Semi-natural grasslands as population sources for pollinating insects in agricultural landscapes


Erik Öckinger, Department of Ecology, Animal Ecology, Lund University, SE-223 62 Lund, Sweden (fax + 46 46 2224716; e-mail


  • 1In intensively farmed agricultural landscapes, many species are confined to very small uncultivated areas such as field margins. However, it has been suggested that these small habitat elements cannot support viable populations of all the species observed there. Instead, species richness and abundance in these small habitat fragments may, at least partly, be dependent on dispersal from larger semi-natural grassland fragments.
  • 2We tested this hypothesis for butterflies and bumble bees in 12 independent landscapes in a region of intense agriculture in southern Sweden. In each landscape we surveyed abundance and species richness in one semi-natural grassland, one linear habitat (uncultivated field margin) adjacent to this (called proximate) and one similar linear habitat (called distant) situated at least 1000 m from the semi-natural grassland patch.
  • 3Both species richness and density (individuals per unit area) of butterflies and bumble bees were significantly higher in proximate linear habitats than in distant ones. Moreover, butterfly species richness was higher for a given area in grasslands than in any of the linear habitat types. Butterfly density in grasslands did not differ from that in proximate linear habitats but was lower in distant linear habitats. The effect of isolation on density was stronger for less mobile butterfly species. For bumble bees there was no difference in species richness between grasslands and proximate linear habitats.
  • 4For at least some of the butterfly species even these relatively small fragments of semi-natural grasslands act as population sources from which individuals disperse to the surrounding habitats and thereby contribute to higher densities and species richness in adjacent areas. For bumble bees, it is more likely that the grasslands contain a higher density of nests than the surrounding intensively cultivated landscape, and that the density of foraging bumble bees decreases with increasing distance from the nest.
  • 5Synthesis and application. Habitat fragmentation and intensified agricultural practices are considered to be a threat against services provided by pollinators. In order to sustain the abundance and diversity of insect pollinators in intensively farmed agricultural landscapes, we suggest that preservation of the remaining semi-natural grasslands or re-creation of flower-rich grasslands is essential.


The persistence of plant and animal populations in agricultural landscapes is important for both maintaining ecosystem services and the conservation of threatened species (Tscharntke et al. 2005). Pollinating insects such as butterflies and bumble bees have suffered from agricultural intensification and landscape change and many species show negative population trends (Maes & van Dyck 2001; Warren et al. 2001; Thomas et al. 2004; Goulson et al. 2005; Öckinger et al. 2006). The maintenance of viable populations of pollinators in farmland may depend on the preservation of more or less permanent semi-natural habitats in agricultural landscapes that are otherwise subjected to repeated disturbances (Tscharntke et al. 2005). In intensively farmed agricultural landscapes only small fragments of such semi-natural habitats remain, typically as linear elements such as field borders and road verges. However, often such small uncultivated habitat fragments contain impoverished faunas compared with larger areas of grassland (Weibull, Östman & Granqvist 2003). It is important to understand the relative roles of local and regional factors in regulating the species richness and abundances of insects in these habitats if we are to maintain viable pollinator populations in farmland.

The importance for species richness and abundance of insects of habitat quality and management of small remnant habitats in agricultural landscapes has been investigated in several studies. For example, the width of uncultivated strips (Munguira & Thomas 1992; Clausen, Holbreck & Reddersen 2001), the abundance of nectar resources (Munguira & Thomas 1992; Clausen, Holbreck & Reddersen 2001; Pywell et al. 2004; Saarinen et al. 2005), insolation (Clausen, Holbreck & Reddersen 2001), presence of shelter (Pywell et al. 2004), vegetation height (Feber, Smith & Macdonald 1996; Saarinen et al. 2005), adjacent land use (Saarinen et al. 2005), the timing of cutting (Feber, Smith & Macdonald 1996) and whether spraying has occurred (Feber, Smith & Macdonald 1996) have all been shown to affect the species richness and abundance of butterflies. For bumble bees, a high abundance of nectar and pollen resources (Bäckman & Tiainen 2005; Pywell et al. 2005) and a continuous availability of nectar throughout the season (Carvell et al. 2004) have positive effects on species richness and abundance.

In contrast, relatively few studies have addressed the importance of landscape composition for species richness and abundance of insects in these habitats (but see Tscharntke et al. 2002b), even though it has been suggested (Clausen, Holbreck & Reddersen 2001; Tscharntke et al. 2005) that the species richness and abundance of insects in otherwise intensively managed agricultural areas may at least partly be dependent on dispersal from nearby semi-natural grasslands. If the small uncultivated habitat fragments are of low quality, they may contain sink populations that depend on immigration from source populations elsewhere (cf. Pulliam 1988). Larger fragments of semi-natural grassland could contain such source populations. If the small habitat fragments are able to sustain low-density populations, they could still have augmented population sizes as a result of immigration from sources, and thus be so-called pseudo-sinks (Watkinson & Sutherland 1995). Given that the small uncultivated habitat fragments are of equal quality compared with larger semi-natural grasslands, their populations may still be dependent on immigration if they are too small to support viable populations of the species in focus. A grassland with surrounding small uncultivated habitat fragments would then function as a mainland–island metapopulation system (cf. Harrison 1991), where the populations in the small uncultivated habitat fragments persist because of the rescue effect (Brown & Kodric-Brown 1977).

Both butterflies and bumble bees feed on nectar as adults, but they have very different life histories and population structures and are therefore likely to be affected by landscape composition through different mechanisms. Many butterfly species are highly specialized with respect to larval host-plants (Stoltze 1996) and also exhibit sedentary behaviour where a high proportion of the individuals stay in their natal patch (Wilson & Thomas 2002). These species are dependent on a sufficient density of larval host-plants as well as nectar resources within the same patch (Ouin et al. 2004). Several relatively sedentary butterfly species have been shown to exist in metapopulation systems where their habitat is highly fragmented (Thomas, Thomas & Warren 1992; Hanski et al. 1995; Lewis et al. 1997; Gutiérrez, Thomas & León-Cortés 1999) while others are highly mobile and able to utilize resources separated by large distances (Wilson & Thomas 2002).

In highly fragmented landscapes, not only can populations become isolated from each other but also the distances between different resources needed by single individuals increase. For bumble bees, being social and colonial central place foragers, nest sites and nectar and pollen sources are the two essential resources (Goulson 2003). If these resources are too isolated from each other, they may become unavailable for the focal species, with negative consequences for population persistence in the landscape. Thus bumble bees may depend on a high abundance of nectar resources in the surrounding landscape at an appropriate scale (Westphal, Steffan-Dewenter & Tscharntke 2003; Bäckman & Tiainen 2005).

Even though butterflies and bumble bees may be affected by landscape composition in different ways, the resulting distribution of species and individuals in the landscape may be similar. We tested the hypothesis that the presence of semi-natural grasslands in the vicinity contributes to higher species richness and population sizes of butterflies and bumble bees in small habitat patches, such as field margins in intensively farmed agricultural areas. We also tested whether this effect is stronger for butterfly species with a more restricted dispersal capacity and for bumble bee species with more specialized feeding and nesting preferences.

Materials and methods

study system

We surveyed species richness and densities of butterflies, burnet moths and bumble bees in 12 non-overlapping landscapes in a region dominated (> 80%) by intensively cultivated fields in southern Sweden. In all landscapes, semi-natural grasslands constituted less than 5% of the area. In each landscape, we surveyed one semi-natural grassland (5–12 ha) and two smaller linear habitat fragments. These were uncultivated strips of perennial grassland vegetation situated either between two cultivated fields (n = 10) or between a cultivated field and a road (n = 14). None of these linear habitats was grazed but some of the road verges (n = 6) were cut in late summer. Semi-natural grasslands were defined as grasslands with no signs of recent fertilization or cultivation, and were identified in a nation-wide survey of valuable grasslands performed in 1988–92. All semi-natural grasslands included in the study were grazed by cattle (n = 11) or horses (n = 1) and were chosen based on their situation in the landscape, using a combination of geographical data from the Swedish Board of Agriculture and from official Swedish maps analysed using a geographical information system (GIS) (ArcView 3·2). In each landscape, one linear habitat situated with one end within 100 m from the border of the grassland, hereafter referred to as ‘proximate linear habitat’, and one linear habitat situated at least 1 km from the study grassland (and from any other permanent grassland patch larger than 0·5 ha), hereafter referred to as ‘distant linear habitat’, were chosen. We avoided linear habitats situated near other biotopes that could have served as population sources for butterflies or bumble bees, such as flower-rich set-aside fields, cultivated grasslands and leys.


At each site butterflies (Rhopalocera), burnet moths (Zygaenidae) and bumble bees (Bombus spp.) were surveyed using a standardized transect count method developed for butterflies (Pollard 1977; Pollard & Yates 1993). In grasslands the transect length was proportional to the area, with 150 m transect/ha area, but rounded off to the nearest 100 m. In linear habitats the transect length was 300 m. Each transect was divided into segments of 100 m, and the number of individuals per species was recorded for each segment. Burnet moths are diurnal and similar to butterflies in most ecological aspects (Naumann, Tarmann & Tremewan 1999). Unless otherwise stated, ‘butterflies’ refer to both groups. Only species regarded to be dependant on grasslands were included in the analyses (see Ecological classification below). All butterflies observed within 5 m ahead and on both sides of the observer and all bumble bees observed within 1 m from the observer were recorded. In linear habitats narrower than 10 m only butterfly individuals observed within the linear habitat were recorded for the respective segments.

Both butterfly and bumble bee species were identified in the field, with two exceptions. First, Bombus terrestris and Bombus lucorum workers are very difficult to distinguish (Prys-Jones & Corbet 1991; cf. Saville et al. 1997; Pywell et al. 2005) and were therefore collectively referred to as B. terrestris. Secondly, parasitic bumble bee species (subgenus Psithyrus) were not separated from their respective hosts and hence were recorded collectively with these. Each site was surveyed six times at regular intervals during 2004, with the first visit between 28 May and 4 June and the last visit between 5 August and 8 August.

habitat quality

We estimated several potential habitat quality variables in each grassland and linear habitat. Vegetation height was recorded on a three-grade scale (1, less than 10 cm; 2, 10–25 cm; 3, higher than 25 cm) using a grassland ruler (Ekstam & Forshed 1996). Vegetation height is the height where 50% of the width of the ruler is hidden behind the vegetation when observed at a distance of 5 m. Vegetation height was recorded for each transect segment on each insect survey occasion. In statistical analyses at the segment level we used vegetation height per segment and visit number as factors. In analyses at site level we used mean vegetation height per site as a covariate.

The floral measures were sampled using 50 × 50-cm plots, each divided into 25 squares of 10 × 10 cm. The plots were placed at regular intervals along each transect and sampling was conducted twice during the season, the first in June and the second in the second half of July We placed 16 plots in each grassland and six in each linear habitat. In each sample plot we measured total floral abundance as the number of 10 × 10-cm squares in which any nectar-producing flower occurred. Floral diversity was the number of species of selected nectar-rich plants (see the Supplementary material). In the statistical analyses the mean values per site of these measures were used.

As a measure of habitat quality for butterflies we included the number of potential larval host-plants along each transect and recorded the presence of each such plant species for each transect segment of 100 m. Herbaceous host-plants, except grasses, listed as host-plants for any butterfly in our region by Henriksen & Kreutzer (1982), Stoltze (1996) (butterflies) or Naumann, Tarmann & Tremewan (1999) (burnet moths) were regarded as potential host-plants. We excluded grasses because grass-feeding butterflies are thought to be unspecific in their choice of host-plants (Henriksen & Kreutzer 1982; Stoltze 1996) and grasses were common at all sites. We included mean host-plant species richness per site only in the statistical analyses concerning butterflies.

We measured the breadth of the linear habitats at the start, mid- and end-point of each transect, and calculated the mean value. The transect breadth was 10 m for butterflies and 2 m for bumble bees in grasslands and in linear habitats wider than this. In narrower linear habitats the transect breadth was equal to the breadth of the linear habitat. In analyses of species richness we included both mean transect breadth and the squared value of mean breadth as a covariate, as a non-linear effect of area could be expected. Densities of groups of species were calculated as the number of individuals per transect segment divided by the breadth of the segment (i.e. 10 m for butterflies and 2 m for bumble bees, or, for linear habitats narrower than this, the mean breadth of the linear habitat).

ecological classification

We excluded butterfly species that are not dependent on any type of grassland vegetation (see the Supplementary material), which included Pieris brassicae, Pieris rapae and Pieris napi, whose larvae feed on both cultivated species and weed species within Brassicaceae, and Aglais urticae, Inachis io and Araschnia levana, whose larvae feed on Urtica spp. Furthermore, we excluded the two migratory species Vanessa atalanta and Vanessa cardui. However, our results were not sensitive to whether non-grassland species were excluded or not.

To examine the effect of butterfly mobility we classified British butterfly species as sedentary, intermediate or wide-ranging following Pollard & Yates (1993). As only one species (Issoria lathonia) observed during this study was classified as wide-ranging, we pooled the categories of intermediate and wide-ranging species into one category, called ‘mobile’. Species not classified by Pollard & Yates (1993) were fitted into these categories using data from Naumann, Tarmann & Tremewan (1999) for Zygaenidae and Henriksen & Kreutzer (1982) and Stoltze (1996) for butterflies (cf. Öckinger et al. 2006).

Even though there are a number of studies addressing the mobility of bumble bees (Walther-Hellwig & Frankl 2000a, 2000b; Darvill, Knight & Goulson 2004), it was not possible to classify bumble bee species according to their mobility as no data were available for most species. Instead, we classified bumble bees according to their proboscis length and nest-site preferences. The length of the proboscis of different bumble bee species is important in determining resource utilization (Ranta & Lundberg 1980). We classified bumble bee species as short-tongued (< 8·0 mm) or long-tongued (> 8·0 mm) based on data from Goulson et al. (2005). Long-tongued species are generally more specific in their flower utilization, with preferences for flowers with deep corollas, while most short-tongued species are generalists, as they are able to rob nectar by biting a hole in the corolla (Goulson 2003). The bumble bees were also classified as preferentially nesting below ground or at the ground surface, according to data from Alford (1975). Two of the species, Bombus hypnorum and Bombus lapidarius, can be categorized as nest site generalists (Alford 1975) and were omitted from analyses based on this classification. The classifications of all recorded butterfly and bumble bee species are given in the Supplementary material.

statistical analyses

We analysed species richness and total densities for butterflies and bumble bees separately per site and visit at the level of transect segments, in order to account for unequal sampling intensity in grasslands and linear habitats, using mixed models. Species richness of butterflies and bumble bees was analysed separately using generalized linear mixed models (SAS macro Glimmix) with Poisson error distribution and log-link function (Littell et al. 1996). Densities of species were analysed using general linear mixed models (SAS proc Mixed) with normal error distribution. In all analyses, landscape identity (a number identifying the matched set of one grassland, one proximate and one distant linear habitat) and the interactions between landscape identity and habitat type and landscape identity, habitat type and transect segment were included as random factors, habitat type and visit number as fixed factors and habitat quality variables as covariates. We also included interactions between habitat type and visit number, habitat type and habitat quality variables and between the different habitat quality variables. We calculated densities (individuals per 100 m2) separately for the various ecological groups described earlier and included mobility, tongue length and nest-site preferences as fixed factors and also the interactions between these factors and habitat type in the analyses of density. If the interactions between ecological groups and habitat type were significant, we performed separate analyses of densities of each group. In these analyses the mean value of densities across all segments was analysed instead of analysing each segment separately. Densities were square-root transformed before analyses in order to achieve normally distributed residuals.

Each respective final model was determined by sequentially removing non-significant fixed factors and covariates until only those with P < 0·05 remained (random factors were kept). To detect differences between the three habitat types we used linear contrasts. Denominator degrees of freedom were estimated with the Satterthwaite method (Littell et al. 1996).

To test for differences in quality between habitat types we used general linear models (SAS proc GLM) with the respective habitat quality variable as dependent variables and habitat type as a factor. We applied linear contrasts to test for differences between grasslands and linear habitats and between distant and proximate linear habitats, respectively. All statistical analyses were performed in SAS 8·2 for Windows (SAS Institute Inc., Cary, NC).


differences in quality between habitat types

None of the measured habitat quality variables, breadth, vegetation height, floral diversity, floral abundance or number of butterfly host-plant species, differed between proximate and distant linear habitats (all P > 0·1; Table 1). In contrast, all the measured aspects of habitat quality differed between grasslands and linear habitats. Grasslands had on average shorter vegetation, higher floral diversity and abundance and higher numbers of butterfly host-plant species than linear habitats (Table 1).

Table 1.  Differences in habitat quality between grasslands and linear habitats. Both the main effect and the results of linear contrasts are shown. In all cases the denominator degrees of freedom are 33 and the nominator degrees of freedom are 1
VariableMain effectProximate–distantGrassland–linear habitats
Breadth68·2< 0·0010·140·71136< 0·001
Vegetation height15·5< 0·0010·370·5530·6< 0·001
Floral diversity6·940·0031·340·2612·50·001
Floral abundance6·340·0052·630·1110·50·003
Host-plant species44·8< 0·0010·030·8589·6< 0·001

species richness and total density

In total, 24 grassland butterfly (including burnet moths) and 11 bumble bee species were recorded. Species richness and total densities of both butterflies and bumble bees differed between habitat types (Table 2 and Figs 1 and 2). For butterflies (F1,14·8 = 8·75, P= 0·010; Fig. 1) but not for bumble bees (F1,25·5 = 2·35, P= 0·14; Fig. 2) there was significantly higher species richness per segment in grasslands than in proximate linear habitats. Both butterflies (F1,38·7 = 23·9, P < 0·001; Fig. 1) and bumble bees (F1,33·3 = 6·15, P= 0·018; Fig. 2) had significantly lower species richness in distant linear habitats than in proximate ones. For butterflies, the species richness was higher in grasslands than in distant linear habitats (F1,22·6 = 9·64, P= 0·005; Fig. 1) while this difference was not significant for bumble bees (F1,26 = 0·07, P= 0·80; Fig. 2). The number of butterfly species was also positively related to increasing vegetation height, and both the number of butterfly and bumble bee species increased with increasing total floral abundance (Table 2).

Table 2.  Results of the effects of habitat and landscape factors on species richness and densities of butterflies and bumble bees. The d.f. column gives both the nominator (first) and denominator degrees of freedom (second). All significant effects are positive, i.e. increasing species richness and densities with increasing vegetation height, floral abundance and floral diversity
 Species richnessDensity
Habitat type24·92,22·5< 0·00120·02,29·7< 0·001
Vegetation height21·01,10·10·0012·411,30·10·13
Floral diversity1·9920·50·170·191,30·30·67
Floral abundance13·01,10·70·0048·341,31·60·007
Host-plant species0·011,15·30·920·031,34·40·86
MobilityNot tested  471,16·1< 0·001
Habitat type × mobilityNot tested  6·732,19·40·006
Visit46·35,1041< 0·00120·65356< 0·001
Visit × type5·2910,1073< 0·0012·3110 3360·012
Bumble bees
Habitat type3·462,28·20·0454·802,23·30·018
Vegetation height0·101,170·760·581,50·80·45
Floral diversity1·801,14·60·200·141,500·71
Floral abundance14·31,200·0015·071,29·10·032
Tongue lengthNot tested  10·21,11·90·008
Nest preferencesNot tested  4·881,11·20·048
Habitat type × tongue lengthNot tested  3·092,22·70·066
Habitat type × nest preferencesNot tested  1·782,42·20·18
Visit11·85,1061< 0·0018·975345< 0·001
Visit × type0·7410,10830·740·7910 3370·63
Figure 1.

Mean number of butterfly species per 100-m transect segment and mean density (number of individuals per 100 m2 and visit) of butterflies at each visit in semi-natural grasslands, proximate (Prox) and distant (Dist) linear habitats. Each bar represents one visit in each habitat. There were larger differences both in species richness and density between visits in the linear habitats than in the grasslands.

Figure 2.

Mean number of bumble bee species per 100-m transect segment and mean density (number of individuals per 100 m2 and visit) of bumble bees in semi-natural grasslands, proximate (Prox) and distant (Dist) linear habitats.

The density of bumble bees was higher in proximate linear habitats than in grasslands (F1,25·1 = 8·36, P= 0·008; Fig. 2), while there was no difference in density of butterflies between proximate linear habitats and grasslands (F1,28·7 = 0·14, P= 0·71; Fig. 1). For butterflies the density was higher in both proximate linear habitats (F1,28·4 = 25·0, P < 0·001; Fig. 1) and grasslands (F1,27·8 = 25·6, P < 0·001; Fig. 1) than in distant linear habitats. For bumble bees there was a non-significant tendency to higher density in proximate linear habitats than in distant ones (F1,20·9 = 3·78, P= 0·065; Fig. 2) and the density in grasslands and distant linear habitats did not differ significantly (F1,25·2 = 2·04, P= 0·17; Fig. 2). The density of both bumble bees and butterflies increased with increasing floral abundance (Table 2).

temporal patterns

For both butterflies and bumble bees, the number of species and densities differed between visits (Table 2). The numbers of butterflies were generally lower in the beginning of the season, but there were also significant interaction effects between habitat type and visit number for both species richness and density (Table 2), with a more constant number of species in the grasslands and a much more pronounced ‘peak’ in species richness in the linear habitats (Fig. 1). Bumble bee densities and numbers of bumble bee species were lowest at the first visit in all habitats (data not shown) and there were no interactions between habitat type and visit number (Table 2).

density of butterflies differing in mobility

There was a significant interaction effect between mobility and habitat type (Table 2), indicating larger differences between habitat types for sedentary than mobile species. In separate analyses, the densities of both sedentary (F2,31 = 21·9, P < 0·001) and mobile (F2,29·8 = 4·23, P= 0·024) species differed significantly between habitat types. For both groups there was an apparent isolation effect, i.e. higher densities in proximate than in distant linear habitats (sedentary F1,30·9 = 27·8, P < 0·001; mobile F1,29·6 = 6·83, P= 0·014) and also higher densities in grasslands than in distant linear habitats (sedentary F1,30·1 = 32·8, P < 0·001; mobile F1,29·9 = 4·74, P= 0·038). For sedentary (F1,31 = 3·46, P= 0·072) but not mobile (F1,30 = 0·01, P= 0·91) species there was also a tendency to higher densities in grasslands than in proximate linear habitats. This difference was nearly statistically significant (F1,11 = 4·63, P= 0·054).

density of bumble bees differing in tongue length and nest-site preferences

Bumble bee density was not significantly affected by an interaction between habitat type and nest-site preferences (Table 2) but tended to be affected by an interaction between habitat type and tongue length (Table 2). In separate analyses of densities of short-tongued and long-tongued bees, the density of short-tongued individuals (F2,21·5 = 5·46, P= 0·012), but not of long-tongued ones (F2,21·8 = 1·97, P= 0·16), differed between habitat types. The density of the short-tongued species was higher in proximate linear habitats than in grasslands (F2,24·4 = 10·5, P= 0·004) and there was a tendency to higher density in distant linear habitats than in grasslands (F2,24·4 = 3·97, P= 0·058) but no difference between the two linear habitat types (F2,20·6 = 2·63, P= 0·12).


This study clearly demonstrates that the presence of semi-natural grasslands has a positive effect on species richness and abundance of butterflies and bumble bees in linear habitat elements in agricultural landscapes. We found significant differences in both species richness and density of individuals between proximate and distant linear habitats, despite the fact that they did not differ in any of the aspects of habitat quality that we measured. Thus this study adds to other recent studies suggesting that habitat heterogeneity in the form of semi-natural grasslands is a key for maintaining farmland biodiversity (Steffan-Dewenter et al. 2002; Tscharntke et al. 2002a, 2002b; Benton, Vickery & Wilson 2003; Öckinger & Smith 2006; Holzschuh et al. 2007).

The mechanisms explaining our results may differ between butterflies and bumble bees. For the butterflies, grasslands may act as source habitat (cf. Pulliam 1988; Dias 1996) and the linear habitats as either true sink habitat (cf. Pulliam 1988) or pseudo-sink habitat (cf. Watkinson & Sutherland 1995). The grasslands are larger, contain higher abundances of potential larval host-plants for many of the butterfly species than the linear habitats and thus are more likely to contain viable populations of these species. The more accelerated increase in both density and species numbers over the season in the linear habitats, especially the distant ones (Fig. 1), is consistent with a successive dispersal of butterflies from grasslands to the linear habitats, even though this pattern might also be an effect of differences in species composition between grasslands and linear habitats.

For both the sedentary and mobile butterfly species monitored, the distant linear habitats were at least 1 km from the nearest potential population source, resulting in significant isolation effects, with lower numbers of species and fewer individuals compared with proximate linear habitats. Densities of sedentary but not of mobile species differed between grasslands and proximate linear habitats, which may be caused by an isolation effect if most of the sedentary butterflies do not utilize resources within more than a few hundred metres. However, an alternative explanation could be that mobility is correlated with habitat specialization (Southwood 1977), such that sedentary species are more specialized on certain resources mainly found in semi-natural grasslands.

For bumble bees, we observed similar isolation effects when comparing proximate and distant linear habitats. However, as bumble bees are central place foragers (Goulson 2003), other mechanisms may act on them. The majority of individuals observed in the field were foraging, and the abundance of bumble bees in different habitats is therefore likely to reflect mainly the amount of food resources available. Even though some species can fly long distances during foraging (Walther-Hellwig & Frankl 2000b; Darvill, Knight & Goulson 2004), and hence are capable of utilizing resources over large areas (Steffan-Dewenter et al. 2002), the density of foragers must be assumed to decrease with increasing distance from the nest. Bumble bee species differ in nest-site preferences, some nesting mainly subterraneously while others nest close to the ground surface or in cavities above ground (Alford 1975; Kells & Goulson 2003). In our landscapes, the small linear habitats may contain suitable nest sites for some bumble bee species but it is likely that the semi-natural grasslands, which are often comparably rich in stones, tussocks and similar structures, provide a larger number of suitable nest sites than both the linear habitat elements and surrounding cultivated fields, and hence probably have a higher density of bumble bee nests. This would explain the higher species richness and density of bumble bees in proximate than distant linear habitats.

We found no effect of nest-site preferences and only a (near-significant) tendency to effect of tongue length on bumble bee densities in different habitat types. In separate analyses we found differences in density of short-tongued but not of long-tongued species between proximate and distant linear habitats. We had expected the opposite pattern, as long-tongued bumble bees are generally more specialized than short-tongued species (Ranta & Lundberg 1980; Goulson 2003) and hence could be expected to be more dependent on larger areas of semi-natural grassland. However, the degree of specialization of food resources does not necessarily reflect specialization in habitat choice.

The fact that the density of bumble bees was higher in proximate linear habitats than in grasslands may seem surprising, as both the abundance and diversity of flowers were higher in the grasslands. However, the composition of plant species differed between grasslands and linear habitats and the latter usually contained a higher abundance of tall plant species with large, nectar-rich inflorescences that are attractive to bees, such as Knautia arvensis, Cirsium spp. and Centaurea spp. (E. Öckinger & H.G. Smith, unpublished data). It is also possible that the timing of the peak in flower or nectar abundance differed in grasslands and linear habitats, and our floral survey might have matched the peak in the grasslands better than the peak in the linear habitats.


Several previous studies have demonstrated a positive effect of landscape complexity and diversity on insect species richness and abundance in agricultural landscapes (Steffan-Dewenter, Münzenberg & Tscharntke 2001, 2002; Tscharntke et al. 2002b; Krauss, Steffan-Dewenter & Tscharntke 2003; Weibull, Östman & Granqvist 2003; Bengtsson, Ahnström & Weibull 2005). Typically, at least in western Europe, the landscape-scale percentage of grasslands is correlated with mean field size, the total length of field margins and the diversity of habitat types, implying that the effects of these factors may be difficult to disentangle (Roschewitz, Thies & Tscharntke 2005). However, we have shown that, in landscapes dominated by intensive agriculture in which the habitat for grassland species is highly fragmented, the presence of even small semi-natural grassland fragments leads to higher abundance and species richness of butterflies and bumble bees in adjacent small habitat fragments.

Our findings have implications for pollination in agricultural landscapes. Habitat fragmentation and intensified agricultural practices pose a threat to services provided by pollinators (Kearns, Inouye & Waser 1998; Wilcock & Neiland 2002; Steffan-Dewenter, Potts & Parker 2005; but see Ghazoul 2005). Bumble bees are important pollinators of both wild plants and crops (Goulson 2003) and, even though the pollination performed by butterflies is of small economic value, they are important pollinators for some plant species (Jennersten 1984). A lower density of pollinators at larger distances from semi-natural grasslands may lead to fewer visits per flower (Steffan-Dewenter, Münzenberg & Tscharntke 2001) and, at least for some plant species, reduced seed set (Jennersten 1988; Knight et al. 2005). Moreover, our results show that community composition in linear habitats may be affected by loss of grasslands. Mobile butterflies and generalist bumble bees increased in dominance with increasing isolation from semi-natural grasslands. Patterns similar to those found in this study may also be expected for other groups of pollinating insects (Steffan-Dewenter et al. 2002). This may have cascading effects on the composition of the plant community (Steffan-Dewenter et al. 2006). However, a more complete understanding of the effect of loss of grassland also requires additional studies on the effects on pollinator behaviour and herbivory (Steffan-Dewenter, Münzenberg & Tscharntke 2001; Knight et al. 2005).

In order to maintain a high diversity and viable populations of pollinator species that are fairly common today, we suggest that it is essential to preserve the few remaining fragments of semi-natural grasslands. Also, restoration and re-creation of patches of flower-rich grassland vegetation would increase the species richness and abundance of insect pollinators in surrounding intensively farmed areas. Agri-environment schemes including measures to promote the preservation and restoration of semi-natural grasslands in intensively farmed agricultural landscapes would not only preserve species in the focal grasslands but also contribute to higher insect diversity and abundance and thereby probably more efficient pollination in the wider agricultural landscape.


Sven G. Nilsson, Maj Rundlöf and two anonymous referees gave valuable comments on the manuscript. The study was supported by the Swedish Environmental Protection Agency through the research program ‘The Conservation Chain’. H. G. Smith was supported by a grant from the Swedish Research Council for Environment, Agricultural Sciences and Spatial Planning (Formas).