The influence of grazing intensity and landscape composition on the diversity and abundance of flower-visiting insects

Authors


Correspondence author. E-mail: Erik.sjodin@ekol.slu.se

Summary

  • 1The loss of semi-natural grasslands in agro-ecosystems has increased the importance of adequate management of remaining grasslands. Recommendations for intensive grazing have been debated because the effects of different management practices may differ between taxa and species. The increased fragmentation of grasslands suggests that the influence of management practices should be studied in a landscape context.
  • 2We studied four groups of flower visitors, many of which are pollinators, bees (Apoidea), butterflies (Lepidoptera), hoverflies (Syrphidae) and beetles (Coleoptera), in semi-natural grasslands managed at three intensity levels in eight areas in central Sweden. Local characteristics of the grasslands were recorded and landscape diversity was quantified. Vegetation height was correlated with grazing intensity: intensive grazing with the shortest vegetation and abandoned grassland with the tallest.
  • 3The insect groups responded differently to grazing intensity. Species richness and abundance differed between management regimes for beetles and hoverflies but not for bees and butterflies.
  • 4The effects of local habitat and landscape composition on species richness, abundance and composition differed between groups. Bee diversity responded to both local and landscape factors. Butterflies were mainly affected by local vegetation height and linear elements in the landscape. More species of hoverflies were recorded in tall vegetation and in landscapes with high forest cover. Beetles responded only to local environment characteristics.
  • 5Synthesis and applications. We demonstrate the importance of studying different insect groups simultaneously when evaluating habitat and landscape qualities for diversity. The results suggest that planning for conservation of biodiversity at landscape scales may be better than implementing grazing guidelines for individual grasslands. Grazing intensity should vary within or between landscapes to preserve pollinator diversity. Conservation management to encourage flower visitors cannot be generalized to include all groups simultaneously.

Introduction

Some of the most diverse communities in agricultural landscapes are found in semi-natural habitats, shaped over centuries by traditional management practices such as burning, mowing and grazing. In Sweden and central Europe, semi-natural habitats previously covered large areas and they were managed at variable but often low intensity, i.e. grazing animals foraged over large areas in mainly forested landscapes (Ihse 1995; Dahlström & Cousins 2006). The conservation value of semi-natural habitats is particularly high because the natural habitats for a large number of plants have been lost and today they are mainly present in traditionally managed grasslands (Eriksson, Cousins & Bruun 2002). Large areas of these grasslands have been fertilized and transformed into arable land. Many pastures have been abandoned and successively transformed into forest. The remaining semi-natural grasslands comprise only a fraction of their past area, and have a much more fragmented distribution in the landscape. As a result, there has been a reduction in the diversity and abundance of different organism groups, such as mammals and birds (Donald, Green & Heath 2001), plants (Luoto et al. 2003) and insects (Maes & van Dyck 2001; Biesmeijer et al. 2006), within semi-natural grasslands.

Insects make up a large part of the biodiversity of semi-natural grasslands, which are among the most species-rich habitats in European landscapes. The quality of grasslands varies according to natural factors, management history and present management (Morris 1969, 2000; Sjödin 2006). However, differences in diversity among semi-natural grasslands are not only caused by local conditions but also by qualities in the surrounding landscape (Tscharntke et al. 2005; Steffan-Dewenter et al. 2006). Several studies have shown the importance of semi-natural habitats in the landscape for insects (Kleijn et al. 2001; Steffan-Dewenter et al. 2002; Moradin et al. 2007; Öckinger & Smith 2007); others have pointed out the importance of the composition of the surrounding landscape (Weibull, Östman & Granqvist 2003; Ouin et al. 2004). In Sweden, as in most of western Europe, management is important for preventing forest vegetation from invading semi-natural grasslands (Ihse 1995). Adequate management and landscape planning is necessary to preserve diversity in grasslands (Opdam, Steingröver & van Rooij 2006). Official advice for the preservation of plant diversity has recommended intensive grazing of semi-natural pastures (Bakker 1998; Ekstam & Forshed 2000; Klimes & Klimesova 2001). It has been argued, however, that this is a threat to insect diversity (Völkl et al. 1993; Carvell 2002; Steffan-Dewenter & Leschke 2003). Alternative, less intensive, grazing strategies have been proposed (Bignal & McCracken 1996) but there are few studies examining the effects of management intensity on several insect groups simultaneously (but see Söderström et al. 2001; Kruess & Tscharntke 2002b).

Studies of diversity and abundance of the insect fauna in semi-natural grasslands have demonstrated the importance of low-intensity management for certain insect groups, for example bees and wasps (Carvell 2002; Steffan-Dewenter & Leschke 2003), beetles (Völkl et al. 1993; Dennis et al. 1997), butterflies (Erhardt 1985; Kruess & Tscharntke 2002a) and leafhoppers (Morris & Plant 1983; Morris 2000). However, it is unlikely that the same local and landscape factors are important for different taxa (Söderström et al. 2001; Kruess & Tscharntke 2002b; Vessby et al. 2002; Wolters, Bengtsson & Zaitsev 2006). Flower visitors have received little attention in relation to management (Morris 2000; Carvell 2002). In the present study, four groups of flower visitors were considered: bees, butterflies, hoverflies and beetles. Within these groups, many flower visitors may function as pollinators, although beetles are mainly predators, seed predators or herbivores on the plants. If different groups of flower visitors respond differently to the environment, this may suggest that, rather than treat flower visitors as a collective functional group, they should instead be evaluated and managed according to their ability to pollinate a certain target crop or plant community.

In the present study, grassland management intensity was investigated in south-central Sweden in relation to the surrounding landscape. We predicted the following. (i) Abandoned grasslands support a low diversity of flowering plants and therefore also a low diversity of flower-visiting insects, because of the dominance of competitive, often tall, plant species and shading from invading trees and bushes. (ii) Intensive grazing results in fewer flowers and a lower diversity of flower-visiting insects than low-intensity grazing, i.e. low-intensity grazing is optimal for many flowering plants and insects. (iii) The diversity of different taxonomic groups shows different patterns in response to local habitat and landscape composition.

Materials and methods

experimental design

The study was performed in eight areas in central Sweden, in the counties of Södermanland, Västmanland and Uppland, situated around Lake Mälaren (between 59°05′–59°50′N, 16°28′–17°40′E). All areas had similar amounts of forest (38·4 ± 3·4%; mean ± SE) and arable land (42·3 ± 4·3%). Semi-natural grasslands (10·9 ± 0·9%) and areas covered by water (7·2 ± 2·6%) were patchily distributed in the landscape. Three management regimes were studied: intensively grazed, grazed at low intensity, and abandoned grasslands that had not been grazed for at least 10 years. Cattle grazed all the intensively managed pastures. All farmers who kept cattle on intensively managed pastures were included in agri-environmental schemes with management contracts and economic compensation for conservation of biodiversity. Two of the pastures with low-intensity grazing were grazed by horses and these farmers also had contracts for 4 years. The other six grasslands with low-intensity grazing were grazed by cattle but farmers did not have agreements, or grasslands were specifically managed at low intensity for research or conservation.

The nation-wide inventory of Swedish semi-natural meadows and pastures in 1988–93 was used to find comparable grassland regimes (Söderström et al. 1993). In the inventory, grasslands were classified based on floral species richness and composition. The eight areas were situated far enough (at least 10 km) from each other to not have the same species pools. Within each area the three pastures were situated close enough to have the same species pool but separated enough to ensure that individual insects would not fly readily between sites (> 2 km). Potential triplets were examined in autumn 2003, when grazing intensity levels were confirmed by examination of litter cover.

In each of the eight areas, the three grazing intensity regimes were visited on the same day (if weather allowed) and in random order. Four observation plots (5 × 5 m) were chosen randomly and established for the duration of the study in open parts of the grasslands in dry to moist vegetation (wet areas were avoided). Insects were censused by observing study plots for 10 min, four times during the summer, between 7 June and 20 August 2004. It may not always be clear how representative the plots were but, compared with walking transects, it was easier to detect movement of small insects when walking around one spot and watching the same flowers from several directions. Comparing fixed plots also gives increased control for the effect of seasonal variation.

insect groups

Flower-visiting bees (Apoidea), butterflies (true butterflies, Rhopalocera, and burnet moths, Zygaenidae), hoverflies (Syrphidae) and beetles (Coleoptera) were included in the study (see Appendix S1, Supplementary material). These groups differ in life histories and may respond differently to management and other habitat and landscape characteristics. Some important differences between groups are larval habitat and utilization of plants. Beetle and butterfly larvae are mainly herbivores. Hoverfly larvae are mainly predators, often on aphids (Torp 1994; Sommaggio 1999). Bee larvae are totally dependent on pollen collected by adult bees. Adult beetles and hoverflies are mainly pollen eaters. Both groups feed mostly on shallow flowers presenting a lot of pollen (Kevan & Baker 1983; Wyatt 1983). Butterflies mainly feed on nectar and therefore often visit deep flowers. Adult bees feed both on pollen and nectar, most often from plants from which pollen is collected (Westrich 1990).

local and landscape characteristics

The following vegetation characteristics were measured as indicators of management intensity (Morris 1969): vegetation height, litter cover, microstructures, vertical temperature differences, flower abundance and number of flowering plants (Table 1). Vegetation height and litter accumulation both correspond directly to grazing intensity (Morris 1969, 2000) and vegetation height was therefore used as a continuous variable indicating grazing intensity. Site-specific characteristics were estimated at the grassland level and included: pasture size, tree and bush cover, and ground structure (Morris 2000). All sites contained bushes and trees that shaded parts of the study grasslands. Two measures of ground structure were estimated: sand-cover and stones. Cover of vegetation indicating eutrophication was estimated using indicator plant species and vegetation structures (Ellenberg et al. 1992) (Table 1).

Table 1.  Description of environmental variables at two scales: (a) local scale, management (grazing-related) characteristics measured in plots within grasslands and pasture characteristics; and (b) landscape scale, measured at three landscape scales in the grassland surroundings (600, 1200, 1800 m radius around each grassland)
VariableDescription
(a) Local scale
 Plot characteristic
 Vegetation heightMean of nine measurements using a rising plate meter (Sanderson et al. 2001). The measure combines effects of vegetation height and density
 Litter coverThe proportion cover (in 5% intervals) of visual dead plant material as means from four plots at the end of the season
 MicrostructuresThe number of cattle dung sites, distinct tussocks, bare ground surfaces, visual ant hills, visual stones and small grazed bushes or shrubs (Morris 1969)
 Vertical temperature variationThe ratio between air and ground temperature. The temperature in shadow was measured 10 cm above and at ground level
 Plant species richnessNumber of flowering plant species
 Flower abundanceCounts of flowers per species for all herbs presenting pollen and/or nectar. For the plant families of Asteraceae, Fabaceae, Plantaginaceae and Dipsacaceae it was more convenient to count inflorescences than flowers, and for Apiaceae and Rubicaceae whole plants were counted
 Pasture characteristics
 StonesEstimated from stone 1 (poor < 0·1 stone m−2) to 5 (rich > 1 stone m−2)
 Sandy soilEstimated as percentage cover (in 5% intervals)
 EutrophicationProportion of the grassland with vegetation affected by eutrophication
 Tree coverThe proportion (in 5% classes) of the grassland covered by trees and bushes
 Thick treesThe number of large trees (> 2·0 m perimeter at breast height) in the whole grassland
 Bush coverProportion of the grassland covered with bushes or shrubs (in 5% intervals)
 Pasture sizeEstimated as proportion of grassland cover from map component at 300 m around the mid-point of four study plots
(b) Landscape scale
 Arable landProportion (in 1% classes) arable land cover
 GrasslandProportion (in 1% classes) grassland cover
 WaterProportion (in 1% classes) water cover
 ForestProportion (in 1% classes) forest cover
 BuildingsNumber of buildings per circular area in the landscape
 Road lengthThe length of roads per circular area in the landscape
 Edge lengthThe length per circle area of edges between two map components

The surroundings of the pastures were analysed with ArcGIS 9, using the Swedish terrain map (vector map) obtained from the National Land Survey of Sweden (Lantmäteriet, Gävle, Sweden). GIS analyses were made within circles at three different radii (600, 1200 and 1800 m) surrounding each site. Land-cover classes included arable land, forest, grassland and water. The total edge length per area of these land-use categories was used as one measure of landscape heterogeneity. The total length of roads per area was used as a measure of landscape connectivity. This measure also reflected human activity, and both edge length and road length represented linear elements in the landscape. Density of buildings was used as an indirect measure of human presence (Table 1).

insect censuses

All insect and other visitors to the plots with typical flower-visitor attributes, such as an elongated proboscis and pollen collection hairs (Proctor 1978), were counted and determined to species (see below for exceptions). Three variables were calculated for each site for each of the four visitor groups, bees, hoverflies, butterflies and beetles: (i) total abundance (the number of individual visitors observed per plot), (ii) species richness (the number of species observed per plot) and (iii) species composition (calculated using the total abundance of each species on each site). Species were usually recognized in the field but samples were collected for a few problematic species. Vouchers were stored at the Department of Ecology, Swedish University of Agricultural Sciences, Uppsala, Sweden (see Appendix S1, Supplementary material). Some common species pairs not distinguishable in the field were grouped: Lepidoptera, Plebejus idas (Linné 1761) and P. argus (Linné 1758); Syrphidae, Melanostoma scalare (Fabricius 1794) and M. mellinum (Linné 1758). Within Sphaerophoria (Lepeletier et Serville 1828) only S. scripta (Linné 1758) was determined to species level (see Appendix S1, Supplementary material).

statistics

Means per grassland were calculated for all environmental variables as well as species richness and abundance.

Transformations and correlations

All environmental count data were log transformed, and all proportions were arcsine square-root transformed (Sokal & Rohlf 2000). Spearman rank correlations were calculated for all combinations of environmental variables. These correlations were used to select the variables to be retained in the following analyses (see below). All variables that were significantly correlated with vegetation height (used as an indirect measure of grazing intensity) and other variables with strong correlations (r > 0·5) were excluded, unless there were particular reasons to retain a variable.

Influence of management on environmental variables and insect groups

The influence of grassland management on environmental variables, as well as species richness and abundance, was tested in a complete randomized block anova procedure-mixed model (Littell et al. 1996) using management category as a class variable. Area was treated as a random block.

Analyses of group-specific responses to vegetation height and flower abundance

An ancova was used to evaluate the responses of the four insect groups. Insect species richness and abundance were tested with insect group as a factor and using the covariates vegetation height and flower abundance, respectively. Different responses between groups were indicated by a significant interaction between groups and vegetation height or flower abundance.

Local vs. landscape effects on richness and abundance

The influence of local and landscape measures (at 600, 1200 and 1800 m radii) on species richness and abundance for each insect group was analysed separately using stepwise forward regression analysis at each scale.

Local and landscape effects on species composition

Species composition of the four insect groups and flowering plants was analysed in relation to local habitat and landscape composition using multivariate ordination analyses. Species with single observations were excluded. In a detrended correspondence analysis (DCA), gradient length varied as follows: plants 1·00, Apoidea 2·35, Lepidoptera 1·49, Syrphidae 2·18 and Coleoptera 5·33. Because of the short to intermediate gradient length, a redundancy analysis (RDA) was used for plants and the first three insect groups, whereas for beetles (with a long gradient) canonical correspondence analysis (CCA) was used in the next step (Oksanen 2006). Biplot scaling was used for all taxonomic groups. Each taxonomic group was analysed separately in a partial RDA/CCA, with environmental variables at four spatial scales, local and landscape variables at radii 600 m, 1200 m and 1800 m. In each analysis, study area was used as covariate. Forward manual selection with Monte Carlo permutations (999 unrestricted permutations) was used for explanatory variables (inflation factor < 20) for each scale. This analysis first ordinates species and then correlates their distribution with environmental factors. Bees and hoverflies, which showed a response to both local and landscape factors, were re-analysed in a new ordination including local and landscape (at 1800 m) factors. Multivariate statistics were carried out using canoco 4·5 for Windows (ter Braak & Smilauer 2002). All other statistical analyses were done using SAS for Windows 8·02 (SAS Institute Inc., Cary, NC, USA).

Results

A total of 3613 insects carrying out 9467 visits was observed, covering six orders, 54 families and 294 species. The number of species per order was Coleoptera 35, Diptera 92, Hymenoptera 121, Lepidoptera 29 and, additionally, Heteroptera 16 and Neuroptera 1.

correlations among environmental variables

Vegetation height was correlated with litter cover, bush cover and thick trees; these three variables were therefore excluded in the following regression analyses (Table 2). The number of stones correlated positively with microstructures and the latter was excluded from the analyses of species richness, abundance and composition. There were only weak correlations between sandy soil and thick trees plus tree cover and eutrophication, and all these variables were retained for the analyses of species richness, abundance and composition (Table 2; see Appendix S2, Supplementary material).

Table 2.  Correlations (rs) between environmental variables at local and landscape scales measured at three radii around each grassland (1, 600 m; 2, 1200 m; 3, 1800 m radius). P-values are: *< 0·05, **< 0·01, ***< 0·001.
ScaleEnvironment variableCorrelated var.rs
Local internal
 Vegetation heightLitter cover0·88***
Vegetation heightBush cover0·53**
Vegetation heightThick trees–0·53**
StonesMicrostructures0·62**
Sandy soilThick trees0·41*
EutrophicationTree cover0·48*
Landscape internal
 Arable landWater123–0·66***
Arable landForest123–0·71***
Arable landEdge length1–0·46*
GrasslandWater3–0·48*
GrasslandForest1–0·56*
GrasslandEdge length230·45*
BuildingsRoad length0·52**
Local vs. landscape
 Vertical temperature variationForest30·42*
Vertical temperature variationEdge length230·44*
Plant species richnessEdge length230·46*
Flower abundanceArable land3–0·41*
Flower abundanceForest230·41*
StonesArable land1230·53**
StonesWater1–0·42*
Sandy soilRoad length230·40*
EutrophicationWater23–0·42*
Tree coverBuildings2–0·53**
Tree coverBuildings3–0·45*
Tree coverEdge length2–0·48*
Pasture sizeGrassland1230·68***
Pasture sizeForest1–0·64***
Pasture sizeForest3–0·43*

At the landscape level, arable land cover was excluded because of strong negative correlations with both forest and water cover at all scales. Pasture size was excluded because it was correlated with grassland and forest cover in the landscape (Table 2; see Appendix S2, Supplementary material). All other variables were retained (Table 2; see Appendix S2, Supplementary material).

management influence on environmental variables and species diversity

Vegetation was shorter and there was less litter in the intensive-grazing regime compared with low-intensity grazing. Both these categories had a lower mean vegetation height and less litter than abandoned grasslands (Table 3). Microstructures, temperature, flower abundance and plant species richness were not influenced by management regime (Table 3).

Table 3.  Pasture characteristics and species richness and abundance of four insect groups in relation to grassland management at three intensity levels. The analysis was a complete randomized block anova (mixed model, with landscape as random block). Means followed by different letters are significantly different in t-tests of least square-means (LSM). Means ± SE. Significance levels (P) are *< 0·05, **< 0·01, ***< 0·001
VariableIntensiveLow intenseAbandonedd.f.FUnit
Grazing characteristics
Vegetation height4·63 ± 1·07C9·14 ± 1·07B16·51 ± 1·07A2131·53***cm
Litter accumulation0·37 ± 0·07C0·72 ± 0·07B1·50 ± 0·07A2167·12***Arcsin square-root prop. plot cover
Microstructures14·53 ± 3·437·37 ± 3·437·22 ± 3·4321 1·48 NSNumbers plot−1
Vertical temperature variation1·09 ± 0·011·10 ± 0·011·09 ± 0·0121 0·12 NS°C
Flower abundance3286 ± 5574230 ± 5573591 ± 55721 0·75 NSFlowers site−1
Plant species richness34·63 ± 2·1333·75 ± 2·1331·38 ± 2·1321 0·63 NSSpp. site−1
Diversity measures
Bee species richness20·88 ± 1·5321·25 ± 1·5317·25 ± 1·5321 2·08 NSSpp. site−1 40 min−1
Bee abundance51·75 ± 7·4362·50 ± 7·4343·38 ± 7·4321 2·62 NSIndividuals site−1 40 min−1
Butterfly species richness9·37 ± 0·729·13 ± 0·7210·38 ± 0·7221 1·09 NSSpp. site−1 40 min−1
Butterfly abundance22·75 ± 3·5322·87 ± 3·5329·50 ± 3·5321 1·23 NSIndividuals site−1 40 min−1
Hoverfly species richness7·50 ± 1·22B9·88 ± 1·22A10·63 ± 1·22A21 5·51*Spp. site−1 40 min−1
Hoverfly abundance28·25 ± 5·47B35·38 ± 5·47AB44·38 ± 5·47A21 4·70*Individuals site−1 40 min−1
Beetle species richness2·75 ± 0·60C3·75 ± 0·60B5·13 ± 0·60A21 4·28*Spp. site−1 40 min−1
Beetle abundance4·50 ± 2·36B6·88 ± 2·36B15·00 ± 2·36A21 5·19*Individuals site−1 40 min−1

Within insect groups, species richness and abundance were significantly correlated. Bees and butterflies did not differ significantly in abundance or species richness between management categories. Hoverfly species richness and abundance were highest in low-intensity and abandoned grasslands. Beetle species richness was highest in the abandoned grasslands (Table 3).

effects of vegetation height and flower abundance on different groups

The variation (R2) in species richness explained by group identity and vegetation height in the ancova model was 81% (P < 0·0001). There was a significant interaction between vegetation height (VH) and species richness in different groups (ancova, group × VH, F= 5·01, P= 0·0030), indicating that insect groups were influenced in different ways by vegetation height (Fig. 1a). Species richness of hoverflies and beetles was highest in tall vegetation (i.e. low-intensity and abandoned grasslands), whereas species richness of butterflies and bees was not influenced by vegetation height (Fig. 1a).

Figure 1.

(a) Mean number of species per site of four insect groups in relation to vegetation height (VH). Significant regression lines are shown; significance levels are indicated as P= *< 0·05, **< 0·01. Linear regressions for bees and butterflies were non-significant. Hoverflies (thick solid line) = 5·63 + 0·67 × VH (R2 = 0·36, F= 12·11**), beetles (dashed line) = 2·36 + 0·15 × VH (R2 = 0·21, F= 5·86*). (b) Abundance of four insect groups in relation to flower abundance (FlwAb). Bees (thin solid line) = 22·93 + 0·008 × FlwAb (R2 = 0·33, F = 9·35**). Linear regressions for butterflies, hoverflies and beetles were non-significant. Insect and vegetation measures from 24 grasslands grazed at three intensity levels.

When analysing abundance in the four insect groups, an interaction between flower abundance (FlwAb) and the different groups was found (ancova, group × FlwAb, F= 4·08, P= 0·0093; Fig. 1b). Only bees showed a significant increase in abundance in response to flower abundance (Fig. 1b).

local and landscape effects on species richness and abundance

Bee species richness was positively associated with cover of sandy soil and, at larger landscape scales, road length. Bee abundance increased with an increase in the number of flowers, grassland cover and the edge length or length of roads in the landscape (Table 4).

Table 4.  Stepwise regression analyses on species richness and abundance of four insect groups in relation to local and landscape (600, 1200 and 1800 m) variables, analysed separately for each scale. Minus signs (–) following environmental variables indicate negative relations
Insect groupScaleEnvironmental variableCUM R2d.f.Model FModel P
  1. Lep., Lepidoptera; Syrph., Syrphidae.

Bee sp. richnessLocalSandy soil0·215226·040·0224
600 m    NS
1200 mRoad length0·3152210·130·0043
1800 mRoad length0·3162210·160·0043
Bee abundanceLocalFlower abundance0·3322210·950·0032
600 mEdge length0·231226·590·0175
600 mGrassland0·362215·960·0089
1200 mRoad length0·3472211·690·0025
1800 mRoad length0·300229·420·0056
Lep. sp. richnessLocal    NS
600 mRoad length (–)0·248227·240·0133
1200 mRoad length (–)0·3162210·150·0043
1800 mRoad length (–)0·214225·980·0229
1800 mEdge length0·364216·000·0087
Lep. abundanceLocal    NS
600 mRoad length (–)0·270228·140·0092
1200 m    NS
1800 m    NS
Syrph. sp. richnessLocalVegetation height0·3552212·110·0021
LocalEutrophication (–)0·451218·640·0018
600 mForest0·299229·360·0057
1200 mForest0·3592212·320·0020
1800 mForest0·3652212·640·0018
Syrph. abundanceLocalVegetation height0·4572218·480·0003
600 m    NS
1200 mForest0·228226·500·0183
1800 mForest0·303229·570·0053
Beetle sp. richnessLocalVegetation height0·210225·860·0242
LocalPlant sp. richness (–)0·386216·600·0060
600 m    NS
1200 m    NS
1800 m    NS
Beetle abundanceLocalVegetation height0·5052222·46< 0·0001
LocalSandy soil (–)0·5962115·49< 0·0001
600 m    NS
1200 m    NS
1800 m    NS

Butterfly species richness was negatively correlated with road length (at all scales) and positively correlated with edge length (1800 m). Butterfly abundance responded negatively to road length. Neither richness nor abundance was influenced by local factors (Table 4).

Hoverfly species richness was positively correlated with vegetation height and negatively with cover of vegetation, indicating eutrophication. At the landscape scale, proportion of forest was the strongest predictor of species richness at all radii. Hoverfly abundance was also related to vegetation height and forest cover in the landscape (Table 4). Beetle species richness was positively correlated with vegetation height and negatively with plant species richness. Fewer beetle individuals were observed in grasslands containing short vegetation and in grasslands with sandy soils (Table 4).

species composition

Plant species composition was influenced only by local variables: vegetation height and the grassland cover of sandy soil (Fig. 2). Species composition of the different insect groups responded to local and landscape variables in different ways (Fig. 2). For all groups, local factors explained a large part of the variation. Bee species composition mainly varied in relation to flower abundance. For butterflies, hoverflies and beetles, vegetation height was the main influence. Only bees and hoverflies responded to both local and landscape variables, and approximately 66% of the total bee variation and 55% of the hoverfly variation was explained by the two RDA axes (Figs 2 and 3). For bees, the proportions of land covered by grassland and water at the largest scale (1800 m) were most important (Fig. 2) and different bee species responded to different sets of factors (Fig. 3). For hoverflies, vegetation height was important (Fig. 2) and, when landscape factors were analysed separately, road length was most influential (Fig. 2).

Figure 2.

Explained amount of variance (%) in species composition of four insect groups and plants in relation to environmental variables at three grazing intensity levels in eight areas compared with a partial RDA (CCA for beetles), with area as covariable. Each spatial level, local (Loc) and landscape (Lsc; 600, 1200 and 1800 m), was analysed separately and the results are shown for the scale with the highest r2 (1800 m).

Figure 3.

Species composition of bees in relation to local and landscape environment variability analysed with an RDA using area as the covariable. Only species with at least a 15% correlation with the first two axes are shown. Grasslands with different grazing intensities are shown, the inclusion order (for significant environment variables) is indicated by numbers. Apimel, Apis mellifera; Osmbic, Osmia bicolor; Bomlap, Bombus lapidarius; Bomluc, Bombus lucorum; Macaeu, Macropis aeuropaea; Bomsub, Bombus subterraneus; Andsem, Andrena semilaevis; Haltum, Halictus tumulorum; Bomsyl, Bombus sylvarum; Andhel, Andrena helvola; Andnig, Andrena nigroaenea; Bomhor, Bombus hortorum; Nomfla, Nomada flavoguttata; Bomsor; Bombus soroëensis; Andfal, Andrena falsifica; Lasleu, Lasioglossum leucopus; Lascal, Lasioglossum calceatum; Checam, Chelostoma campanularum; Andhae, Andrena haemorrhoa; Psibor, Psithyrus barbutellus; Andsub, Andrena subopaca; Psiurup, Pisthyrus rupestris; Meglig, Megachile ligniseca; Cheflo, Chelostoma florisomne.

Discussion

Our study design (with replicates at the landscape level) allows separation of local, landscape and regional factors affecting insect communities. In contrast to our expectations, plant and insect species richness as well as insect abundance were not lower in abandoned grasslands compared with the other grassland management regimes. Intensively grazed grasslands did not have significantly fewer flowers, lower plant species richness or lower diversity of bees and butterflies than low-intensity grasslands. But species richness of beetles and hoverflies in intensive management was significantly lower than in low-intensity regimes. Mean vegetation height was closely linked to management regime, with the highest vegetation in abandoned grasslands and the lowest in intensively grazed grasslands. Because vegetation height was also the most important local determinant of species composition for butterflies, hoverflies and beetles, management intensity does play an important role regarding the identity of species found in these insect groups.

Diversity of the four insect groups responded differently to local habitat and landscape composition. Beetle richness, abundance and composition were solely related to local factors, mainly vegetation height. Hoverflies were more species rich and abundant in tall vegetation and where the landscape had high forest cover. Hoverfly species composition was also influenced by both local (vegetation height) and landscape (road length) variables. In contrast, bee species richness and abundance were not significantly correlated with vegetation height. Instead sandy soil and flower abundance influenced bee species richness and abundance at the local scale, whereas at a landscape scale linear elements were important. Bee species composition was related to flower abundance at a local scale and cover of water and grassland at a landscape scale. Butterfly species richness and abundance were only influenced by linear element landscape variables. In contrast, butterfly species composition was only influenced by a local-scale variable, vegetation height.

flower-visitor response to local variability

Vegetation height and accumulated litter were the two characteristics that best differentiated the three management categories (Table 3), as was also found in a study on insect behaviour in relation to grazing intensity (Sjödin 2006). Hoverflies and beetles were more abundant and species rich in grasslands with tall vegetation. One possible explanation is that increased biomass (indicated by both tall vegetation and high litter accumulation) provides more food for herbivorous beetles and more prey for predatory beetles and juvenile stages of hoverflies. Bees and butterflies, on the other hand, were neither more abundant nor species rich in tall vegetation. Species composition varied for butterflies in response to vegetation height, suggesting that some species were favoured by tall vegetation whereas others were more common in short vegetation. Previous studies have shown that some butterflies (e.g. Lycaenidae) prefer recently grazed pastures whereas others are associated with tall vegetation (Balmer & Erhardt 2000). Other studies have shown that butterfly diversity in general increases with tall vegetation (Kruess & Tscharntke 2002a).

For bees, community composition was mainly determined by flower abundance. Many small bee species are dependent on floral resources close to the nest (Gathmann & Tscharntke 2002; Vulliamy, Potts & Willmer 2006), whereas large species may often fly quite far from their nests, built either in tall vegetation or sparsely vegetated soils (Westrich 1990).

Flower abundance should be important to all flower visitors, but only bee abundance was significantly related to this variable. It is possible that beetles, butterflies and hoverflies may aggregate in relation to specific plant species associated with adult feeding sites or larval habitats rather than total flower abundance. Local and landscape factors have been shown to be related to species body size (Gathmann & Tscharntke 2002; Tscharntke et al. 2002). Large bees may aggregate in flower-rich sites in the landscape, whereas larger populations of small bees are mainly supported by flower-rich local habitats.

flower-visitor response to landscape variability

All groups except beetles responded to landscape composition. Flower-visiting beetles are a very heterogeneous group containing species with very different life histories, and many species may only sporadically appear on flowers. Because of this diversity it is probably difficult to find general landscape trends for beetles. In accordance with previous studies, forest cover was a good predictor of hoverfly diversity (Haslett 2001; Ouin et al. 2004). At sites in open landscapes, species that can be associated with trees were less common. Some hoverfly species’ larvae eat aphids specialized on tree species growing in forests (Torp 1994; Sommaggio 1999).

Bees (positive association) and butterflies (negative association) were both influenced by road length. Roads constitute linear elements that function as corridors connecting foraging patches in the landscape (Beier & Noss 1998; Tewksbury et al. 2002; Damschen et al. 2006). This function should be important for bees and butterflies, as both groups contain large-bodied species with good flight abilities (Neiminen 1996; Gathmann & Tscharntke 2002), qualities that have previously been found to respond to features at large landscape scales (Steffan-Dewenter, Münzenberg & Tscharntke 2001; Tscharntke et al. 2002; Bergman et al. 2004).

A possible explanation for the opposite responses by bees and butterflies might be that roadsides are linked to human activity and this may cause direct mortality for some insects. Roadsides may be used as habitat for larvae in both groups. In the case of bees, sloping roadsides often constitute preferred nest sites, as erosion keeps vegetation sparse. Moreover, road length is also related to grassland cover of sandy soils, which may partly explain the increased bee species richness, because many species use such soils for their nests. Roadsides may, on the other hand, constitute sink habitats for butterflies (Battin 2004), even though they appear to be attractive breeding habitats. The roadside is often flower-rich and sun-exposed, i.e. qualities preferred by many butterfly species (Erhardt 1985; Balmer & Erhardt 2000), but dust and removal of vegetation by mowing might be detrimental to butterfly larvae (cf. Öckinger & Smith 2007). Mortality from vehicles might be higher for butterflies, possessing large and fragile wings, than for bees. Perhaps the fact that proportionally fewer butterfly individuals were found flying along roads (Ouin et al. 2004) can be explained by such direct mortality. Clearly, the influence of roads on insects merits more attention.

The strong influence on bees of the proportion of water cover in the landscape (Fig. 2) was unexpected. Water is a useless habitat for bees, although the shore may be flower-rich and contain nectar-rich flowers (Leiss & Klinkhamer 2005). Proximity to water may also indicate the existence of preferred flowers. The most probable explanation is, however, that water cover is correlated with other landscape characteristics. For example, low water cover was related to high cover of arable land. The latter is often associated with eutrophication and pesticide use (which is directly detrimental to insects). Water cover was also positively related to landscape diversity because water cover was not present in all areas.

management implications

The four insect groups studied here differed significantly in their responses to grazing, as indicated by vegetation height. Grasslands containing tall vegetation, i.e. low intensity or abandoned grasslands, showed higher visitor richness than intensive grazing for two of the four insect groups, and no insect group was more species rich or abundant in intensively managed pastures. For both bees and butterflies, the weak relationship with grazing intensity can be explained by a large variation in abundance responses of different species. Some bee species build nests in tall vegetation (e.g. some bumblebees) whereas others dig holes in the soil (Vulliamy, Potts & Willmer 2006). These groups need grasslands in different successional stages or grasslands on different types of soil to maintain diversity.

Although diversity was not higher in intensively grazed grasslands, our results suggest that many species may be lost if no sites with intensive grazing are present at the landscape level, particularly among bees and butterflies (Figs 2 and 3). If the goal is to preserve species, variation in management at the landscape level will be preferable to managing for variation within grasslands, as it ensures that local populations are large enough to survive. Another option may be to vary management between landscapes to favour different groups. Traditional land use, knowledge of residents and the preferences of stakeholders may form the basis for decisions on management in different landscapes (Opdam, Steingröver & van Rooij 2006).

Acknowledgements

We thank all farmers for providing access to their grasslands, county council staff for helping locate comparable grasslands, Göran Adelskjöld for help with GIS, Sandra Öberg and Ulf Grandin for valuable discussions about multivariate statistics, Birgitta Vegerfors for statistical assistance, Åke Berg, Johan Ahnström, Karin Ahrné and Anna-Karin Kuusk for valuable comments on the manuscript, and for their help with species determination Gunnar Sjödin (beetles), Ruth Hobro (bugs), L. Anders Nilsson (bees) and Hans Bartsch (hoverflies). The project was funded by MISTRA as a part of the HagmarksMISTRA project.

Ancillary