How does landscape context contribute to effects of habitat fragmentation on diversity and population density of butterflies?


*Jochen Krauss, Agroecology, University of Göttingen, Waldweg 26, D-37073 Göttingen, Germany. E-mail:


Aim  Studies on habitat fragmentation of insect communities mostly ignore the impact of the surrounding landscape matrix and treat all species equally. In our study, on habitat fragmentation and the importance of landscape context, we expected that habitat specialists are more affected by area and isolation, and habitat generalists more by landscape context.

Location and methods  The study was conducted in the vicinity of the city of Göttingen in Germany in the year 2000. We analysed butterfly communities by transect counts on thirty-two calcareous grasslands differing in size (0.03–5.14 ha), isolation index (2100–86,000/edge-to-edge distance 55–1894 m), and landscape diversity (Shannon–Wiener: 0.09–1.56), which is correlated to percentage grassland in the landscape.

Results  A total of 15,185 butterfly specimens belonging to fifty-four species are recorded. In multiple regression analysis, the number of habitat specialist (n = 20) and habitat generalist (n = 34) butterfly species increased with habitat area, but z-values (slopes) of the species–area relationships for specialists (z = 0.399) were significantly steeper compared with generalists (z = 0.096). Generalists, but not specialists, showed a marginally significant increase with landscape diversity. Effects of landscape diversity were scale-dependent and significant only at the smallest scale (landscape context within a 250 m radius around the habitat). Habitat isolation was not related to specialist and generalist species numbers. In multiple regression analysis the density of specialists increased significantly with habitat area, whereas generalist density increased only marginally. Habitat isolation and landscape diversity did not show any effects.

Main conclusions  Habitat area was the most important predictor of butterfly community structure and influenced habitat specialists more than habitat generalists. In contrast to our expectations, habitat isolation had no effect as most butterflies could cope with the degree of isolation in our study region. Landscape diversity appeared to be important for generalist butterflies only.


Most species live in habitat islands of fragmented landscapes (Wiens, 1997; Hanski, 1999). Habitat area (Connor et al., 2000; Zschokke et al., 2000), habitat isolation (Hanski et al., 1994; Kruess & Tscharntke, 1994; Steffan-Dewenter & Tscharntke, 1999; Ricketts et al., 2001; Thomas et al., 2001) and habitat quality including effects of management and age (Feber et al., 1996; Gonzalez, 2000; Swengel & Swengel, 2001; Thomas et al., 2001) are known to affect species richness and population densities. Unlike oceanic islands, where the marine environment is a clear isolation barrier, the matrix of the surrounding landscape may contribute to the survival of populations in terrestrial habitat islands (Marino & Landis, 1996; Jonsen & Fahrig, 1997; Thies & Tscharntke, 1999; Weibull et al., 2000; Steffan-Dewenter et al., 2002).

Species–area relationships have been shown for many taxonomic groups (Rosenzweig, 1995), including butterflies (Wilcox et al., 1986; Baz & Garcia-Boyero, 1995; Robertson et al., 1995; Wettstein & Schmid, 1999; Steffan-Dewenter & Tscharntke, 2000; Zschokke et al., 2000). However, comparative studies on species groups differing in ecological traits are rare (Harrison & Bruna, 1999). Population density studies for insects are also not common, and show that densities can decrease or increase with increasing habitat area (Connor et al., 2000; Matter, 2000; Steffan-Dewenter & Tscharntke, 2000).

Most studies on fragmentation of terrestrial habitats ignore the potential influence of the surrounding landscape on species richness and population density, and focus on area and isolation only (Wiens, 1997; Hanski, 1999; Vandermeer & Carvajal, 2001; Tscharntke et al., 2002). However, many grassland butterflies exist as metapopulations in fragmented habitats and may benefit from adjacent vegetation depending on the matrices of the surrounding landscapes (Thomas & Hanski, 1997). Habitat specialists have been predicted to be particularly affected by habitat fragmentation (Thomas et al., 1992; Warren et al., 2001), whereas generalists should be more strongly affected by the surrounding landscape diversity (Jonsen & Fahrig, 1997). Landscape analyses have found negative effects for species richness with increasing habitat isolation (Wettstein & Schmid, 1999; Ricketts et al., 2001) and variable effects with landscape context, as different species may respond at different spatial scales (e.g. Jonsen & Fahrig, 1997; Weibull et al., 2000; Steffan-Dewenter et al., 2001, 2002).

In the temperate zone, butterfly species richness is known to be high for grasslands (Oates, 1995; Pfaff & Wolters, 1999). Calcareous grasslands rank as the most species-rich habitat for butterflies in Europe (Van Swaay, 2002). In this study, we tested whether habitat area, habitat isolation, habitat quality or landscape context affect species richness and population density of specialist and generalist butterflies and burnets on calcareous grasslands. The following predictions were tested:

  • 1Species richness increases with increasing habitat area, decreasing habitat isolation and increasing habitat quality.
  • 2Species richness increases with increasing habitat diversity of the surrounding landscape.
  • 3Total butterfly and burnet densities, and densities of single species, depend on habitat area, habitat isolation and habitat quality, and are affected by landscape context.
  • 4Habitat specialists are more sensitive to habitat area and isolation, whereas generalists are more sensitive to landscape context.

Materials and methods

Study region and study sites

A total of thirty-two calcareous grasslands in the vicinity of the city of Göttingen in Lower Saxony (Germany) were studied (Fig. 1). The study sites were chosen to cover the full gradient of habitat area and isolation in the study region. Calcareous grasslands cover only 0.26% of the area in the study region and are sharply delimited from the surrounding landscape matrix. The natural fragmentation of the semi-natural calcareous grasslands increased greatly because of intensification of agricultural development during the last decades in Germany (WallisDeVries et al., 2002). In Lower Saxony the total area of calcareous grasslands is much lower and probably more fragmented than in southern Germany (WallisDeVries et al., 2002). Compared with northern Europe, our study region might be less fragmented. The study sites were neither grazed nor mown during the sampling period in the year 2000 and belong to the plant association Gentiano-Koelerietum. Our study region includes a total of 285 calcareous grasslands with an area of 5.01 km2 (unpublished data following official maps from the ‘Niedersächsisches Landesamt für Ökologie’, ‘Untere Naturschutzbehörden Stadt und Landkreis Göttingen, Landkreis Northeim und Landkreis Heiligenstadt’, and ‘Regierungspräsidium Kassel’).

Figure 1.

Location of the thirty-two calcareous grasslands studied, and all further calcareous grasslands around the city of Göttingen (the city is in the centre of the map). Each grassland is marked by a circle (with 500 m radius), and the study sites are numbered from the largest (1) to the smallest (32).


Butterflies (Lepidoptera: Hesperioidea and Papilionoidea) and burnets (Lepidoptera: Zygaenidae) were sampled in the year 2000 (26 April–24 August) by visual counts along five randomized transect walks through each of the thirty-two study sites. In the following study, butterflies always included burnets. The number of species and individuals were recorded within a 5 m corridor when weather conditions were suitable for butterfly activity (see Pollard, 1977; Erhardt, 1985). All study sites were sampled in a randomized sequence every 3–4 weeks within the total sampling time. During each of the five sampling periods all thirty-two study sites were sampled within 6–11 days, to minimize phenology effects. To achieve adequate sample sizes, transect time varied from 20 to 60 min, depending on the size of the grassland. Transect time for twelve small grasslands (314–1326 m2) was 20 min per transect walk, for twelve intermediate grasslands (1914–7887 m2) 40 min, and for eight large grasslands (11,528–51,395 m2) 60 min. Counts were conducted in 5 min intervals to calculate species accumulation curves (following Colwell, 1997, see below). Transect distance was measured during each transect walk with an electronic step counter allowing calculations of butterfly densities per 100 m2 for all thirty-two study sites. Species numbers and population densities from the five transect walks were pooled for each study site over the sampling period.

We defined butterfly species as habitat specialists (n = 20) for calcareous grassland when they are found in Lower Saxony almost exclusively on calcareous grasslands (based on unpublished data from the known distribution of butterflies of Lower Saxony, personal communication Hans Joger ‘Niedersächsisches Landesamt für Ökologie’, Hildesheim). Most of these species were also recorded in Zub (1996) and Settele et al. (1999) as calcareous grassland specialists or with strong preferences for this habitat type. However, some of these specialists can inhabit additional habitats in other regions of Germany or Europe; although in Lower Saxony they are restricted to calcareous grasslands because of larval food plant limitation or climatic tolerances. We defined butterflies as habitat generalists (n = 34) when they are ubiquitous species or species with preferences for other habitat types, following Zub (1996) and Settele et al. (1999). Similarly Warren et al. (2001) classified British butterflies into two groups as wider-countryside or habitat specialist species.

Habitat area, isolation, quality and the landscape context

Habitat area

The area of the thirty-two calcareous grasslands was measured in the year 2000 with a differential GPS GEOmeter 12L (GEOsat GmbH, Wuppertal, Germany) and ranged from 314–51,395 m2. The area of former grassland, which was completely covered by shrubs, was excluded in this measurement.

Habitat isolation

To calculate the isolation of grasslands we checked the 285 officially mapped calcareous grasslands in the region. Of these, 222 of them were still grasslands, whereas sixty-three were covered by shrubs and trees because of lack of management, and were excluded for isolation calculations in this study. Estimation of habitat area from the official maps with exclusion of the area covered by shrubs was closely correlated with our GPS measurements (n = 32, r = 0.936, P < 0.0001) for the thirty-two study sites. Habitat isolation (I) of each study site (i) was measured from edge-to-edge and took into account all known calcareous grasslands in a radius of 8 km around our study sites, using the following formula:


where Aj is the size (in m2) of neighbouring calcareous grasslands and dij the distance (in km) from the neighbouring grassland j to the study site i. The parameter a determines the effect of distance on isolation and the parameter b, the scaling of immigration. The formula is based on Hanski et al. (2000), and was used in a simpler form with a = 1 and b = 1 by Hanski et al. (1994) and Steffan-Dewenter & Tscharntke (2000). Larger values of the isolation index I indicate lower isolation (and better connectivity) than smaller values. We tested dij multiplied with the factors a = 1 and a = 3, and Aj with the exponents of b = 1, 0.2 and 0.5 (plus all combinations) to take into account the possibility of different dispersal distances and immigration rates of butterflies (see Moilanen & Nieminen, 2002). The resulting habitat isolation indices were similar and closely correlated (r > 0.6). We also tested the distance to the nearest calcareous grassland, following Thomas et al. (1992, 2001). Isolation index (a = 1, b = 1) and distance were correlated (r = −0.44, P = 0.012) and showed very similar results with respect to the butterfly communities. Log10-transformed indices and distances were tested for specialist, generalist and total butterfly species richness and density and always gave non-significant results. Therefore, we only show results of the isolation index with a = 1 and b = 1.

Landscape context

Landscape context was analysed using modified digital thematic maps (ATKIS®-DLM 25/1 Landesvermessung & Geobasisinformationen Niedersachsen 1991–1996, Hannover, Germany, and ATKIS®-DLM 25/2 Hessisches Landesvermessungsamt 1996, Kassel, Germany). In total, the study region covered an area of c. 1944 km2. The mainly semi-natural land-use types grassland (12.14%), garden land (0.31%), hedgerows (0.30%), calcareous grasslands (0.26%), orchard meadows (0.20%) and fen (0.05%) were pooled and defined as ‘grasslands’ (13.26%), because these habitats were assumed to be the most suitable nectar habitats for adult butterflies. The remaining land-use types contained arable land (42.15%), forest (36.80%), built-up area (6.24%), other habitats (1.48%) and plantations (0.06%) (see Steffan-Dewenter et al., 2001, 2002). The proportion of the defined grassland habitats showed no significant relationships and was correlated with landscape diversity (see Results).

We used the Shannon–Wiener index to calculate landscape diversity following Thies & Tscharntke (1999) and Steffan-Dewenter et al. (2002) as


where pi is the proportion of each of the mentioned eleven land-use types (Krebs, 1989). Landscape context was quantified for each of the thirty-two grassland fragments using a nested set of circles around the center of the study sites from 0.25 to 3.00 km in 0.25 km steps. As a result of a few missing ATKIS values, we could test the landscape context for a radius of 0.25 and 0.50 km for all thirty-two study sites, whereas 0.75–2.00 km scales were tested for only thirty-one sites, 2.25 km for thirty sites, and 2.50–3.00 km for twenty-nine sites. For each landscape analysis, habitat area of the study site was excluded to reduce the habitat area effect.

Habitat quality

Habitat quality was quantified after each transect walk (i) by estimating the proportion of area covered by plants in flower within the transect area to quantify nectar resources and (ii) by identifying all plant species in flower inside the transect area as an indicator of larval food plant availability. The number of plant species in flower per study site was highly correlated with the number of all vascular plant species per study site (n = 31, r = 0.922, P < 0.0001) (J. Krauss, unpublished data).

Statistical analysis

The statistical analyses were performed using the software Statgraphics Plus for Windows 3.0 (Statgraphics, 1995). All data were tested to see if they satisfy the assumption of normality. We calculated simple and multiple regressions, Pearson and Spearman rank correlations, multiple logistic regressions and comparison of regression lines (Sokal & Rohlf, 1995). We chose stepwise backward elimination for multiple regressions. The independent variables habitat area and isolation were always log10-transformed and cover of plant species in flower was arcsine-square root transformed, whereas the proportion of grassland habitats in the surrounding landscape was normally distributed. Species numbers in regressions were also log10-transformed to calculate slopes of z-values to compare them with other studies. Arithmetic means of non-transformed values ± one standard error are given in the text.

We calculated the species richness estimator Abundance-based Coverage Estimator (ACE) of species richness, computed by EstimateS, Version 5 (Colwell, 1997) to indicate the percentage of sampled species in relation to estimated species richness per habitat, showing the saturation of species richness. To avoid effects of season-dependent species turnover, we pooled the first 5 min of all five transect walks per habitat to a first 25 min interval (first step), the next 5 min of the transects were pooled to a second 25 min interval (second step) and so on. Small habitats with 20 min transect walks have therefore four steps, intermediate habitats eight steps and large habitats twelve steps to calculate the estimated species richness.


In total, we recorded forty-eight butterfly species and six species of burnets, comprising a total of 15,185 individuals on all thirty-two calcareous grasslands. Only Pieris rapae (L.) and P. napi (L.) could be found on all calcareous grassland fragments (Appendix). The most abundant species were the satyrid Maniola jurtina (L.) (22.9% of all individuals), and the lycaenids Polyommatus coridon (Poda) (18.5%) and P. icarus (Rottemburg) (8.1%).

Table Appendix.  Number of individuals (NoI) of the fifty-four butterfly species and the number of inhabited habitat fragments (NoH) out of thirty-two calcareous grasslands (sampled in the year 2000). Multiple logistic regression for presence–absence data and Spearman rank correlations for species density are shown. The three independent habitat factors were tested: habitat area (A), habitat isolation (I), and landscape diversity (L) as the Shannon–Wiener Index calculated at a 250-m radius level. M (e) is the maximum likelihood estimate, M (P) is the significance level, M (%) is the percentage of deviance explaining the model, (e) stands for maximum likelihood estimate, and (r) for the Spearman rank correlation coefficient
TaxonNoINoHPresence or absence data logistic regression modelsPopulation densities Spearman rank correlations
A (e)I (e)L (e)M (e)M (p)M (%)A (r)I (r)L (r)
  1. Significance levels are: ***P < 0.001, **P < 0.01, *P < 0.05, (*) P < 0.1, n.s. = not significant.
    Note: Positive values of A (e, r) mean that the species occurrence is positively related to larger habitats, of I (e, r) that they profit from less isolation (better connectivity), and L (e, r) that they profit from higher landscape diversity. Only species with occurrence in >1 and <32 habitats were used for the multiple logistic regressions, and only species with a minimum of ten individuals for the Spearman rank correlations. Habitat area and habitat isolation are log10-transformed. In order to avoid too many type II errors, we did not apply Bonferroni corrections, thereby accepting to fall below the α level of significance by chance in some cases.

Polyommatus coridon (Poda, 1761)2815233.70***3.32*n.s.−25.050.000240.810.63***n.s.n.s.
Coenonympha arcania (Linnaeus, 1761)36417n.s.n.s.3.73*−4.040.01413.69n.s.n.s.0.31(*)
Zygaena carniolica (Scopoli, 1763)344194.92***3.60*n.s.−31.33<0.000154.050.59***n.s.n.s.
Erynnis tages (Linnaeus, 1758)315243.97***n.s.n.s.−11.540.000142.610.50**n.s.n.s.
Zygaena purpuralis (Brünnich, 1763)283131.79**n.s.n.s.−6.740.00518.570.46**n.s.n.s.
Cupido minimus (Fuessly, 1775)24982.32**n.s.n.s.−9.700.00225.710.50**n.s.n.s.
Zygaena viciae ([Denis & Schiffermüller], 1775167141.44*n.s.n.s.−5.350.01613.220.34(*)n.s.n.s.
Thymelicus acteon (Rottemburg, 1775)133181.18*n.s.n.s.−3.830.0459.17n.s.n.s.n.s.
Polyommatus agestis ([Denis & Schiffermüller], 1775)115172.47***n.s.n.s.−8.430.000329.020.49**n.s.n.s.
Melitaea aurelia Nickerl, 18505492.17**n.s.n.s.−8.930.00323.930.52**n.s.n.s.
Erebia medusa ([Denis & Schiffermüller], 1775)425n.s.n.s.n.s.−1.69n.s.n.s.n.s.n.s.
Argynnis aglaja (Linnaeus, 1758)3772.74**n.s.n.s.−11.620.00131.180.50**n.s.n.s.
Spialia sertorius (Hoffmannsegg, 1804)3064.79***n.s.8.42*−30.150.000258.410.56**n.s.0.37*
Hesperia comma (Linnaeus, 1758)3041.83*n.s.n.s.−8.840.04816.220.35(*)n.s.n.s.
Colias alfacarensis Ribbe, 19052352.54**n.s.n.s.−11.370.00626.810.46**n.s.n.s.
Zygaena hippocrepidis (Hübner, [1799])2141.72(*)n.s.n.s.−8.420.05914.800.32(*)n.s.n.s.
Boloria euphrosyne (Linnaeus, 1758)183n.s.n.s.n.s.−2.27n.s.n.s.n.s.n.s.
Leptidea reali Reissinger, 19891381.37*n.s.n.s.−6.090.04211.53n.s.n.s.n.s.
Zygaena lonicerae (Scheven, 1777)103n.s.n.s.n.s.−2.27n.s.n.s.n.s.n.s.
Euphydryas aurinia (Rottemburg, 1775)11
Maniola jurtina (Linnaeus, 1758)347931n.s.n.s.28.67**−7.800.00399.710.47**n.s.n.s.
Polyommatus icarus (Rottemburg, 1775)12263014.78***n.s.−6.30*−31.210.00285.800.64***n.s.n.s.
Coenonympha pamphilus (Linnaeus, 1758)894315.62(*)n.s.n.s.−12.840.09930.58n.s.n.s.n.s.
Pieris rapae (Linnaeus, 1758)636320.32(*)n.s.n.s.
Pieris napi (Linnaeus, 1758)42832−0.49**n.s.n.s.
Zygaena filipendulae (Linnaeus, 1758)378272.14*n.s.n.s.−5.220.02318.64n.s.n.s.n.s.
Melanargia galathea (Linnaeus, 1758)361284.51**n.s.n.s.−11.720.00338.050.51**n.s.n.s.
Pyrgus malvae (Linnaeus, 1758)328272.60*n.s.n.s.−6.590.01023.840.48**n.s.n.s.
Aphantopus hyperantus (Linnaeus, 1758)29231n.s.n.s.n.s.3.43n.s.n.s.n.s.n.s.
Callophrys rubi (Linnaeus, 1758)254202.41***n.s.n.s.−7.670.000727.330.30(*)n.s.n.s.
Nymphalis io (Linnaeus, 1758)24731n.s.−12.86*n.s.63.360.01961.50n.s.n.s.n.s.
Nymphalis urticae (Linnaeus, 1758)23828n.s.n.s.n.s.1.95n.s.n.s.n.s.n.s.
Argynnis paphia (Linnaeus, 1758)20823n.s.n.s.n.s.0.94n.s.n.s.n.s.n.s.
Pieris brassicae (Linnaeus, 1758)16628n.s.−5.42*5.79**20.010.00445.75n.s.n.s.0.34(*)
Ochlodes sylvanus (Esper, [1778])14028n.s.−5.41*3.06(*)22.620.01634.56n.s.n.s.n.s.
Gonepteryx rhamni (Linnaeus, 1758)14028n.s.n.s.3.49*−1.520.02919.69n.s.n.s.0.31(*)
Araschnia levana (Linnaeus, 1758)13229n.s.n.s.2.90(*)−0.590.08714.76n.s.n.s.n.s.
Anthocharis cardamines (Linnaeus, 1758)11026n.s.n.s.n.s.1.46n.s.n.s.n.s.n.s.
Carterocephalus palaemon (Pallas, 1771)9523n.s.n.s.2.68*−1.930.04810.31n.s.n.s.n.s.
Pararge aegeria (Linnaeus, 1758)9226n.s.n.s.n.s.1.47n.s.n.s.n.s.n.s.
Hamearis lucina (Linnaeus, 1758)62172.13**n.s.n.s.−7.260.00123.800.41*n.s.n.s.
Thymelicus sylvestris (Poda, 1761)5618n.s.n.s.n.s.0.25n.s.n.s.n.s.n.s.
Plebeius argus (Linnaeus, 1758)4051.85(*)−3.19*n.s.4.650.03623.960.30(*)−0.40*n.s.
Thymelicus lineola (Ochsenheimer, 1808)197n.s.n.s.n.s.−1.27n.s.n.s.n.s.0.30(*)
Lycaena phlaeas (Linnaeus,1761)1971.18(*)n.s.n.s.−5.560.0888.680.33(*)n.s.n.s.
Vanessa cardui (Linnaeus, 1758)1581.74*n.s.n.s.−7.490.01317.030.37*n.s.n.s.
Lasiommata megera (Linnaeus, 1767)157n.s.n.s.n.s.−1.27n.s.n.s.n.s.n.s.
Celastrina argiolus (Linnaeus, 1758)1382.39**n.s.n.s.−9.960.00226.680.47**n.s.n.s.
Papilio machaon Linnaeus, 1758128n.s.2.35(*)n.s.−11.230.0947.81n.s.n.s.n.s.
Nymphalis c-album (Linnaeus, 1758)1181.49*n.s.n.s.−6.540.02913.280.31(*)n.s.n.s.
Issoria lathonia (Linnaeus, 1758)53n.s.−6.45*17.37**1.740.00358.07
Thecla betulae (Linnaeus, 1758)42n.s.n.s.n.s.−2.71n.s.
Limenitis camilla (Linnaeus, 1764)432.57*n.s.n.s.−12.240.02724.61
Lycaena tityrus (Poda, 1761)22n.s.n.s.n.s.−2.71n.s.

The comparison of sampled with estimated species, which were calculated with the species richness estimator ACE, showed that 90.24 ± 1.18% (minimum 74.73%, maximum 98.91%) of the species were found in each study site. No correlation between this proportion and the independent factors habitat area (Spearman rank correlation n = 32: r = 0.212, P = 0.237), isolation (r = 0.023, P = 0.899), number (r = 0.168, P = 0.351) and cover (r = 0.129, P = 0.471) of plant species in flower, landscape diversity (r = 0.076, P = 0.671) and the percentage of grasslands in the surrounding landscape (r = 0.132, P = 0.463) were found. Smoothed sample size–butterfly species number accumulation curves reach asymptote (J. Krauss, unpublished data). These results justify the usage of sampled species numbers instead of estimated species numbers.

Habitat area (0.03–5.14 ha) was not correlated with habitat isolation (index: 2100–86,000), and only marginally with landscape diversity (at a 250 m scale; Shannon–Wiener: 0.09–1.56) (Table 1). Distance to the nearest calcareous grassland differed from 55 to 1894 m. The proportion of grassland habitats (0.42–58.37%) was correlated to landscape diversity and the two characteristics of habitat quality, the number (average: 11.2–24.8 species) and cover (2.10–14.09%) of plant species in flower were highly correlated with habitat area (Table 1). For further analysis, we excluded (i) the proportion of grassland habitats and (ii) the two habitat quality characteristics.

Table 1.  Pearson correlation coefficients (r) are shown for the relationships between the independent variables of the thirty-two calcareous grasslands. Habitat area and habitat isolation are log10-transformed, cover of plant species in flower is arcsine-square root transformed
 Habitat isolationLandscape diversity (radius: 250 m)Grassland habitats (radius: 250 m)Number of blooming plant speciesCover of blooming plant species
  1. Significance levels are: ***P < 0.001, **P < 0.01, *P < 0.05, (*)P < 0.1, n.s. = not significant.

Habitat area−0.2 n.s.0.32(*)0.40*0.69***0.46**
Habitat isolation−0.01 n.s.0.15 n.s.−0.30(*)−0.28 n.s.
Landscape diversity (radius: 250 m)0.64***0.17 n.s.0.09 n.s.
Grassland habitats (radius: 250 m)0.20 n.s.0.01 n.s.
Number of plant species in flower0.39*

Species richness of butterflies

The species–area relationship for all butterflies was highly significant. Species richness of both the specialists and generalists increased significantly with increasing habitat area (Fig. 2, Table 2). Comparison of these regressions showed no differences in slopes (F = 0.88, P = 0.353). In contrast, the z-value (slope of log–log regressions) for all butterfly species was z = 0.164, and was significantly higher for habitat specialists (z = 0.399) than for generalists (z = 0.096; comparison of slopes: F = 31.53, P < 0.0001). The presence of fourteen specialist (70%) and ten generalist (29%) butterfly species was determined by habitat area (Appendix).

Figure 2.

Relationship between the number of specialist and generalist butterfly species and grassland area (n = 32 fragments). Specialist butterflies (twenty species): y = −12.18 + 5.33  log10x, generalist butterflies (thirty-four species): y = 5.39 + 4.37  log10x. Statistics see Table 2.

Table 2.  Simple regressions between the number of species and the density of all butterflies (fifty-four species), specialists (twenty species) and generalists (thirty-four species) and the three independent factors habitat area, habitat isolation and landscape diversity of the thirty-two calcareous grasslands. Habitat area and habitat isolation are log10-transformed
 Habitat areaHabitat isolationLandscape diversity (radius: 250 m)
Species numbers
 All species69.960.836<0.00010.24−0.0890.6267.050.4360.013
 All species7.820.4550.0090.12−0.0630.7310.860.1670.361

Neither number of species of all butterflies nor that of only the specialists or generalists showed a significant relationship with the habitat isolation index (Fig. 3, Table 2). Additionally, all other isolation calculations did not show any significant effect of habitat isolation (results not shown). Only the presence of the two specialist species Polyommatus coridon and Zygaena carniolica (Scopoli) appeared to increase with increasing habitat connectivity (Appendix).

Figure 3.

Relationship between the number of specialist and generalist butterfly species and habitat isolation (n = 32 fragments). Specialist butterflies (twenty species), generalist butterflies (thirty-four species). Statistics see Table 2.

Species number of all butterflies, specialists and generalists, increased significantly with increasing diversity of the surrounding landscape, using a 250 m radius of the landscape sector for the calculation (Fig. 4, Table 2). Comparison of regressions between specialists and generalists showed no differences in slopes (F = 0.15, P = 0.702). The presence of two specialist and five generalist butterfly species could be explained by increasing landscape diversity (Appendix).

Figure 4.

Relationship between the number of specialist and generalist butterfly species and landscape diversity, calculated as Shannon–Wiener Index at the 250 m landscape scale (n = 32 fragments). Specialist butterflies (twenty species): y = 1.11 + 4.89x, generalist butterflies (thirty-four species): y = 13.93 + 6.14x. Statistics see Table 2.

In multiple regression models, habitat area was the only explanatory factor for the species number of all butterflies (70.0%), the specialists (69.8%) and generalists (49.5%; see the simple regressions; Table 2). When we also included marginally significant factors (P < 0.1) in the multiple regression model, landscape diversity explained a further 3.1% for all butterfly species and 5.8% for generalists. None of the isolation measurements had a negative effect for species richness in multiple regressions (results not shown).

We analysed the effects of landscape context at twelve nested spatial scales. Landscape diversity and proportion of grassland habitats correlated best with the number of butterfly species at the smallest scale of 250 m for both the specialists and generalists (Fig. 5). Generalist butterflies showed higher correlations with landscape diversity than specialist butterflies (Fig. 5). In contrast, the proportion of grassland habitats showed no significant relationships, neither for specialists (250 m: r = 0.22, P = 0.129) nor for generalists (250 m: r = 0.28, P = 0.117).

Figure 5.

Pearson correlation coefficients between number of specialist (20) and generalist (34) butterfly species and landscape diversity (Shannon–Wiener Index). Landscape diversity is calculated for different spatial scales ranging from a radius of 0.25–3.00 km. **P < 0.01, *P < 0.05.

Density of butterflies

The density of all butterflies and of specialists increased significantly with increasing habitat area, whereas generalist density increased only with marginal significance (Fig. 6, Table 2). Ten of twenty specialist species (50%) showed significantly increasing population densities with increasing habitat area, while only seven of thirty-four generalists (21%) showed this trend. Density of P. napi was significantly negatively correlated with increasing habitat area (Appendix). Butterfly densities on calcareous grasslands did not correlate with the habitat isolation index, neither for all butterflies nor specialists or generalists (Table 2). All other isolation calculations also showed no significant effects of habitat isolation (results not shown). Except for Plebeius argus (L.), which showed decreasing densities with decreasing isolation, none of the other fifty-three butterfly species showed any significant relation to habitat isolation (Appendix). No correlations between densities of all butterflies, specialists or generalists, with landscape diversity were found (Table 2). Only the density of Spialia sertorius (Hoffmannsegg) increased significantly with increasing landscape diversity (Appendix).

Figure 6.

Relationship between the total population density of specialist and generalist butterflies and grassland area (n = 32 fragments). Specialist butterflies (twenty species): y = −1.09 + 0.51 log10x, generalist butterflies (thirty-four species): y = 0.55 + 0.31 log10x. Statistics see Table 2.

In multiple regression models, habitat area was the only factor related to the density of all butterflies (explaining 20.7% of the variation) and specialists (22.4%), whereas generalists showed only a marginally significant relationship (9.1%) (see simple regressions Table 2). Correlations with habitat isolation and landscape diversity were not even marginally significant. All other isolation calculations also had no significant effects on species density (results not shown).

With an inclusion of cover of plant species in flower in the multiple model, this factor remains as the only explanatory factor (16.5%, P = 0.021) for generalist density, whereas specialist and total density were not correlated (results not shown).


Species richness of butterflies

Habitat area

Species numbers increased significantly with increasing habitat area for both habitat specialist and habitat generalist butterflies, confirming the general validity of species–area relationships, as previously shown in similar studies of butterfly communities (Wilcox et al., 1986; Baz & Garcia-Boyero, 1995; Robertson et al., 1995; Wettstein & Schmid, 1999; Steffan-Dewenter & Tscharntke, 2000; Zschokke et al., 2000). Range of habitat area in our study was 0.03–5.14 ha, but species–area relationships for butterflies can be found from very small scales (0.25–20.25 m2; Zschokke et al., 2000) to large scales (from 100–3000 km2; Wilcox et al., 1986). Not all species depend on habitat area equally as shown by Thomas et al. (1992, 2001). Habitat area influenced species richness of specialists more than that of generalists, indicating that generalists additionally use other habitat types in the surrounding landscape matrix, whereas specialists completely depend on calcareous grasslands. The z-value for specialists was 0.40 and was therefore higher than expected from published values for oceanic islands or isolated mainland habitats (z = 0.25–0.33), while the z-value for generalists was only 0.09 and even lower than those known for non-isolated mainland habitats (z = 0.13–0.18) (Rosenzweig, 1995). This result is in support of the findings of Steffan-Dewenter & Tscharntke (2000) that monophagous habitat specialists have the highest z-values in a butterfly community.

Habitat isolation

In contrast to our predictions, habitat isolation did not have a consistent negative effect on butterfly species richness. This result is in support of published studies, which also could not show effects of habitat isolation on butterfly species richness (Wilcox et al., 1986; Baz & Garcia-Boyero, 1995; Steffan-Dewenter & Tscharntke, 2000), but Ricketts et al. (2001) found fewer moth species in habitats with distances of > 3.5 km from a tropical forest reserve compared with distances < 1.0 km. Wettstein & Schmid (1999) found more wetland indicator butterfly species when large additional wetland areas occurred nearby. In our study, only two specialist butterfly species responded negatively to habitat isolation, while Thomas et al. (1992, 2001) found that isolation affected the occurrence of all six studied butterfly species. This difference may be the result of the greater range of habitat isolation in their study and the relatively low isolation in our study (< 2 km). In addition, single species with low dispersal abilities may suffer from even comparatively low isolation distances. Habitat isolation of calcareous grasslands as found in our study appeared to be typical for many regions in central Europe, but the regions of northern Europe might be more fragmented thereby causing clear isolation effects. Further, patterns found in one region may not hold for other regions because of a wealth of geographical factors that change between regions (Tscharntke et al., 2001). Evidence of habitat isolation (or connectivity) may also depend on the indices used (Moilanen & Nieminen, 2002). We tested two isolation indices, the distance to the nearest grassland and an index (including a comparison of different values for the parameters a and b) based on the proportion of calcareous grassland area around the habitat. None of these calculations gave evidence for any negative influence of isolation on species numbers of specialist or generalist butterflies. Summarizing, habitat connectivity of calcareous grasslands appeared to be sufficient for the butterfly communities of our German study region, although isolation has been reported to greatly affect some butterfly species in the UK.

Landscape context

Effects of landscape context on species–area relationships are mostly unknown (Wiens, 1997; Hanski, 1999; Vandermeer & Carvajal, 2001; Tscharntke et al., 2002). In our study, landscape diversity marginally influenced species number of generalists, but not of specialists. Therefore, generalists appeared to be more affected than specialists. The influence of landscape context was most important at a small spatial scale in that landscape diversity within a 250 m radius turned out to best predict species richness. Our findings are in support of the results of Weibull et al. (2000) for butterflies and Steffan-Dewenter et al. (2002) for bees, where small spatial scales best predicted species richness. In metapopulation models, based on habitat area and isolation, no improvement was achieved by adding landscape context and habitat quality (Moilanen & Hanski, 1998), whereas the models of Vandermeer & Carvajal (2001) showed that matrix quality can be of major importance for metapopulations. The weak landscape effects documented for calcareous grasslands in our study region might be explained by the relatively complex landscapes that surrounded our calcareous grasslands. Although landscape context played a minor role for butterfly species richness, we found evidence for the expected tendency that generalists were more affected than specialists.

Habitat quality

Habitat quality for requirements of larvae has been predicted to be the main factor for presence or absence of butterfly species (Thomas et al., 2001). In our study, we did not discriminate between the factor habitat area and the factor habitat quality – which we characterized by the number of plant species in flower and the cover of plants in flower – because of correlations between these variables. Higher habitat heterogeneity, measured as richness of plants in flower, increased with habitat area and butterfly species richness, which supports the habitat heterogeneity hypothesis (Stevens, 1986; Rosenzweig, 1995).

Density of butterflies

Most population density studies have been conducted for mammals and birds (Bowers & Matter, 1997; Connor et al., 2000; Matter, 2000; Gaston & Matter, 2002). Empirical studies for insects are rare, particularly for insect communities (Connor et al., 2000; Steffan-Dewenter & Tscharntke, 2000).

In our study, population densities increased with increasing habitat area for all butterfly and specialist butterfly species, but only marginally for generalists. Steffan-Dewenter & Tscharntke (2000) found a similar result in that specialists were more affected by habitat area than other species. In our study, population densities of most specialist and some generalist species increased with increasing habitat area including the three most abundant species Maniola jurtina, Polyommatus coridon and Polyommatus icarus. This is in support of the meta-analysis of Connor et al. (2000), who assume an average increase of population densities with increasing habitat area. However, Thomas et al. (2001) found no correlations between population densities and habitat area for three butterfly species, whereas Thomas et al. (1992) found in a four-species study increasing densities for Plebeius argus and Hesperia comma (L.). Decreasing population densities with area have been recorded for Melitaea cinxia (L.) (Hanski et al., 1994). These inconsistent results may have been the result of numerous difficulties in calculating relationships between population density and habitat area, like the problems of including or excluding empty habitat patches and the difference between patch (PIARs) or generalized (GIARs) approaches of individual–area relationships (Gaston & Matter, 2002). The range of habitat area and the number of replicates are also important (Bowers & Matter, 1997). Matter (2000) reported for the red milkweed beetle Tetraopes tetraophthalmus (Forster) increasing, decreasing or constant densities with habitat area in different years.

We did not find correlations between population densities for butterfly communities or single species and habitat isolation, thereby confirming the analysis of Steffan-Dewenter & Tscharntke (2000) and Thomas et al. (2001). However, increasing densities with decreasing isolation were reported for Melitaea cinxia (Hanski et al., 1994).

Similarly, landscape diversity had no consistent effect, and was related to only a few species. We are aware of only one other study testing butterfly densities in a landscape context. Weibull et al. (2000) reported increasing or decreasing abundances for butterflies depending on the spatial scale where landscape diversity was measured.

Summarizing, butterfly population densities were mainly affected by habitat area and not by isolation or landscape context in our study region, but generalizations are difficult because of limited and inconsistent data in the literature.


In conclusion, our data emphasize the importance of habitat area on the butterfly community, while only a small impact of landscape diversity and no impact of habitat isolation was observed. Habitat specialist butterflies were more affected by habitat area than habitat generalists, whereas generalists were also affected by landscape diversity. Therefore, protection of large calcareous grasslands would contribute substantially to the conservation of endangered habitat specialists. Contrary to expectations, landscape context and habitat connectivity appeared to be sufficient to support species-rich butterfly communities in calcareous grasslands in our study region.


We thank Konrad Fiedler, Steve Matter, Marko Nieminen, Josef Settele, Thomas Schmitt, Christian H. Schulze and one anonymous referee for helpful discussions and comments on the manuscript. We thank Hans Joger and Thomas Meineke for their butterfly advice and the following regional authorities: Niedersächsisches Landesamt für Ökologie (Reinhard Altmüller), Untere Naturschutzbehörden Stadt Göttingen (Roland Dillenburger), Landkreis Göttingen (Bertram Preuschoff), Landkreis Northeim, Landkreis Heiligenstadt, and Regierungspräsidium Kassel. This work was financially supported by the German Science Foundation (Deutsche Forschungsgemeinschaft).


Jochen Krauss is a PhD student of agroecology at the University of Göttingen. He finished his master degree in biology on Collembola and Protura. He worked on chamois in a tourism and wildlife project in Switzerland and now investigates effects of landscape context and habitat fragmentation on butterflies and plants for his PhD thesis. His research interests include community and population ecology and conservation biology.

Ingolf Steffan-Dewenter is Assistant Professor of agroecology at the University of Göttingen. His main research interests are in the community ecology of plants and insects and in the effects of habitat fragmentation and succession on plant–pollinator and plant–herbivore interactions.

Teja Tscharntke is Professor of agroecology at the University of Göttingen. His research focuses is on plant–insect interactions including parasitism, predation and pollination, insect communities at a landscape scale, and temperate–tropical comparisons.