Interaction diversity within quantified insect food webs in restored and adjacent intensively managed meadows

Authors

  • MATTHIAS ALBRECHT,

    1. Swiss Federal Institute for Forest, Snow and Landscape Research, Zurcherstrasse 111, CH-8903 Birmensdorf, Switzerland, Institute of Environmental Sciences, University of Zurich, Winterthurerstrasse 190, 8057 Zurich,Switzerland
    Search for more papers by this author
  • PETER DUELLI,

    1. Swiss Federal Institute for Forest, Snow and Landscape Research, Zurcherstrasse 111, CH-8903 Birmensdorf, Switzerland, Institute of Environmental Sciences, University of Zurich, Winterthurerstrasse 190, 8057 Zurich,Switzerland
    Search for more papers by this author
  • BERNHARD SCHMID,

    1. Swiss Federal Institute for Forest, Snow and Landscape Research, Zurcherstrasse 111, CH-8903 Birmensdorf, Switzerland, Institute of Environmental Sciences, University of Zurich, Winterthurerstrasse 190, 8057 Zurich,Switzerland
    Search for more papers by this author
  • CHRISTINE B. MÜLLER

    1. Swiss Federal Institute for Forest, Snow and Landscape Research, Zurcherstrasse 111, CH-8903 Birmensdorf, Switzerland, Institute of Environmental Sciences, University of Zurich, Winterthurerstrasse 190, 8057 Zurich,Switzerland
    Search for more papers by this author

Christine Müller, Institute of Environmental Sciences, University of Zurich, Winterthurerstrasse 190, 8057 Zurich, Switzerland. Tel.: +41 1 635 48 06. Fax: +41 44 635 61 29. E-mail: cbm@uwinst.uzh.ch

Summary

  • 1We studied the community and food-web structure of trap-nesting insects in restored meadows and at increasing distances within intensively managed grassland at 13 sites in Switzerland to test if declining species diversity correlates with declining interaction diversity and changes in food-web structure.
  • 2We analysed 49 quantitative food webs consisting of a total of 1382 trophic interactions involving 39 host/prey insect species and 14 parasitoid/predator insect species. Species richness and abundance of three functional groups, bees and wasps as the lower trophic level and natural enemies as the higher trophic level, were significantly higher in restored than in adjacent intensively managed meadows. Diversity and abundance of specific trophic interactions also declined from restored to intensively managed meadows.
  • 3The proportion of attacked brood cells and the mortality of bees and wasps due to natural enemies were significantly higher in restored than in intensively managed meadows. Bee abundance and the rate of attacked brood cells of bees declined with increasing distance from restored meadows. These findings indicate that interaction diversity declines more rapidly than species diversity in our study system.
  • 4Quantitative measures of food-web structure (linkage density, interaction diversity, interaction evenness and compartment diversity) were higher in restored than in intensively managed meadows. This was reflected in a higher mean number of host/prey species per consumer species (degree of generalism) in restored than in intensively managed meadows.
  • 5The higher insect species and interaction diversity was related to higher plant species richness in restored than in intensively managed meadows. In particular, bees and natural enemies reacted positively to increased plant diversity.
  • 6Our findings provide empirical evidence for the theoretical prediction that decreasing species richness at lower trophic levels should reduce species richness at higher trophic levels, and in addition lead to even stronger reductions in interaction diversity at these higher levels. Species at higher trophic levels may thus benefit relatively more than species at lower trophic levels from habitat restoration in the grassland ecosystems studied. We also demonstrate enhanced compartment diversity and lower interaction evenness in restored than in intensively managed meadows, both of which are theoretically positively associated with increased ecosystem stability in restored meadows.

Introduction

Agricultural intensification during the past decades has been identified as the major cause of the worldwide loss of biodiversity (Robinson & Sutherland 2002; Foley et al. 2005). Empirical evidence is growing that in simplified, intensively managed ecosystems, biotic interactions can become disrupted as a consequence of this biodiversity loss (Östman, Ekbom & Bengtsson 2001; Tscharntke et al. 2005), and associated ecosystem services such as natural pest control and pollination of crops will be at risk (Matson et al. 1997; Balvanera et al. 2006). Currently there is limited knowledge on how the loss of biodiversity affects the functioning of whole food webs in human-influenced ecosystems (van der Putten et al. 2004; Kremen 2005; Klein, Steffan-Dewenter & Tscharntke 2006).

Quantitative food webs describe the feeding links and interaction strength among species at different trophic positions (basal resource species, consumers and top predators or parasites), and can explore the factors that structure and maintain ecological communities and ultimately biological diversity (Müller et al. 1999; Lewis et al. 2002; Tylianakis, Tscharntke & Lewis 2007). Food-web structure altered by human intervention may have important consequences for biodiversity conservation and ecosystem functioning (Pimm 1979, 1991), because decreased asymmetries in energy flow and interaction strength are theoretically identified as factors jeopardizing ecosystem stability and function (McCann, Hastings & Huxel 1998; McCann 2000; Rooney et al. 2006). As species diversity and abundance decrease, the complexity of biotic interactions will also decrease and leave systems more prone to extinctions of important key elements (Mélian & Bascompte 2002; Ives & Cardinale 2004). Recent progress in the analysis of quantitative food webs (Cohen et al. 1993; Müller et al. 1999; Bersier, Banasek & Cattin 2002; Lewis et al. 2002) can now be brought into contact with applied problems, such as the effects of intensified land use or the effectiveness of agri-environment schemes (AES) that restore part of the agricultural land to increase biological diversity.

Recognition of the negative impacts of modern intensive agriculture on agro-biodiversity and associated ecosystem functioning has led to the implementation of AES. These schemes provide financial incentives to farmers for creating benefits to environment and biodiversity by the adoption of moderate farming practices, and are considered the most important policy instruments to conserve and restore biodiversity in European agricultural landscapes (EU 2005). The effectiveness of AES in restoring the functioning of communities and ecosystems by enhancing biotic interactions has rarely been tested (Kleijn & Sutherland 2003; but see Östman et al. 2001; Albrecht et al. 2007). The need for studies aimed at a better understanding of how AES shape whole communities and drive their functioning within agroecosystems has been repeatedly stressed (van der Putten et al. 2004; Tscharntke et al. 2005; Klein et al. 2006).

Diversity pattern, community structure and biotic interactions are often driven by processes that are not confined to a local habitat patch (e.g. Tscharntke et al. 2005). Most species living in agroecosystems depend on complementary resources in different habitats to successfully fulfil all their resource requirements and life cycles (Dunning, Danielson & Pulliam 1992; Klein, Steffan-Dewenter & Tscharntke 2004). Increasing isolation from natural source habitats is often associated with a decline in biodiversity (Duelli & Obrist 2003; Klein, Steffan-Dewenter & Tscharntke 2003). In particular, higher trophic species of parasites and predators may suffer disproportionately more from isolation than their host or prey (Pimm 1991; Lawton 1995; Davies, Margules & Lawrence 2000) as a consequence of their limited dispersal and colonization ability, and their tendency to occur in small and variable populations (Kruess & Tscharntke 1994; Holt et al. 1999).

We analysed 49 quantitative insect food webs of trap nests at 13 sites in Switzerland to test if a decline in species richness at lower trophic levels was correlated with that at higher trophic levels, and if interaction diversity declined more rapidly than species diversity in intensively managed meadows as compared with meadows restored under an AES. Trap-nesting Hymenoptera are attacked by an assemblage of parasitoid and predator species and hence allow analysis of interactions among trophic levels (host–parasitoid and predator–prey), ecosystem functioning (mortality by parasitoids and predators), and quantitative food-web metrics (e.g. linkage density, interaction diversity, interaction evenness, compartment diversity, generality, vulnerability). The trap-nesting communities include three important functional groups: bees provide pollination services to both wild plants and crops (Corbet, Williams & Osborne 1991), and predatory wasps control certain pest species (Harris 1994) or act as top predators. We also tested for distance effects between restored and adjacent intensively managed meadows to answer the question whether the trophically higher ranked parasitoids and predators responded less or more strongly to increasing distance from restored habitats.

Materials and methods

study sites

In Switzerland at least 7% of the farmland is managed as ecological compensation areas (ECAs) that represent a variety of specifically defined biotopes. In 2003, ECAs covered 13% of the cultivated area of Switzerland (Herzog et al. 2005). As approximately 80% of the Swiss agricultural area is grassland, ECA meadows constitute by far the most widespread type, with a share of 78% in 2004 (BLW 2004). Postponed mowing (after 15 June in the Swiss lowland) and prohibition of fertilizer application are the major prescriptions for the management of ECA meadows.

In 2003 and 2004, 13 ECA meadows (= restored meadows) and adjacent intensively managed meadows (IMM) identical in slope and exposition were selected in the eastern part of the Swiss lowland (four in 2003 and nine in 2004; all sites between 370 and 710 m a.s.l.), ranging in size from 0·48 to 2·15 ha (mean = 1·05 ± 0·13 ha). The region (30 × 50 km) is characterized by intensive agriculture embedded in a small-scale mosaic of grasslands, arable fields and forests, and the average distance between our sites was 5·8 ± 0·9 km. No seminatural habitats or other ECAs occurred within a perimeter of 200 m around the selected meadows, and the distances to orchard trees did not differ significantly among exposed trap nests.

insect species

Wooden posts, each fitted with two trap nests at a height of 1·5 m, were set up in the centre of each of the 13 restored meadows and in the adjacent IMMs at distances of 25, 50 or 100 m (= 52 ‘stations’). All trap nests consisted of a plastic tube (diameter 11 cm), containing ≈ 200 internodes of common reed Phragmites australis (Cav.) Trin. with diameters ranging from 2 to 10 mm and a length of 23 cm (Tscharntke, Gathman & Steffan-Dewenter 1998). The trap nests were protected against rain by wooden roofs (50 × 50 cm). Trap nests were set up in the field in mid-April and collected in mid-October. Three trap-nest stations had to be omitted from analyses because they were damaged by agricultural machines. The collected trap nests were stored at 5 °C for 14 weeks to mimic a hibernation phase before all reed internodes were opened in the laboratory. Each nest was then reared separately in glass tubes closed by pieces of cotton to let imagines emerge for identification. Species richness was the total number of species observed, and abundance of trap-nesting Hymenoptera was the number of brood cells per station. Abundance of natural enemies was the number of parasitized brood cells. The rate of parasitism was the number of brood cells attacked divided by the total number of brood cells of the host/prey species (not all cells attacked resulted in host death: in 2·1% of cases, hosts emerged despite parasitization). Mortality rate by natural enemies was the number of individuals killed divided by the number of brood cells per host/prey species.

quantifying food-web structure

Pooled quantitative food webs for each site were plotted. The quantitative, weighted measures of linkage density (LDq), interaction diversity (IDq), interaction evenness (IEq), compartment diversity (CDq), generality (Gq, the mean number of host/prey species per consumer species) and vulnerability (Vq, the mean number of consumer species per host/prey species) were calculated following Bersier et al. (2002), Lewis et al. (2002) and Tylianakis et al. (2007). The metrics to measure the diversity of prey (HN, the diversity of inflows) and consumer individuals (HP, the diversity of outflows) for each taxon k were based on the Shannon index:

image
image

where b•kis the number of host/prey individuals (brood cells) attacked by the natural enemy species k, and bk is the number of natural enemies attacking host/prey species k. The web metrics are based on the ‘reciprocals’ of HN,k and HP,k:

image
image

The quantitative linkage density (LDq) was calculated per trap nest station as:

image

where b.. is the total number of attacked individuals (brood cells) per station. Weighted generality (Gq) and vulnerability (Vq) were calculated as:

image
image

The quantitative diversity of links (IDq) and the dominance structure of the links (IEq) were calculated using the Shannon index:

image
image

where pi is the proportion of the interaction i of the total number of different trophic interactions (S).

The quantitative compartment diversity (CDq) was calculated as

image

where pi is the fraction of all species in the ith of n compartments. The number of compartments in a web was defined as the number of subwebs with no link to any other subweb.

plant species

All vascular plant species were mapped and counted within five 2 × 5-m strips arranged on both sides of each trap-nest station and parallel to the border of the ECA at an interval of 2·5 m. The vegetation surveys were carried out between May and June. One ECA meadow was cut before the vegetation survey could be done, and was omitted from all analyses including plant variables. Plant species richness was the cumulative number of plant species of all five sampling plots.

statistical analysis

General linear models with site as blocking factor, nested within year, and distance (ECA, 25, 50 and 100 m) as treatment factor, nested within site, were fitted to analyse variation in species richness and abundance, and in quantitative food-web properties (linkage density, interaction diversity, interaction evenness, compartment diversity, generality and vulnerability). As year had no significant effect on any of the response variables tested, data were pooled over the 2 years before further analysis. Distance was partitioned into a management contrast (ECAs vs. IMMs), a log-distance contrast (increasing distance within IMMs) and the remaining deviation from log-linearity (within IMMs). The log-distance contrast provided a better fit to the data than a linear-distance contrast. The insect species were divided into three functional groups: bees, wasps and natural enemies. The number of brood cells was closely correlated with the number of nests (bees, F1,40 = 157·33, P < 0·001; wasps, F1,47 = 376·05, P < 0·001). Therefore only the number of brood cells was analysed and reported.

To test the effects of management and log-distance from the ECA on the rate of parasitism and mortality by natural enemies, the same contrasts were used as described above for the general linear model, but fitting logistic models to binary response variables (attacked brood cells = 1, brood cells not attacked = 0; mortality by natural enemies = 1, not attacked brood cells or attacked brood cells with living progeny = 0). To account for statistical overdispersion, mean deviance changes were compared with F-tests (Crawley 2005). When analysing the response variables richness and abundance of natural enemies, rate of parasitism and mortality by natural enemies, only stations with host/prey species present were included in the analyses.

In further analyses, the covariables plant species richness and ECA area were included in the models to test for possible management effects on richness and abundance of bees, wasps, natural enemies and interactions, and mortality by natural enemies. For the response variables of natural enemies, the covariable number of brood cells was tested too. For the response variable mortality by natural enemies, the explanatory variables natural enemy species richness and abundance were included in the test. For the quantitative food web metrics as response variables, the explanatory variables number of brood cells and total species richness per food web were included in the model. A stepwise procedure with backward elimination was used to detect the minimum adequate model. Referring to the arguments of Moran (2003), we did not apply sequential Bonferroni, but carefully interpret results of multiple separate analyses (see Discussion). All statistical analyses were done with the statistical software r (R Development Core Team 2004). Where required, response variables were square-root transformed (abundance of all analysed groups of trap-nesting Hymenoptera, natural enemy abundance, generality, interaction evenness) to achieve normality and homoscedasticity (Crawley 2005). Arithmetic means ± 1 SE are reported in the Results section.

Results

insect food webs

The quantified food webs at the four distances (ECA and 25, 50 and 100 m from ECA within adjacent intensively managed meadows) were pooled and are depicted in Fig. 1 (see Appendix S1 in Supplementary material for species names and abundances). In total, 1941 nests and 8361 brood cells of 39 hymenopteran host species were found in the trap nests: 10 species of bees (Apidae; pollinators), 14 species of digger wasps (Sphecidae; predators of caterpillars, flies and spiders), 12 species of mason wasps (Eumenidae; predators of flies and caterpillars) and three species of spider wasps (Pompilidae; predators of spiders). Digger wasps were the most abundant family with a share of 55·5% of all brood cells, followed by bees (28·4%), mason wasps (13·5%) and spider wasps (2·6%). The digger wasp Trypoxylon figulus L. (Fig. 1; code 46) was the most abundant species and occupied 40·3% of all brood cells. Five other trap-nesting species were found in more than 5% of all brood cells: the digger wasp Passaloecus gracilis Curtis (10·7%; code 26), the bees Chelostoma florisomne L. (8·6%; code 11), Hylaeus communis Nylander (7·7%; code 18) and Osmia rufa L. (7·3%; code 15), and the mason wasp Ancistrocerus gazella Panzer (5·3%; code 5).

Figure 1.

Quantitative host–natural enemy food webs for (a) ecological compensation area (ECA) meadows (distance = 0), and at distances of (b) 25, (c) 50 and (d) 100 m from the ECA within intensively managed grasslands. Lower bars represent host abundance (wasps, darker grey and bees, lighter grey; pooled number of brood cells); upper bars, abundance of natural enemies (Hymenoptera, darker grey, Diptera, lighter grey and Coleoptera, lightest grey; pooled number of attacked brood cells), drawn at different scales. The width at the basis of the wedges represents the pooled frequency of each host–natural enemy interaction. Taxa are coded by numbers (see Appendix S1 for names).

Fourteen species of natural enemies attacked an average of 16·3% of all brood cells (bees, 16·9%; sphecid wasps, 17·1%; eumenid wasps, 15·6%; pompilid wasps, 1·8%). We recovered only parasitic wasps, flies and beetles, but no parasitic bees from our trap nests (Fig. 1; Appendix S1). The generalist parasitoid Melittobia acasta Walker (Hymenoptera: Eulophidae; code 62) was the most abundant natural enemy accounting for 46·8% of all attacks, parasitizing 7·7% of all brood cells. Chrysis cyanea L. (Hymenoptera: Chrysididae; code 53) and Gasteruption assectator L. (Hymenoptera: Gasteruptionidae; code 60) together caused 20·3% of total parasitism and together attacked 3·3% of all brood cells. Other species of natural enemies separately accounted for less than 5% of the total parasitism.

management and distance effects on insect species and interaction diversity

The size of the interaction webs in terms of species number of trap-nesting insects and their natural enemies decreased with distance, which is caused by a decrease in abundance of brood cells of bees with log-distance (Fig. 1; Table 1). Apart from this distance effect on brood cell number of bees, all diversity measures and mortality rates were affected more strongly by the factors site and management than by distance, with overall higher richness and abundance in ECAs than in adjacent IMMs (Table 1; Fig. 2). Only mortality rate by natural enemies showed a statistical trend to decrease with distance (= 0·053; Fig. 2c). Despite the increased mortality of trap-nesting insects by natural enemies in ECAs (F1,12 = 7·04, P = 0·021), the number of unattacked brood cells was still significantly higher in ECAs than in IMMs (F1,21 = 11·50, P = 0·003). Parasitism of the most abundant and widespread trap-nesting species, T. figulus (code 46) was caused by C. cyanea (code 53) and M. acasta (code 62) (together 88% of the number of parasitized brood cells of T. figulus). Here, neither an effect of management nor of distance from ECAs was found for the rate of parasitism by M. acasta, the most generalized parasitoid (management, F1,8 = 0·73, P = 0·418; distance, F1,8 = 0·51, P = 0·498). In contrast, the rate of parasitism of T. figulus by the specialist C. cyanea was significantly higher in ECAs and decreased with increasing distance (management, F1,8 = 37·23, P < 0·001; distance, F1,8 = 8·47, P = 0·020).

Table 1.  Results of nested general linear model analyses testing for the effects of study site, management (ECA vs. IMM), log-distance within the IMM on the species richness and abundance of trap-nesting Hymenoptera (bees and wasps), bees and wasps separately, natural enemies and weighted, quantitative food web properties
Response variablesdfSitedfManagementLog-distance*
FPFPFP
  • Deviation from log-distance was significant for none of the response variables and is not shown. The result for the rate of mortality due to parasitism and predation was obtained by fitting a log-linear (logistic) model to a binary data set (1 = attacked brood cell, 0 = not attacked brood cell). Significant interactions were found for the response variable compartment diversity (site–management interaction: F11,9 = 4·56, P = 0·015; site–log–distance interaction: F11,9 = 5·07, P = 0·011). Interaction terms for all other response variables were not significant and are not shown.

  • *

    For the response variable interaction evenness, linear distance was used instead of log-distance because it fitted the data better.

  • ECA, ecological compensation area; IMM, intensively managed meadow.

Species richness of bees12,2115·34< 0·0011,2116·92< 0·0010·020·882
Species richness of wasps12,33 2·56  0·0161,33 8·34  0·0071·870·181
Species richness of natural enemies12,33 4·53< 0·0011,33 6·51  0·0162·520·122
Abundance of bees (brood cell number)12,21 7·84< 0·0011,2115·28  0·0015·120·034
Abundance of wasps (brood cell number)12,33 3·21  0·0041,33 0·35  0·5580·420·524
Abundance of natural enemies (Number of attacked brood cells)12,33 4·02  0·0011,3315·24< 0·0010·060·802
Rate of mortality due to natural enemies12,33 3·61  0·0311,33 7·12  0·0214·620·053
Linkage density12,31 2·37  0·0271,31 4·13  0·0510·080·783
Interaction diversity12,9 6·06  0·0061,9 7·07  0·0261·810·212
Interaction evenness*12,9 5·67  0·0071,9 5·37  0·0464·960·053
Compartment diversity12,9 5·42  0·0081,918·71  0·0021·040·335
Generality12,31 1·66  0·1261,31 4·33  0·0460·780·385
Vulnerability12,31 2·13  0·0451,31 0·72  0·4040·030·864
Figure 2.

Mean abundance (a) and mean species richness (b) of trap-nesting bees, wasps and their natural enemies; (c) percentage brood cells attacked by natural enemies in ecological compensation areas (ECAs) (0 m) and at distances of 25, 50 and 100 m from the ECAs within intensively managed meadows (n = 13). Untransformed data are shown.

Linkage density, interaction diversity and compartment diversity were all influenced by site and were significantly higher in restored than in adjacent meadows (Table 1; Fig. 3). In addition, there was a significant interaction between site and log-distance (F11,9 = 5·07, P = 0·011) for compartment diversity, indicating exponentially decreasing compartment diversity with increasing distance from ECAs (Fig. 3b). The only web property that tended to increase with log-distance was interaction evenness (Table 1; Fig. 3c). Generality (mean number of host species per natural enemy species) was higher in restored meadows (ECA, 1·5 ± 0·2, IMM, 1·2 ± 0·1) but did not change with distance (Table 1). Vulnerability (mean number of attacking enemies for each host species) in turn was influenced by neither management (ECA, 1·4 ± 0·1, IMM, 1·3 ± 0·1) nor distance, but only by site (Table 1).

Figure 3.

Mean quantitative interaction diversity (a), mean quantitative compartment diversity (b) and mean quantitative interaction evenness (c) in ecological compensation areas (ECAs) (0 m) and at distances of 25, 50 and 100 m from the ECAs within intensively managed meadows (n = 13). Untransformed data are shown.

effects of covariables on insect species and interaction diversity

To detect potential mechanisms by which restoration management increased most of the measured diversity variables, we fitted the covariables site, plant richness and ECA area with stepwise linear models (ignoring the management and distance contrasts). Plant species richness explained a large amount of the total variation in species richness and abundance of bees and natural enemies (< 0·001; Fig. 4a). Because plant species richness was increased in ECAs (Albrecht et al. in press), a mechanistic link restoration → plant species richness → insect species and interaction diversity is a plausible explanation for the management effects reported in the previous section. Indeed, if the management contrast was fitted after plant species richness, it no longer affected bee and natural enemy abundances significantly. In a similar way, the covariate host/prey abundance explained a large amount of the total variation in species richness and abundance of natural enemies (< 0·001; Fig. 4b), indicating a mechanistic link host/prey insect diversity → natural enemy insect diversity. Finally, natural enemy richness and abundance explained a large amount of the total variation in the mortality of trap-nesting insects by natural enemies (< 0·001), indicating a link natural enemy diversity → interaction diversity. In contrast, there was no clear relationship between the number of brood cells and the proportion of cells attacked by natural enemies (Fig. 4c).

Figure 4.

(a) Relationship between plant species richness and species richness of trap-nesting bees per trap station; (b) relationship between abundance of trap-nesting insects (number of brood cells) and abundance of natural enemies (number of attacked brood cells, square-root transformed). (c) Relationship between abundance of trap-nesting insects (number of brood cells) and percentage of brood cells attacked by natural enemies (untransformed data).

For wasps, which in contrast to bees do not depend directly on plant resources, plant species richness was weaker in explaining species richness (= 0·02) and abundance was not at all influenced by plant species richness. The total number of brood cells and species per web explained a large amount of the total variance in link diversity and generality of food webs (< 0·001). However, management explained additional variance for the response variable generality (= 0·039). Linkage evenness, compartment diversity and vulnerability were best predicted by the total species richness of a web (< 0·001). For the response variable linkage evenness, plant species richness (= 0·007), site (= 0·001) and distance from ECAs (= 0·015) explained additional variation. Most of the variation for linkage density was explained by the number of brood cells (< 0·001).

Discussion

We showed that restored meadows have increased insect species and interaction diversity, which parallel increases in plant species richness, compared with adjacent intensively managed meadows. These effects are particularly strong for bees and natural enemies, as expected by effects driven up the food chain (Hulot et al. 2000) and predicted by hypotheses on resource heterogeneity and productivity (Strong, Lawton & Southwood 1984). Wasp diversity was less affected by basal plant resources or management than was bee diversity, because the dependence of wasps on plants is indirect: they are mainly consumers of herbivorous insects (Gathmann, Greiler & Tscharntke 1994). The food-web structure, in terms of linkage density, and interaction and compartment diversity, was richer or more complex in ECAs than in IMMs, which indicates successful attraction of species-rich, interacting assemblages by management. In contrast, a study using trap-nesting bees and wasps within tropical systems found that rates of parasitism were generally higher in more modified habitats (Tylianakis et al. 2007). However, in that study very different habitats, from forests to rice paddies, were compared, which reflect shaded habitats with closed tree canopy and open crop fields that might attract different insect communities. The study also found a strong negative effect of management intensity on interaction evenness, while our study detected only a trend (= 0·053) for interaction evenness to increase with distance from restored meadows.

A further desirable effect of restoration management would be that these habitats act as a source of insect diversity in adjacent agricultural habitats (Rand, Tylianakis & Tscharntke 2006). This was only partly supported, as most diversity and food-web measures were not, or only weakly, affected by distance from ECAs. Theoretically, higher levels of compartmentalization should increase the stability of food webs (Pimm 1979; McCann 2000). However, compartmentalization was generally quite low (on average 2·3 ‘true compartments’ in ECAs and 1·7 in IMMs) in our study, which may have resulted from relatively few highly specialized host/prey–natural enemy interactions. The generally high interconnectance of subwebs was mainly caused by the most abundant and highly generalized parasitoid M. acasta (code 62) and the high vulnerability of the most abundant host T. figulus (code 46).

Parasitism and predation decreased in intensively managed grasslands and with distance from the restored meadows. This is reflected in a significant effect of management on generality: as diversity and abundance of potential host/prey species decrease, the number of species attacked must decrease, and natural enemies become more restricted to single host or prey species, which results in fewer links between resource and consumer species. However, the partial release of trap-nesting Hymenoptera from parasitism with increasing distance from ECAs was caused not by a decrease in natural enemy abundance, but by an increased number of brood cells of wasps, the dominant host group; here, the most abundant wasp species T. figulus accounted for the observed pattern. A possible explanation may be interspecific competition with bees, such that wasps avoided nesting in the traps near ECAs, which contained more nests occupied by bees. However, on average only 10% of all reed stems per trap in the IMMs were occupied. According to theoretical predictions, increased generality may stabilize food webs in restored meadows (McCann et al. 1998; McCann 2000).

The multiple analyses with partly intercorrelated variables may represent some caveats on the interpretation of our results. By chance, 5% of all our tests should have produced significant results even if there would have been no ‘true’ effects. However, we found 67% of the explanatory variables shown in Table 1 to be significant, and all significances were consistent between tests and could be interpreted in biologically meaningful ways. Thus we believe it is unlikely that the observed results are spurious. We did not apply sequential Bonferroni, because this procedure can lead to very low statistical power and either an overemphasis on not rejecting null hypotheses (Moran 2003), or an unwarranted rejection of covariables in statistical analyses. A further caveat in our study is that some of the reared individuals could not be determined to species, and we cannot exclude the possibility that this may have biased some food-web metrics. We tried to reduce this problem by assigning undetermined individuals to morphospecies and to one of the three functional groups. The undetermined individuals represented only a small proportion (2·8%) of all individuals. A third caveat is that the availability of alternative nesting opportunities in cavities or dead wood could affect communities of trap-nesting insects (Steffan-Dewenter 2003; Potts et al. 2005). In our study, the distances to orchard trees with possible nesting sites did not differ significantly between trap nest stations. Hence it is unlikely that this factor has influenced our results. Finally, there could have been inter- or intraspecific competition for nesting holes among trap-nesting insects. However, the fact that on average 85% of the reed internodes per trap nest were not occupied suggests that competition for nesting sites was minimal.

The higher richness and abundance of insect species in trap nests is in line with previous studies, which examined the effectiveness of ECAs in promoting bee diversity (Knop et al. 2006; Albrecht et al. 2007). Studying the effects of grazing intensity on trap-nesting communities, Kruess & Tscharntke (2002) found higher abundance, but not species richness, of insects on extensively grazed vs. intensively grazed pastures. In agricultural landscapes in Germany, increasing distance of up to several hundred metres from species-rich chalk grasslands was associated with a decline of both species richness and number of trap-nesting bees and wasps, species richness of natural enemies and mortality by natural enemies (Tscharntke et al. 1998). In an Indonesian tropical landscape, the number of bee individuals, wasp and natural enemy species and percentage parasitized brood cells decreased along isolation gradients from natural forests into adjacent agro-forest systems (Klein et al. 2006). Empirical evidence for an increased susceptibility of species at higher trophic levels to isolation was found in several herbivore–parasitoid systems (Kareiva 1987; Kruess & Tscharntke 1994; Roland & Taylor 1997) but not in others (van Nouhuys & Hanski 2002; Esch, Klinkhamer & van der Meijden 2005). Although we found some evidence that natural enemies may have been less likely to colonize more isolated nests, distances of 100 m from restored meadows appear to affect food-web structure only weakly, as indicated by the tendency of interaction evenness to increase with distance. On average, diversity metrics appeared to decrease with increasing distance from restored meadows (Fig. 3), but this general pattern was statistically not significant.

In our study, both plant species richness and host density explained most of the variance in the species richness and abundance of natural enemies. However, the higher number of attacked brood cells was proportional to the total number of brood cells, and in contrast to Steffan-Dewenter (2003) we found no significant positive relationship between the rate of parasitism and the abundance of trap-nesting hosts. Other studies on the impacts of host density on rate of parasitism found evidence for both density dependence and density independence (reviewed by Stiling 1987; Esch et al. 2005; Tylianakis, Tscharntke & Klein 2006). The highest trophic level of the trap-nest community – parasitoids and predators – benefited relatively more from the restored meadows than their host and prey species, which was supported by the higher attack rates in ECAs. This result is in line with theoretical predictions of the trophic level hypothesis of island biogeography, where the susceptibility of populations to disturbance increases with their trophic rank (Southwood 1988; Holt et al. 1999). Populations on higher trophic levels are generally smaller and more variable, and thus more at risk of stochastic extinction (Pimm 1991; Lawton 1995; Fagan et al. 2001), and this is particularly true for parasitoids (Kruess & Tscharntke 1994). Indeed, our results on brood parasitism of the most abundant species, T. figulus, show significant effects of management intensity and isolation from ECAs only for the specialized parasitoid C. cyanea, but not for the generalized M. acasta.

For bee communities, our results indicated that management of restored meadows enhanced two counteracting factors that drive local bee diversity: plant species richness and mortality by natural enemies. However, although mortality by natural enemies was higher in restored habitats, we found a higher number of brood cells that were not attacked.

Conclusion

Our study demonstrates that meadows restored for increased plant diversity support insect food webs with higher species and interaction diversity than adjacent intensively managed meadows. The restored meadows have increased abundances of beneficial insect species that provide pollination services to crop plants. The highest trophic level, parasitoids and predators, appear to be more threatened by a general loss of biodiversity. Similarly, interaction diversity seems to decline more rapidly than species diversity with agricultural intensification. The more diverse community interactions of natural enemies in restored habitats may be a positive attribute of structurally and functionally more stable food webs. The mechanism by which restored meadows attract higher insect interaction diversity is increased plant species diversity after restoration.

Acknowledgements

We thank Louis-Felix Bersier, Jason Tylianakis and an anonymous reviewer for valuable comments on the manuscript. Andrea Holzschuh gave valuable advice on the construction, preparation and analysis of trap nests, and Mike Herrmann confirmed identifications of bees and wasps. We are grateful to Peter Wirz, Yvonne Schwarzenbach and Miriam Schädler for assistance in the field and laboratory, Martin Obrist for help with the calculations of quantitative web metrics, and Roland Schmid for help with drawing the webs. The study was part of the European research project ‘Evaluating current European agri-environment schemes to quantify and improve nature conservation efforts in agricultural landscapes’ (EASY). The project was funded by the European commission (QLRT-2001-01495), the Swiss Federal Office for Science and Technology (01·0524-2) and the Swiss National Science Foundation (grant to C.B.M.; 631-065950).

Ancillary

Advertisement