecting the overall plant species richness ( Table 2 , Fig. 3 ), and there were no interactions between matrix and area or isolation. Species richness was highest in patches surrounded by forest matrix, intermediate in mixed and lowest in patches surrounded by arable matrix, but only the difference between forest and arable matrix was statistically significant (t 39 = 2.97, p = 0.013). However, when including only species specialized on semi‐natural grasslands, there was no difference between landscapes ( Table 2 ). 2 Results of general linear models relating patch area, isolation, matrix type and their interactions to species richness of plants, butterflies, bees and hoverflies. Only variables included in the final models are shown. Significant values are marked in bold. Plants Butterflies Bees Hoverflies F DF p F DF p F DF p F DF p All species Area 0.94 1,40 0.339 8.07 1,40 0.007 4.121 1,35 0.050 12.0 1,36 0.001 Isolation 0.05 1,40 0.830 0.06 1,40 0.800 0.84 1,35 0.366 4.83 1,36 0.035 Matrix 4.05 2,40 0.025 5.20 2,40 0.010 2.69 2,35 0.082 1.90 2,36 0.164 Area × matrix N.S. N.S. 2.99 2,35 0.063 5.53 2,36 0.008 Isolation × matrix N.S. N.S. 2.96 2,35 0.065 2.73 2,36 0.079 Specialists Area 0.63 1,40 0.433 4.23 1,40 0.046 4.18 1,39.6 0.031 3.23 1,37.9 0.080 Isolation 2.02 1,40 0,163 0.96 1,40 0.334 0.00 1,40 0.983 0.06 1,35.1 0.803 Matrix 0.31 2,40 0.732 5.55 1,40 0.008 1.55 2,39.2 0.225 3.65 2,37.2 0.019 Area × matrix N.S. N.S. N.S. 4.64 2,37.3 0.015 Isolation × matrix N.S. N.S. N.S. N.S. 3 Species richness in relation to matrix type for plants, butterflies, bees and hoverflies. Black bars = all species, white bars = grassland specialists. Letters indicate significant (at the 0.05 level) differences in mean values. Butterflies The total species richness of butterflies was positively related to patch area and was also affected by matrix type ( Table 2 , Fig. 3 ). The number of species was lower in arable landscapes than in forest (t 40 = 3.19, p = 0.008) but no difference between forest and mixed (t 40 = 1.06, p = 0.54) or arable and mixed (t 40 = 2.22, p = 0.08) landscapes, ( Fig. 4 ). The pattern was similar when only grassland specialist species were included in the analysis. 4 Total species richness and richness of grassland specialist species in relation to patch area for plants (not significant), butterflies (total richness: p = 0.007, slope (z) = 0.24; specialists: p = 0.046, z = 0.24), bees (total richness: p = 0.05, z = 0.40; specialists: p = 0.031, z = 0.35) and hoverflies (total richness: p = 0.001, z = 0.49; specialists: p = 0.08, z = 0.35). Regression lines are shown only when p < 0.1. Bees Species richness of bees was positively related to patch area ( Table 2 , Fig. 4 ). There was also a marginally non‐significant effect of matrix type, with highest species richness in patches with arable matrix and lowest in patches with mixed matrix, and marginally non‐significant interactions between matrix type and patch area and matrix type and isolation, respectively ( Table 2 , Fig. 3 ). Separate analyses for each matrix typed revealed that there were no effects of patch area or isolation in patches surrounded by an arable matrix. In the forest matrix, there was a negative effect of isolation (F 1,11 = 6.90, p = 0.024) but not area. In the mixed matrix, there was a positive effect of patch area (F 1,11 = 18.4, p = 0.001) but not isolation. The positive effect of patch area remained for grassland specialist bees, but there were no matrix effects. Hoverflies Hoverfly species richness increased with increasing patch area ( Fig. 4 ) and decreasing isolation ( Table 2 ). There was also a significant interaction between patch area and matrix type ( Fig. 5 ), and a marginally non‐significant interaction between isolation and matrix type ( Table 2 ). Separate analyses for each matrix type revealed that the effects of patch area (F 1,11 = 30.0, p < 0.001) and isolation (F 1,11 = 11.9, p = 0.006) were only significant in the forest matrix, not in the arable or mixed matrix. The number of specialist hoverflies was not related to patch area or isolation ( Table 2 ), but the area × matrix interaction, with a an effect of patch area only in the forested matrix (F 1,13 = 14.3, p = 0.002), remained significant for the specialist hoverflies. For the specialist hoverflies there was also a difference in richness between matrix types, with the highest species richness in arable matrix. 5 Interactive effect of patch area and matrix type on hoverfly species richness. The effect of patch area is only statistically significant for grasslands situated in a forested matrix. Discussion There is a growing body of evidence that matrix land use influences species in fragmented habitats in addition to the effects of patch area and isolation, but few studies have investigated the effects of matrix land use on multiple taxa simultaneously ( Prevedello and Vieira 2010 , but see Lomolino and Smith 2003 ). In this study we demonstrate that the composition of landscape surrounding patches of semi‐natural grassland has direct effects on the species richness of plants and butterflies, and modifies the species–area relationship for hoverflies, but in different directions. Understanding the relative importance of habitat area, isolation and landscape matrix for species distributions and community composition is crucial for biodiversity conservation in human‐dominated landscapes ( Pöyry et al. 2009 ). According to the island biogeography and metapopulation paradigms, the isolation of habitat patches is the most important property a landscape, but empirical studies show variable results regarding the importance of isolation for species distributions and diversity patterns ( Hanski and Pöyry 2007 , Prugh et al. 2008 ). In our study, we show that at least three of the four taxa (plants, butterflies and hoverflies) are influenced by landscape‐scale factors (i.e. either isolation or matrix land use), but these factors differ among the studied taxa. Only hoverflies are influenced by geographic isolation, whereas other landscape level processes than dispersal between habitat patches appear to be limiting plant and butterfly distributions in our studied landscapes. This highlights the importance of a more holistic landscape approach for understanding species distributions in human‐modified landscapes ( Fischer and Lindenmayer 2007 ). The contrasting patterns for the studied species assemblages suggest that they are influenced by matrix land use through different mechanisms. Species‐specific responses to matrix land use appear to be common ( Prevedello and Vieira 2010 ). For example, Lomolino and Smith (2003) found opposite responses of mammals and reptiles to forest cover in the landscape matrix surrounding prairie dog towns, and Ricketts (2001) found different responses in dispersal to inter‐patch matrix land use between butterfly species. Such differences can be due to species‐specific differences in the ability to use the matrix as a secondary habitat or in the ability to disperse through the matrix ( Prevedello and Vieira 2010 ). Our focus on species richness patterns does not allow a complete assessment of the mechanisms behind the observed patterns, but the predictions summarized in Table 1 give some suggestions. For plants, the positive effect of forest matrix was significant for total species richness, and disappeared when grassland specialists were analysed. This suggests that grassland patches in forest‐dominated landscapes are enriched by immigration of matrix‐inhabiting species ( Table 1 ). Few studies have assessed matrix effects on plants ( Prevedello and Vieira 2010 ), but similarly to our results Söderström et al. (2001) found a positive effect of forest cover in the surrounding landscape on plant species richness. Indeed, even if most of the forest is intensively managed, the forest‐dominated matrix is likely to be more species‐rich than the arable matrix, and hence there are more potential immigrant species ( Cousins and Aggemyr 2008 ). An alternative explanation for the flat species–area relationship and positive effect of forest matrix for plants is that land use conversion took place later in forest dominated landscapes than in more productive regions ( Cousins 2009 ). Since many plants are long‐lived, it is possible that population extinctions are lagging behind habitat loss, i.e. there is an extinction debt ( Lindborg and Eriksson 2004 , Kuussaari et al. 2009 , Krauss et al. 2010 ). We suggest that the contrasting results with respect to matrix and isolation among the insect groups are a result of differences in life histories and resource use. The availability of larval host plants ( Krauss et al. 2004 ) and nectar ( Franzén and Nilsson 2008 ) are major limiting factors for butterflies at a landscape scale. In forest‐dominated landscapes, these resources can be abundant in small open areas such as glades and road verges and temporarily in clear‐cuts ( Bergman et al. 2008 ) but largely absent from landscapes dominated by intensively farmed arable fields. Hence, the higher number of butterfly species in patches in forest‐dominated landscapes is likely explained by resource supplementation and complementation ( Dunning et al. 1992 ). Alternatively, the higher number of plant species in these patches imply a wider range of larval host plants for the butterflies, and thus enables higher butterfly species richness. For hoverflies, forest matrix cover had the opposite effect compared to plants and butterflies, and also opposite to previous observations ( Sjödin et al. 2008 ). A possible explanation is that forest inhibits hoverfly dispersal (cf. Wratten et al. 2003 ). Also the observation that hoverflies was the only group which was affected by isolation suggests that dispersal limitation is an important factor influencing hoverfly communities in fragmented landscapes. Alternatively, resource supplementation and complementation is the main mechanism also for hoverflies, but their complementary resources are more abundant in arable land than in forest, e.g. in the form of mass‐flowering crops ( Westphal et al. 2003 ) and larval food for predatory hover flies. Bee diversity was only significantly related to patch area. Many bees are linked to permanent grasslands, and a possible mechanisms for observed strong species–area relationships ( Bommarco et al. 2010 ), is that there is evidence that bee populations are often primarily more limited by the availability of nesting sites rather than by flower resources ( Potts et al. 2005 , Steffan‐Dewenter and Schiele 2008 ), and nest sitesthese are probably much more abundant in semi‐natural grasslands than in managed forest or arable land. For some of the insect species, parts of their habitat might in fact be situated outside the grassland patches, in the matrix. This relates to the problem of the anthropocentric way of defining habitats as biotopes or landscape elements. A resource‐based habitat definition ( Dennis et al. 2003 ) would be more biologically relevant, but is complicated when assemblages of species are considered. Landscape heterogeneity is generally assumed to promote high species richness, because heterogeneous landscapes contain habitats for a larger number of species ( Benton et al. 2003 ). In contrast to our expectations, grasslands surrounded by a mixed matrix did not have the highest species richness or the flattest species–area relationship for any of the taxa. Instead, patches embedded in a mixed matrix tended to take an intermediate position regardless of whether the forest or the arable matrix had the most positive effect. However, landscape heterogeneity measures depend on the resolution of the landscape data, and both the relevant resolution and landscape categories are likely to differ between taxa. In our case we cannot exclude the possibility that for example forest dominated landscapes where more heterogeneous than the mixed ones with respect to the resolution and categories relevant for plants and butterflies. It may also be that the effect of a certain matrix land use is taxon‐specific such that species richness of that group will not change with landscape heterogeneity, but will instead be related to the proportion in the landscape of a certain land cover such as forest or arable land. It has been debated whether or not the landscape matrix land use influences species–area relationships. In a review, Watling and Donnelly (2006) found that terrestrial ‘habitat islands’ generally have a shallower species–area‐curve compared to true islands, but this was contradicted by a meta‐analysis by Drakare et al. (2006) who instead found the greatest differences in species–area relationships among habitat types, but no general difference between terrestrial habitats and islands. More recently, Prugh et al. (2008) , found differences in species sensitive to decreasing patch area between matrix types, but could not separate between effects of matrix type and habitat type. In our study, matrix land use modified the species–area relationship only in one group, hoverflies. In plants and butterflies matrix land use influenced the level of species richness in grassland patches, but not the slope of the species–area relationship. We demonstrate that matrix type can modify the effects on species richness of area and isolation in remnant grassland patches and that the responses and likely underlying mechanisms vary among taxa. This has important implications for landscape‐scale biodiversity conservation strategies. Increasing the quality of the matrix has been suggested as an alternative to habitat restoration or re‐creation ( Donald and Evans 2006 ). Our results imply that modifying the matrix could compensate for decreasing area of high‐quality habitat for some, but not all organisms. Also, since we found very different responses to matrix land use for plants and butterflies on one hand and hoverflies on the other hand, completely different actions to increase matrix quality would be needed for different taxa and there is a clear risk that management actions to increase matrix quality for one group would have negative consequences for other taxa. Most likely, there is also a large variation with respect to responses to matrix land use among species in each of the studied species groups, depending on their life‐history traits as observed for area and isolation effects ( Bommarco et al. 2010 , Öckinger et al. 2010 ). Hence, if actions to modify matrix land use should be applied in compensation for loss of natural and semi‐natural habitats we would also need to know the consequences of habitat reduction and matrix modification on the community composition, not only species richness. In order to provide more detailed recommendations for biodiversity conservation in anthropogenic landscapes, we need better understanding of the links between species life history traits, the mechanisms through which species interact with the surrounding landscape and their consequences for population dynamics and extinction risk. Acknowledgements We thank Iria Soto Embodas, Madeleine Arnqvist, Magnus Granbom and Johan Björklind Möllegård for assistance with the field work and L. Anders Nilsson and Hans D. Bartsch for help with identification of insect specimen. Lorenzo Marini and two anonymous reviewers gave valuable comments on the text. Funding for this study was provided by the EU‐projects ‘COCONUT – Understanding effects of land use changes on ecosystems to halt loss of biodiversity’ (FP6 SSPI‐CT‐2006‐044343) and ‘SCALES – Securing the Conservation of biodiversity across Administrative Levels and spatial, temporal and Ecological Scales’ (FP7 226852), and by the Swedish research council for environment, agricultural sciences and spatial planning (FORMAS). http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Ecography Wiley

Landscape matrix modifies richness of plants and insects in grassland fragments

Loading next page...
 
/lp/wiley/landscape-matrix-modifies-richness-of-plants-and-insects-in-grassland-73CU0411N7

References (50)

Publisher
Wiley
Copyright
© 2011 The Authors
ISSN
0906-7590
eISSN
1600-0587
DOI
10.1111/j.1600-0587.2011.06870.x
Publisher site
See Article on Publisher Site

Abstract

Traditionally, studies on habitat loss and fragmentation and effects on biodiversity have regarded habitat fragments as isolated islands surrounded by a hostile matrix, in analogy with the ocean in the theory of island biogeography ( MacArthur and Wilson 1967 ). There is, however, an increasing recognition that this may be an oversimplified assumption, and that the composition of the surrounding landscape can indeed influence species abundance or diversity in fragmented landscapes ( Ricketts 2001 , Cook et al. 2002 , Brotons et al. 2003 , Debinski 2006 , Ewers and Didham 2006 , Prugh et al. 2008 ). The composition and land use of the surrounding landscape might influence organisms through several mechanisms. Land use in the matrix can influence the quality of embedded habitat patches positively or negatively by altering conditions such as micro‐climate, nutrient status or biotic interactions near habitat edges ( Ries et al. 2004 ). The permeability of the matrix can influence dispersal between habitat fragments, and hence the functional connectivity of the landscape ( Roland et al. 2000 , Ricketts 2001 , Baum et al. 2004 ) and as a result, also the colonization‐extinction dynamics in the habitat fragments ( Vandermeer and Carvajal 2001 , Cronin and Haynes 2004 ). Mobile animals can benefit from complementary or supplementary resources available outside the habitat fragments, and hence suffer less from reduced habitat area ( Dunning et al. 1992 , Ås 1999 , Brotons et al. 2003 , Debinski 2006 ). A large species pool in the surrounding landscape, e.g. due to a high heterogeneity ( Benton et al. 2003 ), can also influence communities in embedded habitat fragments, by immigration of species for which the studied habitat type constitute sink habitat ( Leibold et al. 2004 , Rand et al. 2006 ). Hence, in general matrix land use can modify, positively or negatively, the negative effects of small patch area and high isolation on species richness remnant habitat patches ( Anderson and Wait 2001 , Watson et al. 2005 ). In central and northern Europe, extensive areas of semi‐natural grasslands have been transformed either into forest or arable land during the last century ( Cousins 2009 ). As a consequence, remnant patches of semi‐natural grassland typically occur as small and isolated patches surrounded either by arable land or forest or a mix of these. Until now, however, most empirical studies of the effect of matrix quality in fragmented landscapes have either focused on single species ( Roland et al. 2000 , Schooley and Wiens 2005 ), or examined diversity effects in only one replicate of each matrix type ( Watson et al. 2005 , Wethered and Lawes 2005 ). However, not all organisms are likely to be influenced by the mentioned mechanisms to the same extent, and we can expect diverging responses to matrix land use for different taxa ( Prevedello and Vieira 2010 ). In real landscapes, matrix quality is often confounded with patch area, quality, isolation or all of these ( Haynes and Cronin 2004 ). To separate the potential effects of patch area, isolation and matrix quality, we set up a study system consisting of semi‐natural grassland patches along gradients of patch area and isolation replicated in three standardized types of surrounding landscapes (hereafter referred to as ‘matrix types’): dominated by forest, arable land or a mixture of these. In each grassland patch we sampled species richness of four taxa, vascular plants, butterflies, bees and hoverflies. The four groups represent different life‐histories, trophic levels and feeding niches, where butterflies are herbivores (larvae) and nectar feeders (adults), bees are nectar and pollen feeders and the hoverflies are nectar feeders as adults but contain species that are herbivores, predators and detrivores as larvae. For all four groups, we hypothesize that species richness increases with increasing area and decreasing isolation, and in addition is influenced by matrix type. We also predict matrix type to interact with patch area and isolation, i.e. the matrix can either amplify or reduce the negative effects of small patch area and high isolation on species richness. Matrix land use can influence species richness patterns through several different, but not mutually exclusive mechanisms, which would probably result in different patterns in species richness ( Table 1 ). If matrix influences species richness in grassland patches by immigration from adjacent matrix habitats, we expect the highest species richness in patches surrounded by a mixed matrix which is more heterogeneous and likely to contain a wider range of source habitats for immigrants (cf. Benton et al. 2003 ). We would also expect the matrix effect to disappear if only habitat specialist species are included in the analysis ( Table 1 ). If matrix land use influences dispersal between habitat patches, we would expect a negative effect of forest matrix, because previous studies have found that forest inhibit the dispersal of grassland insects ( Roland et al. 2000 , Ricketts 2001 , Wratten et al. 2003 ). If matrix land use influences dispersal this would also result in matrix‐dependent effects of isolation (i.e. an interaction between matrix and isolation, Table 1 ). If the matrix influences species richness in the habitat patches through negative edge effects, we would expect stronger matrix effect in small grassland patches because these have a larger perimeter/area ratio. Often, matrix effects are more pronounced for species on higher trophic levels ( Ewers and Didham 2006 ), so we would expect a stronger matrix effect on insects than on plants. The insect groups could also benefit from complementary or supplementary resources in the matrix ( Dunning et al. 1992 ). This would reduce the negative effect of small patch area for the total species richness of insects, but probably not for the richness of grassland specialists. 1 Predictions on the responses of the studied plant and insect taxa depending on the mechanisms through which matrix land use influences species richness. Predictions Mechanism Matrix types Type of matrix effect (main effect, interaction with area or isolation) Total species richness vs richness of grassland specialists Taxa Immigration of matrix‐ inhabiting species Positive effect of mixed matrix Main + interaction with area Effect only on total richness No difference Dispersal Negative effect of forest Main + interaction with isolation No difference Strongest effect on plants (low mobility) Edge effects Mixed effects Main + interaction with area No difference Possibly stronger effect on higher trophic levels (i.e. insects) Complementation/supplementation Positive effect of mixed matrix Main + interaction with area Effect only on total richness Insects only Material and methods Study system Our study system consisted of 45 semi‐natural grassland patches, grazed by cattle or horses, in two regions (the counties of Östergötland and Uppland) in south‐eastern Sweden. We define semi‐natural grasslands as grasslands that have not been re‐sown or improved by inorganic fertilizers, but depend on continuous management by grazing or mowing. We selected grassland patches so that they were situated along similar gradients of area and isolation, but embedded in three contrasting matrix types: arable, forest and mixed. There was no correlation between patch area and isolation (log‐transformed values; r = 0.18, p = 0.22), and neither mean log(area) (F 2,45 = 2.16, p = 0.13) nor log(isolation) (F 2,45 = 1.66, p = 0.20) differed between landscape types. The study design was balanced, with 15 patches in each matrix type, and with 9 patches with each matrix type in Östergötland and 6 patches with each matrix type in Uppland ( Fig. 1 ). Based on a number of pre‐defined criteria the patches were selected from a national data base on semi‐natural grasslands ( ). We selected dry to mesic grassland patches with an area of 1 10 ha The minimum distance between any pair of patches was 3.5 km. Land use (%) within the landscapes surrounding each patch was measured from a topographic map (scale 1:50 000; Lantmäteriet, Gävle, Sweden) in a circle with 2 km radius around the focal grassland patch. Both insect and plant species richness has previously been found to respond to landscape composition at this spatial scale ( Lindborg and Eriksson 2004 , Öckinger et al. 2010 ). We arbitrarily set the maximum forest cover in landscapes categorized as arable to 20%. The forest landscapes were allowed to contain a maximum of 20% of arable land and the mixed landscape a minimum of 30% each of forest and arable land. In all landscapes, no more than 5% of the area was allowed to consist of water, urban area or land cover types other than arable land, forest and grassland. Aerial photos of examples of landscapes in each category are shown in Fig. 2 . Local grassland quality factors such as soil type, grazing pressure and cover of trees and shrubs were standardized as far as possible across matrix types. 1 Map showing the location of study sites of the three matrix types in Sweden and Europe. 2 Aerial photographs showing examples of landscapes classified as arable, mixed and forest. The studied sites are situated the centre of the circles, which have a radius of 2 km. Copyright Lantmäteriet Gävle 2010. Permission I 2010/0055. We used geographical data from a national data base on semi‐natural grasslands ( ) to identify all patches of semi‐natural grassland within 5 km from each studied habitat patch. We measured the isolation of each grassland patch i as I i =−Σ exp(−αd ij ) A j where d ij is the distance between patches i and j, A j is the area of patch j (in ha), and α is a variable describing the migration range. Since we are dealing with multiple species differing in dispersal capacity, we set α= 1 (cf. Hanski 1994 ). The rank order of isolation between patches is not sensitive to the value of α ( Moilanen and Nieminen 2002 ). The log‐transformed values were used in the analyses to improve normality. Species richness data In each grassland patch, we sampled the species richness of vascular plants, butterflies and burnet moths (hereafter collectively referred to as ‘butterflies’), bees and hoverflies. Plant species richness was estimated by recording all vascular plants in 10 randomly placed m 2 plots, and thereafter walking through the grassland patch at a standardized pace (30 min ha −1 ) recording all additional plants not found in the small scale inventory (cf. Sutherland 2000 ). Insects were sampled by standardized transect counts with a length proportional to patch area. For butterflies, we walked 200 m ha −1 , and the breadth of the transects were 5 m to each side and in front of the recorder. When necessary, butterflies were caught by a hand‐held net for identification in the field. Bees and hoverflies were sampled along transects with a length of 50 m ha −1 and a breadth of 1 m to each side of the recorder. Bumblebees could usually be identified in the field. All other bees and hoverflies were caught by a hand‐held net, killed, pinned and labeled for later identification in the lab. Transect walks were performed only on days with temperature ≥17°C, no precipitation, dry vegetation and low wind speeds. For each group of organisms, we identified a subset of all species as grassland specialists, based on independent data. Plants were regarded as specialists if they occur only in grasslands and disappear within 15 yr after ceased management ( Ekstam and Forshed 1997 ) or have a very high abundance in grasslands. For butterflies, the classification was done by an external expert (M. Franzén, Lund Univ.) based on data from Eliasson et al. (2005) . For bees, grassland specialists were defined as species recorded in the category ‘natural/semi‐natural grasslands’ and maximum three other major habitat types (out of 10) in the ALARM bee trait data base held by S. P. M. Roberts, Univ. of Reading, UK. For hoverflies we used the classification in the database ‘Syrph the Net’ ( Speight et al. 2008 ). Statistical analyses To test whether the response of species richness of each organism group to area and isolation differed depending on matrix type, we applied general linear models (GLM (SAS proc Mixed, Littell et al. 2006 )), separately for each organism group. Log‐transformed species richness of each group (plants, butterflies, bees, hoverflies) were included as response variables. Matrix type and log‐transformed patch area and isolation measures were predictor variables. To test whether the responses of species richness to patch area or isolation were affected by matrix type, we also included the interaction terms matrix × log(area) and matrix × log(isolation). Non‐significant interactions (> 0.1), but not main terms, were sequentially removed from the model. We also estimated the matrix‐specific responses to patch area and isolation, and ran a separate GLM for each matrix type. Mean values for the matrix categories were separated using the LSmeans statement and Pdiff option in SAS proc Mixed. P‐values are adjusted according to the Tukey method. To test the hypothesis that any matrix effects are not only due to immigration of matrix‐inhabiting species, we repeated all analyses with the richness of grassland specialists instead of total species richness for each group. Results Plants Matrix type was the only factor affecting the overall plant species richness ( Table 2 , Fig. 3 ), and there were no interactions between matrix and area or isolation. Species richness was highest in patches surrounded by forest matrix, intermediate in mixed and lowest in patches surrounded by arable matrix, but only the difference between forest and arable matrix was statistically significant (t 39 = 2.97, p = 0.013). However, when including only species specialized on semi‐natural grasslands, there was no difference between landscapes ( Table 2 ). 2 Results of general linear models relating patch area, isolation, matrix type and their interactions to species richness of plants, butterflies, bees and hoverflies. Only variables included in the final models are shown. Significant values are marked in bold. Plants Butterflies Bees Hoverflies F DF p F DF p F DF p F DF p All species Area 0.94 1,40 0.339 8.07 1,40 0.007 4.121 1,35 0.050 12.0 1,36 0.001 Isolation 0.05 1,40 0.830 0.06 1,40 0.800 0.84 1,35 0.366 4.83 1,36 0.035 Matrix 4.05 2,40 0.025 5.20 2,40 0.010 2.69 2,35 0.082 1.90 2,36 0.164 Area × matrix N.S. N.S. 2.99 2,35 0.063 5.53 2,36 0.008 Isolation × matrix N.S. N.S. 2.96 2,35 0.065 2.73 2,36 0.079 Specialists Area 0.63 1,40 0.433 4.23 1,40 0.046 4.18 1,39.6 0.031 3.23 1,37.9 0.080 Isolation 2.02 1,40 0,163 0.96 1,40 0.334 0.00 1,40 0.983 0.06 1,35.1 0.803 Matrix 0.31 2,40 0.732 5.55 1,40 0.008 1.55 2,39.2 0.225 3.65 2,37.2 0.019 Area × matrix N.S. N.S. N.S. 4.64 2,37.3 0.015 Isolation × matrix N.S. N.S. N.S. N.S. 3 Species richness in relation to matrix type for plants, butterflies, bees and hoverflies. Black bars = all species, white bars = grassland specialists. Letters indicate significant (at the 0.05 level) differences in mean values. Butterflies The total species richness of butterflies was positively related to patch area and was also affected by matrix type ( Table 2 , Fig. 3 ). The number of species was lower in arable landscapes than in forest (t 40 = 3.19, p = 0.008) but no difference between forest and mixed (t 40 = 1.06, p = 0.54) or arable and mixed (t 40 = 2.22, p = 0.08) landscapes, ( Fig. 4 ). The pattern was similar when only grassland specialist species were included in the analysis. 4 Total species richness and richness of grassland specialist species in relation to patch area for plants (not significant), butterflies (total richness: p = 0.007, slope (z) = 0.24; specialists: p = 0.046, z = 0.24), bees (total richness: p = 0.05, z = 0.40; specialists: p = 0.031, z = 0.35) and hoverflies (total richness: p = 0.001, z = 0.49; specialists: p = 0.08, z = 0.35). Regression lines are shown only when p < 0.1. Bees Species richness of bees was positively related to patch area ( Table 2 , Fig. 4 ). There was also a marginally non‐significant effect of matrix type, with highest species richness in patches with arable matrix and lowest in patches with mixed matrix, and marginally non‐significant interactions between matrix type and patch area and matrix type and isolation, respectively ( Table 2 , Fig. 3 ). Separate analyses for each matrix typed revealed that there were no effects of patch area or isolation in patches surrounded by an arable matrix. In the forest matrix, there was a negative effect of isolation (F 1,11 = 6.90, p = 0.024) but not area. In the mixed matrix, there was a positive effect of patch area (F 1,11 = 18.4, p = 0.001) but not isolation. The positive effect of patch area remained for grassland specialist bees, but there were no matrix effects. Hoverflies Hoverfly species richness increased with increasing patch area ( Fig. 4 ) and decreasing isolation ( Table 2 ). There was also a significant interaction between patch area and matrix type ( Fig. 5 ), and a marginally non‐significant interaction between isolation and matrix type ( Table 2 ). Separate analyses for each matrix type revealed that the effects of patch area (F 1,11 = 30.0, p < 0.001) and isolation (F 1,11 = 11.9, p = 0.006) were only significant in the forest matrix, not in the arable or mixed matrix. The number of specialist hoverflies was not related to patch area or isolation ( Table 2 ), but the area × matrix interaction, with a an effect of patch area only in the forested matrix (F 1,13 = 14.3, p = 0.002), remained significant for the specialist hoverflies. For the specialist hoverflies there was also a difference in richness between matrix types, with the highest species richness in arable matrix. 5 Interactive effect of patch area and matrix type on hoverfly species richness. The effect of patch area is only statistically significant for grasslands situated in a forested matrix. Discussion There is a growing body of evidence that matrix land use influences species in fragmented habitats in addition to the effects of patch area and isolation, but few studies have investigated the effects of matrix land use on multiple taxa simultaneously ( Prevedello and Vieira 2010 , but see Lomolino and Smith 2003 ). In this study we demonstrate that the composition of landscape surrounding patches of semi‐natural grassland has direct effects on the species richness of plants and butterflies, and modifies the species–area relationship for hoverflies, but in different directions. Understanding the relative importance of habitat area, isolation and landscape matrix for species distributions and community composition is crucial for biodiversity conservation in human‐dominated landscapes ( Pöyry et al. 2009 ). According to the island biogeography and metapopulation paradigms, the isolation of habitat patches is the most important property a landscape, but empirical studies show variable results regarding the importance of isolation for species distributions and diversity patterns ( Hanski and Pöyry 2007 , Prugh et al. 2008 ). In our study, we show that at least three of the four taxa (plants, butterflies and hoverflies) are influenced by landscape‐scale factors (i.e. either isolation or matrix land use), but these factors differ among the studied taxa. Only hoverflies are influenced by geographic isolation, whereas other landscape level processes than dispersal between habitat patches appear to be limiting plant and butterfly distributions in our studied landscapes. This highlights the importance of a more holistic landscape approach for understanding species distributions in human‐modified landscapes ( Fischer and Lindenmayer 2007 ). The contrasting patterns for the studied species assemblages suggest that they are influenced by matrix land use through different mechanisms. Species‐specific responses to matrix land use appear to be common ( Prevedello and Vieira 2010 ). For example, Lomolino and Smith (2003) found opposite responses of mammals and reptiles to forest cover in the landscape matrix surrounding prairie dog towns, and Ricketts (2001) found different responses in dispersal to inter‐patch matrix land use between butterfly species. Such differences can be due to species‐specific differences in the ability to use the matrix as a secondary habitat or in the ability to disperse through the matrix ( Prevedello and Vieira 2010 ). Our focus on species richness patterns does not allow a complete assessment of the mechanisms behind the observed patterns, but the predictions summarized in Table 1 give some suggestions. For plants, the positive effect of forest matrix was significant for total species richness, and disappeared when grassland specialists were analysed. This suggests that grassland patches in forest‐dominated landscapes are enriched by immigration of matrix‐inhabiting species ( Table 1 ). Few studies have assessed matrix effects on plants ( Prevedello and Vieira 2010 ), but similarly to our results Söderström et al. (2001) found a positive effect of forest cover in the surrounding landscape on plant species richness. Indeed, even if most of the forest is intensively managed, the forest‐dominated matrix is likely to be more species‐rich than the arable matrix, and hence there are more potential immigrant species ( Cousins and Aggemyr 2008 ). An alternative explanation for the flat species–area relationship and positive effect of forest matrix for plants is that land use conversion took place later in forest dominated landscapes than in more productive regions ( Cousins 2009 ). Since many plants are long‐lived, it is possible that population extinctions are lagging behind habitat loss, i.e. there is an extinction debt ( Lindborg and Eriksson 2004 , Kuussaari et al. 2009 , Krauss et al. 2010 ). We suggest that the contrasting results with respect to matrix and isolation among the insect groups are a result of differences in life histories and resource use. The availability of larval host plants ( Krauss et al. 2004 ) and nectar ( Franzén and Nilsson 2008 ) are major limiting factors for butterflies at a landscape scale. In forest‐dominated landscapes, these resources can be abundant in small open areas such as glades and road verges and temporarily in clear‐cuts ( Bergman et al. 2008 ) but largely absent from landscapes dominated by intensively farmed arable fields. Hence, the higher number of butterfly species in patches in forest‐dominated landscapes is likely explained by resource supplementation and complementation ( Dunning et al. 1992 ). Alternatively, the higher number of plant species in these patches imply a wider range of larval host plants for the butterflies, and thus enables higher butterfly species richness. For hoverflies, forest matrix cover had the opposite effect compared to plants and butterflies, and also opposite to previous observations ( Sjödin et al. 2008 ). A possible explanation is that forest inhibits hoverfly dispersal (cf. Wratten et al. 2003 ). Also the observation that hoverflies was the only group which was affected by isolation suggests that dispersal limitation is an important factor influencing hoverfly communities in fragmented landscapes. Alternatively, resource supplementation and complementation is the main mechanism also for hoverflies, but their complementary resources are more abundant in arable land than in forest, e.g. in the form of mass‐flowering crops ( Westphal et al. 2003 ) and larval food for predatory hover flies. Bee diversity was only significantly related to patch area. Many bees are linked to permanent grasslands, and a possible mechanisms for observed strong species–area relationships ( Bommarco et al. 2010 ), is that there is evidence that bee populations are often primarily more limited by the availability of nesting sites rather than by flower resources ( Potts et al. 2005 , Steffan‐Dewenter and Schiele 2008 ), and nest sitesthese are probably much more abundant in semi‐natural grasslands than in managed forest or arable land. For some of the insect species, parts of their habitat might in fact be situated outside the grassland patches, in the matrix. This relates to the problem of the anthropocentric way of defining habitats as biotopes or landscape elements. A resource‐based habitat definition ( Dennis et al. 2003 ) would be more biologically relevant, but is complicated when assemblages of species are considered. Landscape heterogeneity is generally assumed to promote high species richness, because heterogeneous landscapes contain habitats for a larger number of species ( Benton et al. 2003 ). In contrast to our expectations, grasslands surrounded by a mixed matrix did not have the highest species richness or the flattest species–area relationship for any of the taxa. Instead, patches embedded in a mixed matrix tended to take an intermediate position regardless of whether the forest or the arable matrix had the most positive effect. However, landscape heterogeneity measures depend on the resolution of the landscape data, and both the relevant resolution and landscape categories are likely to differ between taxa. In our case we cannot exclude the possibility that for example forest dominated landscapes where more heterogeneous than the mixed ones with respect to the resolution and categories relevant for plants and butterflies. It may also be that the effect of a certain matrix land use is taxon‐specific such that species richness of that group will not change with landscape heterogeneity, but will instead be related to the proportion in the landscape of a certain land cover such as forest or arable land. It has been debated whether or not the landscape matrix land use influences species–area relationships. In a review, Watling and Donnelly (2006) found that terrestrial ‘habitat islands’ generally have a shallower species–area‐curve compared to true islands, but this was contradicted by a meta‐analysis by Drakare et al. (2006) who instead found the greatest differences in species–area relationships among habitat types, but no general difference between terrestrial habitats and islands. More recently, Prugh et al. (2008) , found differences in species sensitive to decreasing patch area between matrix types, but could not separate between effects of matrix type and habitat type. In our study, matrix land use modified the species–area relationship only in one group, hoverflies. In plants and butterflies matrix land use influenced the level of species richness in grassland patches, but not the slope of the species–area relationship. We demonstrate that matrix type can modify the effects on species richness of area and isolation in remnant grassland patches and that the responses and likely underlying mechanisms vary among taxa. This has important implications for landscape‐scale biodiversity conservation strategies. Increasing the quality of the matrix has been suggested as an alternative to habitat restoration or re‐creation ( Donald and Evans 2006 ). Our results imply that modifying the matrix could compensate for decreasing area of high‐quality habitat for some, but not all organisms. Also, since we found very different responses to matrix land use for plants and butterflies on one hand and hoverflies on the other hand, completely different actions to increase matrix quality would be needed for different taxa and there is a clear risk that management actions to increase matrix quality for one group would have negative consequences for other taxa. Most likely, there is also a large variation with respect to responses to matrix land use among species in each of the studied species groups, depending on their life‐history traits as observed for area and isolation effects ( Bommarco et al. 2010 , Öckinger et al. 2010 ). Hence, if actions to modify matrix land use should be applied in compensation for loss of natural and semi‐natural habitats we would also need to know the consequences of habitat reduction and matrix modification on the community composition, not only species richness. In order to provide more detailed recommendations for biodiversity conservation in anthropogenic landscapes, we need better understanding of the links between species life history traits, the mechanisms through which species interact with the surrounding landscape and their consequences for population dynamics and extinction risk. Acknowledgements We thank Iria Soto Embodas, Madeleine Arnqvist, Magnus Granbom and Johan Björklind Möllegård for assistance with the field work and L. Anders Nilsson and Hans D. Bartsch for help with identification of insect specimen. Lorenzo Marini and two anonymous reviewers gave valuable comments on the text. Funding for this study was provided by the EU‐projects ‘COCONUT – Understanding effects of land use changes on ecosystems to halt loss of biodiversity’ (FP6 SSPI‐CT‐2006‐044343) and ‘SCALES – Securing the Conservation of biodiversity across Administrative Levels and spatial, temporal and Ecological Scales’ (FP7 226852), and by the Swedish research council for environment, agricultural sciences and spatial planning (FORMAS).

Journal

EcographyWiley

Published: Mar 1, 2012

There are no references for this article.