Butterfly and plant specialists suffer from reduced connectivity in fragmented landscapes

Authors


Correspondence author: sabrina.brueckmann@uni-bayreuth.de

Summary

1. Calcareous grasslands are diversity hotspots for plant and butterfly species in Europe, but connectivity of these grasslands has been reduced by habitat loss and fragmentation in recent decades. Reduced habitat area leads to biodiversity loss, but the significance of habitat connectivity independent of patch size and habitat quality is unclear.

2. Here, we focus on the impact of habitat connectivity on both butterfly and plant species richness and compare (i) a connectivity index, (ii) percentage habitat cover and (iii) distance to the next suitable habitat patch as alternative measures of connectivity. Species were recorded in 2007 in northern Bavaria (Germany) in paired large and small study sites in 31 independent landscapes.

3. We found that total loss of grassland connectivity would reduce species richness of specialist butterflies (38–69%) and specialist plants (24–37%). A connectivity index combining patch size and distance in the surrounding landscape was a better measure of connectivity than percentage habitat cover or distance to the next suitable habitat.

4. Species richness, species density and abundance of habitat-specialist butterflies and plants were significantly higher in large compared with small study sites.

5.Synthesis and applications. We conclude that connectivity is highly relevant for conservation of butterfly and plant species with specialized habitat requirements, but the decision as to which connectivity measure is most appropriate depends on patch configuration, landscape context and study design. We suggest that management strategies should aim to increase connectivity by active restoration of former calcareous grasslands to ensure long-term survival of habitat-specialist species.

Introduction

Habitat loss and fragmentation are known to be major threats for local, regional and global biodiversity (Saunders, Hobbs & Margules 1991; Debinski & Holt 2000). In Europe, natural and semi-natural grasslands have faced significant reduction in area as a consequence of agricultural intensification and of the abandonment of historic land use practices, such as extensive grazing, within the last century (Krebs et al. 1999; Benton, Vickery & Wilson 2003). Today, semi-natural calcareous grasslands are highly fragmented but still of high conservation value, as they rank as the most species-rich habitat of flora and invertebrate fauna in central Europe (WallisDeVries, Poschlod & Willems 2002). Due to habitat loss and fragmentation, many species have suffered serious declines (Ewers & Didham 2006). The relationship between habitat area and species richness is called the species–area relationship, and has been demonstrated for numerous species guilds (Rosenzweig 1995). However, some guilds are particularly affected by habitat loss and fragmentation, such as species at high trophic levels, food specialists, species with poor dispersal abilities, rare species and habitat specialists (Steffan-Dewenter & Tscharntke 2000; Krauss, Steffan-Dewenter & Tscharntke 2003; Ewers & Didham 2006).

Butterflies and vascular plants are excellent model organisms for fragmentation studies, as many species are restricted to specific habitats like calcareous grasslands and persist as metapopulations (Hanski & Thomas 1994; Petit et al. 2001; WallisDeVries, Poschlod & Willems 2002; Helm, Hanski & Pärtel 2006). For these habitat specialists, the surrounding matrix is hostile. Increasing habitat fragmentation should therefore reduce species richness of specialists more than that of generalists (Krauss, Steffan-Dewenter & Tscharntke 2003; Krauss et al. 2004), but responses to habitat isolation have been inconsistent (Bruun 2000).

Habitat fragmentation can be defined in a broad sense including habitat loss, connectivity effects and edge effects, or in a stricter sense covering mainly habitat connectivity and edge effects (Fahrig 2003; Ewers & Didham 2006). According to Fahrig (2003), habitat fragmentation per se should be defined independently of habitat loss. However, in real landscapes, fragmentation is generally accompanied by habitat loss, in contrast to modelled landscapes, which allow the effects of habitat loss and fragmentation to be distinguished. Therefore, it is debatable whether habitat fragmentation stricto sensu, which mainly refers to reduced connectivity but also increased edge effects irrespective of habitat area, has a clear negative effect on biodiversity (Fahrig 2003; Ries et al. 2004). Many field studies have not found significant effects of connectivity on species richness in fragmented landscapes (e.g. Wilcox et al. 1986; Steffan-Dewenter & Tscharntke 2000; Krauss, Steffan-Dewenter & Tscharntke 2003; Krauss et al. 2004; Bisteau & Mahy 2005), although connectivity has been shown to play a key role for single butterfly species occurrence (Thomas et al. 2001; Öckinger 2006) and to be an important factor in addition to habitat quality (Dennis & Eales 1999). One reason for the lack of significant effects of connectivity on species richness might be an unsuitable study design. Connectivity may not have been the main focus of studies or may not have been independent from other factors, such as patch size, edge effects or habitat quality. Furthermore, it remains unclear whether species richness depends more on patch connectivity in small patches than in large patches. Small patches often only maintain populations close to extinction thresholds, which need to be rescued by immigration from nearby source habitats (Hanski, Moilanen & Gyllenberg 1996; Kuussaari et al. 2009). Other factors like edge effects and landscape matrix might also contribute to changes in species interactions and community dynamics (Fagan, Cantrell & Cosner 1999; Krauss, Steffan-Dewenter & Tscharntke 2003; Ries et al. 2004; Fletcher et al. 2007).

In this study, we used a landscape-scale design, where small and large sites were paired within a connectivity gradient in order to assess connectivity effects independently of patch size and other factors, similar to a Swedish study (Öckinger & Smith 2006). Moreover, we considered the whole gradient of connectivity in our study region, whereas in other studies an incomplete cover of regional variation in connectivity might have contributed to non-significant results. Finally, there are different possibilities to measure connectivity, such as the Hanski’s Connectivity Index (Hanski 1994), the percentage habitat cover or the distance to the next suitable patch (Moilanen & Nieminen 2002; Fahrig 2003; Winfree et al. 2005; Kindlmann & Burel 2008). In our study, we tested these three measurement methods in parallel, to answer the questions: ‘What are the relative merits of different indices of habitat connectivity? Which of them best predict conservation value?’ currently considered by British ecologists as one of the 100 most important and policy relevant research questions in ecology (Sutherland et al. 2006).

The hypotheses addressed in this study are:

  • 1 Large study sites have a higher species richness and abundance of butterflies and plants than small sites.
  • 2 Habitat specialists suffer more from decreasing habitat area than habitat generalists.
  • 3 Habitat-specialist butterfly and plant species benefit from increasing habitat connectivity.
  • 4 Reduced connectivity has stronger effects on species from small than from large sites, as small sites are less likely to maintain viable populations and depend more on immigration from surrounding patches.

Material and methods

Study region and study sites

The study region ‘Fränkische Schweiz’ is located in the vicinity of the town Bayreuth in northern Bavaria (southern Germany). It is characterized by a bedrock of White Jura including numerous cliffs and semi-natural grasslands (Böhmer 1994). Since the middle of the 19th century, massive losses of calcareous grasslands (up to 95%) occurred in the study region mainly due to forestation, fertilization, the abandonment of historic land use and passive succession (Böhmer 1994). In southern Germany, the most severe loss of calcareous grasslands occurred after 1960, with a reduction of more than 50% by 1990 (WallisDeVries, Poschlod & Willems 2002).

A total of 62 calcareous grasslands were selected as study sites, arranged as pairs (large and small patches) within 31 non-overlapping landscapes. The study sites were chosen to cover the whole connectivity gradient in the study region, ranging from isolated study sites to highly connected study sites within a 2-km radius. One large study site in the landscape centre (area: 2·4 ± 0·2 ha; perimeter: 1158 ± 88 m) and one small study site close by (area: 0·12 ± 0·02 ha; perimeter: 232 ± 17 m, with an average distance to the large sites of 488 ± 76 m; range 13–1400 m) were located in each landscape (Fig. 1). Landscapes were dominated by agricultural land or pastures (open habitat: 58·4 ± 2·7%) and forest (38·2 ± 2·9%) (CORINE Land Cover data 2000; http://www.eea.europa.eu/themes/landuse/clc-download). Landscapes varied slightly in the amount of open habitat and forest, but increasing area of open habitat (or decreasing forest cover) did not increase species richness of plants or butterflies (results not shown).

Figure 1.

 Illustration of the three connectivity measures used in this study; black are the focal study sites: one large and one small study sites per landscape, where the large site is in the centre of the landscape: (a) Hanski’s connectivity index (see equation in the text), d = distance from focal study site to another habitat patch (km), A = size of the habitat patch (m2); (b) the total amount of habitat in the landscape (grey) is calculated as percentage habitat cover (habitat cover for the small sites always includes the area of the large study site and habitat cover for the large sites includes the area of the small study sites; the focal study site is always excluded); (c) distance to next patch is measured.

For the landscapes where distances between large and small study sites were <50 m (n = 4), the patches were additionally separated by a barrier (e.g. forest). Patch size, patch perimeter and distances between sites were calculated with the software ArcView gis 3.2 (ESRI 1995) using orthorectified digital aerial photos from 2005 for interpretation. Within the study site categories (small or large sites), the gradient of patch size and perimeter were small (Table S1). Patch size and patch perimeter were highly correlated within small sites (r = 0·60, < 0·001) and within large sites (r = 0·87, < 0·001). Connectivity did not correlate with patch size within the categories, small or large sites (Table S2), so that connectivity could be tested independently from size and perimeter in the two habitat area categories (small vs. large sites). As small sites have small perimeters and large sites have large perimeters we cannot distinguish between area and edge effects in this study.

Connectivity and habitat predictors

We calculated and tested three different and often used connectivity measures: (i) Hanski’s Connectivity Index (=CI) (Hanski 1994); (ii) the amount/percentage of calcareous grassland in the landscapes on a 2 km radius (=%habitat cover); and (iii) the distance from the study site to the nearest calcareous grassland patch in the surrounding landscape (=distance), independently for all 62 small and large study sites. Thereby CI and % habitat cover are less precise for small compared with large patches, as only the large patches were exactly in the centre of the 2-km radius for which detailed land cover data were available.

Hanski’s Connectivity Index of each study site was calculated by measuring edge-to-edge distances between study site (separately for large and small study sites) and all other habitat patches within the 2-km radius of each landscape (Fig. 1a) using the equation:

image

where Aj is the size (in m2) of neighbouring calcareous grasslands and dij the distance (in km) from neighbouring calcareous grasslands j to the study site i (Hanski 1994). The parameter α is a measure of the dispersal ability (1/average migration distance in km) and b is a parameter, which scales the size of the surrounding habitat patches. We chose α = 0·5, as we expect an average migration distance of 2 km for our butterfly and plant communities (e.g. Moilanen & Nieminen 2002; Adriaens, Honnay & Hermy 2006). We further tested values for α = 0·3 and α = 1 (similar to Moilanen & Nieminen 2002; Krauss, Steffan-Dewenter & Tscharntke 2003; Helm, Hanski & Pärtel 2006; Adriaens, Honnay & Hermy 2006; Pöyry et al. 2009), but they resulted in highly correlated connectivity measures (pairwise Pearson correlation, all r > 0·995, all < 0·001). For the scaling parameter b, we chose = 0·5 according to the assumptions of Moilanen & Nieminen (2002), who suggested that the ratio of patch edge to patch size decreases with A0·5 when patch size increases. The connectivity gradient of our landscapes ranged from 18·9 to 1899·6 for large sites and from 103·2 to 1731·7 for small sites.

Percentage habitat cover (Fig. 1b) was measured as the amount of calcareous grassland within each landscape, and calculated separately for each large and small study site. The patch area of each respective focal study site (either large or small site) was excluded. Percentage habitat cover is a landscape connectivity measure especially suitable for landscapes with a high cover of the focal habitat (Winfree et al. 2005; Cozzi, Müller & Krauss 2008). Percentage habitat cover in our landscapes ranged from 0·01% to 2·02% calcareous grasslands for large sites and 0·20% to 2·16% for small sites.

A third measure of connectivity is the Euclidean distance of each study site to the next habitat patch (calcareous grassland) in the surrounding landscape (Fig. 1c), a method regularly used for recolonization events (Boughton 1999), for migration rates (Roland, Keyghobadi & Fownes 2000) or the importance of isolation (Bauerfeind, Theisen & Fischer 2009). The distance to the next habitat patch ranged from 4 to 1400 m, and 35% (n = 22) of the 62 study sites were <50 m from the next suitable habitat patches with a minimum area of 50 m². We therefore reanalysed the data using only study sites with distances >50 m without a barrier or 20 m with a barrier (e.g. forest), but results did not change consistently.

The Hanski’s Connectivity Index was positively correlated with percentage habitat cover for large and small study sites. Also the distance to the next patch was significantly negatively correlated with the Hanski’s Connectivity Index and percentage habitat cover (Table S2). We used habitat specialists as well as generalists in the statistical analyses with connectivity measures. Even though generalists were not expected to be restricted to the focal habitat type, or closely linked with measured connectivity values, the comparison of the two species groups provides a useful reference point.

All study sites were selected to be similar in habitat quality within their habitat category (small/large). Nevertheless, factors influencing habitat quality such as management practice, percentage bush cover and percentage flower cover (estimated by pooling the cover of flowering plants per site for the eight transects) were recorded (Table S1). Of the large study sites, 21 were managed, either by sheep grazing (20 sites) or by mowing (1 site), whereas 10 sites were not managed in the study year 2007. Management of the small study sites was not always clear, as some sites close to roads were incidentally mown and others were grazed by sheep or were fallows. Therefore, management could not be considered for small sites. None of the habitat-quality factors was correlated with connectivity measures or showed a significant effect on either butterfly or plant species richness (Tables S2 and S3).

Butterflies

Butterflies (Lepidoptera: Hesperioidea and Papilionoidea) and burnet moths (Lepidoptera: Zygaenidae) were sampled from April to the end of August in 2007 by variable visual transect walks (Krauss, Steffan-Dewenter & Tscharntke 2003; Westphal et al. 2008). Species were recorded within a 5-m corridor when weather conditions were suitable for butterfly activity (Pollard 1977). The length of transect walks was c. 800 m and the duration was 40 min on large study sites, and c. 200 m and 10 min on small study sites. Transect distance and transect time were measured with a GPS (eTrex Vista; Garmin, Múnchen, Germany). All 62 study sites were sampled eight times during summer 2007 in every second to third week. Butterfly counts were separated in 50 m sub-transects to calculate accumulation curves and species saturation. Butterfly individuals were netted and either released immediately or collected for genitalization when necessary for identification. Genitalization was necessary to distinguish between Zygaena minos and Zygaena purpuralis, between Zygaena filipendulae and Zygaena angelicae, between Melitaea britomartis, Melitaea aurelia and Melitaea athalia, and between Polyommatus icarus and Polyommatus thersites as well as between rare individuals of the family Hesperiidae. We did not distinguish between Leptidea reali and Leptidea sinapis or between Colias alfacariensis and Colias hyale. Identification and nomenclature followed Naumann, Tarmann & Tremewan (1999) for burnet moths and Settele et al. (2005) for butterflies. In the following, butterflies always include burnet moths.

Butterflies were defined as strict habitat specialists (n = 27) when the species are only found in calcareous grasslands in Bavaria, according to Stettmer et al. (2007) and confirmed by Weidemann (1995). Butterfly species not specialized for calcareous grasslands were defined as generalists (n = 62), even though this category includes other grassland or forest specialists (Table S4). Butterflies were summed up from the eight conducted butterfly surveys per site on the basis of the transect walks and abundance was calculated as density per 100 m2.

Plants

Vascular plants were recorded twice, in May and August 2007. Surveys were conducted by two different botanists: one covering the spring period, and one the summer period. On the large sites, plant species were recorded in 16 random 2-m2 plots, whereas plant species on the small sites were recorded in 4 random 2-m2 plots to cope with the two different size classes of the study sites. Species identification and nomenclature followed Rothmaler (1999). Plants were divided into strict habitat specialists and generalists by a local expert, in agreement with Gerstberger & Vollrath (2007). As with butterflies, plant specialists were defined as strict specialist species when restricted to calcareous grasslands (n = 102) and as generalists when they have no or other habitat preferences (n = 306) (Table S4). Plants were summed from the two plant surveys for each site and plant abundance was calculated as the density of plants per 1 m2.

Statistical analysis

The statistical analyses were performed using the software r 2.9.1 (R Development Core Team 2004). The explanatory variables were: connectivity, habitat area and factors of habitat quality (bush cover, flower cover and management) plus patch size within the two habitat area categories. Transformation of the explanatory variables was not necessary. The response variables were: species richness (area adjusted sample size), species richness (estimated), species density (equal sample size) and species abundance of habitat specialized butterflies and plants as well as butterfly and plant generalists. The response variables met the assumptions of normality and homoscedasticity in the statistical models and were therefore not transformed. In linear mixed effects models, landscape identity was included as a random factor, as large and small study sites were nested within the 31 landscapes. Explanatory variables entered the mixed effects models in the following sequence: (i) connectivity predictor (Hanski’s Connectivity Index, percentage habitat cover or distance to next patch); (ii) habitat area (small vs. large sites); and (iii) interaction of habitat area (small vs. large sites) with the connectivity predictor (Hanski’s Connectivity Index, percentage habitat cover or distance to next patch). As interactions were not significant, the interaction terms were removed from the final models. To compare the importance of the three connectivity measures, we used Akaike Information Criterion for small sample sizes AICc (library bbmle in r) and ranked models with small AICc as better than models with large AICc values (Burnham & Anderson 2002).

Species accumulation curves and species richness estimators were calculated using the software EstimateS version 8 (Colwell 2006). Accumulation curves were asymptotic for large and small patches indicating sufficient sample size for all study sites (Fig  S1). With our study design using subunits for species records, we calculated an estimated species richness per study site separately for butterflies and plants. We used the estimator ACE and divided the recorded species richness by estimated species richness to obtain the species saturation per site. A paired t-test revealed that butterflies, but not plants, showed a significantly higher saturation on small sites compared with large sites. Therefore, we also present estimated species richness for butterflies. For species density relationships, the sample size of large sites was reduced to the same sampling effort as small sites (first 4 transect units or first 4 plot units).

Multivariate ordination methods with nonmetric multidimensional scaling (NMDS) were used to reveal community patterns in the study sites using the r package Vegan (Oksanen 2009). We used ‘Bray-Curtis’ dissimilarities as dissimilarity indices between sites. Arithmetic means ± 1 standard error are given in the text.

Results

In total, 89 butterfly and 408 plant species were identified on the 62 calcareous grasslands. On the large patches, a total of 88 butterfly and 379 plant species were found with 22 870 individual butterflies and 3741 plant records; on the small patches, we recorded 73 butterfly and 296 plant species with 3805 individual butterflies and 2185 plant records.

On average, 45·3 ± 1·1 butterfly species (range: 31–57) and 120·7 ± 2·8 plant species (range: 96–150) were found on the large patches and 25·2 ± 1·4 butterfly species (range: 12–42) and 70·5 ± 1·9 plant species (range: 48–89) were found on the small patches. Large study sites contained 31% specialized butterfly species and 69% generalists respectively, whereas small patches supported just 19% specialists and 81% generalists. The differences were less obvious for plant species with 26% specialist species in large patches and 25% specialists in small patches.

Connectivity

None of the interactions between the connectivity measures and habitat area were significant in the mixed effects models (Table 1) indicating that butterflies and plants from large patches reacted similarly to those from small patches (equal slopes). We therefore excluded the interaction terms from all models (Table 1). As species-richness patterns were not significantly related to any of the habitat-quality measurements (Table S2), connectivity effects could be tested independently of habitat quality.

Table 1.   Mixed effect models for butterfly and plant species separately for habitat specialists and generalist species
  ButterfliesPlants
 Mixed effects modelMixed effects model
 F1,29PAICc F1,29PAICc
  1. +, Relationship: large patches > small patches or positive slope; −, relationship: small patches > large patches (occurs only for generalist plant species); CI, connectivity index; distance, distance to next suitable patch; habitat cover, % of calcareous grassland cover; habitat area, small or large study site.

Habitat specialists
Species richness (area adjusted sample size)CI11·820·002 +5·30·029 +
Habitat area119·06<0·001295·77+135·53<0·001429·03+
CI × habitat area ns   ns  
Habitat cover0·20·659  1·020·32  
Habitat area121·25<0·001303·37+133·34<0·001429·71+
Habitat cover × habitat area ns   ns  
Distance3·090·09  3·120·088  
Habitat area115·91<0·001306·68+130·2<0·001434·47+
Distance × habitat area ns   ns  
Species richness (estimated, ACE)CI10·140·004 +    
Habitat area24·84<0·001386·45+    
CI × habitat area ns      
Habitat cover1·920·177      
Habitat area26·18<0·001392·03+    
Habitat cover × habitat area ns      
Distance3·870·059      
Habitat area22·75<0·001393·36+    
Distance × habitat area ns      
Species density (equal sample size)CI15·09<0·001 +4·810·037 +
Habitat area37·84<0·001286·56+5·480·026435·91+
CI × habitat area ns   ns  
Habitat cover2·40·132  6·370·017 +
Habitat area38·22<0·001295·14+7·140·012433·13+
Habitat cover × habitat area0ns   ns  
Distance1·980·17  1·090·306  
Habitat area33·46<0·001299·88+4·830·036440·08+
Distance × habitat area ns   ns  
AbundanceCI4·290·047 +2·80·105  
Habitat area14·83<0·00151·4+8·590·007442·44+
CI × habitat area ns   ns  
Habitat generalists
Species richness (area adjusted sample size)CI4·990·033 +0·040·847  
Habitat area127·51<0·001379·27+116·22<0·001478·85+
CI × habitat area ns   ns  
Habitat cover1·730·198  3·430·074  
Habitat area120·49<0·001386·61+112·28<0·001479·04+
Habitat cover × habitat area ns   ns  
Distance1·270·268  0·020·902  
Habitat area123·4<0·001386·02+115·77<0·001479·01+
Distance × habitat area ns   ns  
Species richness (estimated, ACE)CI7·420·011 +    
Habitat area58·77<0·001424·4+    
CI × habitat area ns      
Habitat cover0·010·905      
Habitat area54·55<0·001432·4+    
Habitat cover × habitat area ns      
Distance10·326      
Habitat area52·78<0·001432·92+    
Distance × habitat area ns      
Species density (equal sample size)CI4·040·054 +2·380·134  
Habitat area7·420·011372·53+4·380·045478·27
CI × habitat area ns   ns  
Habitat cover0·030·867  2·490·126  
Habitat area6·730·015377·09+5·340·028477·27
Habitat cover × habitat area ns   ns  
Distance1·090·306  0·920·347  
Habitat area6·770·015376·09+3·970·056480·14 
Distance × habitat area ns   ns  
AbundanceCI1·710·201  1·870·182  
Habitat area15·470·00181·29+6·130·019444·24
CI × habitat area ns   ns  

Increasing connectivity, calculated as Hanski’s Connectivity Index, had a significantly positive effect on species richness of specialized butterflies and plants (Fig. 2a,c) and also increased the species richness of generalist butterflies (Table 1). Based on this relationship, on large study sites, 38% of specialized butterfly species and 24% of specialized plant species would be lost if all surrounding habitat patches at a 2-km scale were destroyed. For small study sites, up to 69% of specialist butterfly and 37% of specialist plant species could be lost (calculations based on regressions from Fig. 2a,c). Connectivity effects were similar for species richness, estimated species richness and species density (Table 1). The abundance of specialized butterfly species increased with increasing connectivity, but neither the abundance of generalist butterfly species nor that of plants was affected by connectivity (Fig. 2b, Table 1).

Figure 2.

 Butterfly specialist species richness; butterfly specialist abundance and plant specialist species richness in relation to Hanski’s connectivity index: (a) butterfly specialist species richness from large (y = 0·002962x + 9·04) and small sites (y = 0·002962x + 2·28); (b) butterfly specialist abundance from large (y = 0·000229x + 0·47) and small sites (y = 0·000229x + 0·13); (c) plant specialist species richness from large (y = 0·006410x + 39·66) and small sites (y = 0·006410x + 18·66).

Comparing the three alternative connectivity measures, only Hanski’s Connectivity Index was significant for species-richness patterns for specialists: these models also had a lower AICc value than models with habitat cover or distance to next patch (Table 1). For plant species density, habitat cover was a similar good predictor as Hanski’s Connectivity Index (Table 1).

Habitat area

Species richness, estimated species richness, species densities, and abundance of both specialist and generalist butterflies were significantly higher in large compared with small study sites (Table 1, Fig. 3; Fig. S2). Species richness and species density of plant specialists as well as species richness of plant generalists were also significantly higher in large sites, but species density of plant generalists and abundance of plant generalists were higher on small sites (Table 1, Fig. 3; Fig. S2).

Figure 3.

 Species richness, estimated species richness (only for butterflies), and species density of butterfly and plant specialists and generalists in large (grey) vs. small (white) study sites. Species richness: (a) butterfly specialists and (b) butterfly generalists. Estimated species richness: (c) butterfly specialists and (d) butterfly generalists. Species density: (e) butterfly specialists and (f) butterfly generalists. Species richness: (g) plant specialists and (h) plant generalists. Species density: (i) plant specialists and (j) plant generalists. All figures show significantly different means (statistics see Table 1).

We found 64% fewer butterfly specialists and 40% fewer butterfly generalists as well as 47% fewer plant specialists and 38% fewer plant generalists on small compared with large sites (Fig. 3a,b). Patterns were similar for estimated species richness for butterflies (Fig. 3c,d). Also species density patterns were similar, showing 45% lower butterfly specialist density and a 12% lower butterfly generalist density as well as a 16% lower plant specialist density (Fig. 3e,f,i). However, plant generalist density was 14 % higher on small compared with large sites (Fig. 3j).

Nonmetric multidimensional scaling ordinations confirmed our results and showed that large patches and small patches differed in community patterns. Thereby, small patches were more dissimilar than large patches and the patterns were clearer for butterflies than for plants. For community composition, the habitat area (small vs. large sites) was more relevant than connectivity (Fig. S3).

Discussion

It is important to know which local and landscape factors are necessary for the conservation of habitat-specialist species. For example, the arrangement and size of patches might be important for species persistence. It has been suggested that habitat quality and area are relevant, as well as connectivity and landscape context (e.g. Dennis & Eales 1999; Stefanescu, Herrando & Paramo 2004; Dover & Settele 2009). Our study makes three relevant contributions: first, we show that habitat connectivity is of particular importance for specialized butterfly and plant species. Secondly, a comparison of different connectivity measures reveals that an index combining neighbouring patch area and distance is more suitable than using cover of habitat or distance to the next habitat. Thirdly, we find clear differences in the species richness, species composition and proportion of habitat specialists in small vs. large habitat patches, suggesting that small patches have already lost most butterfly and plant specialists.

Effects of connectivity

Connectivity of habitats is an important determinant of species richness and abundance of specialized species in our study. These findings are in contrast to previous studies on plant and butterfly species richness (Wilcox et al. 1986; Steffan-Dewenter & Tscharntke 2000; Krauss, Steffan-Dewenter & Tscharntke 2003; Krauss et al. 2004; Bisteau & Mahy 2005). This might be due to the focus of these studies on species–area relationships, to intercorrelations between area and connectivity, or to a partial coverage of regional connectivity gradients. Other studies showed marginally significant effects of connectivity on plant species richness (Adriaens, Honnay & Hermy 2006), or an impact of historical landscape connectivity on present day plant diversity (Lindborg & Eriksson 2004), indicating a slow response of plant species richness to fragmentation (Helm, Hanski & Pärtel 2006). Thus, delayed extinction of local plant populations might partly mask the relevance of habitat connectivity for long-living species like plants (Kuussaari et al. 2009). Indeed, we found more severe species loss for butterflies than for plants with decreasing connectivity.

This might also be caused by different dispersal mechanisms between plants and butterflies. Seed dispersal to new habitats takes place passively by wind or animal transport but this might be lacking if calcareous grasslands are not regularly grazed by sheep (Wessels et al. 2008). In contrast, butterflies are active dispersers, have a visual orientation and can choose where to fly (Hambäck et al. 2007).

Even though small sites might not be able to maintain viable populations and thus rely more on immigration from surrounding habitat patches than large sites (Öckinger & Smith 2006), we found no differences between large and small study sites in species–connectivity relationships (no significant interactions). The generally strong connectivity effects in our data underline the importance of immigration events for the long-term survival of species with specialized habitat requirements, not only in small but also in relatively larger patches.

Connectivity measures

We found strong effects of Hanski’s Connectivity Index on species richness and species density of specialized butterflies and plants. Percentage habitat cover was also a suitable connectivity measure for specialized plant species density, whereas distance to the next habitat patch was generally a weak predictor. In our landscapes, many small habitat patches were scattered around the focal study sites so that distances to the next habitat patch were often small, and furthermore these small patches might not represent sources for immigration (Boughton 1999). Therefore, it is reasonable that a connectivity index that combines distance and patch area of neighbouring habitats is more suitable to quantify differences in habitat connectivity. However, for extremely fragmented landscapes or landscapes with a clear mainland–island situation, the Euclidian distance to next habitat patch can be useful and has been successfully applied (Winfree et al. 2005). Habitat cover is a good connectivity measure for situations with large proportions of habitats nearby (Winfree et al. 2005). In a study in the Swiss Alps, habitat cover ranged from 0% to 27% on landscape scales between 500 and 4000 m radii, and was a good predictor for butterfly species occurrence (Cozzi, Müller & Krauss 2008). However, habitat cover includes neither the distances from the focal study sites to all the habitat patches in the landscape, nor a scaling of size for the adjacent habitat patches, giving all habitats within a certain radius the same weight (Moilanen & Nieminen 2002). In our study region with a habitat cover gradient from 0·01% to 2·2%, Hanski’s Connectivity Index was the best predictor. Further connectivity measures exist, taking functional responses of species into account, such as dispersal ability and the probability of dispersing between habitat patches (Kindlmann & Burel 2008) or the spatial arrangement of habitat patches in the surrounding landscape (Matisziw & Murray 2009). These new indices will definitely contribute to connectivity research for single species studies. However, for community studies, we suggest the use of somewhat simpler connectivity measures depending on patch configuration, landscape context and study design.

Specialists and generalists

In contrast to habitat specialists, there was little impact of connectivity on butterfly and plant species that are not specialized on calcareous grasslands. For reasons of simplicity, we call these species ‘generalists’. As generalists can occur in the matrix, an effect of connectivity was not expected.

Habitat area was a strong predictor for not only habitat specialist but also generalist species. However, specialists lose a higher proportion of species, when habitat area is lost. This might be explained by edge effects where generalists profit from a higher proportion of habitat edges when habitat patch size is decreasing (Laurance & Yensen 1991), whereas edges are less-preferred by habitat specialists (Ries & Sisk 2008). Habitat-specialist butterflies have been shown to suffer more from habitat loss than generalists (Warren et al. 2001; Krauss, Steffan-Dewenter & Tscharntke 2003). The seeds of habitat-specialist plant species need specific conditions to survive. For example, an influx of nitrogen from habitat edges might decrease species richness of specialized plants, whereas generalist species might survive better at edges and consequently show a higher dominance in small habitat patches (Jacquemyn, Brys & Hermy 2003). Such edge effects might explain why generalist plants in our study showed higher species densities and abundances in small compared with large patches.

Habitat area and quality

The species richness of both specialists and generalists depended on habitat area, which confirms the general validity of the species–area relationship (e.g. Wilcox et al. 1986; Rosenzweig 1995; Wettstein & Schmid 1999; Bruun 2000; Zschokke et al. 2000; Krauss, Alfert & Steffan-Dewenter 2009). The positive relationship between the abundance of butterflies and increasing habitat area has also been shown before (e.g. Krauss, Steffan-Dewenter & Tscharntke 2003). However, our results show that specialist species density (equal sample size) is higher on large compared with small patches. Sampling effort should therefore play a minor role, giving more importance to ‘area per se’ and ‘habitat heterogenity’ explanations for our species–area relationship (Ouin et al. 2006; Krauss, Alfert & Steffan-Dewenter 2009). Species density–area relationships are less frequently studied than species–area relationships (but see e.g. Wettstein & Schmid 1999; Krauss et al. 2004; Lindborg & Eriksson 2004).

Nonmetric multidimensional scaling ordination confirmed the finding that large sites differ from small sites in their community composition and that habitat area played a major role. Butterfly communities differed clearly between large and small sites whereas this difference was less distinct for plant species as smaller sites exhibited very heterogeneous species composition.

Habitat-quality factors did not have a significant effect on plant and butterfly species richness, but the study design was chosen to keep quality between study sites as similar as possible. However, large and small sites did differ in habitat quality and at least some of the variance explained by the area of small vs. large sites could be related to differences in habitat quality. Habitat quality was not the main focus of this study, even though its importance has been often highlighted (e.g. Dennis & Eales 1999; Adriaens, Honnay & Hermy 2006; Kuussaari et al. 2007; Raatikainen, Heikkinen & Pykälä 2007). We also found no effect of management (grazed/mown vs. fallows) on species richness. Similarly in Swiss grasslands, butterfly species diversity did not differ between early fallows (no management for 2–3 years) and managed calcareous pastures (Balmer & Erhardt 2000). Unmanaged grasslands in our region were similar to early fallows. In addition, as our study focused on the community level and not on single species, habitat quality for species communities is difficult to define and might be of less overall importance due to variability in habitat requirements.

Conclusions

Habitat-specialist butterfly and plant species were highly dependent on connectivity and habitat area. The best connectivity measure in our study region was a connectivity index, which takes into account the area of all habitat patches with their distances to the focal habitat. In contrast, (i) distance to the next habitat patch and the slightly better (ii) amount of habitat in a landscape were both unsuitable connectivity indicators for our study region. We suggest that distance to the next habitat might be a good connectivity measure for distinct mainland–island situations and in highly fragmented landscapes with few remaining habitat patches, which contain viable populations (source habitats). The proportion of habitat within a landscape of a specific radius is a reasonable measure, if the proportion of habitat is relatively high. However, for most landscapes with intermediate to high fragmentation levels, Hanski’s Connectivity Index is probably the best approach to detect connectivity effects on species richness (Winfree et al. 2005; Cozzi, Müller & Krauss 2008). We found that if all habitat patches in our landscape were lost except for the focal study site, there would be a loss of 38–69% of specialized butterfly species and 24–37% of specialized plant species. Thus, disruption of habitat connectivity would lead to significant future extinctions of species in addition to extinctions due to habitat loss per se. Conservation management should therefore seek to improve connectivity at a landscape scale in addition to the protection and adequate management of conservation areas at a patch scale. We recommend conserving large grasslands and suggest regular grazing or mowing of these to keep the habitat quality suitable for butterfly and plant species that are specialized on calcareous grasslands. We also encourage active restoration of patches that have once been calcareous grasslands (e.g. by removing trees and bushes) in order to increase habitat connectivity. This will be particularly important for mitigation of possible extinction debt and the long-term survival of habitat specialists in highly fragmented semi-natural grasslands in Europe.

Acknowledgements

We thank the editors, Isabell Karl, Annette Leingärtner, Kathrin Wagner and four anonymous reviewers for helpful comments on the work; Emily Martin for checking the English, Pedro Gerstberger, Gerhard Hübner, Manfred Rauh, Jonathan Heubes and Julia Laube for taxonomic support; Björn Reineking and Andrea Holzschuh for statistical support; and Thomas Nauss for help with Corine Land Cover analysis. This project was supported by the EU 6FP project COCONUT (Understanding effects of land use changes on ecosystems to halt loss of biodiversity due to habitat destruction, fragmentation and degradation; Contract Number 2006-044346 to ISD and JK; http://www.coconut-project.net).

Ancillary